2022-01-22 01:28:58

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 00/11] media: atmel: atmel-isc: implement media controller


This series is the v4 series that attempts to support media controller in the
atmel ISC and XISC drivers.
The CSI2DC driver was accepted thus removed from the patch series, together with
other patches.

Important note: this series applies on top of current media_staging tree, as it
relies on previous patches in the series which were accepted.

Changes in v4:
-> as reviewed by Hans, added new patch to remove the 'stop' variable and reworked
one patch that was using it
-> as reviewed by Jacopo, reworked some parts of the media controller implementation

Thanks to everyone who reviewed my work !

Eugen


Previous cover letter from v3:

This series is the v3 of the series that attempts to support media controller
in the atmel ISC and XISC drivers.
This series also includes the csi2dc driver which was previously sent in a
separate series:
https://www.spinics.net/lists/linux-media/msg181042.html
https://www.spinics.net/lists/linux-media/msg181044.html
The driver now addresses comments received in latest v5 series from last year.

The series includes some minor changes and fixes that improve the isc common
code base, like removing the enum frameintervals VIDIOC, fixing bytesperline
for planar formats, etc.

Many thanks to folks from libcamera who helped a lot with understanding
how a media controller driver should behave.

Feedback is welcome !

Changes in v3:
- change in bindings, small fixes in csi2dc driver and conversion to mc
for the isc-base.
- removed some MAINTAINERS patches and used patterns in MAINTAINERS

Changes in v2:
- integrated many changes suggested by Jacopo in the review of the v1 series.
- add a few new patches


Eugen Hristev (11):
media: atmel: atmel-isc: replace 'stop' variable with vb2 calls
media: atmel: atmel-isc-base: use streaming status when queueing
buffers
media: atmel: atmel-isc: implement media controller
media: atmel: atmel-sama5d2-isc: fix wrong mask in YUYV format check
media: atmel: atmel-isc-base: use mutex to lock awb workqueue from
streaming
media: atmel: atmel-isc: compact the controller formats list
media: atmel: atmel-isc: change format propagation to subdev into only
verification
dt-bindings: media: microchip,xisc: add bus-width of 14
ARM: dts: at91: sama7g5: add nodes for video capture
ARM: configs: at91: sama7: add xisc and csi2dc
ARM: multi_v7_defconfig: add atmel video pipeline modules

.../bindings/media/microchip,xisc.yaml | 2 +-
arch/arm/boot/dts/sama7g5.dtsi | 49 ++
arch/arm/configs/multi_v7_defconfig | 3 +
arch/arm/configs/sama7_defconfig | 2 +
drivers/media/platform/atmel/Makefile | 2 +-
drivers/media/platform/atmel/atmel-isc-base.c | 455 +++++++++++-------
.../media/platform/atmel/atmel-isc-scaler.c | 250 ++++++++++
drivers/media/platform/atmel/atmel-isc.h | 43 +-
.../media/platform/atmel/atmel-sama5d2-isc.c | 67 ++-
.../media/platform/atmel/atmel-sama7g5-isc.c | 72 ++-
10 files changed, 694 insertions(+), 251 deletions(-)
create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c

--
2.25.1


2022-01-22 01:29:14

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Implement the support for media-controller.
This means that the capabilities of the driver have changed and now
it also advertises the IO_MC .
The driver will register it's media device, and add the video entity to this
media device. The subdevices are registered to the same media device.
The ISC will have a base entity which is auto-detected as atmel_isc_base.
It will also register a subdevice that allows cropping of the incoming frame
to the maximum frame size supported by the ISC.
The ISC will create a link between the subdevice that is asynchronously
registered and the atmel_isc_scaler entity.
Then, the atmel_isc_scaler and atmel_isc_base are connected through another
link.

Signed-off-by: Eugen Hristev <[email protected]>
---
Changes in v4:
As suggested by Jacopo:
- renamed atmel_isc_mc to atmel_isc_scaler.c
- moved init_mc/clean_mc to isc_base file

Changes in v2:
- implement try formats

drivers/media/platform/atmel/Makefile | 2 +-
drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
.../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
drivers/media/platform/atmel/atmel-isc.h | 37 +++
.../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
.../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
6 files changed, 375 insertions(+), 8 deletions(-)
create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c

diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
index 794e8f739287..f02d03df89d6 100644
--- a/drivers/media/platform/atmel/Makefile
+++ b/drivers/media/platform/atmel/Makefile
@@ -1,7 +1,7 @@
# SPDX-License-Identifier: GPL-2.0-only
atmel-isc-objs = atmel-sama5d2-isc.o
atmel-xisc-objs = atmel-sama7g5-isc.o
-atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
+atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o

obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
index 6b0005987a17..6b482270eb93 100644
--- a/drivers/media/platform/atmel/atmel-isc-base.c
+++ b/drivers/media/platform/atmel/atmel-isc-base.c
@@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
struct isc_device, v4l2_dev);
struct isc_subdev_entity *subdev_entity =
container_of(notifier, struct isc_subdev_entity, notifier);
+ int pad;

if (video_is_registered(&isc->video_dev)) {
v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
@@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,

subdev_entity->sd = subdev;

+ pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
+ MEDIA_PAD_FL_SOURCE);
+ if (pad < 0) {
+ v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
+ subdev->name);
+ return pad;
+ }
+
+ isc->remote_pad = pad;
+
return 0;
}

@@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
v4l2_ctrl_handler_free(&isc->ctrls.handler);
}

-static struct isc_format *find_format_by_code(struct isc_device *isc,
- unsigned int code, int *index)
+struct isc_format *isc_find_format_by_code(struct isc_device *isc,
+ unsigned int code, int *index)
{
struct isc_format *fmt = &isc->formats_list[0];
unsigned int i;
@@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,

return NULL;
}
+EXPORT_SYMBOL_GPL(isc_find_format_by_code);

static int isc_formats_init(struct isc_device *isc)
{
@@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
NULL, &mbus_code)) {
mbus_code.index++;

- fmt = find_format_by_code(isc, mbus_code.code, &i);
+ fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
if (!fmt) {
v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
mbus_code.code);
@@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
vdev->queue = q;
vdev->lock = &isc->lock;
vdev->ctrl_handler = &isc->ctrls.handler;
- vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
+ vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
+ V4L2_CAP_IO_MC;
video_set_drvdata(vdev, isc);

ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
@@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
goto isc_async_complete_err;
}

+ ret = isc_scaler_link(isc);
+ if (ret < 0)
+ goto isc_async_complete_unregister_device;
+
+ ret = media_device_register(&isc->mdev);
+ if (ret < 0)
+ goto isc_async_complete_unregister_device;
return 0;

+isc_async_complete_unregister_device:
+ video_unregister_device(vdev);
+
isc_async_complete_err:
mutex_destroy(&isc->lock);
return ret;
@@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
}
EXPORT_SYMBOL_GPL(isc_pipeline_init);

+int isc_mc_init(struct isc_device *isc, u32 ver)
+{
+ const struct of_device_id *match;
+ int ret;
+
+ isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
+ isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
+ isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
+
+ ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
+ isc->pads);
+ if (ret < 0) {
+ dev_err(isc->dev, "media entity init failed\n");
+ return ret;
+ }
+
+ isc->mdev.dev = isc->dev;
+
+ match = of_match_node(isc->dev->driver->of_match_table,
+ isc->dev->of_node);
+
+ strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
+ sizeof(isc->mdev.driver_name));
+ strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
+ snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
+ isc->v4l2_dev.name);
+ isc->mdev.hw_revision = ver;
+
+ media_device_init(&isc->mdev);
+
+ isc->v4l2_dev.mdev = &isc->mdev;
+
+ return isc_scaler_init(isc);
+}
+EXPORT_SYMBOL_GPL(isc_mc_init);
+
+void isc_mc_cleanup(struct isc_device *isc)
+{
+ media_entity_cleanup(&isc->video_dev.entity);
+}
+EXPORT_SYMBOL_GPL(isc_mc_cleanup);
+
/* regmap configuration */
#define ATMEL_ISC_REG_MAX 0xd5c
const struct regmap_config isc_regmap_config = {
diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
new file mode 100644
index 000000000000..ec95c9665883
--- /dev/null
+++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
@@ -0,0 +1,245 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Microchip Image Sensor Controller (ISC) Scaler entity support
+ *
+ * Copyright (C) 2021 Microchip Technology, Inc.
+ *
+ * Author: Eugen Hristev <[email protected]>
+ *
+ */
+
+#include <media/media-device.h>
+#include <media/media-entity.h>
+#include <media/v4l2-device.h>
+#include <media/v4l2-subdev.h>
+
+#include "atmel-isc-regs.h"
+#include "atmel-isc.h"
+
+static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
+ struct v4l2_subdev_state *sd_state,
+ struct v4l2_subdev_format *format)
+{
+ struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
+ struct v4l2_mbus_framefmt *v4l2_try_fmt;
+
+ if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
+ v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
+ format->pad);
+ format->format = *v4l2_try_fmt;
+
+ return 0;
+ }
+
+ format->format = isc->scaler_format;
+
+ return 0;
+}
+
+static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
+ struct v4l2_subdev_state *sd_state,
+ struct v4l2_subdev_format *req_fmt)
+{
+ struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
+ struct v4l2_mbus_framefmt *v4l2_try_fmt;
+ struct isc_format *fmt;
+ unsigned int i;
+
+ if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
+ v4l_bound_align_image
+ (&req_fmt->format.width, 16, isc->max_width, 0,
+ &req_fmt->format.height, 16, isc->max_height, 0, 0);
+ else
+ v4l_bound_align_image
+ (&req_fmt->format.width, 16, 10000, 0,
+ &req_fmt->format.height, 16, 10000, 0, 0);
+
+ req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
+ req_fmt->format.field = V4L2_FIELD_NONE;
+ req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
+ req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
+ req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
+
+ fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
+
+ if (!fmt)
+ fmt = &isc->formats_list[0];
+
+ req_fmt->format.code = fmt->mbus_code;
+
+ if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
+ v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
+ req_fmt->pad);
+ *v4l2_try_fmt = req_fmt->format;
+ /* Trying on the pad sink makes the source sink change too */
+ if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
+ v4l2_try_fmt =
+ v4l2_subdev_get_try_format(sd, sd_state,
+ ISC_SCALER_PAD_SOURCE);
+ *v4l2_try_fmt = req_fmt->format;
+
+ v4l_bound_align_image(&v4l2_try_fmt->width,
+ 16, isc->max_width, 0,
+ &v4l2_try_fmt->height,
+ 16, isc->max_height, 0, 0);
+ }
+ /* if we are just trying, we are done */
+ return 0;
+ }
+
+ isc->scaler_format = req_fmt->format;
+
+ return 0;
+}
+
+static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
+ struct v4l2_subdev_state *sd_state,
+ struct v4l2_subdev_mbus_code_enum *code)
+{
+ struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
+ int supported_index = 0;
+ int i;
+
+ for (i = 0; i < isc->formats_list_size; i++) {
+ if (!isc->formats_list[i].sd_support)
+ continue;
+ if (supported_index == code->index) {
+ code->code = isc->formats_list[i].mbus_code;
+ return 0;
+ }
+ supported_index++;
+ }
+
+ return -EINVAL;
+}
+
+static int isc_scaler_g_sel(struct v4l2_subdev *sd,
+ struct v4l2_subdev_state *sd_state,
+ struct v4l2_subdev_selection *sel)
+{
+ struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
+
+ if (sel->pad == ISC_SCALER_PAD_SOURCE)
+ return -EINVAL;
+
+ if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
+ sel->target != V4L2_SEL_TGT_CROP)
+ return -EINVAL;
+
+ sel->r.height = isc->max_height;
+ sel->r.width = isc->max_width;
+
+ sel->r.left = 0;
+ sel->r.top = 0;
+
+ return 0;
+}
+
+static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
+ struct v4l2_subdev_state *sd_state)
+{
+ struct v4l2_mbus_framefmt *v4l2_try_fmt =
+ v4l2_subdev_get_try_format(sd, sd_state, 0);
+ struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
+
+ *v4l2_try_fmt = isc->scaler_format;
+
+ return 0;
+}
+
+static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
+ .enum_mbus_code = isc_scaler_enum_mbus_code,
+ .set_fmt = isc_scaler_set_fmt,
+ .get_fmt = isc_scaler_get_fmt,
+ .get_selection = isc_scaler_g_sel,
+ .init_cfg = isc_scaler_init_cfg,
+};
+
+static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
+ .pad = &isc_scaler_pad_ops,
+};
+
+int isc_scaler_init(struct isc_device *isc)
+{
+ int ret;
+
+ v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
+
+ isc->scaler_sd.owner = THIS_MODULE;
+ isc->scaler_sd.dev = isc->dev;
+ snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
+ "atmel_isc_scaler");
+
+ isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
+ isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
+ isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
+ isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
+
+ isc->scaler_format.height = isc->max_height;
+ isc->scaler_format.width = isc->max_width;
+ isc->scaler_format.code = isc->formats_list[0].mbus_code;
+ isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
+ isc->scaler_format.field = V4L2_FIELD_NONE;
+ isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
+ isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
+ isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
+
+ ret = media_entity_pads_init(&isc->scaler_sd.entity,
+ ISC_SCALER_PADS_NUM,
+ isc->scaler_pads);
+ if (ret < 0) {
+ dev_err(isc->dev, "scaler sd media entity init failed\n");
+ return ret;
+ }
+ ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
+ if (ret < 0) {
+ dev_err(isc->dev, "scaler sd failed to register subdev\n");
+ return ret;
+ }
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(isc_scaler_init);
+
+int isc_scaler_link(struct isc_device *isc)
+{
+ int ret;
+
+ ret = media_create_pad_link(&isc->current_subdev->sd->entity,
+ isc->remote_pad, &isc->scaler_sd.entity,
+ ISC_SCALER_PAD_SINK,
+ MEDIA_LNK_FL_ENABLED |
+ MEDIA_LNK_FL_IMMUTABLE);
+
+ if (ret < 0) {
+ v4l2_err(&isc->v4l2_dev,
+ "Failed to create pad link: %s to %s\n",
+ isc->current_subdev->sd->entity.name,
+ isc->scaler_sd.entity.name);
+ return ret;
+ }
+
+ dev_dbg(isc->dev, "link with %s pad: %d\n",
+ isc->current_subdev->sd->name, isc->remote_pad);
+
+ ret = media_create_pad_link(&isc->scaler_sd.entity,
+ ISC_SCALER_PAD_SOURCE,
+ &isc->video_dev.entity, ISC_PAD_SINK,
+ MEDIA_LNK_FL_ENABLED |
+ MEDIA_LNK_FL_IMMUTABLE);
+
+ if (ret < 0) {
+ v4l2_err(&isc->v4l2_dev,
+ "Failed to create pad link: %s to %s\n",
+ isc->scaler_sd.entity.name,
+ isc->video_dev.entity.name);
+ return ret;
+ }
+
+ dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
+ ISC_SCALER_PAD_SOURCE);
+
+ return ret;
+}
+EXPORT_SYMBOL_GPL(isc_scaler_link);
+
diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
index 5fbf52a9080b..c9234c90ae58 100644
--- a/drivers/media/platform/atmel/atmel-isc.h
+++ b/drivers/media/platform/atmel/atmel-isc.h
@@ -183,6 +183,17 @@ struct isc_reg_offsets {
u32 his_entry;
};

+enum isc_mc_pads {
+ ISC_PAD_SINK = 0,
+ ISC_PADS_NUM = 1,
+};
+
+enum isc_scaler_pads {
+ ISC_SCALER_PAD_SINK = 0,
+ ISC_SCALER_PAD_SOURCE = 1,
+ ISC_SCALER_PADS_NUM = 2,
+};
+
/*
* struct isc_device - ISC device driver data/config struct
* @regmap: Register map
@@ -257,6 +268,12 @@ struct isc_reg_offsets {
* be used as an input to the controller
* @controller_formats_size: size of controller_formats array
* @formats_list_size: size of formats_list array
+ * @pads: media controller pads for isc video entity
+ * @mdev: media device that is registered by the isc
+ * @remote_pad: remote pad on the connected subdevice
+ * @scaler_sd: subdevice for the scaler that isc registers
+ * @scaler_pads: media controller pads for the scaler subdevice
+ * @scaler_format: current format for the scaler subdevice
*/
struct isc_device {
struct regmap *regmap;
@@ -344,6 +361,19 @@ struct isc_device {
struct isc_format *formats_list;
u32 controller_formats_size;
u32 formats_list_size;
+
+ struct {
+ struct media_pad pads[ISC_PADS_NUM];
+ struct media_device mdev;
+
+ u32 remote_pad;
+ };
+
+ struct {
+ struct v4l2_subdev scaler_sd;
+ struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
+ struct v4l2_mbus_framefmt scaler_format;
+ };
};

extern const struct regmap_config isc_regmap_config;
@@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
void isc_subdev_cleanup(struct isc_device *isc);
void isc_clk_cleanup(struct isc_device *isc);

+int isc_scaler_link(struct isc_device *isc);
+int isc_scaler_init(struct isc_device *isc);
+int isc_mc_init(struct isc_device *isc, u32 ver);
+void isc_mc_cleanup(struct isc_device *isc);
+
+struct isc_format *isc_find_format_by_code(struct isc_device *isc,
+ unsigned int code, int *index);
#endif
diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
index c5b9563e36cb..c244682ea22f 100644
--- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
+++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
@@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
break;
}

+ regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
+
+ ret = isc_mc_init(isc, ver);
+ if (ret < 0)
+ goto isc_probe_mc_init_err;
+
pm_runtime_set_active(dev);
pm_runtime_enable(dev);
pm_request_idle(dev);
@@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
ret = clk_prepare_enable(isc->ispck);
if (ret) {
dev_err(dev, "failed to enable ispck: %d\n", ret);
- goto cleanup_subdev;
+ goto isc_probe_mc_init_err;
}

/* ispck should be greater or equal to hclock */
@@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
goto unprepare_clk;
}

- regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
dev_info(dev, "Microchip ISC version %x\n", ver);

return 0;
@@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
unprepare_clk:
clk_disable_unprepare(isc->ispck);

+isc_probe_mc_init_err:
+ isc_mc_cleanup(isc);
+
cleanup_subdev:
isc_subdev_cleanup(isc);

@@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)

pm_runtime_disable(&pdev->dev);

+ isc_mc_cleanup(isc);
+
isc_subdev_cleanup(isc);

v4l2_device_unregister(&isc->v4l2_dev);
diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
index 07a80b08bc54..9dc75eed0098 100644
--- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
+++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
@@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
break;
}

+ regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
+
+ ret = isc_mc_init(isc, ver);
+ if (ret < 0)
+ goto isc_probe_mc_init_err;
+
pm_runtime_set_active(dev);
pm_runtime_enable(dev);
pm_request_idle(dev);

- regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
dev_info(dev, "Microchip XISC version %x\n", ver);

return 0;

+isc_probe_mc_init_err:
+ isc_mc_cleanup(isc);
+
cleanup_subdev:
isc_subdev_cleanup(isc);

@@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)

pm_runtime_disable(&pdev->dev);

+ isc_mc_cleanup(isc);
+
isc_subdev_cleanup(isc);

v4l2_device_unregister(&isc->v4l2_dev);
--
2.25.1

2022-01-22 01:29:50

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 02/11] media: atmel: atmel-isc-base: use streaming status when queueing buffers

During experiments with libcamera, it looks like vb2_is_streaming returns
true before our start streaming is called.
Order of operations is streamon -> queue -> start_streaming
ISC would have started the DMA immediately when a buffer is being added
to the vbqueue if the queue is streaming.
It is more safe to start the DMA after the start streaming of the driver is
called.
Thus, even if vb2queue is streaming, add the buffer to the dma queue of the
driver instead of actually starting the DMA process, if the start streaming
has not been called yet.
Tho achieve this, we have to use vb2_start_streaming_called instead of
vb2_is_streaming.

Signed-off-by: Eugen Hristev <[email protected]>
---
Changes in v4:
- changed to using vb2_start_streaming_called instead of stop variable

drivers/media/platform/atmel/atmel-isc-base.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
index 9c62d0ae7887..6b0005987a17 100644
--- a/drivers/media/platform/atmel/atmel-isc-base.c
+++ b/drivers/media/platform/atmel/atmel-isc-base.c
@@ -439,7 +439,7 @@ static void isc_buffer_queue(struct vb2_buffer *vb)

spin_lock_irqsave(&isc->dma_queue_lock, flags);
if (!isc->cur_frm && list_empty(&isc->dma_queue) &&
- vb2_is_streaming(vb->vb2_queue)) {
+ vb2_start_streaming_called(vb->vb2_queue)) {
isc->cur_frm = buf;
isc_start_dma(isc);
} else
@@ -1532,7 +1532,7 @@ static int isc_s_awb_ctrl(struct v4l2_ctrl *ctrl)

isc_update_awb_ctrls(isc);

- if (vb2_is_streaming(&isc->vb2_vidq)) {
+ if (vb2_start_streaming_called(&isc->vb2_vidq)) {
/*
* If we are streaming, we can update profile to
* have the new settings in place.
@@ -1549,7 +1549,7 @@ static int isc_s_awb_ctrl(struct v4l2_ctrl *ctrl)

/* if we have autowhitebalance on, start histogram procedure */
if (ctrls->awb == ISC_WB_AUTO &&
- vb2_is_streaming(&isc->vb2_vidq) &&
+ vb2_start_streaming_called(&isc->vb2_vidq) &&
ISC_IS_FORMAT_RAW(isc->config.sd_format->mbus_code))
isc_set_histogram(isc, true);

--
2.25.1

2022-01-22 01:30:30

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 07/11] media: atmel: atmel-isc: change format propagation to subdev into only verification

As a top MC video driver, the atmel-isc should not propagate the format to the
subdevice.
It should rather check at start_streaming() time if the subdev is properly
configured with a compatible format.
Removed the whole format finding logic, and reworked the format verification
at start_streaming time, such that the ISC will return an error if the subdevice
is not properly configured. To achieve this, media_pipeline_start
is called and a link_validate callback is created to check the formats.
With this being done, the module parameter 'sensor_preferred' makes no sense
anymore. The ISC should not decide which format the sensor is using. The
ISC should only cope with the situation and inform userspace if the streaming
is possible in the current configuration.
The redesign of the format propagation has also risen the question of the
enumfmt callback. If enumfmt is called with an mbus_code, the enumfmt handler
should only return the formats that are supported for this mbus_code.
To make it more easy to understand the formats, changed the report order
to report first the native formats, and after that the formats that the ISC
can convert to.

Signed-off-by: Eugen Hristev <[email protected]>
---
Changes in v4:
- moved validation code into link_validate and used media_pipeline_start
- merged this patch with the enum_fmt patch which was previously in v3 of
the series

Changes in v3:
- clamp to maximum resolution once the frame size from the subdev is found

drivers/media/platform/atmel/atmel-isc-base.c | 335 ++++++++++--------
.../media/platform/atmel/atmel-isc-scaler.c | 5 +
drivers/media/platform/atmel/atmel-isc.h | 3 +
3 files changed, 191 insertions(+), 152 deletions(-)

diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
index f7a88399bd54..31c79313aadc 100644
--- a/drivers/media/platform/atmel/atmel-isc-base.c
+++ b/drivers/media/platform/atmel/atmel-isc-base.c
@@ -36,11 +36,6 @@ static unsigned int debug;
module_param(debug, int, 0644);
MODULE_PARM_DESC(debug, "debug level (0-2)");

-static unsigned int sensor_preferred = 1;
-module_param(sensor_preferred, uint, 0644);
-MODULE_PARM_DESC(sensor_preferred,
- "Sensor is preferred to output the specified format (1-on 0-off), default 1");
-
#define ISC_IS_FORMAT_RAW(mbus_code) \
(((mbus_code) & 0xf000) == 0x3000)

@@ -337,6 +332,10 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
unsigned long flags;
int ret;

+ ret = media_pipeline_start(&isc->video_dev.entity, &isc->mpipe);
+ if (ret)
+ goto err_pipeline_start;
+
/* Enable stream on the sub device */
ret = v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 1);
if (ret && ret != -ENOIOCTLCMD) {
@@ -384,6 +383,9 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0);

err_start_stream:
+ media_pipeline_stop(&isc->video_dev.entity);
+
+err_pipeline_start:
spin_lock_irqsave(&isc->dma_queue_lock, flags);
list_for_each_entry(buf, &isc->dma_queue, list)
vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_QUEUED);
@@ -420,6 +422,9 @@ static void isc_stop_streaming(struct vb2_queue *vq)
if (ret && ret != -ENOIOCTLCMD)
v4l2_err(&isc->v4l2_dev, "stream off failed in subdev\n");

+ /* Stop media pipeline */
+ media_pipeline_stop(&isc->video_dev.entity);
+
/* Release all active buffers */
spin_lock_irqsave(&isc->dma_queue_lock, flags);
if (unlikely(isc->cur_frm)) {
@@ -496,21 +501,56 @@ static int isc_enum_fmt_vid_cap(struct file *file, void *priv,
u32 index = f->index;
u32 i, supported_index;

- if (index < isc->controller_formats_size) {
- f->pixelformat = isc->controller_formats[index].fourcc;
- return 0;
+ supported_index = 0;
+
+ for (i = 0; i < isc->formats_list_size; i++) {
+ if (!isc->formats_list[i].sd_support)
+ continue;
+ /*
+ * If specific mbus_code is requested, provide only
+ * supported formats with this mbus code
+ */
+ if (f->mbus_code && f->mbus_code !=
+ isc->formats_list[i].mbus_code)
+ continue;
+ if (supported_index == index) {
+ f->pixelformat = isc->formats_list[i].fourcc;
+ return 0;
+ }
+ supported_index++;
}

- index -= isc->controller_formats_size;
+ /*
+ * If the sensor does not support this mbus_code whatsoever,
+ * there is no reason to advertise any of our output formats
+ */
+ if (supported_index == 0)
+ return -EINVAL;

+ /*
+ * If the sensor uses a format that is not raw, then we cannot
+ * convert it to any of the formats that we usually can with a
+ * RAW sensor. Thus, do not advertise them.
+ */
+ if (isc->config.sd_format &&
+ !ISC_IS_FORMAT_RAW(isc->config.sd_format->mbus_code))
+ return -EINVAL;
+
+ /*
+ * Iterate again through the formats that we can convert to.
+ * However, to avoid duplicates, skip the formats that
+ * the sensor already supports directly
+ */
+ index -= supported_index;
supported_index = 0;

- for (i = 0; i < isc->formats_list_size; i++) {
- if (!ISC_IS_FORMAT_RAW(isc->formats_list[i].mbus_code) ||
- !isc->formats_list[i].sd_support)
+ for (i = 0; i < isc->controller_formats_size; i++) {
+ /* if this format is already supported by sensor, skip it */
+ if (find_format_by_fourcc(isc, isc->controller_formats[i].fourcc))
continue;
if (supported_index == index) {
- f->pixelformat = isc->formats_list[i].fourcc;
+ f->pixelformat =
+ isc->controller_formats[i].fourcc;
return 0;
}
supported_index++;
@@ -581,20 +621,30 @@ static int isc_try_validate_formats(struct isc_device *isc)
break;
default:
/* any other different formats are not supported */
+ v4l2_err(&isc->v4l2_dev, "Requested unsupported format.\n");
ret = -EINVAL;
}
v4l2_dbg(1, debug, &isc->v4l2_dev,
"Format validation, requested rgb=%u, yuv=%u, grey=%u, bayer=%u\n",
rgb, yuv, grey, bayer);

- /* we cannot output RAW if we do not receive RAW */
- if ((bayer) && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
+ if ((bayer) &&
+ !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code)) {
+ v4l2_err(&isc->v4l2_dev, "Cannot output RAW if we do not receive RAW.\n");
return -EINVAL;
+ }

- /* we cannot output GREY if we do not receive RAW/GREY */
if (grey && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code) &&
- !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code))
+ !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
+ v4l2_err(&isc->v4l2_dev, "Cannot output GREY if we do not receive RAW/GREY.\n");
+ return -EINVAL;
+ }
+
+ if ((rgb || bayer || yuv) &&
+ ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
+ v4l2_err(&isc->v4l2_dev, "Cannot convert GREY to another format.\n");
return -EINVAL;
+ }

return ret;
}
@@ -822,7 +872,7 @@ static void isc_try_fse(struct isc_device *isc,
* If we do not know yet which format the subdev is using, we cannot
* do anything.
*/
- if (!isc->try_config.sd_format)
+ if (!isc->config.sd_format)
return;

fse.code = isc->try_config.sd_format->mbus_code;
@@ -843,180 +893,141 @@ static void isc_try_fse(struct isc_device *isc,
}
}

-static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f,
- u32 *code)
+static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f)
{
- int i;
- struct isc_format *sd_fmt = NULL, *direct_fmt = NULL;
struct v4l2_pix_format *pixfmt = &f->fmt.pix;
- struct v4l2_subdev_pad_config pad_cfg = {};
- struct v4l2_subdev_state pad_state = {
- .pads = &pad_cfg
- };
- struct v4l2_subdev_format format = {
- .which = V4L2_SUBDEV_FORMAT_TRY,
- };
- u32 mbus_code;
- int ret;
- bool rlp_dma_direct_dump = false;
+ unsigned int i;

if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
return -EINVAL;

- /* Step 1: find a RAW format that is supported */
- for (i = 0; i < isc->num_user_formats; i++) {
- if (ISC_IS_FORMAT_RAW(isc->user_formats[i]->mbus_code)) {
- sd_fmt = isc->user_formats[i];
+ isc->try_config.fourcc = isc->user_formats[0]->fourcc;
+
+ /* find if the format requested is supported */
+ for (i = 0; i < isc->controller_formats_size; i++)
+ if (isc->controller_formats[i].fourcc == pixfmt->pixelformat) {
+ isc->try_config.fourcc = pixfmt->pixelformat;
break;
}
- }
- /* Step 2: We can continue with this RAW format, or we can look
- * for better: maybe sensor supports directly what we need.
- */
- direct_fmt = find_format_by_fourcc(isc, pixfmt->pixelformat);
-
- /* Step 3: We have both. We decide given the module parameter which
- * one to use.
- */
- if (direct_fmt && sd_fmt && sensor_preferred)
- sd_fmt = direct_fmt;
-
- /* Step 4: we do not have RAW but we have a direct format. Use it. */
- if (direct_fmt && !sd_fmt)
- sd_fmt = direct_fmt;
-
- /* Step 5: if we are using a direct format, we need to package
- * everything as 8 bit data and just dump it
- */
- if (sd_fmt == direct_fmt)
- rlp_dma_direct_dump = true;
-
- /* Step 6: We have no format. This can happen if the userspace
- * requests some weird/invalid format.
- * In this case, default to whatever we have
- */
- if (!sd_fmt && !direct_fmt) {
- sd_fmt = isc->user_formats[isc->num_user_formats - 1];
- v4l2_dbg(1, debug, &isc->v4l2_dev,
- "Sensor not supporting %.4s, using %.4s\n",
- (char *)&pixfmt->pixelformat, (char *)&sd_fmt->fourcc);
- }
-
- if (!sd_fmt) {
- ret = -EINVAL;
- goto isc_try_fmt_err;
- }
-
- /* Step 7: Print out what we decided for debugging */
- v4l2_dbg(1, debug, &isc->v4l2_dev,
- "Preferring to have sensor using format %.4s\n",
- (char *)&sd_fmt->fourcc);
-
- /* Step 8: at this moment we decided which format the subdev will use */
- isc->try_config.sd_format = sd_fmt;
-
- /* Limit to Atmel ISC hardware capabilities */
- if (pixfmt->width > isc->max_width)
- pixfmt->width = isc->max_width;
- if (pixfmt->height > isc->max_height)
- pixfmt->height = isc->max_height;
-
- /*
- * The mbus format is the one the subdev outputs.
- * The pixels will be transferred in this format Sensor -> ISC
- */
- mbus_code = sd_fmt->mbus_code;
-
- /*
- * Validate formats. If the required format is not OK, default to raw.
- */
-
- isc->try_config.fourcc = pixfmt->pixelformat;
-
- if (isc_try_validate_formats(isc)) {
- pixfmt->pixelformat = isc->try_config.fourcc = sd_fmt->fourcc;
- /* Re-try to validate the new format */
- ret = isc_try_validate_formats(isc);
- if (ret)
- goto isc_try_fmt_err;
- }
-
- ret = isc_try_configure_rlp_dma(isc, rlp_dma_direct_dump);
- if (ret)
- goto isc_try_fmt_err;
-
- ret = isc_try_configure_pipeline(isc);
- if (ret)
- goto isc_try_fmt_err;

- /* Obtain frame sizes if possible to have crop requirements ready */
- isc_try_fse(isc, &pad_state);
-
- v4l2_fill_mbus_format(&format.format, pixfmt, mbus_code);
- ret = v4l2_subdev_call(isc->current_subdev->sd, pad, set_fmt,
- &pad_state, &format);
- if (ret < 0)
- goto isc_try_fmt_subdev_err;
+ /* If we did not find the requested format, we will fallback here */
+ pixfmt->pixelformat = isc->try_config.fourcc;
+ pixfmt->colorspace = V4L2_COLORSPACE_SRGB;
+ pixfmt->field = V4L2_FIELD_NONE;

- v4l2_fill_pix_format(pixfmt, &format.format);
+ isc_try_configure_rlp_dma(isc, false);

/* Limit to Atmel ISC hardware capabilities */
- if (pixfmt->width > isc->max_width)
- pixfmt->width = isc->max_width;
- if (pixfmt->height > isc->max_height)
- pixfmt->height = isc->max_height;
+ v4l_bound_align_image(&pixfmt->width, 16, isc->max_width, 0,
+ &pixfmt->height, 16, isc->max_height, 0, 0);

pixfmt->field = V4L2_FIELD_NONE;
pixfmt->bytesperline = (pixfmt->width * isc->try_config.bpp_v4l2) >> 3;
pixfmt->sizeimage = ((pixfmt->width * isc->try_config.bpp) >> 3) *
pixfmt->height;

- if (code)
- *code = mbus_code;
+ isc->try_fmt = *f;

return 0;
+}

-isc_try_fmt_err:
- v4l2_err(&isc->v4l2_dev, "Could not find any possible format for a working pipeline\n");
-isc_try_fmt_subdev_err:
- memset(&isc->try_config, 0, sizeof(isc->try_config));
+static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
+{
+ isc_try_fmt(isc, f);

- return ret;
+ /* make the try configuration active */
+ isc->config = isc->try_config;
+ isc->fmt = isc->try_fmt;
+
+ v4l2_dbg(1, debug, &isc->v4l2_dev, "ISC set_fmt to %.4s @%dx%d\n",
+ (char *)&f->fmt.pix.pixelformat,
+ f->fmt.pix.width, f->fmt.pix.height);
+
+ return 0;
}

-static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
+static int isc_validate(struct isc_device *isc)
{
+ int ret;
+ int i;
+ struct isc_format *sd_fmt = NULL;
+ struct v4l2_pix_format *pixfmt = &isc->fmt.fmt.pix;
struct v4l2_subdev_format format = {
.which = V4L2_SUBDEV_FORMAT_ACTIVE,
+ .pad = isc->remote_pad,
+ };
+ struct v4l2_subdev_pad_config pad_cfg = {};
+ struct v4l2_subdev_state pad_state = {
+ .pads = &pad_cfg,
};
- u32 mbus_code = 0;
- int ret;

- ret = isc_try_fmt(isc, f, &mbus_code);
+ /* Get current format from subdev */
+ ret = v4l2_subdev_call(isc->current_subdev->sd, pad, get_fmt, NULL,
+ &format);
if (ret)
return ret;

- v4l2_fill_mbus_format(&format.format, &f->fmt.pix, mbus_code);
- ret = v4l2_subdev_call(isc->current_subdev->sd, pad,
- set_fmt, NULL, &format);
- if (ret < 0)
- return ret;
+ /* Identify the subdev's format configuration */
+ for (i = 0; i < isc->num_user_formats; i++)
+ if (isc->user_formats[i]->mbus_code == format.format.code) {
+ sd_fmt = isc->user_formats[i];
+ break;
+ }
+
+ /* Check if the format is not supported */
+ if (!sd_fmt) {
+ v4l2_err(&isc->v4l2_dev,
+ "Current subdevice is streaming a media bus code that is not supported 0x%x\n",
+ format.format.code);
+ return -EPIPE;
+ }
+
+ /* At this moment we know which format the subdev will use */
+ isc->try_config.sd_format = sd_fmt;
+
+ /* If the sensor is not RAW, we can only do a direct dump */
+ if (!ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
+ isc_try_configure_rlp_dma(isc, true);

/* Limit to Atmel ISC hardware capabilities */
- if (f->fmt.pix.width > isc->max_width)
- f->fmt.pix.width = isc->max_width;
- if (f->fmt.pix.height > isc->max_height)
- f->fmt.pix.height = isc->max_height;
+ v4l_bound_align_image(&format.format.width, 16, isc->max_width, 0,
+ &format.format.height, 16, isc->max_height, 0, 0);
+
+ /* Check if the frame size is the same. Otherwise we may overflow */
+ if (pixfmt->height != format.format.height ||
+ pixfmt->width != format.format.width) {
+ v4l2_err(&isc->v4l2_dev,
+ "ISC not configured with the proper frame size: %dx%d\n",
+ format.format.width, format.format.height);
+ return -EPIPE;
+ }

- isc->fmt = *f;
+ v4l2_dbg(1, debug, &isc->v4l2_dev,
+ "Identified subdev using format %.4s with %dx%d %d bpp\n",
+ (char *)&sd_fmt->fourcc, pixfmt->width, pixfmt->height,
+ isc->try_config.bpp);

+ /* Reset and restart AWB if the subdevice changed the format */
if (isc->try_config.sd_format && isc->config.sd_format &&
isc->try_config.sd_format != isc->config.sd_format) {
isc->ctrls.hist_stat = HIST_INIT;
isc_reset_awb_ctrls(isc);
isc_update_v4l2_ctrls(isc);
}
- /* make the try configuration active */
+
+ /* Validate formats */
+ ret = isc_try_validate_formats(isc);
+ if (ret)
+ return ret;
+
+ /* Obtain frame sizes if possible to have crop requirements ready */
+ isc_try_fse(isc, &pad_state);
+
+ /* Configure ISC pipeline for the config */
+ ret = isc_try_configure_pipeline(isc);
+ if (ret)
+ return ret;
+
isc->config = isc->try_config;

v4l2_dbg(1, debug, &isc->v4l2_dev, "New ISC configuration in place\n");
@@ -1040,7 +1051,7 @@ static int isc_try_fmt_vid_cap(struct file *file, void *priv,
{
struct isc_device *isc = video_drvdata(file);

- return isc_try_fmt(isc, f, NULL);
+ return isc_try_fmt(isc, f);
}

static int isc_enum_input(struct file *file, void *priv,
@@ -1841,7 +1852,7 @@ static int isc_set_default_fmt(struct isc_device *isc)
};
int ret;

- ret = isc_try_fmt(isc, &f, NULL);
+ ret = isc_try_fmt(isc, &f);
if (ret)
return ret;

@@ -2015,6 +2026,24 @@ int isc_pipeline_init(struct isc_device *isc)
}
EXPORT_SYMBOL_GPL(isc_pipeline_init);

+int isc_link_validate(struct media_link *link)
+{
+ struct video_device *vdev =
+ media_entity_to_video_device(link->sink->entity);
+ struct isc_device *isc = video_get_drvdata(vdev);
+ int ret;
+
+ ret = v4l2_subdev_link_validate(link);
+ if (ret)
+ return ret;
+
+ return isc_validate(isc);
+}
+
+static const struct media_entity_operations isc_entity_operations = {
+ .link_validate = isc_link_validate,
+};
+
int isc_mc_init(struct isc_device *isc, u32 ver)
{
const struct of_device_id *match;
@@ -2022,6 +2051,8 @@ int isc_mc_init(struct isc_device *isc, u32 ver)

isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
+ isc->video_dev.entity.ops = &isc_entity_operations;
+
isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;

ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
index ec95c9665883..93375a61aea6 100644
--- a/drivers/media/platform/atmel/atmel-isc-scaler.c
+++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
@@ -155,6 +155,10 @@ static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
.init_cfg = isc_scaler_init_cfg,
};

+static const struct media_entity_operations isc_scaler_entity_ops = {
+ .link_validate = v4l2_subdev_link_validate,
+};
+
static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
.pad = &isc_scaler_pad_ops,
};
@@ -172,6 +176,7 @@ int isc_scaler_init(struct isc_device *isc)

isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
+ isc->scaler_sd.entity.ops = &isc_scaler_entity_ops;
isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;

diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
index 04ea5e340808..e36cf9d3942a 100644
--- a/drivers/media/platform/atmel/atmel-isc.h
+++ b/drivers/media/platform/atmel/atmel-isc.h
@@ -270,6 +270,7 @@ enum isc_scaler_pads {
* @formats_list_size: size of formats_list array
* @pads: media controller pads for isc video entity
* @mdev: media device that is registered by the isc
+ * @mpipe: media device pipeline used by the isc
* @remote_pad: remote pad on the connected subdevice
* @scaler_sd: subdevice for the scaler that isc registers
* @scaler_pads: media controller pads for the scaler subdevice
@@ -295,6 +296,7 @@ struct isc_device {
struct completion comp;

struct v4l2_format fmt;
+ struct v4l2_format try_fmt;
struct isc_format **user_formats;
unsigned int num_user_formats;

@@ -366,6 +368,7 @@ struct isc_device {
struct {
struct media_pad pads[ISC_PADS_NUM];
struct media_device mdev;
+ struct media_pipeline mpipe;

u32 remote_pad;
};
--
2.25.1

2022-01-22 01:30:34

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 04/11] media: atmel: atmel-sama5d2-isc: fix wrong mask in YUYV format check

While this does not happen in production, this check should be done
versus the mask, as checking with the YCYC value may not include
some bits that may be set.
Is it correct and safe to check the whole mask.

Fixes: 123aaf816b95 ("media: atmel: atmel-sama5d2-isc: fix YUYV format")
Signed-off-by: Eugen Hristev <[email protected]>
---
drivers/media/platform/atmel/atmel-sama5d2-isc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
index c244682ea22f..025c3e8a7e95 100644
--- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
+++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
@@ -291,7 +291,7 @@ static void isc_sama5d2_config_rlp(struct isc_device *isc)
* Thus, if the YCYC mode is selected, replace it with the
* sama5d2-compliant mode which is YYCC .
*/
- if ((rlp_mode & ISC_RLP_CFG_MODE_YCYC) == ISC_RLP_CFG_MODE_YCYC) {
+ if ((rlp_mode & ISC_RLP_CFG_MODE_MASK) == ISC_RLP_CFG_MODE_YCYC) {
rlp_mode &= ~ISC_RLP_CFG_MODE_MASK;
rlp_mode |= ISC_RLP_CFG_MODE_YYCC;
}
--
2.25.1

2022-01-22 01:30:55

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 08/11] dt-bindings: media: microchip,xisc: add bus-width of 14

The Microchip XISC supports a bus width of 14 bits.
Add it to the supported bus widths.

Signed-off-by: Eugen Hristev <[email protected]>
---
Documentation/devicetree/bindings/media/microchip,xisc.yaml | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Documentation/devicetree/bindings/media/microchip,xisc.yaml b/Documentation/devicetree/bindings/media/microchip,xisc.yaml
index 086e1430af4f..3be8f64c3e21 100644
--- a/Documentation/devicetree/bindings/media/microchip,xisc.yaml
+++ b/Documentation/devicetree/bindings/media/microchip,xisc.yaml
@@ -67,7 +67,7 @@ properties:
remote-endpoint: true

bus-width:
- enum: [8, 9, 10, 11, 12]
+ enum: [8, 9, 10, 11, 12, 14]
default: 12

hsync-active:
--
2.25.1

2022-01-22 01:30:57

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 09/11] ARM: dts: at91: sama7g5: add nodes for video capture

Add node for the XISC (eXtended Image Sensor Controller) and CSI2DC
(csi2 demux controller).
These nodes represent the top level of the video capture hardware pipeline
and are directly connected in hardware.

Signed-off-by: Eugen Hristev <[email protected]>
---
Changes in v4:
- add mandatory property bus type

Changes in v3:
- change bus width for endpoints to the default 14

arch/arm/boot/dts/sama7g5.dtsi | 49 ++++++++++++++++++++++++++++++++++
1 file changed, 49 insertions(+)

diff --git a/arch/arm/boot/dts/sama7g5.dtsi b/arch/arm/boot/dts/sama7g5.dtsi
index 7039311bf678..d47d62f01895 100644
--- a/arch/arm/boot/dts/sama7g5.dtsi
+++ b/arch/arm/boot/dts/sama7g5.dtsi
@@ -236,6 +236,55 @@ sdmmc2: mmc@e120c000 {
status = "disabled";
};

+ csi2dc: csi2dc@e1404000 {
+ compatible = "microchip,sama7g5-csi2dc";
+ reg = <0xe1404000 0x500>;
+ clocks = <&pmc PMC_TYPE_PERIPHERAL 34>, <&xisc>;
+ clock-names = "pclk", "scck";
+ assigned-clocks = <&xisc>;
+ assigned-clock-rates = <266000000>;
+
+ ports {
+ #address-cells = <1>;
+ #size-cells = <0>;
+ port@0 {
+ reg = <0>;
+ csi2dc_in: endpoint {
+ };
+ };
+
+ port@1 {
+ reg = <1>;
+ csi2dc_out: endpoint {
+ bus-width = <14>;
+ hsync-active = <1>;
+ vsync-active = <1>;
+ remote-endpoint = <&xisc_in>;
+ };
+ };
+ };
+ };
+
+ xisc: xisc@e1408000 {
+ compatible = "microchip,sama7g5-isc";
+ reg = <0xe1408000 0x2000>;
+ interrupts = <GIC_SPI 56 IRQ_TYPE_LEVEL_HIGH>;
+ clocks = <&pmc PMC_TYPE_PERIPHERAL 56>;
+ clock-names = "hclock";
+ #clock-cells = <0>;
+ clock-output-names = "isc-mck";
+
+ port {
+ xisc_in: endpoint {
+ bus-type = <5>; /* Parallel */
+ bus-width = <14>;
+ hsync-active = <1>;
+ vsync-active = <1>;
+ remote-endpoint = <&csi2dc_out>;
+ };
+ };
+ };
+
pwm: pwm@e1604000 {
compatible = "microchip,sama7g5-pwm", "atmel,sama5d2-pwm";
reg = <0xe1604000 0x4000>;
--
2.25.1

2022-01-22 01:31:02

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 05/11] media: atmel: atmel-isc-base: use mutex to lock awb workqueue from streaming

The AWB workqueue runs in a kernel thread and needs to be synchronized
w.r.t. the streaming status.
It is possible that streaming is stopped while the AWB workq is running.
In this case it is likely that the check for vb2_start_streaming_called is done
at one point in time, but the AWB computations are done later, including a call
to isc_update_profile, which requires streaming to be started.
Thus , isc_update_profile will fail if during this operation sequence the
streaming was stopped.
To solve this issue, a mutex is added, that will serialize the awb work and
streaming stopping, with the mention that either streaming is stopped
completely including termination of the last frame is done, and after that
the AWB work can check stream status and stop; either first AWB work is
completed and after that the streaming can stop correctly.
The awb spin lock cannot be used since this spinlock is taken in the same
context and using it in the stop streaming will result in a recursion BUG.

Signed-off-by: Eugen Hristev <[email protected]>
---
drivers/media/platform/atmel/atmel-isc-base.c | 31 ++++++++++++++++---
drivers/media/platform/atmel/atmel-isc.h | 1 +
2 files changed, 28 insertions(+), 4 deletions(-)

diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
index 6b482270eb93..f7a88399bd54 100644
--- a/drivers/media/platform/atmel/atmel-isc-base.c
+++ b/drivers/media/platform/atmel/atmel-isc-base.c
@@ -400,6 +400,7 @@ static void isc_stop_streaming(struct vb2_queue *vq)
struct isc_buffer *buf;
int ret;

+ mutex_lock(&isc->awb_mutex);
v4l2_ctrl_activate(isc->do_wb_ctrl, false);

/* Wait until the end of the current frame */
@@ -407,6 +408,8 @@ static void isc_stop_streaming(struct vb2_queue *vq)
v4l2_err(&isc->v4l2_dev,
"Timeout waiting for end of the capture\n");

+ mutex_unlock(&isc->awb_mutex);
+
/* Disable DMA interrupt */
regmap_write(isc->regmap, ISC_INTDIS, ISC_INT_DDONE);

@@ -1395,10 +1398,6 @@ static void isc_awb_work(struct work_struct *w)
u32 min, max;
int ret;

- /* streaming is not active anymore */
- if (!vb2_start_streaming_called(&isc->vb2_vidq))
- return;
-
if (ctrls->hist_stat != HIST_ENABLED)
return;

@@ -1453,7 +1452,24 @@ static void isc_awb_work(struct work_struct *w)
}
regmap_write(regmap, ISC_HIS_CFG + isc->offsets.his,
hist_id | baysel | ISC_HIS_CFG_RAR);
+
+ /*
+ * We have to make sure the streaming has not stopped meanwhile.
+ * ISC requires a frame to clock the internal profile update.
+ * To avoid issues, lock the sequence with a mutex
+ */
+ mutex_lock(&isc->awb_mutex);
+
+ /* streaming is not active anymore */
+ if (!vb2_start_streaming_called(&isc->vb2_vidq)) {
+ mutex_unlock(&isc->awb_mutex);
+ return;
+ };
+
isc_update_profile(isc);
+
+ mutex_unlock(&isc->awb_mutex);
+
/* if awb has been disabled, we don't need to start another histogram */
if (ctrls->awb)
regmap_write(regmap, ISC_CTRLEN, ISC_CTRL_HISREQ);
@@ -1532,6 +1548,8 @@ static int isc_s_awb_ctrl(struct v4l2_ctrl *ctrl)

isc_update_awb_ctrls(isc);

+ mutex_lock(&isc->awb_mutex);
+
if (vb2_start_streaming_called(&isc->vb2_vidq)) {
/*
* If we are streaming, we can update profile to
@@ -1546,6 +1564,7 @@ static int isc_s_awb_ctrl(struct v4l2_ctrl *ctrl)
*/
v4l2_ctrl_activate(isc->do_wb_ctrl, false);
}
+ mutex_unlock(&isc->awb_mutex);

/* if we have autowhitebalance on, start histogram procedure */
if (ctrls->awb == ISC_WB_AUTO &&
@@ -1738,6 +1757,7 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
{
struct isc_device *isc = container_of(notifier->v4l2_dev,
struct isc_device, v4l2_dev);
+ mutex_destroy(&isc->awb_mutex);
cancel_work_sync(&isc->awb_work);
video_unregister_device(&isc->video_dev);
v4l2_ctrl_handler_free(&isc->ctrls.handler);
@@ -1848,6 +1868,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
isc->current_subdev = container_of(notifier,
struct isc_subdev_entity, notifier);
mutex_init(&isc->lock);
+ mutex_init(&isc->awb_mutex);
+
init_completion(&isc->comp);

/* Initialize videobuf2 queue */
@@ -1927,6 +1949,7 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
video_unregister_device(vdev);

isc_async_complete_err:
+ mutex_destroy(&isc->awb_mutex);
mutex_destroy(&isc->lock);
return ret;
}
diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
index c9234c90ae58..04ea5e340808 100644
--- a/drivers/media/platform/atmel/atmel-isc.h
+++ b/drivers/media/platform/atmel/atmel-isc.h
@@ -305,6 +305,7 @@ struct isc_device {
struct work_struct awb_work;

struct mutex lock; /* serialize access to file operations */
+ struct mutex awb_mutex; /* serialize access to streaming status from awb work queue */
spinlock_t awb_lock; /* serialize access to DMA buffers from awb work queue */

struct regmap_field *pipeline[ISC_PIPE_LINE_NODE_NUM];
--
2.25.1

2022-01-22 01:31:09

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 10/11] ARM: configs: at91: sama7: add xisc and csi2dc

Enable XISC and CSI2DC drivers.

Signed-off-by: Eugen Hristev <[email protected]>
---
arch/arm/configs/sama7_defconfig | 2 ++
1 file changed, 2 insertions(+)

diff --git a/arch/arm/configs/sama7_defconfig b/arch/arm/configs/sama7_defconfig
index 938aae4bd80b..15978f2ab4ea 100644
--- a/arch/arm/configs/sama7_defconfig
+++ b/arch/arm/configs/sama7_defconfig
@@ -126,6 +126,8 @@ CONFIG_MEDIA_SUPPORT_FILTER=y
CONFIG_MEDIA_CAMERA_SUPPORT=y
CONFIG_MEDIA_PLATFORM_SUPPORT=y
CONFIG_V4L_PLATFORM_DRIVERS=y
+CONFIG_VIDEO_ATMEL_XISC=y
+CONFIG_VIDEO_MICROCHIP_CSI2DC=y
CONFIG_VIDEO_IMX219=m
CONFIG_VIDEO_IMX274=m
CONFIG_VIDEO_OV5647=m
--
2.25.1

2022-01-22 01:31:09

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 06/11] media: atmel: atmel-isc: compact the controller formats list

Compact the list array to be more readable.
No other changes, only cosmetic.

Signed-off-by: Eugen Hristev <[email protected]>
Reviewed-by: Jacopo Mondi <[email protected]>
---
.../media/platform/atmel/atmel-sama5d2-isc.c | 51 ++++++----------
.../media/platform/atmel/atmel-sama7g5-isc.c | 60 +++++++------------
2 files changed, 37 insertions(+), 74 deletions(-)

diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
index 025c3e8a7e95..d96ee3373889 100644
--- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
+++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
@@ -60,56 +60,39 @@
static const struct isc_format sama5d2_controller_formats[] = {
{
.fourcc = V4L2_PIX_FMT_ARGB444,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_ARGB555,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_RGB565,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_ABGR32,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_XBGR32,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUV420,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUYV,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUV422P,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_GREY,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_Y10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SBGGR8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGBRG8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGRBG8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SRGGB8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SBGGR10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGBRG10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGRBG10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SRGGB10,
},
};
diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
index 9dc75eed0098..e07ae188c15f 100644
--- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
+++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
@@ -63,65 +63,45 @@
static const struct isc_format sama7g5_controller_formats[] = {
{
.fourcc = V4L2_PIX_FMT_ARGB444,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_ARGB555,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_RGB565,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_ABGR32,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_XBGR32,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUV420,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_UYVY,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_VYUY,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUYV,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_YUV422P,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_GREY,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_Y10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_Y16,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SBGGR8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGBRG8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGRBG8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SRGGB8,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SBGGR10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGBRG10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SGRBG10,
- },
- {
+ }, {
.fourcc = V4L2_PIX_FMT_SRGGB10,
},
};
--
2.25.1

2022-01-22 01:31:26

by Eugen Hristev

[permalink] [raw]
Subject: [PATCH v4 11/11] ARM: multi_v7_defconfig: add atmel video pipeline modules

Add drivers for the atmel video capture pipeline: atmel isc, xisc and
microchip csi2dc.

Signed-off-by: Eugen Hristev <[email protected]>
---
arch/arm/configs/multi_v7_defconfig | 3 +++
1 file changed, 3 insertions(+)

diff --git a/arch/arm/configs/multi_v7_defconfig b/arch/arm/configs/multi_v7_defconfig
index c951aeed2138..92b7749a6df8 100644
--- a/arch/arm/configs/multi_v7_defconfig
+++ b/arch/arm/configs/multi_v7_defconfig
@@ -638,7 +638,10 @@ CONFIG_VIDEO_S5P_MIPI_CSIS=m
CONFIG_VIDEO_EXYNOS_FIMC_LITE=m
CONFIG_VIDEO_EXYNOS4_FIMC_IS=m
CONFIG_VIDEO_RCAR_VIN=m
+CONFIG_VIDEO_ATMEL_ISC=m
+CONFIG_VIDEO_ATMEL_XISC=m
CONFIG_VIDEO_ATMEL_ISI=m
+CONFIG_VIDEO_MICROCHIP_CSI2DC=m
CONFIG_V4L_MEM2MEM_DRIVERS=y
CONFIG_VIDEO_SAMSUNG_S5P_JPEG=m
CONFIG_VIDEO_SAMSUNG_S5P_MFC=m
--
2.25.1

2022-02-07 16:40:25

by Rob Herring (Arm)

[permalink] [raw]
Subject: Re: [PATCH v4 08/11] dt-bindings: media: microchip,xisc: add bus-width of 14

On Fri, 21 Jan 2022 15:14:13 +0200, Eugen Hristev wrote:
> The Microchip XISC supports a bus width of 14 bits.
> Add it to the supported bus widths.
>
> Signed-off-by: Eugen Hristev <[email protected]>
> ---
> Documentation/devicetree/bindings/media/microchip,xisc.yaml | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>

Acked-by: Rob Herring <[email protected]>

2022-02-08 01:10:08

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> Implement the support for media-controller.
> This means that the capabilities of the driver have changed and now
> it also advertises the IO_MC .
> The driver will register it's media device, and add the video entity to this
> media device. The subdevices are registered to the same media device.
> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> It will also register a subdevice that allows cropping of the incoming frame
> to the maximum frame size supported by the ISC.
> The ISC will create a link between the subdevice that is asynchronously
> registered and the atmel_isc_scaler entity.
> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> link.
>
> Signed-off-by: Eugen Hristev <[email protected]>
> ---
> Changes in v4:
> As suggested by Jacopo:
> - renamed atmel_isc_mc to atmel_isc_scaler.c
> - moved init_mc/clean_mc to isc_base file
>
> Changes in v2:
> - implement try formats
>
> drivers/media/platform/atmel/Makefile | 2 +-
> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> 6 files changed, 375 insertions(+), 8 deletions(-)
> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>
> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> index 794e8f739287..f02d03df89d6 100644
> --- a/drivers/media/platform/atmel/Makefile
> +++ b/drivers/media/platform/atmel/Makefile
> @@ -1,7 +1,7 @@
> # SPDX-License-Identifier: GPL-2.0-only
> atmel-isc-objs = atmel-sama5d2-isc.o
> atmel-xisc-objs = atmel-sama7g5-isc.o
> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>
> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> index 6b0005987a17..6b482270eb93 100644
> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> struct isc_device, v4l2_dev);
> struct isc_subdev_entity *subdev_entity =
> container_of(notifier, struct isc_subdev_entity, notifier);
> + int pad;
>
> if (video_is_registered(&isc->video_dev)) {
> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>
> subdev_entity->sd = subdev;
>
> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> + MEDIA_PAD_FL_SOURCE);
> + if (pad < 0) {
> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> + subdev->name);
> + return pad;
> + }
> +
> + isc->remote_pad = pad;
> +
> return 0;
> }
>
> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> }
>
> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> - unsigned int code, int *index)
> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> + unsigned int code, int *index)
> {
> struct isc_format *fmt = &isc->formats_list[0];
> unsigned int i;
> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>
> return NULL;
> }
> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>
> static int isc_formats_init(struct isc_device *isc)
> {
> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> NULL, &mbus_code)) {
> mbus_code.index++;
>
> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> if (!fmt) {
> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> mbus_code.code);
> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> vdev->queue = q;
> vdev->lock = &isc->lock;
> vdev->ctrl_handler = &isc->ctrls.handler;
> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> + V4L2_CAP_IO_MC;
> video_set_drvdata(vdev, isc);
>
> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> goto isc_async_complete_err;
> }
>
> + ret = isc_scaler_link(isc);
> + if (ret < 0)
> + goto isc_async_complete_unregister_device;
> +
> + ret = media_device_register(&isc->mdev);
> + if (ret < 0)
> + goto isc_async_complete_unregister_device;
> return 0;
>
> +isc_async_complete_unregister_device:
> + video_unregister_device(vdev);
> +
> isc_async_complete_err:
> mutex_destroy(&isc->lock);
> return ret;
> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> }
> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>
> +int isc_mc_init(struct isc_device *isc, u32 ver)
> +{
> + const struct of_device_id *match;
> + int ret;
> +
> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;

Should you set entity.ops.link_validate = v4l2_subdev_link_validate
to be able to have the media pipeline validated at
media_pipeline_start() time ?

> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> +
> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> + isc->pads);
> + if (ret < 0) {
> + dev_err(isc->dev, "media entity init failed\n");
> + return ret;
> + }
> +
> + isc->mdev.dev = isc->dev;
> +
> + match = of_match_node(isc->dev->driver->of_match_table,
> + isc->dev->of_node);
> +
> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> + sizeof(isc->mdev.driver_name));
> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> + isc->v4l2_dev.name);
> + isc->mdev.hw_revision = ver;
> +
> + media_device_init(&isc->mdev);
> +
> + isc->v4l2_dev.mdev = &isc->mdev;
> +
> + return isc_scaler_init(isc);
> +}
> +EXPORT_SYMBOL_GPL(isc_mc_init);
> +
> +void isc_mc_cleanup(struct isc_device *isc)
> +{
> + media_entity_cleanup(&isc->video_dev.entity);
> +}
> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> +
> /* regmap configuration */
> #define ATMEL_ISC_REG_MAX 0xd5c
> const struct regmap_config isc_regmap_config = {
> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> new file mode 100644
> index 000000000000..ec95c9665883
> --- /dev/null
> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> @@ -0,0 +1,245 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> + *
> + * Copyright (C) 2021 Microchip Technology, Inc.

Time flies! It's 2022 already :)

> + *
> + * Author: Eugen Hristev <[email protected]>
> + *
> + */
> +
> +#include <media/media-device.h>
> +#include <media/media-entity.h>
> +#include <media/v4l2-device.h>
> +#include <media/v4l2-subdev.h>
> +
> +#include "atmel-isc-regs.h"
> +#include "atmel-isc.h"
> +
> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> + struct v4l2_subdev_state *sd_state,
> + struct v4l2_subdev_format *format)
> +{
> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> +
> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> + format->pad);
> + format->format = *v4l2_try_fmt;
> +
> + return 0;
> + }
> +
> + format->format = isc->scaler_format;

isc->scaler_format is only used inside this file if I'm not mistaken.
I wonder why it lives in the isc_device struct.

> +
> + return 0;
> +}
> +
> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> + struct v4l2_subdev_state *sd_state,
> + struct v4l2_subdev_format *req_fmt)
> +{
> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> + struct isc_format *fmt;
> + unsigned int i;
> +
> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> + v4l_bound_align_image
> + (&req_fmt->format.width, 16, isc->max_width, 0,
> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> + else
> + v4l_bound_align_image
> + (&req_fmt->format.width, 16, 10000, 0,
> + &req_fmt->format.height, 16, 10000, 0, 0);

Where does 10000 come from ?

> +
> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> + req_fmt->format.field = V4L2_FIELD_NONE;
> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> +
> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);

So you rely on the isc format list for the scaler as well ?
I think it's fine as long as they are identical

> +
> + if (!fmt)
> + fmt = &isc->formats_list[0];
> +
> + req_fmt->format.code = fmt->mbus_code;
> +
> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> + req_fmt->pad);
> + *v4l2_try_fmt = req_fmt->format;
> + /* Trying on the pad sink makes the source sink change too */
> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> + v4l2_try_fmt =
> + v4l2_subdev_get_try_format(sd, sd_state,
> + ISC_SCALER_PAD_SOURCE);
> + *v4l2_try_fmt = req_fmt->format;
> +
> + v4l_bound_align_image(&v4l2_try_fmt->width,
> + 16, isc->max_width, 0,
> + &v4l2_try_fmt->height,
> + 16, isc->max_height, 0, 0);
> + }
> + /* if we are just trying, we are done */
> + return 0;
> + }
> +
> + isc->scaler_format = req_fmt->format;

No per-pad format but a global scaler_format ? How do you configure scaling ?

Actually, I would like to know more about the scaler device
capabilities. What functions can this IP perform ? Does it do
cropping, can it also do (down)scaling or even composition ?

I think it is worth to read
https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
where it is reported how cropping and scaling are implemented by using
the selection API.

Figure 4.5 is particularly helpful to explain the simple crop case,
for which you need to implement support to by adding s_selection on
the sink pad TGT_CROP target.

> +
> + return 0;
> +}
> +
> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> + struct v4l2_subdev_state *sd_state,
> + struct v4l2_subdev_mbus_code_enum *code)
> +{
> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> + int supported_index = 0;
> + int i;
> +
> + for (i = 0; i < isc->formats_list_size; i++) {
> + if (!isc->formats_list[i].sd_support)
> + continue;

The sd_support flag still doesn't click in my head.

Shouldn't the enumeration of available formats on the scaler do not
depend on the sensor supproted formats ?

> + if (supported_index == code->index) {
> + code->code = isc->formats_list[i].mbus_code;
> + return 0;
> + }
> + supported_index++;
> + }
> +
> + return -EINVAL;
> +}
> +
> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> + struct v4l2_subdev_state *sd_state,
> + struct v4l2_subdev_selection *sel)
> +{
> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> +
> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> + return -EINVAL;
> +
> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> + sel->target != V4L2_SEL_TGT_CROP)
> + return -EINVAL;
> +
> + sel->r.height = isc->max_height;
> + sel->r.width = isc->max_width;

The CROP_BOUNDS should be set to the same size as the sink pad image format,
as it represents the maximum valid crop rectangle.

TGT_CROP should report the configured crop rectangle which can be
intiialized to the same size as CROP_BOUNDS, if my understanding of
the spec is correct

> +
> + sel->r.left = 0;
> + sel->r.top = 0;
> +
> + return 0;
> +}
> +
> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> + struct v4l2_subdev_state *sd_state)
> +{
> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> +
> + *v4l2_try_fmt = isc->scaler_format;
> +
> + return 0;
> +}
> +
> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> + .set_fmt = isc_scaler_set_fmt,
> + .get_fmt = isc_scaler_get_fmt,
> + .get_selection = isc_scaler_g_sel,
> + .init_cfg = isc_scaler_init_cfg,

.link_validate = v4l2_subdev_link_validate_default,

To have the formats at the end of links that point to this entity
validated (I think the framework already calls it if not set though,
please check v4l2-subdev.c:v4l2_subdev_link_validate())

> +};
> +
> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> + .pad = &isc_scaler_pad_ops,
> +};
> +
> +int isc_scaler_init(struct isc_device *isc)
> +{
> + int ret;
> +
> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> +
> + isc->scaler_sd.owner = THIS_MODULE;
> + isc->scaler_sd.dev = isc->dev;
> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> + "atmel_isc_scaler");

I would drop 'atmel' for brevity, unless other entities have this
prefix set already

> +
> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> +
> + isc->scaler_format.height = isc->max_height;
> + isc->scaler_format.width = isc->max_width;
> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> + isc->scaler_format.field = V4L2_FIELD_NONE;
> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> +
> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> + ISC_SCALER_PADS_NUM,
> + isc->scaler_pads);
> + if (ret < 0) {
> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> + return ret;
> + }
> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> + if (ret < 0) {
> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> + return ret;
> + }
> +
> + return ret;
> +}
> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> +
> +int isc_scaler_link(struct isc_device *isc)
> +{
> + int ret;
> +
> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> + isc->remote_pad, &isc->scaler_sd.entity,
> + ISC_SCALER_PAD_SINK,
> + MEDIA_LNK_FL_ENABLED |
> + MEDIA_LNK_FL_IMMUTABLE);
> +
> + if (ret < 0) {
> + v4l2_err(&isc->v4l2_dev,
> + "Failed to create pad link: %s to %s\n",
> + isc->current_subdev->sd->entity.name,
> + isc->scaler_sd.entity.name);
> + return ret;
> + }
> +
> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> + isc->current_subdev->sd->name, isc->remote_pad);
> +
> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> + ISC_SCALER_PAD_SOURCE,
> + &isc->video_dev.entity, ISC_PAD_SINK,
> + MEDIA_LNK_FL_ENABLED |
> + MEDIA_LNK_FL_IMMUTABLE);
> +
> + if (ret < 0) {
> + v4l2_err(&isc->v4l2_dev,
> + "Failed to create pad link: %s to %s\n",
> + isc->scaler_sd.entity.name,
> + isc->video_dev.entity.name);
> + return ret;
> + }
> +
> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> + ISC_SCALER_PAD_SOURCE);
> +
> + return ret;
> +}
> +EXPORT_SYMBOL_GPL(isc_scaler_link);

From the DT graph point of view, the ISC appears as a single block
with an CSI-2 input port. It is then in charge of parsing the .dts and
add to its notifier the image sensor remote subdevice.

When the sensor subdev registers and the ISC notifier completes, the scaler
entity which was initialized at isc_probe() time is linked in between
the ISC and the image sensor immutably.

I think it is fine for now, but I wonder if you plan to plumb more
components between the ISC video node and the sensor, if it's not
worth changing the DT bindings and their parsing logic to separate the
CSI-2 receiver from the ISC, whcih can create its media graph at probe
time and have the CSI-2 receiver entity as downstream remote. I think
I need to get to know the ISC better to really have an idea. For now,
this seems ok to me, but please check with maintainers if this is fine
with them.

> +
> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> index 5fbf52a9080b..c9234c90ae58 100644
> --- a/drivers/media/platform/atmel/atmel-isc.h
> +++ b/drivers/media/platform/atmel/atmel-isc.h
> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> u32 his_entry;
> };
>
> +enum isc_mc_pads {
> + ISC_PAD_SINK = 0,
> + ISC_PADS_NUM = 1,
> +};
> +
> +enum isc_scaler_pads {
> + ISC_SCALER_PAD_SINK = 0,
> + ISC_SCALER_PAD_SOURCE = 1,
> + ISC_SCALER_PADS_NUM = 2,
> +};
> +
> /*
> * struct isc_device - ISC device driver data/config struct
> * @regmap: Register map
> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> * be used as an input to the controller
> * @controller_formats_size: size of controller_formats array
> * @formats_list_size: size of formats_list array
> + * @pads: media controller pads for isc video entity
> + * @mdev: media device that is registered by the isc
> + * @remote_pad: remote pad on the connected subdevice
> + * @scaler_sd: subdevice for the scaler that isc registers
> + * @scaler_pads: media controller pads for the scaler subdevice
> + * @scaler_format: current format for the scaler subdevice
> */
> struct isc_device {
> struct regmap *regmap;
> @@ -344,6 +361,19 @@ struct isc_device {
> struct isc_format *formats_list;
> u32 controller_formats_size;
> u32 formats_list_size;
> +
> + struct {
> + struct media_pad pads[ISC_PADS_NUM];
> + struct media_device mdev;
> +
> + u32 remote_pad;
> + };
> +
> + struct {
> + struct v4l2_subdev scaler_sd;
> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> + struct v4l2_mbus_framefmt scaler_format;
> + };
> };
>
> extern const struct regmap_config isc_regmap_config;
> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> void isc_subdev_cleanup(struct isc_device *isc);
> void isc_clk_cleanup(struct isc_device *isc);
>
> +int isc_scaler_link(struct isc_device *isc);
> +int isc_scaler_init(struct isc_device *isc);
> +int isc_mc_init(struct isc_device *isc, u32 ver);
> +void isc_mc_cleanup(struct isc_device *isc);
> +
> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> + unsigned int code, int *index);
> #endif
> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> index c5b9563e36cb..c244682ea22f 100644
> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> break;
> }
>
> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);

I am surprised you can read a register before runtime_pm is
intialized!

Thanks
j


> +
> + ret = isc_mc_init(isc, ver);
> + if (ret < 0)
> + goto isc_probe_mc_init_err;
> +
> pm_runtime_set_active(dev);
> pm_runtime_enable(dev);
> pm_request_idle(dev);
> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> ret = clk_prepare_enable(isc->ispck);
> if (ret) {
> dev_err(dev, "failed to enable ispck: %d\n", ret);
> - goto cleanup_subdev;
> + goto isc_probe_mc_init_err;
> }
>
> /* ispck should be greater or equal to hclock */
> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> goto unprepare_clk;
> }
>
> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> dev_info(dev, "Microchip ISC version %x\n", ver);
>
> return 0;
> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> unprepare_clk:
> clk_disable_unprepare(isc->ispck);
>
> +isc_probe_mc_init_err:
> + isc_mc_cleanup(isc);
> +
> cleanup_subdev:
> isc_subdev_cleanup(isc);
>
> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>
> pm_runtime_disable(&pdev->dev);
>
> + isc_mc_cleanup(isc);
> +
> isc_subdev_cleanup(isc);
>
> v4l2_device_unregister(&isc->v4l2_dev);
> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> index 07a80b08bc54..9dc75eed0098 100644
> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> break;
> }
>
> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> +
> + ret = isc_mc_init(isc, ver);
> + if (ret < 0)
> + goto isc_probe_mc_init_err;
> +
> pm_runtime_set_active(dev);
> pm_runtime_enable(dev);
> pm_request_idle(dev);
>
> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> dev_info(dev, "Microchip XISC version %x\n", ver);
>
> return 0;
>
> +isc_probe_mc_init_err:
> + isc_mc_cleanup(isc);
> +
> cleanup_subdev:
> isc_subdev_cleanup(isc);
>
> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>
> pm_runtime_disable(&pdev->dev);
>
> + isc_mc_cleanup(isc);
> +
> isc_subdev_cleanup(isc);
>
> v4l2_device_unregister(&isc->v4l2_dev);
> --
> 2.25.1
>

2022-02-08 13:22:34

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
>> Implement the support for media-controller.
>> This means that the capabilities of the driver have changed and now
>> it also advertises the IO_MC .
>> The driver will register it's media device, and add the video entity to this
>> media device. The subdevices are registered to the same media device.
>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
>> It will also register a subdevice that allows cropping of the incoming frame
>> to the maximum frame size supported by the ISC.
>> The ISC will create a link between the subdevice that is asynchronously
>> registered and the atmel_isc_scaler entity.
>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
>> link.
>>
>> Signed-off-by: Eugen Hristev <[email protected]>
>> ---
>> Changes in v4:
>> As suggested by Jacopo:
>> - renamed atmel_isc_mc to atmel_isc_scaler.c
>> - moved init_mc/clean_mc to isc_base file
>>
>> Changes in v2:
>> - implement try formats
>>
>> drivers/media/platform/atmel/Makefile | 2 +-
>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
>> 6 files changed, 375 insertions(+), 8 deletions(-)
>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>>
>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
>> index 794e8f739287..f02d03df89d6 100644
>> --- a/drivers/media/platform/atmel/Makefile
>> +++ b/drivers/media/platform/atmel/Makefile
>> @@ -1,7 +1,7 @@
>> # SPDX-License-Identifier: GPL-2.0-only
>> atmel-isc-objs = atmel-sama5d2-isc.o
>> atmel-xisc-objs = atmel-sama7g5-isc.o
>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>>
>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>> index 6b0005987a17..6b482270eb93 100644
>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>> struct isc_device, v4l2_dev);
>> struct isc_subdev_entity *subdev_entity =
>> container_of(notifier, struct isc_subdev_entity, notifier);
>> + int pad;
>>
>> if (video_is_registered(&isc->video_dev)) {
>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>
>> subdev_entity->sd = subdev;
>>
>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
>> + MEDIA_PAD_FL_SOURCE);
>> + if (pad < 0) {
>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
>> + subdev->name);
>> + return pad;
>> + }
>> +
>> + isc->remote_pad = pad;
>> +
>> return 0;
>> }
>>
>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
>> }
>>
>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
>> - unsigned int code, int *index)
>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>> + unsigned int code, int *index)
>> {
>> struct isc_format *fmt = &isc->formats_list[0];
>> unsigned int i;
>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>>
>> return NULL;
>> }
>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>>
>> static int isc_formats_init(struct isc_device *isc)
>> {
>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
>> NULL, &mbus_code)) {
>> mbus_code.index++;
>>
>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
>> if (!fmt) {
>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
>> mbus_code.code);
>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>> vdev->queue = q;
>> vdev->lock = &isc->lock;
>> vdev->ctrl_handler = &isc->ctrls.handler;
>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
>> + V4L2_CAP_IO_MC;
>> video_set_drvdata(vdev, isc);
>>
>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>> goto isc_async_complete_err;
>> }
>>
>> + ret = isc_scaler_link(isc);
>> + if (ret < 0)
>> + goto isc_async_complete_unregister_device;
>> +
>> + ret = media_device_register(&isc->mdev);
>> + if (ret < 0)
>> + goto isc_async_complete_unregister_device;
>> return 0;
>>
>> +isc_async_complete_unregister_device:
>> + video_unregister_device(vdev);
>> +
>> isc_async_complete_err:
>> mutex_destroy(&isc->lock);
>> return ret;
>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
>> }
>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>
>> +int isc_mc_init(struct isc_device *isc, u32 ver)
>> +{
>> + const struct of_device_id *match;
>> + int ret;
>> +
>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>
> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> to be able to have the media pipeline validated at
> media_pipeline_start() time ?

Hi,

I am doing that in a subsequent patch. Things are not completely ready
at this moment, because ISC still relies on the old mechanism to call
v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
removes and adds checks to the link validate ).

>
>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>> +
>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>> + isc->pads);
>> + if (ret < 0) {
>> + dev_err(isc->dev, "media entity init failed\n");
>> + return ret;
>> + }
>> +
>> + isc->mdev.dev = isc->dev;
>> +
>> + match = of_match_node(isc->dev->driver->of_match_table,
>> + isc->dev->of_node);
>> +
>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
>> + sizeof(isc->mdev.driver_name));
>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
>> + isc->v4l2_dev.name);
>> + isc->mdev.hw_revision = ver;
>> +
>> + media_device_init(&isc->mdev);
>> +
>> + isc->v4l2_dev.mdev = &isc->mdev;
>> +
>> + return isc_scaler_init(isc);
>> +}
>> +EXPORT_SYMBOL_GPL(isc_mc_init);
>> +
>> +void isc_mc_cleanup(struct isc_device *isc)
>> +{
>> + media_entity_cleanup(&isc->video_dev.entity);
>> +}
>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
>> +
>> /* regmap configuration */
>> #define ATMEL_ISC_REG_MAX 0xd5c
>> const struct regmap_config isc_regmap_config = {
>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>> new file mode 100644
>> index 000000000000..ec95c9665883
>> --- /dev/null
>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>> @@ -0,0 +1,245 @@
>> +// SPDX-License-Identifier: GPL-2.0-only
>> +/*
>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
>> + *
>> + * Copyright (C) 2021 Microchip Technology, Inc.
>
> Time flies! It's 2022 already :)
>
>> + *
>> + * Author: Eugen Hristev <[email protected]>
>> + *
>> + */
>> +
>> +#include <media/media-device.h>
>> +#include <media/media-entity.h>
>> +#include <media/v4l2-device.h>
>> +#include <media/v4l2-subdev.h>
>> +
>> +#include "atmel-isc-regs.h"
>> +#include "atmel-isc.h"
>> +
>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
>> + struct v4l2_subdev_state *sd_state,
>> + struct v4l2_subdev_format *format)
>> +{
>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>> +
>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>> + format->pad);
>> + format->format = *v4l2_try_fmt;
>> +
>> + return 0;
>> + }
>> +
>> + format->format = isc->scaler_format;
>
> isc->scaler_format is only used inside this file if I'm not mistaken.
> I wonder why it lives in the isc_device struct.

isc_device is a placeholder for all isc things.

I would not create a separate struct in this file that would have to be
allocated etc... just for one two things. So I preferred to have it in
the same place as all the other things.

>
>> +
>> + return 0;
>> +}
>> +
>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
>> + struct v4l2_subdev_state *sd_state,
>> + struct v4l2_subdev_format *req_fmt)
>> +{
>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>> + struct isc_format *fmt;
>> + unsigned int i;
>> +
>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
>> + v4l_bound_align_image
>> + (&req_fmt->format.width, 16, isc->max_width, 0,
>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
>> + else
>> + v4l_bound_align_image
>> + (&req_fmt->format.width, 16, 10000, 0,
>> + &req_fmt->format.height, 16, 10000, 0, 0);
>
> Where does 10000 come from ?

It's a random number. Do you have any suggestion for a better one ?
Maybe this would be much more clear with my comments below (where I will
explain what the isc scaler does )
In short, it allows the other entity in the link to 'sink' a huge format
into this 'scaler' . Because the scaler can crop anything basically down
to the biggest format size the ISC could handle.

>
>> +
>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
>> + req_fmt->format.field = V4L2_FIELD_NONE;
>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>> +
>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
>
> So you rely on the isc format list for the scaler as well ?
> I think it's fine as long as they are identical

Yes, the scaler is kind of a dummy entity , but it's required by the
media controller validation of the pipeline.

>
>> +
>> + if (!fmt)
>> + fmt = &isc->formats_list[0];
>> +
>> + req_fmt->format.code = fmt->mbus_code;
>> +
>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>> + req_fmt->pad);
>> + *v4l2_try_fmt = req_fmt->format;
>> + /* Trying on the pad sink makes the source sink change too */
>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
>> + v4l2_try_fmt =
>> + v4l2_subdev_get_try_format(sd, sd_state,
>> + ISC_SCALER_PAD_SOURCE);
>> + *v4l2_try_fmt = req_fmt->format;
>> +
>> + v4l_bound_align_image(&v4l2_try_fmt->width,
>> + 16, isc->max_width, 0,
>> + &v4l2_try_fmt->height,
>> + 16, isc->max_height, 0, 0);
>> + }
>> + /* if we are just trying, we are done */
>> + return 0;
>> + }
>> +
>> + isc->scaler_format = req_fmt->format;
>
> No per-pad format but a global scaler_format ? How do you configure scaling ?
>
> Actually, I would like to know more about the scaler device
> capabilities. What functions can this IP perform ? Does it do
> cropping, can it also do (down)scaling or even composition ?
>
> I think it is worth to read
> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> where it is reported how cropping and scaling are implemented by using
> the selection API.
>
> Figure 4.5 is particularly helpful to explain the simple crop case,
> for which you need to implement support to by adding s_selection on
> the sink pad TGT_CROP target.

The scaler is kind of a dummy entity.
The problem is that different subdevices cannot be connected directly to
a v4l entity, because the ISC does cropping on the incoming frame.
The ISC has a limit on the total number of pixels per frame, thus it
will do a crop.
So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
only handle 3264x2464 , then we have two choices:
1/ crop it and output the 3264x2464 -> cannot, because userspace expects
3280x2464, and the whole pipeline fails to configure, there is a
mismatch between /dev/video output and what the whole pipeline produces
2/ deny the link, and ask the subdevice to produce the 3264x2464 which
the ISC can handle at most -> cannot , because the sensor let's say can
*only* produce 3280x2464 and *any* other frame size returns an error.

The only solution to make both worlds happy, is to have a dummy entity
called 'scaler' which in fact now it only performs a simple crop.
It accepts any frame size at input (hence the 10000x10000 which you saw
earlier), and outputs at most 3264x2464 the isc_max_height and
isc_max_width.

So to answer your question, the isc scaler is a software model for a
simple cropping, that would make media controller happy, and capture
software happy.

Here it how it looks :

Device topology
- entity 1: atmel_isc_scaler (2 pads, 2 links)
type V4L2 subdev subtype Unknown flags 0
device node name /dev/v4l-subdev0
pad0: Sink
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
crop.bounds:(0,0)/3264x2464
crop:(0,0)/3264x2464]
<- "csi2dc":1 [ENABLED,IMMUTABLE]
pad1: Source
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
-> "atmel_isc_common":0 [ENABLED,IMMUTABLE]

- entity 4: csi2dc (2 pads, 2 links)
type V4L2 subdev subtype Unknown flags 0
device node name /dev/v4l-subdev1
pad0: Sink
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
<- "dw-csi.0":1 [ENABLED]
pad1: Source
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
-> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]

- entity 7: dw-csi.0 (2 pads, 2 links)
type V4L2 subdev subtype Unknown flags 0
device node name /dev/v4l-subdev2
pad0: Sink
[fmt:SRGGB10_1X10/3280x2464]
<- "imx219 1-0010":0 [ENABLED]
pad1: Source
[fmt:SRGGB10_1X10/3280x2464]
-> "csi2dc":0 [ENABLED]

- entity 12: imx219 1-0010 (1 pad, 1 link)
type V4L2 subdev subtype Sensor flags 0
device node name /dev/v4l-subdev3
pad0: Source
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
xfer:srgb ycbcr:601 quantization:full-range
crop.bounds:(8,8)/3280x2464
crop:(8,8)/3280x2464]
-> "dw-csi.0":0 [ENABLED]

- entity 24: atmel_isc_common (1 pad, 1 link)
type Node subtype V4L flags 1
device node name /dev/video0
pad0: Sink
<- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]


Scaler does this one cute little thing :

pad0: Sink
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
crop.bounds:(0,0)/3264x2464
crop:(0,0)/3264x2464]
<- "csi2dc":1 [ENABLED,IMMUTABLE]
pad1: Source
[fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
-> "atmel_isc_common":0 [ENABLED,IMMUTABLE]


Which is what we needed.

>
>> +
>> + return 0;
>> +}
>> +
>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
>> + struct v4l2_subdev_state *sd_state,
>> + struct v4l2_subdev_mbus_code_enum *code)
>> +{
>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>> + int supported_index = 0;
>> + int i;
>> +
>> + for (i = 0; i < isc->formats_list_size; i++) {
>> + if (!isc->formats_list[i].sd_support)
>> + continue;
>
> The sd_support flag still doesn't click in my head.
>
> Shouldn't the enumeration of available formats on the scaler do not
> depend on the sensor supproted formats ?

You're right. I will have to check it again.

>
>> + if (supported_index == code->index) {
>> + code->code = isc->formats_list[i].mbus_code;
>> + return 0;
>> + }
>> + supported_index++;
>> + }
>> +
>> + return -EINVAL;
>> +}
>> +
>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
>> + struct v4l2_subdev_state *sd_state,
>> + struct v4l2_subdev_selection *sel)
>> +{
>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>> +
>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
>> + return -EINVAL;
>> +
>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
>> + sel->target != V4L2_SEL_TGT_CROP)
>> + return -EINVAL;
>> +
>> + sel->r.height = isc->max_height;
>> + sel->r.width = isc->max_width;
>
> The CROP_BOUNDS should be set to the same size as the sink pad image format,
> as it represents the maximum valid crop rectangle.
>
> TGT_CROP should report the configured crop rectangle which can be
> intiialized to the same size as CROP_BOUNDS, if my understanding of
> the spec is correct

So you would like to have this differentiated, and report the
CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
reported now, the maximum size of the ISC frame .
My understanding is correct ?

>
>> +
>> + sel->r.left = 0;
>> + sel->r.top = 0;
>> +
>> + return 0;
>> +}
>> +
>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
>> + struct v4l2_subdev_state *sd_state)
>> +{
>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>> +
>> + *v4l2_try_fmt = isc->scaler_format;
>> +
>> + return 0;
>> +}
>> +
>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
>> + .set_fmt = isc_scaler_set_fmt,
>> + .get_fmt = isc_scaler_get_fmt,
>> + .get_selection = isc_scaler_g_sel,
>> + .init_cfg = isc_scaler_init_cfg,
>
> .link_validate = v4l2_subdev_link_validate_default,
>
> To have the formats at the end of links that point to this entity
> validated (I think the framework already calls it if not set though,
> please check v4l2-subdev.c:v4l2_subdev_link_validate())
>
>> +};
>> +
>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>> + .pad = &isc_scaler_pad_ops,
>> +};
>> +
>> +int isc_scaler_init(struct isc_device *isc)
>> +{
>> + int ret;
>> +
>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
>> +
>> + isc->scaler_sd.owner = THIS_MODULE;
>> + isc->scaler_sd.dev = isc->dev;
>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
>> + "atmel_isc_scaler");
>
> I would drop 'atmel' for brevity, unless other entities have this
> prefix set already

The v4l entity takes it's name from the module name, which is
atmel_isc_common, so I thought to keep the prefix

>
>> +
>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>> +
>> + isc->scaler_format.height = isc->max_height;
>> + isc->scaler_format.width = isc->max_width;
>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
>> + isc->scaler_format.field = V4L2_FIELD_NONE;
>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>> +
>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
>> + ISC_SCALER_PADS_NUM,
>> + isc->scaler_pads);
>> + if (ret < 0) {
>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
>> + return ret;
>> + }
>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
>> + if (ret < 0) {
>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
>> + return ret;
>> + }
>> +
>> + return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
>> +
>> +int isc_scaler_link(struct isc_device *isc)
>> +{
>> + int ret;
>> +
>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
>> + isc->remote_pad, &isc->scaler_sd.entity,
>> + ISC_SCALER_PAD_SINK,
>> + MEDIA_LNK_FL_ENABLED |
>> + MEDIA_LNK_FL_IMMUTABLE);
>> +
>> + if (ret < 0) {
>> + v4l2_err(&isc->v4l2_dev,
>> + "Failed to create pad link: %s to %s\n",
>> + isc->current_subdev->sd->entity.name,
>> + isc->scaler_sd.entity.name);
>> + return ret;
>> + }
>> +
>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
>> + isc->current_subdev->sd->name, isc->remote_pad);
>> +
>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
>> + ISC_SCALER_PAD_SOURCE,
>> + &isc->video_dev.entity, ISC_PAD_SINK,
>> + MEDIA_LNK_FL_ENABLED |
>> + MEDIA_LNK_FL_IMMUTABLE);
>> +
>> + if (ret < 0) {
>> + v4l2_err(&isc->v4l2_dev,
>> + "Failed to create pad link: %s to %s\n",
>> + isc->scaler_sd.entity.name,
>> + isc->video_dev.entity.name);
>> + return ret;
>> + }
>> +
>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
>> + ISC_SCALER_PAD_SOURCE);
>> +
>> + return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
>
> From the DT graph point of view, the ISC appears as a single block
> with an CSI-2 input port. It is then in charge of parsing the .dts and
> add to its notifier the image sensor remote subdevice.

Actually it's only a parallel port.
And it can be connected either to a sensor directly or to the csi2dc
bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)

>
> When the sensor subdev registers and the ISC notifier completes, the scaler
> entity which was initialized at isc_probe() time is linked in between
> the ISC and the image sensor immutably.

yes, because this cannot change at runtime, and usually, it can't change
without altering the board hardware. (at least this is what my
understanding of immutability is )

>
> I think it is fine for now, but I wonder if you plan to plumb more
> components between the ISC video node and the sensor, if it's not
> worth changing the DT bindings and their parsing logic to separate the
> CSI-2 receiver from the ISC, whcih can create its media graph at probe
> time and have the CSI-2 receiver entity as downstream remote. I think
> I need to get to know the ISC better to really have an idea. For now,
> this seems ok to me, but please check with maintainers if this is fine
> with them.

In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
the remote subdev).
The ISC will register a scaler, and connect the subdev to this scaler
first, and then, the scaler to the isc itself (the v4l entity).

>
>> +
>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>> index 5fbf52a9080b..c9234c90ae58 100644
>> --- a/drivers/media/platform/atmel/atmel-isc.h
>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
>> u32 his_entry;
>> };
>>
>> +enum isc_mc_pads {
>> + ISC_PAD_SINK = 0,
>> + ISC_PADS_NUM = 1,
>> +};
>> +
>> +enum isc_scaler_pads {
>> + ISC_SCALER_PAD_SINK = 0,
>> + ISC_SCALER_PAD_SOURCE = 1,
>> + ISC_SCALER_PADS_NUM = 2,
>> +};
>> +
>> /*
>> * struct isc_device - ISC device driver data/config struct
>> * @regmap: Register map
>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
>> * be used as an input to the controller
>> * @controller_formats_size: size of controller_formats array
>> * @formats_list_size: size of formats_list array
>> + * @pads: media controller pads for isc video entity
>> + * @mdev: media device that is registered by the isc
>> + * @remote_pad: remote pad on the connected subdevice
>> + * @scaler_sd: subdevice for the scaler that isc registers
>> + * @scaler_pads: media controller pads for the scaler subdevice
>> + * @scaler_format: current format for the scaler subdevice
>> */
>> struct isc_device {
>> struct regmap *regmap;
>> @@ -344,6 +361,19 @@ struct isc_device {
>> struct isc_format *formats_list;
>> u32 controller_formats_size;
>> u32 formats_list_size;
>> +
>> + struct {
>> + struct media_pad pads[ISC_PADS_NUM];
>> + struct media_device mdev;
>> +
>> + u32 remote_pad;
>> + };
>> +
>> + struct {
>> + struct v4l2_subdev scaler_sd;
>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
>> + struct v4l2_mbus_framefmt scaler_format;

Here are the scaler stuff which I added, in the same bulk struct for the
whole isc device


>> + };
>> };
>>
>> extern const struct regmap_config isc_regmap_config;
>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
>> void isc_subdev_cleanup(struct isc_device *isc);
>> void isc_clk_cleanup(struct isc_device *isc);
>>
>> +int isc_scaler_link(struct isc_device *isc);
>> +int isc_scaler_init(struct isc_device *isc);
>> +int isc_mc_init(struct isc_device *isc, u32 ver);
>> +void isc_mc_cleanup(struct isc_device *isc);
>> +
>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>> + unsigned int code, int *index);
>> #endif
>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>> index c5b9563e36cb..c244682ea22f 100644
>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
>> break;
>> }
>>
>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>
> I am surprised you can read a register before runtime_pm is
> intialized!
>

Actually I can read any register after the moment of starting the clock
on the hardware block.
Maybe the starting and stopping of the clock needs to be moved to the
runtime_pm calls, but this is another story, not related to the media
controller.
I moved this line because I had to pass the version to the isc_mc_init call

> Thanks
> j

Thanks for reviewing !
Eugen
>
>
>> +
>> + ret = isc_mc_init(isc, ver);
>> + if (ret < 0)
>> + goto isc_probe_mc_init_err;
>> +
>> pm_runtime_set_active(dev);
>> pm_runtime_enable(dev);
>> pm_request_idle(dev);
>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
>> ret = clk_prepare_enable(isc->ispck);
>> if (ret) {
>> dev_err(dev, "failed to enable ispck: %d\n", ret);
>> - goto cleanup_subdev;
>> + goto isc_probe_mc_init_err;
>> }
>>
>> /* ispck should be greater or equal to hclock */
>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
>> goto unprepare_clk;
>> }
>>
>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>> dev_info(dev, "Microchip ISC version %x\n", ver);
>>
>> return 0;
>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
>> unprepare_clk:
>> clk_disable_unprepare(isc->ispck);
>>
>> +isc_probe_mc_init_err:
>> + isc_mc_cleanup(isc);
>> +
>> cleanup_subdev:
>> isc_subdev_cleanup(isc);
>>
>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>>
>> pm_runtime_disable(&pdev->dev);
>>
>> + isc_mc_cleanup(isc);
>> +
>> isc_subdev_cleanup(isc);
>>
>> v4l2_device_unregister(&isc->v4l2_dev);
>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>> index 07a80b08bc54..9dc75eed0098 100644
>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
>> break;
>> }
>>
>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>> +
>> + ret = isc_mc_init(isc, ver);
>> + if (ret < 0)
>> + goto isc_probe_mc_init_err;
>> +
>> pm_runtime_set_active(dev);
>> pm_runtime_enable(dev);
>> pm_request_idle(dev);
>>
>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>> dev_info(dev, "Microchip XISC version %x\n", ver);
>>
>> return 0;
>>
>> +isc_probe_mc_init_err:
>> + isc_mc_cleanup(isc);
>> +
>> cleanup_subdev:
>> isc_subdev_cleanup(isc);
>>
>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>>
>> pm_runtime_disable(&pdev->dev);
>>
>> + isc_mc_cleanup(isc);
>> +
>> isc_subdev_cleanup(isc);
>>
>> v4l2_device_unregister(&isc->v4l2_dev);
>> --
>> 2.25.1
>>

2022-02-09 06:46:11

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
> >> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> >>> Hi Eugen
> >>>
> >>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> >>>> Implement the support for media-controller.
> >>>> This means that the capabilities of the driver have changed and now
> >>>> it also advertises the IO_MC .
> >>>> The driver will register it's media device, and add the video entity to this
> >>>> media device. The subdevices are registered to the same media device.
> >>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> >>>> It will also register a subdevice that allows cropping of the incoming frame
> >>>> to the maximum frame size supported by the ISC.
> >>>> The ISC will create a link between the subdevice that is asynchronously
> >>>> registered and the atmel_isc_scaler entity.
> >>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> >>>> link.
> >>>>
> >>>> Signed-off-by: Eugen Hristev <[email protected]>
> >>>> ---
> >>>> Changes in v4:
> >>>> As suggested by Jacopo:
> >>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
> >>>> - moved init_mc/clean_mc to isc_base file
> >>>>
> >>>> Changes in v2:
> >>>> - implement try formats
> >>>>
> >>>> drivers/media/platform/atmel/Makefile | 2 +-
> >>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> >>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> >>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> >>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> >>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> >>>> 6 files changed, 375 insertions(+), 8 deletions(-)
> >>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>
> >>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> >>>> index 794e8f739287..f02d03df89d6 100644
> >>>> --- a/drivers/media/platform/atmel/Makefile
> >>>> +++ b/drivers/media/platform/atmel/Makefile
> >>>> @@ -1,7 +1,7 @@
> >>>> # SPDX-License-Identifier: GPL-2.0-only
> >>>> atmel-isc-objs = atmel-sama5d2-isc.o
> >>>> atmel-xisc-objs = atmel-sama7g5-isc.o
> >>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> >>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
> >>>>
> >>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> >>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> >>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>> index 6b0005987a17..6b482270eb93 100644
> >>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>> struct isc_device, v4l2_dev);
> >>>> struct isc_subdev_entity *subdev_entity =
> >>>> container_of(notifier, struct isc_subdev_entity, notifier);
> >>>> + int pad;
> >>>>
> >>>> if (video_is_registered(&isc->video_dev)) {
> >>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> >>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>
> >>>> subdev_entity->sd = subdev;
> >>>>
> >>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> >>>> + MEDIA_PAD_FL_SOURCE);
> >>>> + if (pad < 0) {
> >>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> >>>> + subdev->name);
> >>>> + return pad;
> >>>> + }
> >>>> +
> >>>> + isc->remote_pad = pad;
> >>>> +
> >>>> return 0;
> >>>> }
> >>>>
> >>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> >>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> >>>> }
> >>>>
> >>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>> - unsigned int code, int *index)
> >>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>> + unsigned int code, int *index)
> >>>> {
> >>>> struct isc_format *fmt = &isc->formats_list[0];
> >>>> unsigned int i;
> >>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>
> >>>> return NULL;
> >>>> }
> >>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
> >>>>
> >>>> static int isc_formats_init(struct isc_device *isc)
> >>>> {
> >>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> >>>> NULL, &mbus_code)) {
> >>>> mbus_code.index++;
> >>>>
> >>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> >>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> >>>> if (!fmt) {
> >>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> >>>> mbus_code.code);
> >>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>> vdev->queue = q;
> >>>> vdev->lock = &isc->lock;
> >>>> vdev->ctrl_handler = &isc->ctrls.handler;
> >>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> >>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> >>>> + V4L2_CAP_IO_MC;
> >>>> video_set_drvdata(vdev, isc);
> >>>>
> >>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> >>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>> goto isc_async_complete_err;
> >>>> }
> >>>>
> >>>> + ret = isc_scaler_link(isc);
> >>>> + if (ret < 0)
> >>>> + goto isc_async_complete_unregister_device;
> >>>> +
> >>>> + ret = media_device_register(&isc->mdev);
> >>>> + if (ret < 0)
> >>>> + goto isc_async_complete_unregister_device;
> >>>> return 0;
> >>>>
> >>>> +isc_async_complete_unregister_device:
> >>>> + video_unregister_device(vdev);
> >>>> +
> >>>> isc_async_complete_err:
> >>>> mutex_destroy(&isc->lock);
> >>>> return ret;
> >>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> >>>> }
> >>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>>>
> >>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
> >>>> +{
> >>>> + const struct of_device_id *match;
> >>>> + int ret;
> >>>> +
> >>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >>>
> >>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> >>> to be able to have the media pipeline validated at
> >>> media_pipeline_start() time ?
> >>
> >> Hi,
> >>
> >> I am doing that in a subsequent patch. Things are not completely ready
> >> at this moment, because ISC still relies on the old mechanism to call
> >> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
> >> removes and adds checks to the link validate ).
> >>
> >
> > I see.. the subsequent patches are not part of this series, right ?
>
> No. Patch 7 does this. In that patch, the link validation is created and
> all the logic of format propagation is removed and reworked, and creates
> the need for the link_validate call
> Could you have also a look at that patch ?

Ah ups, 7 was not in my inbox with the rest of the series. Just
noticed. I had even reviewed the previous version, I should have
remembered link_validate was added later

> >
> > I think patches 1, 2, 4, 5 and 6 from this series do not depend on
> > the introduction of media controller or the scaler, right ? (I'm not
> > sure if the DTS patches do).

Do DTS patches depend on media-controller plumbing ?

> >
> > If that's the case, would it make sense to to fast-track them (as some
> > of them are fixes) and then on top plumb the media controller
> > infrastructure with this patch and the ones you have been mentioning ?
>
> It would make sense.
>

Once the other discussion with Hans about the stop state it might make
sense to fastrack the first part of the series ?

> >
> >>>
> >>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>> +
> >>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >>>> + isc->pads);
> >>>> + if (ret < 0) {
> >>>> + dev_err(isc->dev, "media entity init failed\n");
> >>>> + return ret;
> >>>> + }
> >>>> +
> >>>> + isc->mdev.dev = isc->dev;
> >>>> +
> >>>> + match = of_match_node(isc->dev->driver->of_match_table,
> >>>> + isc->dev->of_node);
> >>>> +
> >>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> >>>> + sizeof(isc->mdev.driver_name));
> >>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> >>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> >>>> + isc->v4l2_dev.name);
> >>>> + isc->mdev.hw_revision = ver;
> >>>> +
> >>>> + media_device_init(&isc->mdev);
> >>>> +
> >>>> + isc->v4l2_dev.mdev = &isc->mdev;
> >>>> +
> >>>> + return isc_scaler_init(isc);
> >>>> +}
> >>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
> >>>> +
> >>>> +void isc_mc_cleanup(struct isc_device *isc)
> >>>> +{
> >>>> + media_entity_cleanup(&isc->video_dev.entity);
> >>>> +}
> >>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> >>>> +
> >>>> /* regmap configuration */
> >>>> #define ATMEL_ISC_REG_MAX 0xd5c
> >>>> const struct regmap_config isc_regmap_config = {
> >>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>> new file mode 100644
> >>>> index 000000000000..ec95c9665883
> >>>> --- /dev/null
> >>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>> @@ -0,0 +1,245 @@
> >>>> +// SPDX-License-Identifier: GPL-2.0-only
> >>>> +/*
> >>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> >>>> + *
> >>>> + * Copyright (C) 2021 Microchip Technology, Inc.
> >>>
> >>> Time flies! It's 2022 already :)
> >>>
> >>>> + *
> >>>> + * Author: Eugen Hristev <[email protected]>
> >>>> + *
> >>>> + */
> >>>> +
> >>>> +#include <media/media-device.h>
> >>>> +#include <media/media-entity.h>
> >>>> +#include <media/v4l2-device.h>
> >>>> +#include <media/v4l2-subdev.h>
> >>>> +
> >>>> +#include "atmel-isc-regs.h"
> >>>> +#include "atmel-isc.h"
> >>>> +
> >>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> >>>> + struct v4l2_subdev_state *sd_state,
> >>>> + struct v4l2_subdev_format *format)
> >>>> +{
> >>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>> +
> >>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>> + format->pad);
> >>>> + format->format = *v4l2_try_fmt;
> >>>> +
> >>>> + return 0;
> >>>> + }
> >>>> +
> >>>> + format->format = isc->scaler_format;
> >>>
> >>> isc->scaler_format is only used inside this file if I'm not mistaken.
> >>> I wonder why it lives in the isc_device struct.
> >>
> >> isc_device is a placeholder for all isc things.
> >>
> >> I would not create a separate struct in this file that would have to be
> >> allocated etc... just for one two things. So I preferred to have it in
> >> the same place as all the other things.
> >>
> >>>
> >>>> +
> >>>> + return 0;
> >>>> +}
> >>>> +
> >>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> >>>> + struct v4l2_subdev_state *sd_state,
> >>>> + struct v4l2_subdev_format *req_fmt)
> >>>> +{
> >>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>> + struct isc_format *fmt;
> >>>> + unsigned int i;
> >>>> +
> >>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> >>>> + v4l_bound_align_image
> >>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
> >>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> >>>> + else
> >>>> + v4l_bound_align_image
> >>>> + (&req_fmt->format.width, 16, 10000, 0,
> >>>> + &req_fmt->format.height, 16, 10000, 0, 0);
> >>>
> >>> Where does 10000 come from ?
> >>
> >> It's a random number. Do you have any suggestion for a better one ?
> >
> > An actual HW limit ? :)
>
> There is no hardware limit. The ISC just stops sampling pixels once the
> crop limit is reached. The element that generates pixels could go on
> forever , or until the next HBLANK/VBLANK. So what limit could I place
> here ?
> The cropping is mandatory, otherwise there could be an overflow w.r.t.
> the buffer size if the ISC would sample more pixels than the software
> expects it to.
>
> >
> >> Maybe this would be much more clear with my comments below (where I will
> >> explain what the isc scaler does )
> >> In short, it allows the other entity in the link to 'sink' a huge format
> >> into this 'scaler' . Because the scaler can crop anything basically down
> >> to the biggest format size the ISC could handle.
> >>
> >
> > doesn't it have any documented input size limit ?
>
> As stated above, ISC handles a pixel stream and stores it in an internal
> SRAM. ISC stops sampling when this SRAM is full or when the cropping
> limit is reached.
> After storing the pixels in the SRAM (which is limited to the ISC
> maximum frame size), the ISC will work on the pixels and then DMA the
> frame to another RAM.
>

I understand...

So whatever the input size is, only the first n-th bytes are actually
dumped to memory, the following ones are discarded.

So the crop is always applied on the top-leftmost corner as a
consequence, right ?

I understand why you had difficulties picking a default :)

This is not easy to handle, as the scaler doesn't actually applies a
crop but rather stops capturing after the destination buffer has been
saturated. So the limiting factor is the full image size, not one of
its dimensions, like in example the line length. This makes it hard
to adjust cropping to something meaningful for most use cases.

In your below example you suggest that an image size of (800000 x 10)
would theoretically be captured correctly. What if the input is of
size (800000 x 10 + 1). What size would you set the crop rectangle to ?

I wouldn't however be too much concerned for the moment, but I would
record this in a comment ?

> >
> >>>
> >>>> +
> >>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>> + req_fmt->format.field = V4L2_FIELD_NONE;
> >>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>> +
> >>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
> >>>
> >>> So you rely on the isc format list for the scaler as well ?
> >>> I think it's fine as long as they are identical
> >>
> >> Yes, the scaler is kind of a dummy entity , but it's required by the
> >> media controller validation of the pipeline.
> >>
> >
> > More on this later
> >
> >>>
> >>>> +
> >>>> + if (!fmt)
> >>>> + fmt = &isc->formats_list[0];
> >>>> +
> >>>> + req_fmt->format.code = fmt->mbus_code;
> >>>> +
> >>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>> + req_fmt->pad);
> >>>> + *v4l2_try_fmt = req_fmt->format;
> >>>> + /* Trying on the pad sink makes the source sink change too */
> >>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> >>>> + v4l2_try_fmt =
> >>>> + v4l2_subdev_get_try_format(sd, sd_state,
> >>>> + ISC_SCALER_PAD_SOURCE);
> >>>> + *v4l2_try_fmt = req_fmt->format;
> >>>> +
> >>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
> >>>> + 16, isc->max_width, 0,
> >>>> + &v4l2_try_fmt->height,
> >>>> + 16, isc->max_height, 0, 0);
> >>>> + }
> >>>> + /* if we are just trying, we are done */
> >>>> + return 0;
> >>>> + }
> >>>> +
> >>>> + isc->scaler_format = req_fmt->format;
> >>>
> >>> No per-pad format but a global scaler_format ? How do you configure scaling ?
> >>>
> >>> Actually, I would like to know more about the scaler device
> >>> capabilities. What functions can this IP perform ? Does it do
> >>> cropping, can it also do (down)scaling or even composition ?
> >>>
> >>> I think it is worth to read
> >>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> >>> where it is reported how cropping and scaling are implemented by using
> >>> the selection API.
> >>>
> >>> Figure 4.5 is particularly helpful to explain the simple crop case,
> >>> for which you need to implement support to by adding s_selection on
> >>> the sink pad TGT_CROP target.
> >>
> >> The scaler is kind of a dummy entity.
> >> The problem is that different subdevices cannot be connected directly to
> >> a v4l entity, because the ISC does cropping on the incoming frame.
> >> The ISC has a limit on the total number of pixels per frame, thus it
> >> will do a crop.
> >
> > What is this limit related to ? I don't know the ISC internals, but is
> > the limitation due to the overall image size, or maybe it's due to
> > some line buffers being limited in the size they can transfer to
> > memory ? Is it a limitation on the size of the image in pixels, or is
> > it due to a limitation of the processing bandwidth on the input
> > interface and thus depends on the actual size of the image in bytes ?
>
> Limitation is both on line size and full image size. From my
> understanding, is that the internal SRAM is limited (8 Mpixels with a
> small safe buffer on top in the case of sama7g5 )
> This SRAM could be used as 800000x10 for example, or 3200x2464 for
> another example. However, I am not sure if a real line of 800,000 pixels
> make sense, or if there is another internal mechanism in the ISC that
> stops it.
> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
> the maximum frame that it can handle
>
> >
> >> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
> >> only handle 3264x2464 , then we have two choices:
> >> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
> >> 3280x2464, and the whole pipeline fails to configure, there is a
> >> mismatch between /dev/video output and what the whole pipeline produces
> >
> > I'm confused in first place. If the ISC cannot output anything larger
> > than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
> > this should be adjusted to 3264x2464.
>
> yes, that is true. However, without an entity that says 'I am doing
> cropping', the media pipeline will fail, because previous entity will
> output 3280x2464 .
> The conclusion to create a scaler entitity was done during my last year
> discussions on the IRC. I could not find a way to perform this cropping
> at the v4l entity side, and one reason is that the v4l entity does not
> support what I needed , or at least that was my understanding.
> You claim that the v4l entity could perform the cropping in the same
> entity ?

I thought so, but I haven't tried to be honest

>
> Let's assume for a minute that it could do this. Even so, I would prefer
> to have a separate scaler entity to do this , for several reasons:
> -> I would like to be able to crop the frame arbitrarily if I want to,
> and the scaler would allow me to do this easier
> -> it would be more obvious that the frame is cropped by having this
> entity there
> -> in fact some of the ISC versions really have a down scaler, that I
> have not implemented yet. So it would be useful to already have this in
> place to be able to scale down with it at a later time.

I think this last reason is enough to have an entity plumbed in
already

>
> >
> > Or is the ISC limitation on the -input- side ?
> >
> >> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
> >> the ISC can handle at most -> cannot , because the sensor let's say can
> >> *only* produce 3280x2464 and *any* other frame size returns an error.
> >
> > buggy sensor I would say :)
>
> I cannot tell this to the customers, and I cannot fail the whole
> pipeline because the sensor does not crop 16 pixels, when in fact I can
> do this very easily on the ISC side.
>

Fair enough :)

> >
> >>
> >> The only solution to make both worlds happy, is to have a dummy entity
> >> called 'scaler' which in fact now it only performs a simple crop.
> >> It accepts any frame size at input (hence the 10000x10000 which you saw
> >> earlier), and outputs at most 3264x2464 the isc_max_height and
> >> isc_max_width.
> >
> > I have a maybe dumb question: can the cropping operation be modeled
> > by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
> > the atmel_isc_common entity sink pad and avoid the scaler completely ?
>
> I am not sure. This is what I wanted initially. But then I thought about
> the three reasons stated above for the use of the scaler entity, and
> discussed with folks on the IRC, and come up with the solution for the
> scaler entity
>
> >
> >>
> >> So to answer your question, the isc scaler is a software model for a
> >> simple cropping, that would make media controller happy, and capture
> >> software happy.
> >>
> >> Here it how it looks :
> >>
> >> Device topology
> >> - entity 1: atmel_isc_scaler (2 pads, 2 links)
> >> type V4L2 subdev subtype Unknown flags 0
> >> device node name /dev/v4l-subdev0
> >> pad0: Sink
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >> crop.bounds:(0,0)/3264x2464
> >> crop:(0,0)/3264x2464]
> >> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >> pad1: Source
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>
> >> - entity 4: csi2dc (2 pads, 2 links)
> >> type V4L2 subdev subtype Unknown flags 0
> >> device node name /dev/v4l-subdev1
> >> pad0: Sink
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >> <- "dw-csi.0":1 [ENABLED]
> >> pad1: Source
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
> >>
> >> - entity 7: dw-csi.0 (2 pads, 2 links)
> >> type V4L2 subdev subtype Unknown flags 0
> >> device node name /dev/v4l-subdev2
> >> pad0: Sink
> >> [fmt:SRGGB10_1X10/3280x2464]
> >> <- "imx219 1-0010":0 [ENABLED]
> >> pad1: Source
> >> [fmt:SRGGB10_1X10/3280x2464]
> >> -> "csi2dc":0 [ENABLED]
> >>
> >> - entity 12: imx219 1-0010 (1 pad, 1 link)
> >> type V4L2 subdev subtype Sensor flags 0
> >> device node name /dev/v4l-subdev3
> >> pad0: Source
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >> xfer:srgb ycbcr:601 quantization:full-range
> >> crop.bounds:(8,8)/3280x2464
> >> crop:(8,8)/3280x2464]
> >> -> "dw-csi.0":0 [ENABLED]
> >>
> >> - entity 24: atmel_isc_common (1 pad, 1 link)
> >> type Node subtype V4L flags 1
> >> device node name /dev/video0
> >> pad0: Sink
> >> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
> >>
> >>
> >> Scaler does this one cute little thing :
> >>
> >> pad0: Sink
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >> crop.bounds:(0,0)/3264x2464
> >> crop:(0,0)/3264x2464]
> >> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >> pad1: Source
> >> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >
> > Shouldn't this be 3264x2464 as that's what the entity outputs after
> > the cropping ? And shouldn't then be 3264x2464 the size output from
> > the video device too ?
>
> That's right.
> I don't know why the source format for the scaler is still 3280x2464.
> Maybe there is a bug on g_fmt for it.. have to check it.

Thanks, I think this should be fixed ten


>
> Anyway, the video format is like this :
>
> # v4l2-ctl --get-fmt-video
> Format Video Capture:
> Width/Height : 3264/2464
> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
> Field : None
> Bytes per Line : 6528
> Size Image : 16084992
> Colorspace : sRGB
> Transfer Function : Default (maps to sRGB)
> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
> Quantization : Default (maps to Full Range)
> Flags :
> #
>
> and initializing the pipeline looks fine from user perspective.
> The scaler is in fact the solution to make this pipeline work with
> libcamera.
> I found this cropping to be an issue in media controller when trying it
> with libcamera. Otherwise, the other user space apps which I was using
> never complained that anything was wrong
> Libcamera simply refuses to acknowledge the pipeline if the video output
> is 3264x2464 but there is no entity that changes the format from
> 3280x2464 down

Not sure why libcamera should play a role there. Isn't it the media
pipeline validation that complains and returns an -EPIPE ?

>
> >
> >
> >> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>
> >>
> >> Which is what we needed.
> >>
> >>>
> >>>> +
> >>>> + return 0;
> >>>> +}
> >>>> +
> >>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> >>>> + struct v4l2_subdev_state *sd_state,
> >>>> + struct v4l2_subdev_mbus_code_enum *code)
> >>>> +{
> >>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>> + int supported_index = 0;
> >>>> + int i;
> >>>> +
> >>>> + for (i = 0; i < isc->formats_list_size; i++) {
> >>>> + if (!isc->formats_list[i].sd_support)
> >>>> + continue;
> >>>
> >>> The sd_support flag still doesn't click in my head.
> >>>
> >>> Shouldn't the enumeration of available formats on the scaler do not
> >>> depend on the sensor supproted formats ?
> >>
> >> You're right. I will have to check it again.
> >>
> >>>
> >>>> + if (supported_index == code->index) {
> >>>> + code->code = isc->formats_list[i].mbus_code;
> >>>> + return 0;
> >>>> + }
> >>>> + supported_index++;
> >>>> + }
> >>>> +
> >>>> + return -EINVAL;
> >>>> +}
> >>>> +
> >>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> >>>> + struct v4l2_subdev_state *sd_state,
> >>>> + struct v4l2_subdev_selection *sel)
> >>>> +{
> >>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>> +
> >>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> >>>> + return -EINVAL;
> >>>> +
> >>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> >>>> + sel->target != V4L2_SEL_TGT_CROP)
> >>>> + return -EINVAL;
> >>>> +
> >>>> + sel->r.height = isc->max_height;
> >>>> + sel->r.width = isc->max_width;
> >>>
> >>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
> >>> as it represents the maximum valid crop rectangle.
> >>>
> >>> TGT_CROP should report the configured crop rectangle which can be
> >>> intiialized to the same size as CROP_BOUNDS, if my understanding of
> >>> the spec is correct
> >>
> >> So you would like to have this differentiated, and report the
> >> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
> >> reported now, the maximum size of the ISC frame .
> >> My understanding is correct ?
> >>
> >
> > I didn't know you have an HW limitation, so your _BOUNDS is not the
> > input image size but rather 3264x2464 ( == max_width x max_height).
> >
> > What I meant is that _BOUNDS should report the maximum rectangle size
> > that can be applied to the _CROP target. In you case you have an HW
> > limitation 3264x2464 and that's the largest rectangle you can apply.
> So the CROP should be at 3264x2464
> > TGT_CROP can be initialized to the same as _BOUND, but if you
> > implement s_selection it should report what has been there applied.
> and BOUND to actual frame size ?
> > But as you don't implement s_selection yet, I think this is fine for
> > now. Maybe a little comment ?
>
> It could also be the case where e.g. the sensor is outputting 1920x1080,
> in this case the scaler would do nothing.
> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
> If the sensor is outputting 1920x1080, this format comes directly from
> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
> sensor's problem)
> >
> > Also, is set->r zeroed by the framework before getting here ?
> > Otherwise you should set r.left and r.top to 0
>
> If these are redundant, no problem to remove them

I was actually confused. I was suggesting you to add...

> >
> >>>
> >>>> +
> >>>> + sel->r.left = 0;
> >>>> + sel->r.top = 0;

These ^

So nothing to change. Sorry I've missed them

Thanks
j


> >>>> +
> >>>> + return 0;
> >>>> +}
> >>>> +
> >>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> >>>> + struct v4l2_subdev_state *sd_state)
> >>>> +{
> >>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> >>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> >>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>> +
> >>>> + *v4l2_try_fmt = isc->scaler_format;
> >>>> +
> >>>> + return 0;
> >>>> +}
> >>>> +
> >>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> >>>> + .set_fmt = isc_scaler_set_fmt,
> >>>> + .get_fmt = isc_scaler_get_fmt,
> >>>> + .get_selection = isc_scaler_g_sel,
> >>>> + .init_cfg = isc_scaler_init_cfg,
> >>>
> >>> .link_validate = v4l2_subdev_link_validate_default,
> >>>
> >>> To have the formats at the end of links that point to this entity
> >>> validated (I think the framework already calls it if not set though,
> >>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
> >>>
> >>>> +};
> >>>> +
> >>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >>>> + .pad = &isc_scaler_pad_ops,
> >>>> +};
> >>>> +
> >>>> +int isc_scaler_init(struct isc_device *isc)
> >>>> +{
> >>>> + int ret;
> >>>> +
> >>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> >>>> +
> >>>> + isc->scaler_sd.owner = THIS_MODULE;
> >>>> + isc->scaler_sd.dev = isc->dev;
> >>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> >>>> + "atmel_isc_scaler");
> >>>
> >>> I would drop 'atmel' for brevity, unless other entities have this
> >>> prefix set already
> >>
> >> The v4l entity takes it's name from the module name, which is
> >> atmel_isc_common, so I thought to keep the prefix
> >>
> >
> > Ack!
> >
> >>>
> >>>> +
> >>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >>>> +
> >>>> + isc->scaler_format.height = isc->max_height;
> >>>> + isc->scaler_format.width = isc->max_width;
> >>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> >>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
> >>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>> +
> >>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> >>>> + ISC_SCALER_PADS_NUM,
> >>>> + isc->scaler_pads);
> >>>> + if (ret < 0) {
> >>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> >>>> + return ret;
> >>>> + }
> >>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> >>>> + if (ret < 0) {
> >>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> >>>> + return ret;
> >>>> + }
> >>>> +
> >>>> + return ret;
> >>>> +}
> >>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> >>>> +
> >>>> +int isc_scaler_link(struct isc_device *isc)
> >>>> +{
> >>>> + int ret;
> >>>> +
> >>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> >>>> + isc->remote_pad, &isc->scaler_sd.entity,
> >>>> + ISC_SCALER_PAD_SINK,
> >>>> + MEDIA_LNK_FL_ENABLED |
> >>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>> +
> >>>> + if (ret < 0) {
> >>>> + v4l2_err(&isc->v4l2_dev,
> >>>> + "Failed to create pad link: %s to %s\n",
> >>>> + isc->current_subdev->sd->entity.name,
> >>>> + isc->scaler_sd.entity.name);
> >>>> + return ret;
> >>>> + }
> >>>> +
> >>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> >>>> + isc->current_subdev->sd->name, isc->remote_pad);
> >>>> +
> >>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> >>>> + ISC_SCALER_PAD_SOURCE,
> >>>> + &isc->video_dev.entity, ISC_PAD_SINK,
> >>>> + MEDIA_LNK_FL_ENABLED |
> >>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>> +
> >>>> + if (ret < 0) {
> >>>> + v4l2_err(&isc->v4l2_dev,
> >>>> + "Failed to create pad link: %s to %s\n",
> >>>> + isc->scaler_sd.entity.name,
> >>>> + isc->video_dev.entity.name);
> >>>> + return ret;
> >>>> + }
> >>>> +
> >>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> >>>> + ISC_SCALER_PAD_SOURCE);
> >>>> +
> >>>> + return ret;
> >>>> +}
> >>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
> >>>
> >>> From the DT graph point of view, the ISC appears as a single block
> >>> with an CSI-2 input port. It is then in charge of parsing the .dts and
> >>> add to its notifier the image sensor remote subdevice.
> >>
> >> Actually it's only a parallel port.
> >> And it can be connected either to a sensor directly or to the csi2dc
> >> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
> >>
> >>>
> >>> When the sensor subdev registers and the ISC notifier completes, the scaler
> >>> entity which was initialized at isc_probe() time is linked in between
> >>> the ISC and the image sensor immutably.
> >>
> >> yes, because this cannot change at runtime, and usually, it can't change
> >> without altering the board hardware. (at least this is what my
> >> understanding of immutability is )
> >>
> >>>
> >>> I think it is fine for now, but I wonder if you plan to plumb more
> >>> components between the ISC video node and the sensor, if it's not
> >>> worth changing the DT bindings and their parsing logic to separate the
> >>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
> >>> time and have the CSI-2 receiver entity as downstream remote. I think
> >>> I need to get to know the ISC better to really have an idea. For now,
> >>> this seems ok to me, but please check with maintainers if this is fine
> >>> with them.
> >>
> >> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
> >> the remote subdev).
> >> The ISC will register a scaler, and connect the subdev to this scaler
> >> first, and then, the scaler to the isc itself (the v4l entity).
> >>
> >>>
> >>>> +
> >>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >>>> index 5fbf52a9080b..c9234c90ae58 100644
> >>>> --- a/drivers/media/platform/atmel/atmel-isc.h
> >>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> >>>> u32 his_entry;
> >>>> };
> >>>>
> >>>> +enum isc_mc_pads {
> >>>> + ISC_PAD_SINK = 0,
> >>>> + ISC_PADS_NUM = 1,
> >>>> +};
> >>>> +
> >>>> +enum isc_scaler_pads {
> >>>> + ISC_SCALER_PAD_SINK = 0,
> >>>> + ISC_SCALER_PAD_SOURCE = 1,
> >>>> + ISC_SCALER_PADS_NUM = 2,
> >>>> +};
> >>>> +
> >>>> /*
> >>>> * struct isc_device - ISC device driver data/config struct
> >>>> * @regmap: Register map
> >>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> >>>> * be used as an input to the controller
> >>>> * @controller_formats_size: size of controller_formats array
> >>>> * @formats_list_size: size of formats_list array
> >>>> + * @pads: media controller pads for isc video entity
> >>>> + * @mdev: media device that is registered by the isc
> >>>> + * @remote_pad: remote pad on the connected subdevice
> >>>> + * @scaler_sd: subdevice for the scaler that isc registers
> >>>> + * @scaler_pads: media controller pads for the scaler subdevice
> >>>> + * @scaler_format: current format for the scaler subdevice
> >>>> */
> >>>> struct isc_device {
> >>>> struct regmap *regmap;
> >>>> @@ -344,6 +361,19 @@ struct isc_device {
> >>>> struct isc_format *formats_list;
> >>>> u32 controller_formats_size;
> >>>> u32 formats_list_size;
> >>>> +
> >>>> + struct {
> >>>> + struct media_pad pads[ISC_PADS_NUM];
> >>>> + struct media_device mdev;
> >>>> +
> >>>> + u32 remote_pad;
> >>>> + };
> >>>> +
> >>>> + struct {
> >>>> + struct v4l2_subdev scaler_sd;
> >>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> >>>> + struct v4l2_mbus_framefmt scaler_format;
> >>
> >> Here are the scaler stuff which I added, in the same bulk struct for the
> >> whole isc device
> >>
> >>
> >>>> + };
> >>>> };
> >>>>
> >>>> extern const struct regmap_config isc_regmap_config;
> >>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> >>>> void isc_subdev_cleanup(struct isc_device *isc);
> >>>> void isc_clk_cleanup(struct isc_device *isc);
> >>>>
> >>>> +int isc_scaler_link(struct isc_device *isc);
> >>>> +int isc_scaler_init(struct isc_device *isc);
> >>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
> >>>> +void isc_mc_cleanup(struct isc_device *isc);
> >>>> +
> >>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>> + unsigned int code, int *index);
> >>>> #endif
> >>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>> index c5b9563e36cb..c244682ea22f 100644
> >>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>> break;
> >>>> }
> >>>>
> >>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>
> >>> I am surprised you can read a register before runtime_pm is
> >>> intialized!
> >>>
> >>
> >> Actually I can read any register after the moment of starting the clock
> >> on the hardware block.
> >
> > Ah right then
> >
> >> Maybe the starting and stopping of the clock needs to be moved to the
> >> runtime_pm calls, but this is another story, not related to the media
> >> controller.
> >
> > It's fine, but maybe worth recording with a todo ?
> >
> > Thanks
> > j
> >
> >> I moved this line because I had to pass the version to the isc_mc_init call
> >>
> >>> Thanks
> >>> j
> >>
> >> Thanks for reviewing !
> >> Eugen
> >>>
> >>>
> >>>> +
> >>>> + ret = isc_mc_init(isc, ver);
> >>>> + if (ret < 0)
> >>>> + goto isc_probe_mc_init_err;
> >>>> +
> >>>> pm_runtime_set_active(dev);
> >>>> pm_runtime_enable(dev);
> >>>> pm_request_idle(dev);
> >>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>> ret = clk_prepare_enable(isc->ispck);
> >>>> if (ret) {
> >>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
> >>>> - goto cleanup_subdev;
> >>>> + goto isc_probe_mc_init_err;
> >>>> }
> >>>>
> >>>> /* ispck should be greater or equal to hclock */
> >>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>> goto unprepare_clk;
> >>>> }
> >>>>
> >>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>> dev_info(dev, "Microchip ISC version %x\n", ver);
> >>>>
> >>>> return 0;
> >>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>> unprepare_clk:
> >>>> clk_disable_unprepare(isc->ispck);
> >>>>
> >>>> +isc_probe_mc_init_err:
> >>>> + isc_mc_cleanup(isc);
> >>>> +
> >>>> cleanup_subdev:
> >>>> isc_subdev_cleanup(isc);
> >>>>
> >>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
> >>>>
> >>>> pm_runtime_disable(&pdev->dev);
> >>>>
> >>>> + isc_mc_cleanup(isc);
> >>>> +
> >>>> isc_subdev_cleanup(isc);
> >>>>
> >>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>> index 07a80b08bc54..9dc75eed0098 100644
> >>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> >>>> break;
> >>>> }
> >>>>
> >>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>> +
> >>>> + ret = isc_mc_init(isc, ver);
> >>>> + if (ret < 0)
> >>>> + goto isc_probe_mc_init_err;
> >>>> +
> >>>> pm_runtime_set_active(dev);
> >>>> pm_runtime_enable(dev);
> >>>> pm_request_idle(dev);
> >>>>
> >>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>> dev_info(dev, "Microchip XISC version %x\n", ver);
> >>>>
> >>>> return 0;
> >>>>
> >>>> +isc_probe_mc_init_err:
> >>>> + isc_mc_cleanup(isc);
> >>>> +
> >>>> cleanup_subdev:
> >>>> isc_subdev_cleanup(isc);
> >>>>
> >>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
> >>>>
> >>>> pm_runtime_disable(&pdev->dev);
> >>>>
> >>>> + isc_mc_cleanup(isc);
> >>>> +
> >>>> isc_subdev_cleanup(isc);
> >>>>
> >>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>> --
> >>>> 2.25.1
> >>>>
> >>
>

2022-02-09 11:33:08

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

On 2/8/22 10:59 AM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
>>> Hi Eugen
>>>
>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
>>>> Implement the support for media-controller.
>>>> This means that the capabilities of the driver have changed and now
>>>> it also advertises the IO_MC .
>>>> The driver will register it's media device, and add the video entity to this
>>>> media device. The subdevices are registered to the same media device.
>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
>>>> It will also register a subdevice that allows cropping of the incoming frame
>>>> to the maximum frame size supported by the ISC.
>>>> The ISC will create a link between the subdevice that is asynchronously
>>>> registered and the atmel_isc_scaler entity.
>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
>>>> link.
>>>>
>>>> Signed-off-by: Eugen Hristev <[email protected]>
>>>> ---
>>>> Changes in v4:
>>>> As suggested by Jacopo:
>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
>>>> - moved init_mc/clean_mc to isc_base file
>>>>
>>>> Changes in v2:
>>>> - implement try formats
>>>>
>>>> drivers/media/platform/atmel/Makefile | 2 +-
>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>
>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
>>>> index 794e8f739287..f02d03df89d6 100644
>>>> --- a/drivers/media/platform/atmel/Makefile
>>>> +++ b/drivers/media/platform/atmel/Makefile
>>>> @@ -1,7 +1,7 @@
>>>> # SPDX-License-Identifier: GPL-2.0-only
>>>> atmel-isc-objs = atmel-sama5d2-isc.o
>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>>>>
>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>>>> index 6b0005987a17..6b482270eb93 100644
>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>> struct isc_device, v4l2_dev);
>>>> struct isc_subdev_entity *subdev_entity =
>>>> container_of(notifier, struct isc_subdev_entity, notifier);
>>>> + int pad;
>>>>
>>>> if (video_is_registered(&isc->video_dev)) {
>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>
>>>> subdev_entity->sd = subdev;
>>>>
>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
>>>> + MEDIA_PAD_FL_SOURCE);
>>>> + if (pad < 0) {
>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
>>>> + subdev->name);
>>>> + return pad;
>>>> + }
>>>> +
>>>> + isc->remote_pad = pad;
>>>> +
>>>> return 0;
>>>> }
>>>>
>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
>>>> }
>>>>
>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>> - unsigned int code, int *index)
>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>> + unsigned int code, int *index)
>>>> {
>>>> struct isc_format *fmt = &isc->formats_list[0];
>>>> unsigned int i;
>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>
>>>> return NULL;
>>>> }
>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>>>>
>>>> static int isc_formats_init(struct isc_device *isc)
>>>> {
>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
>>>> NULL, &mbus_code)) {
>>>> mbus_code.index++;
>>>>
>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
>>>> if (!fmt) {
>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
>>>> mbus_code.code);
>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>> vdev->queue = q;
>>>> vdev->lock = &isc->lock;
>>>> vdev->ctrl_handler = &isc->ctrls.handler;
>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
>>>> + V4L2_CAP_IO_MC;
>>>> video_set_drvdata(vdev, isc);
>>>>
>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>> goto isc_async_complete_err;
>>>> }
>>>>
>>>> + ret = isc_scaler_link(isc);
>>>> + if (ret < 0)
>>>> + goto isc_async_complete_unregister_device;
>>>> +
>>>> + ret = media_device_register(&isc->mdev);
>>>> + if (ret < 0)
>>>> + goto isc_async_complete_unregister_device;
>>>> return 0;
>>>>
>>>> +isc_async_complete_unregister_device:
>>>> + video_unregister_device(vdev);
>>>> +
>>>> isc_async_complete_err:
>>>> mutex_destroy(&isc->lock);
>>>> return ret;
>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
>>>> }
>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>>>
>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
>>>> +{
>>>> + const struct of_device_id *match;
>>>> + int ret;
>>>> +
>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>>>
>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
>>> to be able to have the media pipeline validated at
>>> media_pipeline_start() time ?
>>
>> Hi,
>>
>> I am doing that in a subsequent patch. Things are not completely ready
>> at this moment, because ISC still relies on the old mechanism to call
>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
>> removes and adds checks to the link validate ).
>>
>
> I see.. the subsequent patches are not part of this series, right ?

No. Patch 7 does this. In that patch, the link validation is created and
all the logic of format propagation is removed and reworked, and creates
the need for the link_validate call
Could you have also a look at that patch ?
>
> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
> the introduction of media controller or the scaler, right ? (I'm not
> sure if the DTS patches do).
>
> If that's the case, would it make sense to to fast-track them (as some
> of them are fixes) and then on top plumb the media controller
> infrastructure with this patch and the ones you have been mentioning ?

It would make sense.

>
>>>
>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>> +
>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>>>> + isc->pads);
>>>> + if (ret < 0) {
>>>> + dev_err(isc->dev, "media entity init failed\n");
>>>> + return ret;
>>>> + }
>>>> +
>>>> + isc->mdev.dev = isc->dev;
>>>> +
>>>> + match = of_match_node(isc->dev->driver->of_match_table,
>>>> + isc->dev->of_node);
>>>> +
>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
>>>> + sizeof(isc->mdev.driver_name));
>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
>>>> + isc->v4l2_dev.name);
>>>> + isc->mdev.hw_revision = ver;
>>>> +
>>>> + media_device_init(&isc->mdev);
>>>> +
>>>> + isc->v4l2_dev.mdev = &isc->mdev;
>>>> +
>>>> + return isc_scaler_init(isc);
>>>> +}
>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
>>>> +
>>>> +void isc_mc_cleanup(struct isc_device *isc)
>>>> +{
>>>> + media_entity_cleanup(&isc->video_dev.entity);
>>>> +}
>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
>>>> +
>>>> /* regmap configuration */
>>>> #define ATMEL_ISC_REG_MAX 0xd5c
>>>> const struct regmap_config isc_regmap_config = {
>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>> new file mode 100644
>>>> index 000000000000..ec95c9665883
>>>> --- /dev/null
>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>> @@ -0,0 +1,245 @@
>>>> +// SPDX-License-Identifier: GPL-2.0-only
>>>> +/*
>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
>>>> + *
>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
>>>
>>> Time flies! It's 2022 already :)
>>>
>>>> + *
>>>> + * Author: Eugen Hristev <[email protected]>
>>>> + *
>>>> + */
>>>> +
>>>> +#include <media/media-device.h>
>>>> +#include <media/media-entity.h>
>>>> +#include <media/v4l2-device.h>
>>>> +#include <media/v4l2-subdev.h>
>>>> +
>>>> +#include "atmel-isc-regs.h"
>>>> +#include "atmel-isc.h"
>>>> +
>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
>>>> + struct v4l2_subdev_state *sd_state,
>>>> + struct v4l2_subdev_format *format)
>>>> +{
>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>> +
>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>> + format->pad);
>>>> + format->format = *v4l2_try_fmt;
>>>> +
>>>> + return 0;
>>>> + }
>>>> +
>>>> + format->format = isc->scaler_format;
>>>
>>> isc->scaler_format is only used inside this file if I'm not mistaken.
>>> I wonder why it lives in the isc_device struct.
>>
>> isc_device is a placeholder for all isc things.
>>
>> I would not create a separate struct in this file that would have to be
>> allocated etc... just for one two things. So I preferred to have it in
>> the same place as all the other things.
>>
>>>
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
>>>> + struct v4l2_subdev_state *sd_state,
>>>> + struct v4l2_subdev_format *req_fmt)
>>>> +{
>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>> + struct isc_format *fmt;
>>>> + unsigned int i;
>>>> +
>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
>>>> + v4l_bound_align_image
>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
>>>> + else
>>>> + v4l_bound_align_image
>>>> + (&req_fmt->format.width, 16, 10000, 0,
>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
>>>
>>> Where does 10000 come from ?
>>
>> It's a random number. Do you have any suggestion for a better one ?
>
> An actual HW limit ? :)

There is no hardware limit. The ISC just stops sampling pixels once the
crop limit is reached. The element that generates pixels could go on
forever , or until the next HBLANK/VBLANK. So what limit could I place
here ?
The cropping is mandatory, otherwise there could be an overflow w.r.t.
the buffer size if the ISC would sample more pixels than the software
expects it to.

>
>> Maybe this would be much more clear with my comments below (where I will
>> explain what the isc scaler does )
>> In short, it allows the other entity in the link to 'sink' a huge format
>> into this 'scaler' . Because the scaler can crop anything basically down
>> to the biggest format size the ISC could handle.
>>
>
> doesn't it have any documented input size limit ?

As stated above, ISC handles a pixel stream and stores it in an internal
SRAM. ISC stops sampling when this SRAM is full or when the cropping
limit is reached.
After storing the pixels in the SRAM (which is limited to the ISC
maximum frame size), the ISC will work on the pixels and then DMA the
frame to another RAM.

>
>>>
>>>> +
>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>> +
>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
>>>
>>> So you rely on the isc format list for the scaler as well ?
>>> I think it's fine as long as they are identical
>>
>> Yes, the scaler is kind of a dummy entity , but it's required by the
>> media controller validation of the pipeline.
>>
>
> More on this later
>
>>>
>>>> +
>>>> + if (!fmt)
>>>> + fmt = &isc->formats_list[0];
>>>> +
>>>> + req_fmt->format.code = fmt->mbus_code;
>>>> +
>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>> + req_fmt->pad);
>>>> + *v4l2_try_fmt = req_fmt->format;
>>>> + /* Trying on the pad sink makes the source sink change too */
>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
>>>> + v4l2_try_fmt =
>>>> + v4l2_subdev_get_try_format(sd, sd_state,
>>>> + ISC_SCALER_PAD_SOURCE);
>>>> + *v4l2_try_fmt = req_fmt->format;
>>>> +
>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
>>>> + 16, isc->max_width, 0,
>>>> + &v4l2_try_fmt->height,
>>>> + 16, isc->max_height, 0, 0);
>>>> + }
>>>> + /* if we are just trying, we are done */
>>>> + return 0;
>>>> + }
>>>> +
>>>> + isc->scaler_format = req_fmt->format;
>>>
>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
>>>
>>> Actually, I would like to know more about the scaler device
>>> capabilities. What functions can this IP perform ? Does it do
>>> cropping, can it also do (down)scaling or even composition ?
>>>
>>> I think it is worth to read
>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
>>> where it is reported how cropping and scaling are implemented by using
>>> the selection API.
>>>
>>> Figure 4.5 is particularly helpful to explain the simple crop case,
>>> for which you need to implement support to by adding s_selection on
>>> the sink pad TGT_CROP target.
>>
>> The scaler is kind of a dummy entity.
>> The problem is that different subdevices cannot be connected directly to
>> a v4l entity, because the ISC does cropping on the incoming frame.
>> The ISC has a limit on the total number of pixels per frame, thus it
>> will do a crop.
>
> What is this limit related to ? I don't know the ISC internals, but is
> the limitation due to the overall image size, or maybe it's due to
> some line buffers being limited in the size they can transfer to
> memory ? Is it a limitation on the size of the image in pixels, or is
> it due to a limitation of the processing bandwidth on the input
> interface and thus depends on the actual size of the image in bytes ?

Limitation is both on line size and full image size. From my
understanding, is that the internal SRAM is limited (8 Mpixels with a
small safe buffer on top in the case of sama7g5 )
This SRAM could be used as 800000x10 for example, or 3200x2464 for
another example. However, I am not sure if a real line of 800,000 pixels
make sense, or if there is another internal mechanism in the ISC that
stops it.
In any case, the ISC is cropping the incoming frame from aaaXbbb up to
the maximum frame that it can handle

>
>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
>> only handle 3264x2464 , then we have two choices:
>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
>> 3280x2464, and the whole pipeline fails to configure, there is a
>> mismatch between /dev/video output and what the whole pipeline produces
>
> I'm confused in first place. If the ISC cannot output anything larger
> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
> this should be adjusted to 3264x2464.

yes, that is true. However, without an entity that says 'I am doing
cropping', the media pipeline will fail, because previous entity will
output 3280x2464 .
The conclusion to create a scaler entitity was done during my last year
discussions on the IRC. I could not find a way to perform this cropping
at the v4l entity side, and one reason is that the v4l entity does not
support what I needed , or at least that was my understanding.
You claim that the v4l entity could perform the cropping in the same
entity ?

Let's assume for a minute that it could do this. Even so, I would prefer
to have a separate scaler entity to do this , for several reasons:
-> I would like to be able to crop the frame arbitrarily if I want to,
and the scaler would allow me to do this easier
-> it would be more obvious that the frame is cropped by having this
entity there
-> in fact some of the ISC versions really have a down scaler, that I
have not implemented yet. So it would be useful to already have this in
place to be able to scale down with it at a later time.

>
> Or is the ISC limitation on the -input- side ?
>
>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
>> the ISC can handle at most -> cannot , because the sensor let's say can
>> *only* produce 3280x2464 and *any* other frame size returns an error.
>
> buggy sensor I would say :)

I cannot tell this to the customers, and I cannot fail the whole
pipeline because the sensor does not crop 16 pixels, when in fact I can
do this very easily on the ISC side.

>
>>
>> The only solution to make both worlds happy, is to have a dummy entity
>> called 'scaler' which in fact now it only performs a simple crop.
>> It accepts any frame size at input (hence the 10000x10000 which you saw
>> earlier), and outputs at most 3264x2464 the isc_max_height and
>> isc_max_width.
>
> I have a maybe dumb question: can the cropping operation be modeled
> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
> the atmel_isc_common entity sink pad and avoid the scaler completely ?

I am not sure. This is what I wanted initially. But then I thought about
the three reasons stated above for the use of the scaler entity, and
discussed with folks on the IRC, and come up with the solution for the
scaler entity

>
>>
>> So to answer your question, the isc scaler is a software model for a
>> simple cropping, that would make media controller happy, and capture
>> software happy.
>>
>> Here it how it looks :
>>
>> Device topology
>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
>> type V4L2 subdev subtype Unknown flags 0
>> device node name /dev/v4l-subdev0
>> pad0: Sink
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>> crop.bounds:(0,0)/3264x2464
>> crop:(0,0)/3264x2464]
>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>> pad1: Source
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>
>> - entity 4: csi2dc (2 pads, 2 links)
>> type V4L2 subdev subtype Unknown flags 0
>> device node name /dev/v4l-subdev1
>> pad0: Sink
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>> <- "dw-csi.0":1 [ENABLED]
>> pad1: Source
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
>>
>> - entity 7: dw-csi.0 (2 pads, 2 links)
>> type V4L2 subdev subtype Unknown flags 0
>> device node name /dev/v4l-subdev2
>> pad0: Sink
>> [fmt:SRGGB10_1X10/3280x2464]
>> <- "imx219 1-0010":0 [ENABLED]
>> pad1: Source
>> [fmt:SRGGB10_1X10/3280x2464]
>> -> "csi2dc":0 [ENABLED]
>>
>> - entity 12: imx219 1-0010 (1 pad, 1 link)
>> type V4L2 subdev subtype Sensor flags 0
>> device node name /dev/v4l-subdev3
>> pad0: Source
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>> xfer:srgb ycbcr:601 quantization:full-range
>> crop.bounds:(8,8)/3280x2464
>> crop:(8,8)/3280x2464]
>> -> "dw-csi.0":0 [ENABLED]
>>
>> - entity 24: atmel_isc_common (1 pad, 1 link)
>> type Node subtype V4L flags 1
>> device node name /dev/video0
>> pad0: Sink
>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
>>
>>
>> Scaler does this one cute little thing :
>>
>> pad0: Sink
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>> crop.bounds:(0,0)/3264x2464
>> crop:(0,0)/3264x2464]
>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>> pad1: Source
>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>
> Shouldn't this be 3264x2464 as that's what the entity outputs after
> the cropping ? And shouldn't then be 3264x2464 the size output from
> the video device too ?

That's right.
I don't know why the source format for the scaler is still 3280x2464.
Maybe there is a bug on g_fmt for it.. have to check it.

Anyway, the video format is like this :

# v4l2-ctl --get-fmt-video
Format Video Capture:
Width/Height : 3264/2464
Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
Field : None
Bytes per Line : 6528
Size Image : 16084992
Colorspace : sRGB
Transfer Function : Default (maps to sRGB)
YCbCr/HSV Encoding: Default (maps to ITU-R 601)
Quantization : Default (maps to Full Range)
Flags :
#

and initializing the pipeline looks fine from user perspective.
The scaler is in fact the solution to make this pipeline work with
libcamera.
I found this cropping to be an issue in media controller when trying it
with libcamera. Otherwise, the other user space apps which I was using
never complained that anything was wrong
Libcamera simply refuses to acknowledge the pipeline if the video output
is 3264x2464 but there is no entity that changes the format from
3280x2464 down

>
>
>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>
>>
>> Which is what we needed.
>>
>>>
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
>>>> + struct v4l2_subdev_state *sd_state,
>>>> + struct v4l2_subdev_mbus_code_enum *code)
>>>> +{
>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>> + int supported_index = 0;
>>>> + int i;
>>>> +
>>>> + for (i = 0; i < isc->formats_list_size; i++) {
>>>> + if (!isc->formats_list[i].sd_support)
>>>> + continue;
>>>
>>> The sd_support flag still doesn't click in my head.
>>>
>>> Shouldn't the enumeration of available formats on the scaler do not
>>> depend on the sensor supproted formats ?
>>
>> You're right. I will have to check it again.
>>
>>>
>>>> + if (supported_index == code->index) {
>>>> + code->code = isc->formats_list[i].mbus_code;
>>>> + return 0;
>>>> + }
>>>> + supported_index++;
>>>> + }
>>>> +
>>>> + return -EINVAL;
>>>> +}
>>>> +
>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
>>>> + struct v4l2_subdev_state *sd_state,
>>>> + struct v4l2_subdev_selection *sel)
>>>> +{
>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>> +
>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
>>>> + return -EINVAL;
>>>> +
>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
>>>> + sel->target != V4L2_SEL_TGT_CROP)
>>>> + return -EINVAL;
>>>> +
>>>> + sel->r.height = isc->max_height;
>>>> + sel->r.width = isc->max_width;
>>>
>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
>>> as it represents the maximum valid crop rectangle.
>>>
>>> TGT_CROP should report the configured crop rectangle which can be
>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
>>> the spec is correct
>>
>> So you would like to have this differentiated, and report the
>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
>> reported now, the maximum size of the ISC frame .
>> My understanding is correct ?
>>
>
> I didn't know you have an HW limitation, so your _BOUNDS is not the
> input image size but rather 3264x2464 ( == max_width x max_height).
>
> What I meant is that _BOUNDS should report the maximum rectangle size
> that can be applied to the _CROP target. In you case you have an HW
> limitation 3264x2464 and that's the largest rectangle you can apply.
So the CROP should be at 3264x2464
> TGT_CROP can be initialized to the same as _BOUND, but if you
> implement s_selection it should report what has been there applied.
and BOUND to actual frame size ?
> But as you don't implement s_selection yet, I think this is fine for
> now. Maybe a little comment ?

It could also be the case where e.g. the sensor is outputting 1920x1080,
in this case the scaler would do nothing.
CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
If the sensor is outputting 1920x1080, this format comes directly from
the sensor (the sensor is cropping it from 3280x2464 or not... it's the
sensor's problem)
>
> Also, is set->r zeroed by the framework before getting here ?
> Otherwise you should set r.left and r.top to 0

If these are redundant, no problem to remove them
>
>>>
>>>> +
>>>> + sel->r.left = 0;
>>>> + sel->r.top = 0;
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
>>>> + struct v4l2_subdev_state *sd_state)
>>>> +{
>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>> +
>>>> + *v4l2_try_fmt = isc->scaler_format;
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
>>>> + .set_fmt = isc_scaler_set_fmt,
>>>> + .get_fmt = isc_scaler_get_fmt,
>>>> + .get_selection = isc_scaler_g_sel,
>>>> + .init_cfg = isc_scaler_init_cfg,
>>>
>>> .link_validate = v4l2_subdev_link_validate_default,
>>>
>>> To have the formats at the end of links that point to this entity
>>> validated (I think the framework already calls it if not set though,
>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
>>>
>>>> +};
>>>> +
>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>>>> + .pad = &isc_scaler_pad_ops,
>>>> +};
>>>> +
>>>> +int isc_scaler_init(struct isc_device *isc)
>>>> +{
>>>> + int ret;
>>>> +
>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
>>>> +
>>>> + isc->scaler_sd.owner = THIS_MODULE;
>>>> + isc->scaler_sd.dev = isc->dev;
>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
>>>> + "atmel_isc_scaler");
>>>
>>> I would drop 'atmel' for brevity, unless other entities have this
>>> prefix set already
>>
>> The v4l entity takes it's name from the module name, which is
>> atmel_isc_common, so I thought to keep the prefix
>>
>
> Ack!
>
>>>
>>>> +
>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>>>> +
>>>> + isc->scaler_format.height = isc->max_height;
>>>> + isc->scaler_format.width = isc->max_width;
>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>> +
>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
>>>> + ISC_SCALER_PADS_NUM,
>>>> + isc->scaler_pads);
>>>> + if (ret < 0) {
>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
>>>> + return ret;
>>>> + }
>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
>>>> + if (ret < 0) {
>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
>>>> + return ret;
>>>> + }
>>>> +
>>>> + return ret;
>>>> +}
>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
>>>> +
>>>> +int isc_scaler_link(struct isc_device *isc)
>>>> +{
>>>> + int ret;
>>>> +
>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
>>>> + isc->remote_pad, &isc->scaler_sd.entity,
>>>> + ISC_SCALER_PAD_SINK,
>>>> + MEDIA_LNK_FL_ENABLED |
>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>> +
>>>> + if (ret < 0) {
>>>> + v4l2_err(&isc->v4l2_dev,
>>>> + "Failed to create pad link: %s to %s\n",
>>>> + isc->current_subdev->sd->entity.name,
>>>> + isc->scaler_sd.entity.name);
>>>> + return ret;
>>>> + }
>>>> +
>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
>>>> + isc->current_subdev->sd->name, isc->remote_pad);
>>>> +
>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
>>>> + ISC_SCALER_PAD_SOURCE,
>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
>>>> + MEDIA_LNK_FL_ENABLED |
>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>> +
>>>> + if (ret < 0) {
>>>> + v4l2_err(&isc->v4l2_dev,
>>>> + "Failed to create pad link: %s to %s\n",
>>>> + isc->scaler_sd.entity.name,
>>>> + isc->video_dev.entity.name);
>>>> + return ret;
>>>> + }
>>>> +
>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
>>>> + ISC_SCALER_PAD_SOURCE);
>>>> +
>>>> + return ret;
>>>> +}
>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
>>>
>>> From the DT graph point of view, the ISC appears as a single block
>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
>>> add to its notifier the image sensor remote subdevice.
>>
>> Actually it's only a parallel port.
>> And it can be connected either to a sensor directly or to the csi2dc
>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
>>
>>>
>>> When the sensor subdev registers and the ISC notifier completes, the scaler
>>> entity which was initialized at isc_probe() time is linked in between
>>> the ISC and the image sensor immutably.
>>
>> yes, because this cannot change at runtime, and usually, it can't change
>> without altering the board hardware. (at least this is what my
>> understanding of immutability is )
>>
>>>
>>> I think it is fine for now, but I wonder if you plan to plumb more
>>> components between the ISC video node and the sensor, if it's not
>>> worth changing the DT bindings and their parsing logic to separate the
>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
>>> time and have the CSI-2 receiver entity as downstream remote. I think
>>> I need to get to know the ISC better to really have an idea. For now,
>>> this seems ok to me, but please check with maintainers if this is fine
>>> with them.
>>
>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
>> the remote subdev).
>> The ISC will register a scaler, and connect the subdev to this scaler
>> first, and then, the scaler to the isc itself (the v4l entity).
>>
>>>
>>>> +
>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>>>> index 5fbf52a9080b..c9234c90ae58 100644
>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
>>>> u32 his_entry;
>>>> };
>>>>
>>>> +enum isc_mc_pads {
>>>> + ISC_PAD_SINK = 0,
>>>> + ISC_PADS_NUM = 1,
>>>> +};
>>>> +
>>>> +enum isc_scaler_pads {
>>>> + ISC_SCALER_PAD_SINK = 0,
>>>> + ISC_SCALER_PAD_SOURCE = 1,
>>>> + ISC_SCALER_PADS_NUM = 2,
>>>> +};
>>>> +
>>>> /*
>>>> * struct isc_device - ISC device driver data/config struct
>>>> * @regmap: Register map
>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
>>>> * be used as an input to the controller
>>>> * @controller_formats_size: size of controller_formats array
>>>> * @formats_list_size: size of formats_list array
>>>> + * @pads: media controller pads for isc video entity
>>>> + * @mdev: media device that is registered by the isc
>>>> + * @remote_pad: remote pad on the connected subdevice
>>>> + * @scaler_sd: subdevice for the scaler that isc registers
>>>> + * @scaler_pads: media controller pads for the scaler subdevice
>>>> + * @scaler_format: current format for the scaler subdevice
>>>> */
>>>> struct isc_device {
>>>> struct regmap *regmap;
>>>> @@ -344,6 +361,19 @@ struct isc_device {
>>>> struct isc_format *formats_list;
>>>> u32 controller_formats_size;
>>>> u32 formats_list_size;
>>>> +
>>>> + struct {
>>>> + struct media_pad pads[ISC_PADS_NUM];
>>>> + struct media_device mdev;
>>>> +
>>>> + u32 remote_pad;
>>>> + };
>>>> +
>>>> + struct {
>>>> + struct v4l2_subdev scaler_sd;
>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
>>>> + struct v4l2_mbus_framefmt scaler_format;
>>
>> Here are the scaler stuff which I added, in the same bulk struct for the
>> whole isc device
>>
>>
>>>> + };
>>>> };
>>>>
>>>> extern const struct regmap_config isc_regmap_config;
>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
>>>> void isc_subdev_cleanup(struct isc_device *isc);
>>>> void isc_clk_cleanup(struct isc_device *isc);
>>>>
>>>> +int isc_scaler_link(struct isc_device *isc);
>>>> +int isc_scaler_init(struct isc_device *isc);
>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
>>>> +void isc_mc_cleanup(struct isc_device *isc);
>>>> +
>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>> + unsigned int code, int *index);
>>>> #endif
>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>> index c5b9563e36cb..c244682ea22f 100644
>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>> break;
>>>> }
>>>>
>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>
>>> I am surprised you can read a register before runtime_pm is
>>> intialized!
>>>
>>
>> Actually I can read any register after the moment of starting the clock
>> on the hardware block.
>
> Ah right then
>
>> Maybe the starting and stopping of the clock needs to be moved to the
>> runtime_pm calls, but this is another story, not related to the media
>> controller.
>
> It's fine, but maybe worth recording with a todo ?
>
> Thanks
> j
>
>> I moved this line because I had to pass the version to the isc_mc_init call
>>
>>> Thanks
>>> j
>>
>> Thanks for reviewing !
>> Eugen
>>>
>>>
>>>> +
>>>> + ret = isc_mc_init(isc, ver);
>>>> + if (ret < 0)
>>>> + goto isc_probe_mc_init_err;
>>>> +
>>>> pm_runtime_set_active(dev);
>>>> pm_runtime_enable(dev);
>>>> pm_request_idle(dev);
>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>> ret = clk_prepare_enable(isc->ispck);
>>>> if (ret) {
>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
>>>> - goto cleanup_subdev;
>>>> + goto isc_probe_mc_init_err;
>>>> }
>>>>
>>>> /* ispck should be greater or equal to hclock */
>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>> goto unprepare_clk;
>>>> }
>>>>
>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
>>>>
>>>> return 0;
>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>> unprepare_clk:
>>>> clk_disable_unprepare(isc->ispck);
>>>>
>>>> +isc_probe_mc_init_err:
>>>> + isc_mc_cleanup(isc);
>>>> +
>>>> cleanup_subdev:
>>>> isc_subdev_cleanup(isc);
>>>>
>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>>>>
>>>> pm_runtime_disable(&pdev->dev);
>>>>
>>>> + isc_mc_cleanup(isc);
>>>> +
>>>> isc_subdev_cleanup(isc);
>>>>
>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>> index 07a80b08bc54..9dc75eed0098 100644
>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
>>>> break;
>>>> }
>>>>
>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>> +
>>>> + ret = isc_mc_init(isc, ver);
>>>> + if (ret < 0)
>>>> + goto isc_probe_mc_init_err;
>>>> +
>>>> pm_runtime_set_active(dev);
>>>> pm_runtime_enable(dev);
>>>> pm_request_idle(dev);
>>>>
>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
>>>>
>>>> return 0;
>>>>
>>>> +isc_probe_mc_init_err:
>>>> + isc_mc_cleanup(isc);
>>>> +
>>>> cleanup_subdev:
>>>> isc_subdev_cleanup(isc);
>>>>
>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>>>>
>>>> pm_runtime_disable(&pdev->dev);
>>>>
>>>> + isc_mc_cleanup(isc);
>>>> +
>>>> isc_subdev_cleanup(isc);
>>>>
>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>> --
>>>> 2.25.1
>>>>
>>

2022-02-09 12:09:32

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> >> Implement the support for media-controller.
> >> This means that the capabilities of the driver have changed and now
> >> it also advertises the IO_MC .
> >> The driver will register it's media device, and add the video entity to this
> >> media device. The subdevices are registered to the same media device.
> >> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> >> It will also register a subdevice that allows cropping of the incoming frame
> >> to the maximum frame size supported by the ISC.
> >> The ISC will create a link between the subdevice that is asynchronously
> >> registered and the atmel_isc_scaler entity.
> >> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> >> link.
> >>
> >> Signed-off-by: Eugen Hristev <[email protected]>
> >> ---
> >> Changes in v4:
> >> As suggested by Jacopo:
> >> - renamed atmel_isc_mc to atmel_isc_scaler.c
> >> - moved init_mc/clean_mc to isc_base file
> >>
> >> Changes in v2:
> >> - implement try formats
> >>
> >> drivers/media/platform/atmel/Makefile | 2 +-
> >> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> >> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> >> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> >> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> >> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> >> 6 files changed, 375 insertions(+), 8 deletions(-)
> >> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
> >>
> >> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> >> index 794e8f739287..f02d03df89d6 100644
> >> --- a/drivers/media/platform/atmel/Makefile
> >> +++ b/drivers/media/platform/atmel/Makefile
> >> @@ -1,7 +1,7 @@
> >> # SPDX-License-Identifier: GPL-2.0-only
> >> atmel-isc-objs = atmel-sama5d2-isc.o
> >> atmel-xisc-objs = atmel-sama7g5-isc.o
> >> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> >> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
> >>
> >> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> >> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> >> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >> index 6b0005987a17..6b482270eb93 100644
> >> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >> struct isc_device, v4l2_dev);
> >> struct isc_subdev_entity *subdev_entity =
> >> container_of(notifier, struct isc_subdev_entity, notifier);
> >> + int pad;
> >>
> >> if (video_is_registered(&isc->video_dev)) {
> >> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> >> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>
> >> subdev_entity->sd = subdev;
> >>
> >> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> >> + MEDIA_PAD_FL_SOURCE);
> >> + if (pad < 0) {
> >> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> >> + subdev->name);
> >> + return pad;
> >> + }
> >> +
> >> + isc->remote_pad = pad;
> >> +
> >> return 0;
> >> }
> >>
> >> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> >> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> >> }
> >>
> >> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> >> - unsigned int code, int *index)
> >> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >> + unsigned int code, int *index)
> >> {
> >> struct isc_format *fmt = &isc->formats_list[0];
> >> unsigned int i;
> >> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>
> >> return NULL;
> >> }
> >> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
> >>
> >> static int isc_formats_init(struct isc_device *isc)
> >> {
> >> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> >> NULL, &mbus_code)) {
> >> mbus_code.index++;
> >>
> >> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> >> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> >> if (!fmt) {
> >> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> >> mbus_code.code);
> >> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >> vdev->queue = q;
> >> vdev->lock = &isc->lock;
> >> vdev->ctrl_handler = &isc->ctrls.handler;
> >> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> >> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> >> + V4L2_CAP_IO_MC;
> >> video_set_drvdata(vdev, isc);
> >>
> >> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> >> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >> goto isc_async_complete_err;
> >> }
> >>
> >> + ret = isc_scaler_link(isc);
> >> + if (ret < 0)
> >> + goto isc_async_complete_unregister_device;
> >> +
> >> + ret = media_device_register(&isc->mdev);
> >> + if (ret < 0)
> >> + goto isc_async_complete_unregister_device;
> >> return 0;
> >>
> >> +isc_async_complete_unregister_device:
> >> + video_unregister_device(vdev);
> >> +
> >> isc_async_complete_err:
> >> mutex_destroy(&isc->lock);
> >> return ret;
> >> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> >> }
> >> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>
> >> +int isc_mc_init(struct isc_device *isc, u32 ver)
> >> +{
> >> + const struct of_device_id *match;
> >> + int ret;
> >> +
> >> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >
> > Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> > to be able to have the media pipeline validated at
> > media_pipeline_start() time ?
>
> Hi,
>
> I am doing that in a subsequent patch. Things are not completely ready
> at this moment, because ISC still relies on the old mechanism to call
> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
> removes and adds checks to the link validate ).
>

I see.. the subsequent patches are not part of this series, right ?

I think patches 1, 2, 4, 5 and 6 from this series do not depend on
the introduction of media controller or the scaler, right ? (I'm not
sure if the DTS patches do).

If that's the case, would it make sense to to fast-track them (as some
of them are fixes) and then on top plumb the media controller
infrastructure with this patch and the ones you have been mentioning ?

> >
> >> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >> +
> >> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >> + isc->pads);
> >> + if (ret < 0) {
> >> + dev_err(isc->dev, "media entity init failed\n");
> >> + return ret;
> >> + }
> >> +
> >> + isc->mdev.dev = isc->dev;
> >> +
> >> + match = of_match_node(isc->dev->driver->of_match_table,
> >> + isc->dev->of_node);
> >> +
> >> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> >> + sizeof(isc->mdev.driver_name));
> >> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> >> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> >> + isc->v4l2_dev.name);
> >> + isc->mdev.hw_revision = ver;
> >> +
> >> + media_device_init(&isc->mdev);
> >> +
> >> + isc->v4l2_dev.mdev = &isc->mdev;
> >> +
> >> + return isc_scaler_init(isc);
> >> +}
> >> +EXPORT_SYMBOL_GPL(isc_mc_init);
> >> +
> >> +void isc_mc_cleanup(struct isc_device *isc)
> >> +{
> >> + media_entity_cleanup(&isc->video_dev.entity);
> >> +}
> >> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> >> +
> >> /* regmap configuration */
> >> #define ATMEL_ISC_REG_MAX 0xd5c
> >> const struct regmap_config isc_regmap_config = {
> >> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >> new file mode 100644
> >> index 000000000000..ec95c9665883
> >> --- /dev/null
> >> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >> @@ -0,0 +1,245 @@
> >> +// SPDX-License-Identifier: GPL-2.0-only
> >> +/*
> >> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> >> + *
> >> + * Copyright (C) 2021 Microchip Technology, Inc.
> >
> > Time flies! It's 2022 already :)
> >
> >> + *
> >> + * Author: Eugen Hristev <[email protected]>
> >> + *
> >> + */
> >> +
> >> +#include <media/media-device.h>
> >> +#include <media/media-entity.h>
> >> +#include <media/v4l2-device.h>
> >> +#include <media/v4l2-subdev.h>
> >> +
> >> +#include "atmel-isc-regs.h"
> >> +#include "atmel-isc.h"
> >> +
> >> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> >> + struct v4l2_subdev_state *sd_state,
> >> + struct v4l2_subdev_format *format)
> >> +{
> >> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >> +
> >> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> >> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >> + format->pad);
> >> + format->format = *v4l2_try_fmt;
> >> +
> >> + return 0;
> >> + }
> >> +
> >> + format->format = isc->scaler_format;
> >
> > isc->scaler_format is only used inside this file if I'm not mistaken.
> > I wonder why it lives in the isc_device struct.
>
> isc_device is a placeholder for all isc things.
>
> I would not create a separate struct in this file that would have to be
> allocated etc... just for one two things. So I preferred to have it in
> the same place as all the other things.
>
> >
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> >> + struct v4l2_subdev_state *sd_state,
> >> + struct v4l2_subdev_format *req_fmt)
> >> +{
> >> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >> + struct isc_format *fmt;
> >> + unsigned int i;
> >> +
> >> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> >> + v4l_bound_align_image
> >> + (&req_fmt->format.width, 16, isc->max_width, 0,
> >> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> >> + else
> >> + v4l_bound_align_image
> >> + (&req_fmt->format.width, 16, 10000, 0,
> >> + &req_fmt->format.height, 16, 10000, 0, 0);
> >
> > Where does 10000 come from ?
>
> It's a random number. Do you have any suggestion for a better one ?

An actual HW limit ? :)

> Maybe this would be much more clear with my comments below (where I will
> explain what the isc scaler does )
> In short, it allows the other entity in the link to 'sink' a huge format
> into this 'scaler' . Because the scaler can crop anything basically down
> to the biggest format size the ISC could handle.
>

doesn't it have any documented input size limit ?

> >
> >> +
> >> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> >> + req_fmt->format.field = V4L2_FIELD_NONE;
> >> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >> +
> >> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
> >
> > So you rely on the isc format list for the scaler as well ?
> > I think it's fine as long as they are identical
>
> Yes, the scaler is kind of a dummy entity , but it's required by the
> media controller validation of the pipeline.
>

More on this later

> >
> >> +
> >> + if (!fmt)
> >> + fmt = &isc->formats_list[0];
> >> +
> >> + req_fmt->format.code = fmt->mbus_code;
> >> +
> >> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> >> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >> + req_fmt->pad);
> >> + *v4l2_try_fmt = req_fmt->format;
> >> + /* Trying on the pad sink makes the source sink change too */
> >> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> >> + v4l2_try_fmt =
> >> + v4l2_subdev_get_try_format(sd, sd_state,
> >> + ISC_SCALER_PAD_SOURCE);
> >> + *v4l2_try_fmt = req_fmt->format;
> >> +
> >> + v4l_bound_align_image(&v4l2_try_fmt->width,
> >> + 16, isc->max_width, 0,
> >> + &v4l2_try_fmt->height,
> >> + 16, isc->max_height, 0, 0);
> >> + }
> >> + /* if we are just trying, we are done */
> >> + return 0;
> >> + }
> >> +
> >> + isc->scaler_format = req_fmt->format;
> >
> > No per-pad format but a global scaler_format ? How do you configure scaling ?
> >
> > Actually, I would like to know more about the scaler device
> > capabilities. What functions can this IP perform ? Does it do
> > cropping, can it also do (down)scaling or even composition ?
> >
> > I think it is worth to read
> > https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> > where it is reported how cropping and scaling are implemented by using
> > the selection API.
> >
> > Figure 4.5 is particularly helpful to explain the simple crop case,
> > for which you need to implement support to by adding s_selection on
> > the sink pad TGT_CROP target.
>
> The scaler is kind of a dummy entity.
> The problem is that different subdevices cannot be connected directly to
> a v4l entity, because the ISC does cropping on the incoming frame.
> The ISC has a limit on the total number of pixels per frame, thus it
> will do a crop.

What is this limit related to ? I don't know the ISC internals, but is
the limitation due to the overall image size, or maybe it's due to
some line buffers being limited in the size they can transfer to
memory ? Is it a limitation on the size of the image in pixels, or is
it due to a limitation of the processing bandwidth on the input
interface and thus depends on the actual size of the image in bytes ?

> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
> only handle 3264x2464 , then we have two choices:
> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
> 3280x2464, and the whole pipeline fails to configure, there is a
> mismatch between /dev/video output and what the whole pipeline produces

I'm confused in first place. If the ISC cannot output anything larger
than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
this should be adjusted to 3264x2464.

Or is the ISC limitation on the -input- side ?

> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
> the ISC can handle at most -> cannot , because the sensor let's say can
> *only* produce 3280x2464 and *any* other frame size returns an error.

buggy sensor I would say :)

>
> The only solution to make both worlds happy, is to have a dummy entity
> called 'scaler' which in fact now it only performs a simple crop.
> It accepts any frame size at input (hence the 10000x10000 which you saw
> earlier), and outputs at most 3264x2464 the isc_max_height and
> isc_max_width.

I have a maybe dumb question: can the cropping operation be modeled
by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
the atmel_isc_common entity sink pad and avoid the scaler completely ?

>
> So to answer your question, the isc scaler is a software model for a
> simple cropping, that would make media controller happy, and capture
> software happy.
>
> Here it how it looks :
>
> Device topology
> - entity 1: atmel_isc_scaler (2 pads, 2 links)
> type V4L2 subdev subtype Unknown flags 0
> device node name /dev/v4l-subdev0
> pad0: Sink
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> crop.bounds:(0,0)/3264x2464
> crop:(0,0)/3264x2464]
> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> pad1: Source
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>
> - entity 4: csi2dc (2 pads, 2 links)
> type V4L2 subdev subtype Unknown flags 0
> device node name /dev/v4l-subdev1
> pad0: Sink
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> <- "dw-csi.0":1 [ENABLED]
> pad1: Source
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
>
> - entity 7: dw-csi.0 (2 pads, 2 links)
> type V4L2 subdev subtype Unknown flags 0
> device node name /dev/v4l-subdev2
> pad0: Sink
> [fmt:SRGGB10_1X10/3280x2464]
> <- "imx219 1-0010":0 [ENABLED]
> pad1: Source
> [fmt:SRGGB10_1X10/3280x2464]
> -> "csi2dc":0 [ENABLED]
>
> - entity 12: imx219 1-0010 (1 pad, 1 link)
> type V4L2 subdev subtype Sensor flags 0
> device node name /dev/v4l-subdev3
> pad0: Source
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> xfer:srgb ycbcr:601 quantization:full-range
> crop.bounds:(8,8)/3280x2464
> crop:(8,8)/3280x2464]
> -> "dw-csi.0":0 [ENABLED]
>
> - entity 24: atmel_isc_common (1 pad, 1 link)
> type Node subtype V4L flags 1
> device node name /dev/video0
> pad0: Sink
> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
>
>
> Scaler does this one cute little thing :
>
> pad0: Sink
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> crop.bounds:(0,0)/3264x2464
> crop:(0,0)/3264x2464]
> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> pad1: Source
> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]

Shouldn't this be 3264x2464 as that's what the entity outputs after
the cropping ? And shouldn't then be 3264x2464 the size output from
the video device too ?


> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>
>
> Which is what we needed.
>
> >
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> >> + struct v4l2_subdev_state *sd_state,
> >> + struct v4l2_subdev_mbus_code_enum *code)
> >> +{
> >> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >> + int supported_index = 0;
> >> + int i;
> >> +
> >> + for (i = 0; i < isc->formats_list_size; i++) {
> >> + if (!isc->formats_list[i].sd_support)
> >> + continue;
> >
> > The sd_support flag still doesn't click in my head.
> >
> > Shouldn't the enumeration of available formats on the scaler do not
> > depend on the sensor supproted formats ?
>
> You're right. I will have to check it again.
>
> >
> >> + if (supported_index == code->index) {
> >> + code->code = isc->formats_list[i].mbus_code;
> >> + return 0;
> >> + }
> >> + supported_index++;
> >> + }
> >> +
> >> + return -EINVAL;
> >> +}
> >> +
> >> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> >> + struct v4l2_subdev_state *sd_state,
> >> + struct v4l2_subdev_selection *sel)
> >> +{
> >> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >> +
> >> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> >> + return -EINVAL;
> >> +
> >> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> >> + sel->target != V4L2_SEL_TGT_CROP)
> >> + return -EINVAL;
> >> +
> >> + sel->r.height = isc->max_height;
> >> + sel->r.width = isc->max_width;
> >
> > The CROP_BOUNDS should be set to the same size as the sink pad image format,
> > as it represents the maximum valid crop rectangle.
> >
> > TGT_CROP should report the configured crop rectangle which can be
> > intiialized to the same size as CROP_BOUNDS, if my understanding of
> > the spec is correct
>
> So you would like to have this differentiated, and report the
> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
> reported now, the maximum size of the ISC frame .
> My understanding is correct ?
>

I didn't know you have an HW limitation, so your _BOUNDS is not the
input image size but rather 3264x2464 ( == max_width x max_height).

What I meant is that _BOUNDS should report the maximum rectangle size
that can be applied to the _CROP target. In you case you have an HW
limitation 3264x2464 and that's the largest rectangle you can apply.
TGT_CROP can be initialized to the same as _BOUND, but if you
implement s_selection it should report what has been there applied.
But as you don't implement s_selection yet, I think this is fine for
now. Maybe a little comment ?

Also, is set->r zeroed by the framework before getting here ?
Otherwise you should set r.left and r.top to 0

> >
> >> +
> >> + sel->r.left = 0;
> >> + sel->r.top = 0;
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> >> + struct v4l2_subdev_state *sd_state)
> >> +{
> >> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> >> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> >> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >> +
> >> + *v4l2_try_fmt = isc->scaler_format;
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> >> + .set_fmt = isc_scaler_set_fmt,
> >> + .get_fmt = isc_scaler_get_fmt,
> >> + .get_selection = isc_scaler_g_sel,
> >> + .init_cfg = isc_scaler_init_cfg,
> >
> > .link_validate = v4l2_subdev_link_validate_default,
> >
> > To have the formats at the end of links that point to this entity
> > validated (I think the framework already calls it if not set though,
> > please check v4l2-subdev.c:v4l2_subdev_link_validate())
> >
> >> +};
> >> +
> >> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >> + .pad = &isc_scaler_pad_ops,
> >> +};
> >> +
> >> +int isc_scaler_init(struct isc_device *isc)
> >> +{
> >> + int ret;
> >> +
> >> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> >> +
> >> + isc->scaler_sd.owner = THIS_MODULE;
> >> + isc->scaler_sd.dev = isc->dev;
> >> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> >> + "atmel_isc_scaler");
> >
> > I would drop 'atmel' for brevity, unless other entities have this
> > prefix set already
>
> The v4l entity takes it's name from the module name, which is
> atmel_isc_common, so I thought to keep the prefix
>

Ack!

> >
> >> +
> >> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >> +
> >> + isc->scaler_format.height = isc->max_height;
> >> + isc->scaler_format.width = isc->max_width;
> >> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> >> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> >> + isc->scaler_format.field = V4L2_FIELD_NONE;
> >> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >> +
> >> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> >> + ISC_SCALER_PADS_NUM,
> >> + isc->scaler_pads);
> >> + if (ret < 0) {
> >> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> >> + return ret;
> >> + }
> >> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> >> + if (ret < 0) {
> >> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> >> + return ret;
> >> + }
> >> +
> >> + return ret;
> >> +}
> >> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> >> +
> >> +int isc_scaler_link(struct isc_device *isc)
> >> +{
> >> + int ret;
> >> +
> >> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> >> + isc->remote_pad, &isc->scaler_sd.entity,
> >> + ISC_SCALER_PAD_SINK,
> >> + MEDIA_LNK_FL_ENABLED |
> >> + MEDIA_LNK_FL_IMMUTABLE);
> >> +
> >> + if (ret < 0) {
> >> + v4l2_err(&isc->v4l2_dev,
> >> + "Failed to create pad link: %s to %s\n",
> >> + isc->current_subdev->sd->entity.name,
> >> + isc->scaler_sd.entity.name);
> >> + return ret;
> >> + }
> >> +
> >> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> >> + isc->current_subdev->sd->name, isc->remote_pad);
> >> +
> >> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> >> + ISC_SCALER_PAD_SOURCE,
> >> + &isc->video_dev.entity, ISC_PAD_SINK,
> >> + MEDIA_LNK_FL_ENABLED |
> >> + MEDIA_LNK_FL_IMMUTABLE);
> >> +
> >> + if (ret < 0) {
> >> + v4l2_err(&isc->v4l2_dev,
> >> + "Failed to create pad link: %s to %s\n",
> >> + isc->scaler_sd.entity.name,
> >> + isc->video_dev.entity.name);
> >> + return ret;
> >> + }
> >> +
> >> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> >> + ISC_SCALER_PAD_SOURCE);
> >> +
> >> + return ret;
> >> +}
> >> +EXPORT_SYMBOL_GPL(isc_scaler_link);
> >
> > From the DT graph point of view, the ISC appears as a single block
> > with an CSI-2 input port. It is then in charge of parsing the .dts and
> > add to its notifier the image sensor remote subdevice.
>
> Actually it's only a parallel port.
> And it can be connected either to a sensor directly or to the csi2dc
> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
>
> >
> > When the sensor subdev registers and the ISC notifier completes, the scaler
> > entity which was initialized at isc_probe() time is linked in between
> > the ISC and the image sensor immutably.
>
> yes, because this cannot change at runtime, and usually, it can't change
> without altering the board hardware. (at least this is what my
> understanding of immutability is )
>
> >
> > I think it is fine for now, but I wonder if you plan to plumb more
> > components between the ISC video node and the sensor, if it's not
> > worth changing the DT bindings and their parsing logic to separate the
> > CSI-2 receiver from the ISC, whcih can create its media graph at probe
> > time and have the CSI-2 receiver entity as downstream remote. I think
> > I need to get to know the ISC better to really have an idea. For now,
> > this seems ok to me, but please check with maintainers if this is fine
> > with them.
>
> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
> the remote subdev).
> The ISC will register a scaler, and connect the subdev to this scaler
> first, and then, the scaler to the isc itself (the v4l entity).
>
> >
> >> +
> >> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >> index 5fbf52a9080b..c9234c90ae58 100644
> >> --- a/drivers/media/platform/atmel/atmel-isc.h
> >> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> >> u32 his_entry;
> >> };
> >>
> >> +enum isc_mc_pads {
> >> + ISC_PAD_SINK = 0,
> >> + ISC_PADS_NUM = 1,
> >> +};
> >> +
> >> +enum isc_scaler_pads {
> >> + ISC_SCALER_PAD_SINK = 0,
> >> + ISC_SCALER_PAD_SOURCE = 1,
> >> + ISC_SCALER_PADS_NUM = 2,
> >> +};
> >> +
> >> /*
> >> * struct isc_device - ISC device driver data/config struct
> >> * @regmap: Register map
> >> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> >> * be used as an input to the controller
> >> * @controller_formats_size: size of controller_formats array
> >> * @formats_list_size: size of formats_list array
> >> + * @pads: media controller pads for isc video entity
> >> + * @mdev: media device that is registered by the isc
> >> + * @remote_pad: remote pad on the connected subdevice
> >> + * @scaler_sd: subdevice for the scaler that isc registers
> >> + * @scaler_pads: media controller pads for the scaler subdevice
> >> + * @scaler_format: current format for the scaler subdevice
> >> */
> >> struct isc_device {
> >> struct regmap *regmap;
> >> @@ -344,6 +361,19 @@ struct isc_device {
> >> struct isc_format *formats_list;
> >> u32 controller_formats_size;
> >> u32 formats_list_size;
> >> +
> >> + struct {
> >> + struct media_pad pads[ISC_PADS_NUM];
> >> + struct media_device mdev;
> >> +
> >> + u32 remote_pad;
> >> + };
> >> +
> >> + struct {
> >> + struct v4l2_subdev scaler_sd;
> >> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> >> + struct v4l2_mbus_framefmt scaler_format;
>
> Here are the scaler stuff which I added, in the same bulk struct for the
> whole isc device
>
>
> >> + };
> >> };
> >>
> >> extern const struct regmap_config isc_regmap_config;
> >> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> >> void isc_subdev_cleanup(struct isc_device *isc);
> >> void isc_clk_cleanup(struct isc_device *isc);
> >>
> >> +int isc_scaler_link(struct isc_device *isc);
> >> +int isc_scaler_init(struct isc_device *isc);
> >> +int isc_mc_init(struct isc_device *isc, u32 ver);
> >> +void isc_mc_cleanup(struct isc_device *isc);
> >> +
> >> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >> + unsigned int code, int *index);
> >> #endif
> >> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >> index c5b9563e36cb..c244682ea22f 100644
> >> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >> break;
> >> }
> >>
> >> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >
> > I am surprised you can read a register before runtime_pm is
> > intialized!
> >
>
> Actually I can read any register after the moment of starting the clock
> on the hardware block.

Ah right then

> Maybe the starting and stopping of the clock needs to be moved to the
> runtime_pm calls, but this is another story, not related to the media
> controller.

It's fine, but maybe worth recording with a todo ?

Thanks
j

> I moved this line because I had to pass the version to the isc_mc_init call
>
> > Thanks
> > j
>
> Thanks for reviewing !
> Eugen
> >
> >
> >> +
> >> + ret = isc_mc_init(isc, ver);
> >> + if (ret < 0)
> >> + goto isc_probe_mc_init_err;
> >> +
> >> pm_runtime_set_active(dev);
> >> pm_runtime_enable(dev);
> >> pm_request_idle(dev);
> >> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >> ret = clk_prepare_enable(isc->ispck);
> >> if (ret) {
> >> dev_err(dev, "failed to enable ispck: %d\n", ret);
> >> - goto cleanup_subdev;
> >> + goto isc_probe_mc_init_err;
> >> }
> >>
> >> /* ispck should be greater or equal to hclock */
> >> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >> goto unprepare_clk;
> >> }
> >>
> >> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >> dev_info(dev, "Microchip ISC version %x\n", ver);
> >>
> >> return 0;
> >> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >> unprepare_clk:
> >> clk_disable_unprepare(isc->ispck);
> >>
> >> +isc_probe_mc_init_err:
> >> + isc_mc_cleanup(isc);
> >> +
> >> cleanup_subdev:
> >> isc_subdev_cleanup(isc);
> >>
> >> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
> >>
> >> pm_runtime_disable(&pdev->dev);
> >>
> >> + isc_mc_cleanup(isc);
> >> +
> >> isc_subdev_cleanup(isc);
> >>
> >> v4l2_device_unregister(&isc->v4l2_dev);
> >> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >> index 07a80b08bc54..9dc75eed0098 100644
> >> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> >> break;
> >> }
> >>
> >> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >> +
> >> + ret = isc_mc_init(isc, ver);
> >> + if (ret < 0)
> >> + goto isc_probe_mc_init_err;
> >> +
> >> pm_runtime_set_active(dev);
> >> pm_runtime_enable(dev);
> >> pm_request_idle(dev);
> >>
> >> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >> dev_info(dev, "Microchip XISC version %x\n", ver);
> >>
> >> return 0;
> >>
> >> +isc_probe_mc_init_err:
> >> + isc_mc_cleanup(isc);
> >> +
> >> cleanup_subdev:
> >> isc_subdev_cleanup(isc);
> >>
> >> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
> >>
> >> pm_runtime_disable(&pdev->dev);
> >>
> >> + isc_mc_cleanup(isc);
> >> +
> >> isc_subdev_cleanup(isc);
> >>
> >> v4l2_device_unregister(&isc->v4l2_dev);
> >> --
> >> 2.25.1
> >>
>

2022-02-09 13:46:32

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

On 2/8/22 9:30 PM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
>> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
>>> Hi Eugen
>>>
>>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
>>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
>>>>> Hi Eugen
>>>>>
>>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
>>>>>> Implement the support for media-controller.
>>>>>> This means that the capabilities of the driver have changed and now
>>>>>> it also advertises the IO_MC .
>>>>>> The driver will register it's media device, and add the video entity to this
>>>>>> media device. The subdevices are registered to the same media device.
>>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
>>>>>> It will also register a subdevice that allows cropping of the incoming frame
>>>>>> to the maximum frame size supported by the ISC.
>>>>>> The ISC will create a link between the subdevice that is asynchronously
>>>>>> registered and the atmel_isc_scaler entity.
>>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
>>>>>> link.
>>>>>>
>>>>>> Signed-off-by: Eugen Hristev <[email protected]>
>>>>>> ---
>>>>>> Changes in v4:
>>>>>> As suggested by Jacopo:
>>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
>>>>>> - moved init_mc/clean_mc to isc_base file
>>>>>>
>>>>>> Changes in v2:
>>>>>> - implement try formats
>>>>>>
>>>>>> drivers/media/platform/atmel/Makefile | 2 +-
>>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
>>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
>>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
>>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
>>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
>>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
>>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>
>>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
>>>>>> index 794e8f739287..f02d03df89d6 100644
>>>>>> --- a/drivers/media/platform/atmel/Makefile
>>>>>> +++ b/drivers/media/platform/atmel/Makefile
>>>>>> @@ -1,7 +1,7 @@
>>>>>> # SPDX-License-Identifier: GPL-2.0-only
>>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
>>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
>>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
>>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>>>>>>
>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>> index 6b0005987a17..6b482270eb93 100644
>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>> struct isc_device, v4l2_dev);
>>>>>> struct isc_subdev_entity *subdev_entity =
>>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
>>>>>> + int pad;
>>>>>>
>>>>>> if (video_is_registered(&isc->video_dev)) {
>>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
>>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>>
>>>>>> subdev_entity->sd = subdev;
>>>>>>
>>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
>>>>>> + MEDIA_PAD_FL_SOURCE);
>>>>>> + if (pad < 0) {
>>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
>>>>>> + subdev->name);
>>>>>> + return pad;
>>>>>> + }
>>>>>> +
>>>>>> + isc->remote_pad = pad;
>>>>>> +
>>>>>> return 0;
>>>>>> }
>>>>>>
>>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
>>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
>>>>>> }
>>>>>>
>>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>> - unsigned int code, int *index)
>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>> + unsigned int code, int *index)
>>>>>> {
>>>>>> struct isc_format *fmt = &isc->formats_list[0];
>>>>>> unsigned int i;
>>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>>
>>>>>> return NULL;
>>>>>> }
>>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>>>>>>
>>>>>> static int isc_formats_init(struct isc_device *isc)
>>>>>> {
>>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
>>>>>> NULL, &mbus_code)) {
>>>>>> mbus_code.index++;
>>>>>>
>>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
>>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
>>>>>> if (!fmt) {
>>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
>>>>>> mbus_code.code);
>>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>> vdev->queue = q;
>>>>>> vdev->lock = &isc->lock;
>>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
>>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
>>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
>>>>>> + V4L2_CAP_IO_MC;
>>>>>> video_set_drvdata(vdev, isc);
>>>>>>
>>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
>>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>> goto isc_async_complete_err;
>>>>>> }
>>>>>>
>>>>>> + ret = isc_scaler_link(isc);
>>>>>> + if (ret < 0)
>>>>>> + goto isc_async_complete_unregister_device;
>>>>>> +
>>>>>> + ret = media_device_register(&isc->mdev);
>>>>>> + if (ret < 0)
>>>>>> + goto isc_async_complete_unregister_device;
>>>>>> return 0;
>>>>>>
>>>>>> +isc_async_complete_unregister_device:
>>>>>> + video_unregister_device(vdev);
>>>>>> +
>>>>>> isc_async_complete_err:
>>>>>> mutex_destroy(&isc->lock);
>>>>>> return ret;
>>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
>>>>>> }
>>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>>>>>
>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
>>>>>> +{
>>>>>> + const struct of_device_id *match;
>>>>>> + int ret;
>>>>>> +
>>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>>>>>
>>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
>>>>> to be able to have the media pipeline validated at
>>>>> media_pipeline_start() time ?
>>>>
>>>> Hi,
>>>>
>>>> I am doing that in a subsequent patch. Things are not completely ready
>>>> at this moment, because ISC still relies on the old mechanism to call
>>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
>>>> removes and adds checks to the link validate ).
>>>>
>>>
>>> I see.. the subsequent patches are not part of this series, right ?
>>
>> No. Patch 7 does this. In that patch, the link validation is created and
>> all the logic of format propagation is removed and reworked, and creates
>> the need for the link_validate call
>> Could you have also a look at that patch ?
>
> Ah ups, 7 was not in my inbox with the rest of the series. Just
> noticed. I had even reviewed the previous version, I should have
> remembered link_validate was added later
>
>>>
>>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
>>> the introduction of media controller or the scaler, right ? (I'm not
>>> sure if the DTS patches do).
>
> Do DTS patches depend on media-controller plumbing ?

I think not, but, I have not tested them without the media controller.

>
>>>
>>> If that's the case, would it make sense to to fast-track them (as some
>>> of them are fixes) and then on top plumb the media controller
>>> infrastructure with this patch and the ones you have been mentioning ?
>>
>> It would make sense.
>>
>
> Once the other discussion with Hans about the stop state it might make
> sense to fastrack the first part of the series ?

The other three patches (4,5,6) are quite independent and can go faster.
>
>>>
>>>>>
>>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>> +
>>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>>>>>> + isc->pads);
>>>>>> + if (ret < 0) {
>>>>>> + dev_err(isc->dev, "media entity init failed\n");
>>>>>> + return ret;
>>>>>> + }
>>>>>> +
>>>>>> + isc->mdev.dev = isc->dev;
>>>>>> +
>>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
>>>>>> + isc->dev->of_node);
>>>>>> +
>>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
>>>>>> + sizeof(isc->mdev.driver_name));
>>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
>>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
>>>>>> + isc->v4l2_dev.name);
>>>>>> + isc->mdev.hw_revision = ver;
>>>>>> +
>>>>>> + media_device_init(&isc->mdev);
>>>>>> +
>>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
>>>>>> +
>>>>>> + return isc_scaler_init(isc);
>>>>>> +}
>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
>>>>>> +
>>>>>> +void isc_mc_cleanup(struct isc_device *isc)
>>>>>> +{
>>>>>> + media_entity_cleanup(&isc->video_dev.entity);
>>>>>> +}
>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
>>>>>> +
>>>>>> /* regmap configuration */
>>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
>>>>>> const struct regmap_config isc_regmap_config = {
>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>> new file mode 100644
>>>>>> index 000000000000..ec95c9665883
>>>>>> --- /dev/null
>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>> @@ -0,0 +1,245 @@
>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
>>>>>> +/*
>>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
>>>>>> + *
>>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
>>>>>
>>>>> Time flies! It's 2022 already :)
>>>>>
>>>>>> + *
>>>>>> + * Author: Eugen Hristev <[email protected]>
>>>>>> + *
>>>>>> + */
>>>>>> +
>>>>>> +#include <media/media-device.h>
>>>>>> +#include <media/media-entity.h>
>>>>>> +#include <media/v4l2-device.h>
>>>>>> +#include <media/v4l2-subdev.h>
>>>>>> +
>>>>>> +#include "atmel-isc-regs.h"
>>>>>> +#include "atmel-isc.h"
>>>>>> +
>>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>> + struct v4l2_subdev_format *format)
>>>>>> +{
>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>> +
>>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>> + format->pad);
>>>>>> + format->format = *v4l2_try_fmt;
>>>>>> +
>>>>>> + return 0;
>>>>>> + }
>>>>>> +
>>>>>> + format->format = isc->scaler_format;
>>>>>
>>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
>>>>> I wonder why it lives in the isc_device struct.
>>>>
>>>> isc_device is a placeholder for all isc things.
>>>>
>>>> I would not create a separate struct in this file that would have to be
>>>> allocated etc... just for one two things. So I preferred to have it in
>>>> the same place as all the other things.
>>>>
>>>>>
>>>>>> +
>>>>>> + return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>> + struct v4l2_subdev_format *req_fmt)
>>>>>> +{
>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>> + struct isc_format *fmt;
>>>>>> + unsigned int i;
>>>>>> +
>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
>>>>>> + v4l_bound_align_image
>>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
>>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
>>>>>> + else
>>>>>> + v4l_bound_align_image
>>>>>> + (&req_fmt->format.width, 16, 10000, 0,
>>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
>>>>>
>>>>> Where does 10000 come from ?
>>>>
>>>> It's a random number. Do you have any suggestion for a better one ?
>>>
>>> An actual HW limit ? :)
>>
>> There is no hardware limit. The ISC just stops sampling pixels once the
>> crop limit is reached. The element that generates pixels could go on
>> forever , or until the next HBLANK/VBLANK. So what limit could I place
>> here ?
>> The cropping is mandatory, otherwise there could be an overflow w.r.t.
>> the buffer size if the ISC would sample more pixels than the software
>> expects it to.
>>
>>>
>>>> Maybe this would be much more clear with my comments below (where I will
>>>> explain what the isc scaler does )
>>>> In short, it allows the other entity in the link to 'sink' a huge format
>>>> into this 'scaler' . Because the scaler can crop anything basically down
>>>> to the biggest format size the ISC could handle.
>>>>
>>>
>>> doesn't it have any documented input size limit ?
>>
>> As stated above, ISC handles a pixel stream and stores it in an internal
>> SRAM. ISC stops sampling when this SRAM is full or when the cropping
>> limit is reached.
>> After storing the pixels in the SRAM (which is limited to the ISC
>> maximum frame size), the ISC will work on the pixels and then DMA the
>> frame to another RAM.
>>
>
> I understand...
>
> So whatever the input size is, only the first n-th bytes are actually
> dumped to memory, the following ones are discarded.

Yes ! That's the behavior.
>
> So the crop is always applied on the top-leftmost corner as a
> consequence, right ?

Right. Pixels come in order. The ISC does not really know how many
pixels are coming until its memory is full (or counting until the crop
limit is reached). After that it stops sampling pixels, and waits for
vblank to finish the frame.


>
> I understand why you had difficulties picking a default :)
>
> This is not easy to handle, as the scaler doesn't actually applies a
> crop but rather stops capturing after the destination buffer has been
> saturated. So the limiting factor is the full image size, not one of
> its dimensions, like in example the line length. This makes it hard
> to adjust cropping to something meaningful for most use cases.

I didn't know how to name this except crop. It 'crops out' what does not
fit in the internal memory of the ISC

>
> In your below example you suggest that an image size of (800000 x 10)
> would theoretically be captured correctly. What if the input is of
> size (800000 x 10 + 1). What size would you set the crop rectangle to ?
>
> I wouldn't however be too much concerned for the moment, but I would
> record this in a comment ?

I discussed a bit more with designer and looked in datasheet.
It looks like the horizontal line length is not a hard limit, but it
makes the ISC internal ISP stop working, so I would keep the horizontal
limit as well for now, unless an important use case (I am coming back to
that). The vertical limit looks to be a hard requirement.
So in short, we have to impose a vertical limit of 2464 , we cannot have
more lines than these. So if the horizontal is 100 pixels e.g., we can
capture only 100x2464 .
In practice we could capture [16,3264] on the horizontal, and [16,2464]
on the vertical, and in theory horizontal could go more like [16, 4000],
but we have to limit the vertical, and without the internal ISP.
So I would not haphazardly go in this direction. Let's stick to the
datasheet maximum frame 3264x2464 .
According to this , I changed the frame size of the ISC to be a
continuous size from [16,16] to [3264,2464]
About that use case that I talked about earlier, we have one sensor ,
imx274, which is more wide, and outputs something like 3800x2100 .
With the cropping and limitations we have in place now, this will be
limited to 3264x2100 , which is in line with datasheet and what the ISC
supports now.
As in theory we could capture 3800x2100 as long as the SRAM doesn't get
full, for now I would like to avoid capturing this size (and disable the
internal ISP), until we have someone really asking for such kind of
capture and with some good reasons.
In short I would like to have the crop rectangle to 3264x2464 and the
bounds would be limited to the incoming frame but no H>3264 and no
V>2464. Does that make sense ?

>
>>>
>>>>>
>>>>>> +
>>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
>>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>> +
>>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
>>>>>
>>>>> So you rely on the isc format list for the scaler as well ?
>>>>> I think it's fine as long as they are identical
>>>>
>>>> Yes, the scaler is kind of a dummy entity , but it's required by the
>>>> media controller validation of the pipeline.
>>>>
>>>
>>> More on this later
>>>
>>>>>
>>>>>> +
>>>>>> + if (!fmt)
>>>>>> + fmt = &isc->formats_list[0];
>>>>>> +
>>>>>> + req_fmt->format.code = fmt->mbus_code;
>>>>>> +
>>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>> + req_fmt->pad);
>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>> + /* Trying on the pad sink makes the source sink change too */
>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
>>>>>> + v4l2_try_fmt =
>>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>> +
>>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
>>>>>> + 16, isc->max_width, 0,
>>>>>> + &v4l2_try_fmt->height,
>>>>>> + 16, isc->max_height, 0, 0);
>>>>>> + }
>>>>>> + /* if we are just trying, we are done */
>>>>>> + return 0;
>>>>>> + }
>>>>>> +
>>>>>> + isc->scaler_format = req_fmt->format;
>>>>>
>>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
>>>>>
>>>>> Actually, I would like to know more about the scaler device
>>>>> capabilities. What functions can this IP perform ? Does it do
>>>>> cropping, can it also do (down)scaling or even composition ?
>>>>>
>>>>> I think it is worth to read
>>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
>>>>> where it is reported how cropping and scaling are implemented by using
>>>>> the selection API.
>>>>>
>>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
>>>>> for which you need to implement support to by adding s_selection on
>>>>> the sink pad TGT_CROP target.
>>>>
>>>> The scaler is kind of a dummy entity.
>>>> The problem is that different subdevices cannot be connected directly to
>>>> a v4l entity, because the ISC does cropping on the incoming frame.
>>>> The ISC has a limit on the total number of pixels per frame, thus it
>>>> will do a crop.
>>>
>>> What is this limit related to ? I don't know the ISC internals, but is
>>> the limitation due to the overall image size, or maybe it's due to
>>> some line buffers being limited in the size they can transfer to
>>> memory ? Is it a limitation on the size of the image in pixels, or is
>>> it due to a limitation of the processing bandwidth on the input
>>> interface and thus depends on the actual size of the image in bytes ?
>>
>> Limitation is both on line size and full image size. From my
>> understanding, is that the internal SRAM is limited (8 Mpixels with a
>> small safe buffer on top in the case of sama7g5 )
>> This SRAM could be used as 800000x10 for example, or 3200x2464 for
>> another example. However, I am not sure if a real line of 800,000 pixels
>> make sense, or if there is another internal mechanism in the ISC that
>> stops it.
>> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
>> the maximum frame that it can handle
>>
>>>
>>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
>>>> only handle 3264x2464 , then we have two choices:
>>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
>>>> 3280x2464, and the whole pipeline fails to configure, there is a
>>>> mismatch between /dev/video output and what the whole pipeline produces
>>>
>>> I'm confused in first place. If the ISC cannot output anything larger
>>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
>>> this should be adjusted to 3264x2464.
>>
>> yes, that is true. However, without an entity that says 'I am doing
>> cropping', the media pipeline will fail, because previous entity will
>> output 3280x2464 .
>> The conclusion to create a scaler entitity was done during my last year
>> discussions on the IRC. I could not find a way to perform this cropping
>> at the v4l entity side, and one reason is that the v4l entity does not
>> support what I needed , or at least that was my understanding.
>> You claim that the v4l entity could perform the cropping in the same
>> entity ?
>
> I thought so, but I haven't tried to be honest
>
>>
>> Let's assume for a minute that it could do this. Even so, I would prefer
>> to have a separate scaler entity to do this , for several reasons:
>> -> I would like to be able to crop the frame arbitrarily if I want to,
>> and the scaler would allow me to do this easier
>> -> it would be more obvious that the frame is cropped by having this
>> entity there
>> -> in fact some of the ISC versions really have a down scaler, that I
>> have not implemented yet. So it would be useful to already have this in
>> place to be able to scale down with it at a later time.
>
> I think this last reason is enough to have an entity plumbed in
> already
>
>>
>>>
>>> Or is the ISC limitation on the -input- side ?
>>>
>>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
>>>> the ISC can handle at most -> cannot , because the sensor let's say can
>>>> *only* produce 3280x2464 and *any* other frame size returns an error.
>>>
>>> buggy sensor I would say :)
>>
>> I cannot tell this to the customers, and I cannot fail the whole
>> pipeline because the sensor does not crop 16 pixels, when in fact I can
>> do this very easily on the ISC side.
>>
>
> Fair enough :)
>
>>>
>>>>
>>>> The only solution to make both worlds happy, is to have a dummy entity
>>>> called 'scaler' which in fact now it only performs a simple crop.
>>>> It accepts any frame size at input (hence the 10000x10000 which you saw
>>>> earlier), and outputs at most 3264x2464 the isc_max_height and
>>>> isc_max_width.
>>>
>>> I have a maybe dumb question: can the cropping operation be modeled
>>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
>>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
>>
>> I am not sure. This is what I wanted initially. But then I thought about
>> the three reasons stated above for the use of the scaler entity, and
>> discussed with folks on the IRC, and come up with the solution for the
>> scaler entity
>>
>>>
>>>>
>>>> So to answer your question, the isc scaler is a software model for a
>>>> simple cropping, that would make media controller happy, and capture
>>>> software happy.
>>>>
>>>> Here it how it looks :
>>>>
>>>> Device topology
>>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
>>>> type V4L2 subdev subtype Unknown flags 0
>>>> device node name /dev/v4l-subdev0
>>>> pad0: Sink
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>> crop.bounds:(0,0)/3264x2464
>>>> crop:(0,0)/3264x2464]
>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>> pad1: Source
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>
>>>> - entity 4: csi2dc (2 pads, 2 links)
>>>> type V4L2 subdev subtype Unknown flags 0
>>>> device node name /dev/v4l-subdev1
>>>> pad0: Sink
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>> <- "dw-csi.0":1 [ENABLED]
>>>> pad1: Source
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
>>>>
>>>> - entity 7: dw-csi.0 (2 pads, 2 links)
>>>> type V4L2 subdev subtype Unknown flags 0
>>>> device node name /dev/v4l-subdev2
>>>> pad0: Sink
>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>> <- "imx219 1-0010":0 [ENABLED]
>>>> pad1: Source
>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>> -> "csi2dc":0 [ENABLED]
>>>>
>>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
>>>> type V4L2 subdev subtype Sensor flags 0
>>>> device node name /dev/v4l-subdev3
>>>> pad0: Source
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>> xfer:srgb ycbcr:601 quantization:full-range
>>>> crop.bounds:(8,8)/3280x2464
>>>> crop:(8,8)/3280x2464]
>>>> -> "dw-csi.0":0 [ENABLED]
>>>>
>>>> - entity 24: atmel_isc_common (1 pad, 1 link)
>>>> type Node subtype V4L flags 1
>>>> device node name /dev/video0
>>>> pad0: Sink
>>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
>>>>
>>>>
>>>> Scaler does this one cute little thing :
>>>>
>>>> pad0: Sink
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>> crop.bounds:(0,0)/3264x2464
>>>> crop:(0,0)/3264x2464]
>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>> pad1: Source
>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>
>>> Shouldn't this be 3264x2464 as that's what the entity outputs after
>>> the cropping ? And shouldn't then be 3264x2464 the size output from
>>> the video device too ?
>>
>> That's right.
>> I don't know why the source format for the scaler is still 3280x2464.
>> Maybe there is a bug on g_fmt for it.. have to check it.
>
> Thanks, I think this should be fixed ten
>
>
>>
>> Anyway, the video format is like this :
>>
>> # v4l2-ctl --get-fmt-video
>> Format Video Capture:
>> Width/Height : 3264/2464
>> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
>> Field : None
>> Bytes per Line : 6528
>> Size Image : 16084992
>> Colorspace : sRGB
>> Transfer Function : Default (maps to sRGB)
>> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
>> Quantization : Default (maps to Full Range)
>> Flags :
>> #
>>
>> and initializing the pipeline looks fine from user perspective.
>> The scaler is in fact the solution to make this pipeline work with
>> libcamera.
>> I found this cropping to be an issue in media controller when trying it
>> with libcamera. Otherwise, the other user space apps which I was using
>> never complained that anything was wrong
>> Libcamera simply refuses to acknowledge the pipeline if the video output
>> is 3264x2464 but there is no entity that changes the format from
>> 3280x2464 down
>
> Not sure why libcamera should play a role there. Isn't it the media
> pipeline validation that complains and returns an -EPIPE ?

I tried this before the media_pipeline_start .
Perhaps libcamera was doing some similar checks, anyway, it was
complaining and not adding the stream, and reported no usable streams/camera

Once I will rework some of the things according to your review I will
run libcamera again to see what it complains about

>
>>
>>>
>>>
>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>
>>>>
>>>> Which is what we needed.
>>>>
>>>>>
>>>>>> +
>>>>>> + return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>> + struct v4l2_subdev_mbus_code_enum *code)
>>>>>> +{
>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>> + int supported_index = 0;
>>>>>> + int i;
>>>>>> +
>>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
>>>>>> + if (!isc->formats_list[i].sd_support)
>>>>>> + continue;
>>>>>
>>>>> The sd_support flag still doesn't click in my head.
>>>>>
>>>>> Shouldn't the enumeration of available formats on the scaler do not
>>>>> depend on the sensor supproted formats ?
>>>>
>>>> You're right. I will have to check it again.
>>>>
>>>>>
>>>>>> + if (supported_index == code->index) {
>>>>>> + code->code = isc->formats_list[i].mbus_code;
>>>>>> + return 0;
>>>>>> + }
>>>>>> + supported_index++;
>>>>>> + }
>>>>>> +
>>>>>> + return -EINVAL;
>>>>>> +}
>>>>>> +
>>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>> + struct v4l2_subdev_selection *sel)
>>>>>> +{
>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>> +
>>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
>>>>>> + return -EINVAL;
>>>>>> +
>>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
>>>>>> + sel->target != V4L2_SEL_TGT_CROP)
>>>>>> + return -EINVAL;
>>>>>> +
>>>>>> + sel->r.height = isc->max_height;
>>>>>> + sel->r.width = isc->max_width;
>>>>>
>>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
>>>>> as it represents the maximum valid crop rectangle.
>>>>>
>>>>> TGT_CROP should report the configured crop rectangle which can be
>>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
>>>>> the spec is correct
>>>>
>>>> So you would like to have this differentiated, and report the
>>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
>>>> reported now, the maximum size of the ISC frame .
>>>> My understanding is correct ?
>>>>
>>>
>>> I didn't know you have an HW limitation, so your _BOUNDS is not the
>>> input image size but rather 3264x2464 ( == max_width x max_height).
>>>
>>> What I meant is that _BOUNDS should report the maximum rectangle size
>>> that can be applied to the _CROP target. In you case you have an HW
>>> limitation 3264x2464 and that's the largest rectangle you can apply.
>> So the CROP should be at 3264x2464
>>> TGT_CROP can be initialized to the same as _BOUND, but if you
>>> implement s_selection it should report what has been there applied.
>> and BOUND to actual frame size ?
>>> But as you don't implement s_selection yet, I think this is fine for
>>> now. Maybe a little comment ?
>>
>> It could also be the case where e.g. the sensor is outputting 1920x1080,
>> in this case the scaler would do nothing.
>> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
>> If the sensor is outputting 1920x1080, this format comes directly from
>> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
>> sensor's problem)
>>>
>>> Also, is set->r zeroed by the framework before getting here ?
>>> Otherwise you should set r.left and r.top to 0
>>
>> If these are redundant, no problem to remove them
>
> I was actually confused. I was suggesting you to add...
>
>>>
>>>>>
>>>>>> +
>>>>>> + sel->r.left = 0;
>>>>>> + sel->r.top = 0;
>
> These ^
>
> So nothing to change. Sorry I've missed them
>
> Thanks
> j
>
>
>>>>>> +
>>>>>> + return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
>>>>>> + struct v4l2_subdev_state *sd_state)
>>>>>> +{
>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
>>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>> +
>>>>>> + *v4l2_try_fmt = isc->scaler_format;
>>>>>> +
>>>>>> + return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
>>>>>> + .set_fmt = isc_scaler_set_fmt,
>>>>>> + .get_fmt = isc_scaler_get_fmt,
>>>>>> + .get_selection = isc_scaler_g_sel,
>>>>>> + .init_cfg = isc_scaler_init_cfg,
>>>>>
>>>>> .link_validate = v4l2_subdev_link_validate_default,
>>>>>
>>>>> To have the formats at the end of links that point to this entity
>>>>> validated (I think the framework already calls it if not set though,
>>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
>>>>>
>>>>>> +};
>>>>>> +
>>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>>>>>> + .pad = &isc_scaler_pad_ops,
>>>>>> +};
>>>>>> +
>>>>>> +int isc_scaler_init(struct isc_device *isc)
>>>>>> +{
>>>>>> + int ret;
>>>>>> +
>>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
>>>>>> +
>>>>>> + isc->scaler_sd.owner = THIS_MODULE;
>>>>>> + isc->scaler_sd.dev = isc->dev;
>>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
>>>>>> + "atmel_isc_scaler");
>>>>>
>>>>> I would drop 'atmel' for brevity, unless other entities have this
>>>>> prefix set already
>>>>
>>>> The v4l entity takes it's name from the module name, which is
>>>> atmel_isc_common, so I thought to keep the prefix
>>>>
>>>
>>> Ack!
>>>
>>>>>
>>>>>> +
>>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>>>>>> +
>>>>>> + isc->scaler_format.height = isc->max_height;
>>>>>> + isc->scaler_format.width = isc->max_width;
>>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
>>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
>>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>> +
>>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
>>>>>> + ISC_SCALER_PADS_NUM,
>>>>>> + isc->scaler_pads);
>>>>>> + if (ret < 0) {
>>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
>>>>>> + return ret;
>>>>>> + }
>>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
>>>>>> + if (ret < 0) {
>>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
>>>>>> + return ret;
>>>>>> + }
>>>>>> +
>>>>>> + return ret;
>>>>>> +}
>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
>>>>>> +
>>>>>> +int isc_scaler_link(struct isc_device *isc)
>>>>>> +{
>>>>>> + int ret;
>>>>>> +
>>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
>>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
>>>>>> + ISC_SCALER_PAD_SINK,
>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>> +
>>>>>> + if (ret < 0) {
>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>> + isc->current_subdev->sd->entity.name,
>>>>>> + isc->scaler_sd.entity.name);
>>>>>> + return ret;
>>>>>> + }
>>>>>> +
>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
>>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
>>>>>> +
>>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
>>>>>> + ISC_SCALER_PAD_SOURCE,
>>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>> +
>>>>>> + if (ret < 0) {
>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>> + isc->scaler_sd.entity.name,
>>>>>> + isc->video_dev.entity.name);
>>>>>> + return ret;
>>>>>> + }
>>>>>> +
>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>> +
>>>>>> + return ret;
>>>>>> +}
>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
>>>>>
>>>>> From the DT graph point of view, the ISC appears as a single block
>>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
>>>>> add to its notifier the image sensor remote subdevice.
>>>>
>>>> Actually it's only a parallel port.
>>>> And it can be connected either to a sensor directly or to the csi2dc
>>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
>>>>
>>>>>
>>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
>>>>> entity which was initialized at isc_probe() time is linked in between
>>>>> the ISC and the image sensor immutably.
>>>>
>>>> yes, because this cannot change at runtime, and usually, it can't change
>>>> without altering the board hardware. (at least this is what my
>>>> understanding of immutability is )
>>>>
>>>>>
>>>>> I think it is fine for now, but I wonder if you plan to plumb more
>>>>> components between the ISC video node and the sensor, if it's not
>>>>> worth changing the DT bindings and their parsing logic to separate the
>>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
>>>>> time and have the CSI-2 receiver entity as downstream remote. I think
>>>>> I need to get to know the ISC better to really have an idea. For now,
>>>>> this seems ok to me, but please check with maintainers if this is fine
>>>>> with them.
>>>>
>>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
>>>> the remote subdev).
>>>> The ISC will register a scaler, and connect the subdev to this scaler
>>>> first, and then, the scaler to the isc itself (the v4l entity).
>>>>
>>>>>
>>>>>> +
>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>>>>>> index 5fbf52a9080b..c9234c90ae58 100644
>>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
>>>>>> u32 his_entry;
>>>>>> };
>>>>>>
>>>>>> +enum isc_mc_pads {
>>>>>> + ISC_PAD_SINK = 0,
>>>>>> + ISC_PADS_NUM = 1,
>>>>>> +};
>>>>>> +
>>>>>> +enum isc_scaler_pads {
>>>>>> + ISC_SCALER_PAD_SINK = 0,
>>>>>> + ISC_SCALER_PAD_SOURCE = 1,
>>>>>> + ISC_SCALER_PADS_NUM = 2,
>>>>>> +};
>>>>>> +
>>>>>> /*
>>>>>> * struct isc_device - ISC device driver data/config struct
>>>>>> * @regmap: Register map
>>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
>>>>>> * be used as an input to the controller
>>>>>> * @controller_formats_size: size of controller_formats array
>>>>>> * @formats_list_size: size of formats_list array
>>>>>> + * @pads: media controller pads for isc video entity
>>>>>> + * @mdev: media device that is registered by the isc
>>>>>> + * @remote_pad: remote pad on the connected subdevice
>>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
>>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
>>>>>> + * @scaler_format: current format for the scaler subdevice
>>>>>> */
>>>>>> struct isc_device {
>>>>>> struct regmap *regmap;
>>>>>> @@ -344,6 +361,19 @@ struct isc_device {
>>>>>> struct isc_format *formats_list;
>>>>>> u32 controller_formats_size;
>>>>>> u32 formats_list_size;
>>>>>> +
>>>>>> + struct {
>>>>>> + struct media_pad pads[ISC_PADS_NUM];
>>>>>> + struct media_device mdev;
>>>>>> +
>>>>>> + u32 remote_pad;
>>>>>> + };
>>>>>> +
>>>>>> + struct {
>>>>>> + struct v4l2_subdev scaler_sd;
>>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
>>>>>> + struct v4l2_mbus_framefmt scaler_format;
>>>>
>>>> Here are the scaler stuff which I added, in the same bulk struct for the
>>>> whole isc device
>>>>
>>>>
>>>>>> + };
>>>>>> };
>>>>>>
>>>>>> extern const struct regmap_config isc_regmap_config;
>>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
>>>>>> void isc_subdev_cleanup(struct isc_device *isc);
>>>>>> void isc_clk_cleanup(struct isc_device *isc);
>>>>>>
>>>>>> +int isc_scaler_link(struct isc_device *isc);
>>>>>> +int isc_scaler_init(struct isc_device *isc);
>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
>>>>>> +void isc_mc_cleanup(struct isc_device *isc);
>>>>>> +
>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>> + unsigned int code, int *index);
>>>>>> #endif
>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>> index c5b9563e36cb..c244682ea22f 100644
>>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>> break;
>>>>>> }
>>>>>>
>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>
>>>>> I am surprised you can read a register before runtime_pm is
>>>>> intialized!
>>>>>
>>>>
>>>> Actually I can read any register after the moment of starting the clock
>>>> on the hardware block.
>>>
>>> Ah right then
>>>
>>>> Maybe the starting and stopping of the clock needs to be moved to the
>>>> runtime_pm calls, but this is another story, not related to the media
>>>> controller.
>>>
>>> It's fine, but maybe worth recording with a todo ?
>>>
>>> Thanks
>>> j
>>>
>>>> I moved this line because I had to pass the version to the isc_mc_init call
>>>>
>>>>> Thanks
>>>>> j
>>>>
>>>> Thanks for reviewing !
>>>> Eugen
>>>>>
>>>>>
>>>>>> +
>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>> + if (ret < 0)
>>>>>> + goto isc_probe_mc_init_err;
>>>>>> +
>>>>>> pm_runtime_set_active(dev);
>>>>>> pm_runtime_enable(dev);
>>>>>> pm_request_idle(dev);
>>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>> ret = clk_prepare_enable(isc->ispck);
>>>>>> if (ret) {
>>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
>>>>>> - goto cleanup_subdev;
>>>>>> + goto isc_probe_mc_init_err;
>>>>>> }
>>>>>>
>>>>>> /* ispck should be greater or equal to hclock */
>>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>> goto unprepare_clk;
>>>>>> }
>>>>>>
>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
>>>>>>
>>>>>> return 0;
>>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>> unprepare_clk:
>>>>>> clk_disable_unprepare(isc->ispck);
>>>>>>
>>>>>> +isc_probe_mc_init_err:
>>>>>> + isc_mc_cleanup(isc);
>>>>>> +
>>>>>> cleanup_subdev:
>>>>>> isc_subdev_cleanup(isc);
>>>>>>
>>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>>>>>>
>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>
>>>>>> + isc_mc_cleanup(isc);
>>>>>> +
>>>>>> isc_subdev_cleanup(isc);
>>>>>>
>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>> index 07a80b08bc54..9dc75eed0098 100644
>>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
>>>>>> break;
>>>>>> }
>>>>>>
>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>> +
>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>> + if (ret < 0)
>>>>>> + goto isc_probe_mc_init_err;
>>>>>> +
>>>>>> pm_runtime_set_active(dev);
>>>>>> pm_runtime_enable(dev);
>>>>>> pm_request_idle(dev);
>>>>>>
>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
>>>>>>
>>>>>> return 0;
>>>>>>
>>>>>> +isc_probe_mc_init_err:
>>>>>> + isc_mc_cleanup(isc);
>>>>>> +
>>>>>> cleanup_subdev:
>>>>>> isc_subdev_cleanup(isc);
>>>>>>
>>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>>>>>>
>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>
>>>>>> + isc_mc_cleanup(isc);
>>>>>> +
>>>>>> isc_subdev_cleanup(isc);
>>>>>>
>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>> --
>>>>>> 2.25.1
>>>>>>
>>>>
>>

2022-02-10 18:44:51

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 07/11] media: atmel: atmel-isc: change format propagation to subdev into only verification

Hi Eugen

On Fri, Jan 21, 2022 at 03:14:12PM +0200, Eugen Hristev wrote:
> As a top MC video driver, the atmel-isc should not propagate the format to the
> subdevice.
> It should rather check at start_streaming() time if the subdev is properly
> configured with a compatible format.
> Removed the whole format finding logic, and reworked the format verification
> at start_streaming time, such that the ISC will return an error if the subdevice
> is not properly configured. To achieve this, media_pipeline_start
> is called and a link_validate callback is created to check the formats.
> With this being done, the module parameter 'sensor_preferred' makes no sense
> anymore. The ISC should not decide which format the sensor is using. The
> ISC should only cope with the situation and inform userspace if the streaming
> is possible in the current configuration.

This sounds right!

> The redesign of the format propagation has also risen the question of the
> enumfmt callback. If enumfmt is called with an mbus_code, the enumfmt handler
> should only return the formats that are supported for this mbus_code.
> To make it more easy to understand the formats, changed the report order
> to report first the native formats, and after that the formats that the ISC
> can convert to.
>
> Signed-off-by: Eugen Hristev <[email protected]>
> ---
> Changes in v4:
> - moved validation code into link_validate and used media_pipeline_start
> - merged this patch with the enum_fmt patch which was previously in v3 of
> the series
>
> Changes in v3:
> - clamp to maximum resolution once the frame size from the subdev is found
>
> drivers/media/platform/atmel/atmel-isc-base.c | 335 ++++++++++--------
> .../media/platform/atmel/atmel-isc-scaler.c | 5 +
> drivers/media/platform/atmel/atmel-isc.h | 3 +
> 3 files changed, 191 insertions(+), 152 deletions(-)
>
> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> index f7a88399bd54..31c79313aadc 100644
> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> @@ -36,11 +36,6 @@ static unsigned int debug;
> module_param(debug, int, 0644);
> MODULE_PARM_DESC(debug, "debug level (0-2)");
>
> -static unsigned int sensor_preferred = 1;
> -module_param(sensor_preferred, uint, 0644);
> -MODULE_PARM_DESC(sensor_preferred,
> - "Sensor is preferred to output the specified format (1-on 0-off), default 1");
> -
> #define ISC_IS_FORMAT_RAW(mbus_code) \
> (((mbus_code) & 0xf000) == 0x3000)
>
> @@ -337,6 +332,10 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
> unsigned long flags;
> int ret;
>
> + ret = media_pipeline_start(&isc->video_dev.entity, &isc->mpipe);
> + if (ret)
> + goto err_pipeline_start;
> +
> /* Enable stream on the sub device */
> ret = v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 1);
> if (ret && ret != -ENOIOCTLCMD) {
> @@ -384,6 +383,9 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
> v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0);
>
> err_start_stream:
> + media_pipeline_stop(&isc->video_dev.entity);
> +
> +err_pipeline_start:
> spin_lock_irqsave(&isc->dma_queue_lock, flags);
> list_for_each_entry(buf, &isc->dma_queue, list)
> vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_QUEUED);
> @@ -420,6 +422,9 @@ static void isc_stop_streaming(struct vb2_queue *vq)
> if (ret && ret != -ENOIOCTLCMD)
> v4l2_err(&isc->v4l2_dev, "stream off failed in subdev\n");
>
> + /* Stop media pipeline */
> + media_pipeline_stop(&isc->video_dev.entity);
> +
> /* Release all active buffers */
> spin_lock_irqsave(&isc->dma_queue_lock, flags);
> if (unlikely(isc->cur_frm)) {
> @@ -496,21 +501,56 @@ static int isc_enum_fmt_vid_cap(struct file *file, void *priv,
> u32 index = f->index;
> u32 i, supported_index;
>
> - if (index < isc->controller_formats_size) {
> - f->pixelformat = isc->controller_formats[index].fourcc;
> - return 0;
> + supported_index = 0;
> +
> + for (i = 0; i < isc->formats_list_size; i++) {
> + if (!isc->formats_list[i].sd_support)
> + continue;

Sorry to ask again about sd_support, but if the ISC formats
enumeration is not constrained to a specific mbus_code, shouldn't all
the formats be enumerated regardless if they're supported from the
sensor or not ?

> + /*
> + * If specific mbus_code is requested, provide only
> + * supported formats with this mbus code
> + */
> + if (f->mbus_code && f->mbus_code !=
> + isc->formats_list[i].mbus_code)

nit: for clarity
if (f->mbus_code &&
f->mbus_code != isc->formats_list[i].mbus_code)

> + continue;
> + if (supported_index == index) {
> + f->pixelformat = isc->formats_list[i].fourcc;
> + return 0;
> + }
> + supported_index++;
> }
>
> - index -= isc->controller_formats_size;
> + /*
> + * If the sensor does not support this mbus_code whatsoever,
> + * there is no reason to advertise any of our output formats
> + */
> + if (supported_index == 0)
> + return -EINVAL;
>
> + /*
> + * If the sensor uses a format that is not raw, then we cannot
> + * convert it to any of the formats that we usually can with a
> + * RAW sensor. Thus, do not advertise them.
> + */
> + if (isc->config.sd_format &&
> + !ISC_IS_FORMAT_RAW(isc->config.sd_format->mbus_code))
> + return -EINVAL;
> +
> + /*
> + * Iterate again through the formats that we can convert to.
> + * However, to avoid duplicates, skip the formats that
> + * the sensor already supports directly
> + */
> + index -= supported_index;
> supported_index = 0;
>
> - for (i = 0; i < isc->formats_list_size; i++) {
> - if (!ISC_IS_FORMAT_RAW(isc->formats_list[i].mbus_code) ||
> - !isc->formats_list[i].sd_support)
> + for (i = 0; i < isc->controller_formats_size; i++) {
> + /* if this format is already supported by sensor, skip it */
> + if (find_format_by_fourcc(isc, isc->controller_formats[i].fourcc))
> continue;
> if (supported_index == index) {
> - f->pixelformat = isc->formats_list[i].fourcc;
> + f->pixelformat =
> + isc->controller_formats[i].fourcc;

I trust this last part to work correctly, as it's not 100% clear to me :)

> return 0;
> }
> supported_index++;
> @@ -581,20 +621,30 @@ static int isc_try_validate_formats(struct isc_device *isc)
> break;
> default:
> /* any other different formats are not supported */
> + v4l2_err(&isc->v4l2_dev, "Requested unsupported format.\n");
> ret = -EINVAL;
> }
> v4l2_dbg(1, debug, &isc->v4l2_dev,
> "Format validation, requested rgb=%u, yuv=%u, grey=%u, bayer=%u\n",
> rgb, yuv, grey, bayer);
>
> - /* we cannot output RAW if we do not receive RAW */
> - if ((bayer) && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
> + if ((bayer) &&

just 'bayer' ?

> + !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code)) {
> + v4l2_err(&isc->v4l2_dev, "Cannot output RAW if we do not receive RAW.\n");
> return -EINVAL;
> + }
>
> - /* we cannot output GREY if we do not receive RAW/GREY */
> if (grey && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code) &&
> - !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code))
> + !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
> + v4l2_err(&isc->v4l2_dev, "Cannot output GREY if we do not receive RAW/GREY.\n");
> + return -EINVAL;
> + }
> +
> + if ((rgb || bayer || yuv) &&
> + ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
> + v4l2_err(&isc->v4l2_dev, "Cannot convert GREY to another format.\n");
> return -EINVAL;
> + }
>
> return ret;
> }
> @@ -822,7 +872,7 @@ static void isc_try_fse(struct isc_device *isc,
> * If we do not know yet which format the subdev is using, we cannot
> * do anything.
> */
> - if (!isc->try_config.sd_format)
> + if (!isc->config.sd_format)
> return;
>
> fse.code = isc->try_config.sd_format->mbus_code;
> @@ -843,180 +893,141 @@ static void isc_try_fse(struct isc_device *isc,
> }
> }
>
> -static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f,
> - u32 *code)
> +static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f)
> {
> - int i;
> - struct isc_format *sd_fmt = NULL, *direct_fmt = NULL;
> struct v4l2_pix_format *pixfmt = &f->fmt.pix;
> - struct v4l2_subdev_pad_config pad_cfg = {};
> - struct v4l2_subdev_state pad_state = {
> - .pads = &pad_cfg
> - };
> - struct v4l2_subdev_format format = {
> - .which = V4L2_SUBDEV_FORMAT_TRY,
> - };
> - u32 mbus_code;
> - int ret;
> - bool rlp_dma_direct_dump = false;
> + unsigned int i;
>
> if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
> return -EINVAL;
>
> - /* Step 1: find a RAW format that is supported */
> - for (i = 0; i < isc->num_user_formats; i++) {
> - if (ISC_IS_FORMAT_RAW(isc->user_formats[i]->mbus_code)) {
> - sd_fmt = isc->user_formats[i];
> + isc->try_config.fourcc = isc->user_formats[0]->fourcc;
> +
> + /* find if the format requested is supported */
> + for (i = 0; i < isc->controller_formats_size; i++)
> + if (isc->controller_formats[i].fourcc == pixfmt->pixelformat) {
> + isc->try_config.fourcc = pixfmt->pixelformat;
> break;
> }
> - }
> - /* Step 2: We can continue with this RAW format, or we can look
> - * for better: maybe sensor supports directly what we need.
> - */
> - direct_fmt = find_format_by_fourcc(isc, pixfmt->pixelformat);
> -
> - /* Step 3: We have both. We decide given the module parameter which
> - * one to use.
> - */
> - if (direct_fmt && sd_fmt && sensor_preferred)
> - sd_fmt = direct_fmt;
> -
> - /* Step 4: we do not have RAW but we have a direct format. Use it. */
> - if (direct_fmt && !sd_fmt)
> - sd_fmt = direct_fmt;
> -
> - /* Step 5: if we are using a direct format, we need to package
> - * everything as 8 bit data and just dump it
> - */
> - if (sd_fmt == direct_fmt)
> - rlp_dma_direct_dump = true;
> -
> - /* Step 6: We have no format. This can happen if the userspace
> - * requests some weird/invalid format.
> - * In this case, default to whatever we have
> - */
> - if (!sd_fmt && !direct_fmt) {
> - sd_fmt = isc->user_formats[isc->num_user_formats - 1];
> - v4l2_dbg(1, debug, &isc->v4l2_dev,
> - "Sensor not supporting %.4s, using %.4s\n",
> - (char *)&pixfmt->pixelformat, (char *)&sd_fmt->fourcc);
> - }
> -
> - if (!sd_fmt) {
> - ret = -EINVAL;
> - goto isc_try_fmt_err;
> - }
> -
> - /* Step 7: Print out what we decided for debugging */
> - v4l2_dbg(1, debug, &isc->v4l2_dev,
> - "Preferring to have sensor using format %.4s\n",
> - (char *)&sd_fmt->fourcc);
> -
> - /* Step 8: at this moment we decided which format the subdev will use */
> - isc->try_config.sd_format = sd_fmt;
> -
> - /* Limit to Atmel ISC hardware capabilities */
> - if (pixfmt->width > isc->max_width)
> - pixfmt->width = isc->max_width;
> - if (pixfmt->height > isc->max_height)
> - pixfmt->height = isc->max_height;
> -
> - /*
> - * The mbus format is the one the subdev outputs.
> - * The pixels will be transferred in this format Sensor -> ISC
> - */
> - mbus_code = sd_fmt->mbus_code;
> -
> - /*
> - * Validate formats. If the required format is not OK, default to raw.
> - */
> -
> - isc->try_config.fourcc = pixfmt->pixelformat;
> -
> - if (isc_try_validate_formats(isc)) {
> - pixfmt->pixelformat = isc->try_config.fourcc = sd_fmt->fourcc;
> - /* Re-try to validate the new format */
> - ret = isc_try_validate_formats(isc);
> - if (ret)
> - goto isc_try_fmt_err;
> - }
> -
> - ret = isc_try_configure_rlp_dma(isc, rlp_dma_direct_dump);
> - if (ret)
> - goto isc_try_fmt_err;
> -
> - ret = isc_try_configure_pipeline(isc);
> - if (ret)
> - goto isc_try_fmt_err;
>
> - /* Obtain frame sizes if possible to have crop requirements ready */
> - isc_try_fse(isc, &pad_state);
> -
> - v4l2_fill_mbus_format(&format.format, pixfmt, mbus_code);
> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad, set_fmt,
> - &pad_state, &format);
> - if (ret < 0)
> - goto isc_try_fmt_subdev_err;
> + /* If we did not find the requested format, we will fallback here */
> + pixfmt->pixelformat = isc->try_config.fourcc;
> + pixfmt->colorspace = V4L2_COLORSPACE_SRGB;
> + pixfmt->field = V4L2_FIELD_NONE;
>
> - v4l2_fill_pix_format(pixfmt, &format.format);
> + isc_try_configure_rlp_dma(isc, false);
>
> /* Limit to Atmel ISC hardware capabilities */
> - if (pixfmt->width > isc->max_width)
> - pixfmt->width = isc->max_width;
> - if (pixfmt->height > isc->max_height)
> - pixfmt->height = isc->max_height;
> + v4l_bound_align_image(&pixfmt->width, 16, isc->max_width, 0,
> + &pixfmt->height, 16, isc->max_height, 0, 0);
>
> pixfmt->field = V4L2_FIELD_NONE;
> pixfmt->bytesperline = (pixfmt->width * isc->try_config.bpp_v4l2) >> 3;
> pixfmt->sizeimage = ((pixfmt->width * isc->try_config.bpp) >> 3) *
> pixfmt->height;

Seems pixfmt->field is set twice.

Could you move the assignment of all pixfmt to a single block ?

>
> - if (code)
> - *code = mbus_code;
> + isc->try_fmt = *f;
>
> return 0;
> +}
>
> -isc_try_fmt_err:
> - v4l2_err(&isc->v4l2_dev, "Could not find any possible format for a working pipeline\n");
> -isc_try_fmt_subdev_err:
> - memset(&isc->try_config, 0, sizeof(isc->try_config));
> +static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
> +{
> + isc_try_fmt(isc, f);
>
> - return ret;
> + /* make the try configuration active */
> + isc->config = isc->try_config;
> + isc->fmt = isc->try_fmt;
> +
> + v4l2_dbg(1, debug, &isc->v4l2_dev, "ISC set_fmt to %.4s @%dx%d\n",
> + (char *)&f->fmt.pix.pixelformat,
> + f->fmt.pix.width, f->fmt.pix.height);
> +
> + return 0;
> }
>
> -static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
> +static int isc_validate(struct isc_device *isc)
> {
> + int ret;
> + int i;
> + struct isc_format *sd_fmt = NULL;
> + struct v4l2_pix_format *pixfmt = &isc->fmt.fmt.pix;
> struct v4l2_subdev_format format = {
> .which = V4L2_SUBDEV_FORMAT_ACTIVE,
> + .pad = isc->remote_pad,
> + };
> + struct v4l2_subdev_pad_config pad_cfg = {};
> + struct v4l2_subdev_state pad_state = {
> + .pads = &pad_cfg,
> };
> - u32 mbus_code = 0;
> - int ret;
>
> - ret = isc_try_fmt(isc, f, &mbus_code);
> + /* Get current format from subdev */
> + ret = v4l2_subdev_call(isc->current_subdev->sd, pad, get_fmt, NULL,
> + &format);
> if (ret)
> return ret;
>
> - v4l2_fill_mbus_format(&format.format, &f->fmt.pix, mbus_code);
> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad,
> - set_fmt, NULL, &format);
> - if (ret < 0)
> - return ret;
> + /* Identify the subdev's format configuration */
> + for (i = 0; i < isc->num_user_formats; i++)
> + if (isc->user_formats[i]->mbus_code == format.format.code) {
> + sd_fmt = isc->user_formats[i];
> + break;
> + }
> +
> + /* Check if the format is not supported */
> + if (!sd_fmt) {
> + v4l2_err(&isc->v4l2_dev,
> + "Current subdevice is streaming a media bus code that is not supported 0x%x\n",
> + format.format.code);
> + return -EPIPE;
> + }
> +
> + /* At this moment we know which format the subdev will use */
> + isc->try_config.sd_format = sd_fmt;
> +
> + /* If the sensor is not RAW, we can only do a direct dump */
> + if (!ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
> + isc_try_configure_rlp_dma(isc, true);
>
> /* Limit to Atmel ISC hardware capabilities */
> - if (f->fmt.pix.width > isc->max_width)
> - f->fmt.pix.width = isc->max_width;
> - if (f->fmt.pix.height > isc->max_height)
> - f->fmt.pix.height = isc->max_height;
> + v4l_bound_align_image(&format.format.width, 16, isc->max_width, 0,
> + &format.format.height, 16, isc->max_height, 0, 0);
> +
> + /* Check if the frame size is the same. Otherwise we may overflow */
> + if (pixfmt->height != format.format.height ||
> + pixfmt->width != format.format.width) {
> + v4l2_err(&isc->v4l2_dev,
> + "ISC not configured with the proper frame size: %dx%d\n",
> + format.format.width, format.format.height);
> + return -EPIPE;
> + }
>
> - isc->fmt = *f;
> + v4l2_dbg(1, debug, &isc->v4l2_dev,
> + "Identified subdev using format %.4s with %dx%d %d bpp\n",
> + (char *)&sd_fmt->fourcc, pixfmt->width, pixfmt->height,
> + isc->try_config.bpp);
>
> + /* Reset and restart AWB if the subdevice changed the format */
> if (isc->try_config.sd_format && isc->config.sd_format &&
> isc->try_config.sd_format != isc->config.sd_format) {
> isc->ctrls.hist_stat = HIST_INIT;
> isc_reset_awb_ctrls(isc);
> isc_update_v4l2_ctrls(isc);
> }
> - /* make the try configuration active */
> +
> + /* Validate formats */
> + ret = isc_try_validate_formats(isc);
> + if (ret)
> + return ret;
> +
> + /* Obtain frame sizes if possible to have crop requirements ready */
> + isc_try_fse(isc, &pad_state);
> +
> + /* Configure ISC pipeline for the config */
> + ret = isc_try_configure_pipeline(isc);
> + if (ret)
> + return ret;
> +
> isc->config = isc->try_config;
>
> v4l2_dbg(1, debug, &isc->v4l2_dev, "New ISC configuration in place\n");
> @@ -1040,7 +1051,7 @@ static int isc_try_fmt_vid_cap(struct file *file, void *priv,
> {
> struct isc_device *isc = video_drvdata(file);
>
> - return isc_try_fmt(isc, f, NULL);
> + return isc_try_fmt(isc, f);
> }
>
> static int isc_enum_input(struct file *file, void *priv,
> @@ -1841,7 +1852,7 @@ static int isc_set_default_fmt(struct isc_device *isc)
> };
> int ret;
>
> - ret = isc_try_fmt(isc, &f, NULL);
> + ret = isc_try_fmt(isc, &f);
> if (ret)
> return ret;
>
> @@ -2015,6 +2026,24 @@ int isc_pipeline_init(struct isc_device *isc)
> }
> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>
> +int isc_link_validate(struct media_link *link)
> +{
> + struct video_device *vdev =
> + media_entity_to_video_device(link->sink->entity);
> + struct isc_device *isc = video_get_drvdata(vdev);
> + int ret;
> +
> + ret = v4l2_subdev_link_validate(link);
> + if (ret)
> + return ret;

Doesn't v4l2_subdev_link_validate() call
v4l2_subdev_link_validate_default() if you don't define a .link_validate()
pad operation ? v4l2_subdev_link_validate_default() compares format
and sizes of both ends to the links and want them to be identical,
something that defeat the purpose of your custom isc_validate() ?

Thanks
j

> +
> + return isc_validate(isc);
> +}
> +
> +static const struct media_entity_operations isc_entity_operations = {
> + .link_validate = isc_link_validate,
> +};
> +
};
> int isc_mc_init(struct isc_device *isc, u32 ver)
> {
> const struct of_device_id *match;
> @@ -2022,6 +2051,8 @@ int isc_mc_init(struct isc_device *isc, u32 ver)
>
> isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> + isc->video_dev.entity.ops = &isc_entity_operations;
> +
> isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>
> ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> index ec95c9665883..93375a61aea6 100644
> --- a/drivers/media/platform/atmel/atmel-isc-scaler.c
> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> @@ -155,6 +155,10 @@ static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> .init_cfg = isc_scaler_init_cfg,
> };
>
> +static const struct media_entity_operations isc_scaler_entity_ops = {
> + .link_validate = v4l2_subdev_link_validate,
> +};
> +
> static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> .pad = &isc_scaler_pad_ops,
> };
> @@ -172,6 +176,7 @@ int isc_scaler_init(struct isc_device *isc)
>
> isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> + isc->scaler_sd.entity.ops = &isc_scaler_entity_ops;
> isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>
> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> index 04ea5e340808..e36cf9d3942a 100644
> --- a/drivers/media/platform/atmel/atmel-isc.h
> +++ b/drivers/media/platform/atmel/atmel-isc.h
> @@ -270,6 +270,7 @@ enum isc_scaler_pads {
> * @formats_list_size: size of formats_list array
> * @pads: media controller pads for isc video entity
> * @mdev: media device that is registered by the isc
> + * @mpipe: media device pipeline used by the isc
> * @remote_pad: remote pad on the connected subdevice
> * @scaler_sd: subdevice for the scaler that isc registers
> * @scaler_pads: media controller pads for the scaler subdevice
> @@ -295,6 +296,7 @@ struct isc_device {
> struct completion comp;
>
> struct v4l2_format fmt;
> + struct v4l2_format try_fmt;
> struct isc_format **user_formats;
> unsigned int num_user_formats;
>
> @@ -366,6 +368,7 @@ struct isc_device {
> struct {
> struct media_pad pads[ISC_PADS_NUM];
> struct media_device mdev;
> + struct media_pipeline mpipe;
>
> u32 remote_pad;
> };
> --
> 2.25.1
>

2022-02-11 02:55:25

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Wed, Feb 09, 2022 at 06:59:37AM +0000, [email protected] wrote:
> On 2/8/22 9:30 PM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
> >> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
> >>> Hi Eugen
> >>>
> >>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
> >>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> >>>>> Hi Eugen
> >>>>>
> >>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> >>>>>> Implement the support for media-controller.
> >>>>>> This means that the capabilities of the driver have changed and now
> >>>>>> it also advertises the IO_MC .
> >>>>>> The driver will register it's media device, and add the video entity to this
> >>>>>> media device. The subdevices are registered to the same media device.
> >>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> >>>>>> It will also register a subdevice that allows cropping of the incoming frame
> >>>>>> to the maximum frame size supported by the ISC.
> >>>>>> The ISC will create a link between the subdevice that is asynchronously
> >>>>>> registered and the atmel_isc_scaler entity.
> >>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> >>>>>> link.
> >>>>>>
> >>>>>> Signed-off-by: Eugen Hristev <[email protected]>
> >>>>>> ---
> >>>>>> Changes in v4:
> >>>>>> As suggested by Jacopo:
> >>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
> >>>>>> - moved init_mc/clean_mc to isc_base file
> >>>>>>
> >>>>>> Changes in v2:
> >>>>>> - implement try formats
> >>>>>>
> >>>>>> drivers/media/platform/atmel/Makefile | 2 +-
> >>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> >>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> >>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> >>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> >>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> >>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
> >>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>
> >>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> >>>>>> index 794e8f739287..f02d03df89d6 100644
> >>>>>> --- a/drivers/media/platform/atmel/Makefile
> >>>>>> +++ b/drivers/media/platform/atmel/Makefile
> >>>>>> @@ -1,7 +1,7 @@
> >>>>>> # SPDX-License-Identifier: GPL-2.0-only
> >>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
> >>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
> >>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> >>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
> >>>>>>
> >>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> >>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>> index 6b0005987a17..6b482270eb93 100644
> >>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>> struct isc_device, v4l2_dev);
> >>>>>> struct isc_subdev_entity *subdev_entity =
> >>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
> >>>>>> + int pad;
> >>>>>>
> >>>>>> if (video_is_registered(&isc->video_dev)) {
> >>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> >>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>>
> >>>>>> subdev_entity->sd = subdev;
> >>>>>>
> >>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> >>>>>> + MEDIA_PAD_FL_SOURCE);
> >>>>>> + if (pad < 0) {
> >>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> >>>>>> + subdev->name);
> >>>>>> + return pad;
> >>>>>> + }
> >>>>>> +
> >>>>>> + isc->remote_pad = pad;
> >>>>>> +
> >>>>>> return 0;
> >>>>>> }
> >>>>>>
> >>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> >>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> >>>>>> }
> >>>>>>
> >>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>> - unsigned int code, int *index)
> >>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>> + unsigned int code, int *index)
> >>>>>> {
> >>>>>> struct isc_format *fmt = &isc->formats_list[0];
> >>>>>> unsigned int i;
> >>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>>
> >>>>>> return NULL;
> >>>>>> }
> >>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
> >>>>>>
> >>>>>> static int isc_formats_init(struct isc_device *isc)
> >>>>>> {
> >>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> >>>>>> NULL, &mbus_code)) {
> >>>>>> mbus_code.index++;
> >>>>>>
> >>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> >>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> >>>>>> if (!fmt) {
> >>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> >>>>>> mbus_code.code);
> >>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>> vdev->queue = q;
> >>>>>> vdev->lock = &isc->lock;
> >>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
> >>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> >>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> >>>>>> + V4L2_CAP_IO_MC;
> >>>>>> video_set_drvdata(vdev, isc);
> >>>>>>
> >>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> >>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>> goto isc_async_complete_err;
> >>>>>> }
> >>>>>>
> >>>>>> + ret = isc_scaler_link(isc);
> >>>>>> + if (ret < 0)
> >>>>>> + goto isc_async_complete_unregister_device;
> >>>>>> +
> >>>>>> + ret = media_device_register(&isc->mdev);
> >>>>>> + if (ret < 0)
> >>>>>> + goto isc_async_complete_unregister_device;
> >>>>>> return 0;
> >>>>>>
> >>>>>> +isc_async_complete_unregister_device:
> >>>>>> + video_unregister_device(vdev);
> >>>>>> +
> >>>>>> isc_async_complete_err:
> >>>>>> mutex_destroy(&isc->lock);
> >>>>>> return ret;
> >>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> >>>>>> }
> >>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>>>>>
> >>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
> >>>>>> +{
> >>>>>> + const struct of_device_id *match;
> >>>>>> + int ret;
> >>>>>> +
> >>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >>>>>
> >>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> >>>>> to be able to have the media pipeline validated at
> >>>>> media_pipeline_start() time ?
> >>>>
> >>>> Hi,
> >>>>
> >>>> I am doing that in a subsequent patch. Things are not completely ready
> >>>> at this moment, because ISC still relies on the old mechanism to call
> >>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
> >>>> removes and adds checks to the link validate ).
> >>>>
> >>>
> >>> I see.. the subsequent patches are not part of this series, right ?
> >>
> >> No. Patch 7 does this. In that patch, the link validation is created and
> >> all the logic of format propagation is removed and reworked, and creates
> >> the need for the link_validate call
> >> Could you have also a look at that patch ?
> >
> > Ah ups, 7 was not in my inbox with the rest of the series. Just
> > noticed. I had even reviewed the previous version, I should have
> > remembered link_validate was added later
> >
> >>>
> >>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
> >>> the introduction of media controller or the scaler, right ? (I'm not
> >>> sure if the DTS patches do).
> >
> > Do DTS patches depend on media-controller plumbing ?
>
> I think not, but, I have not tested them without the media controller.
>
> >
> >>>
> >>> If that's the case, would it make sense to to fast-track them (as some
> >>> of them are fixes) and then on top plumb the media controller
> >>> infrastructure with this patch and the ones you have been mentioning ?
> >>
> >> It would make sense.
> >>
> >
> > Once the other discussion with Hans about the stop state it might make
> > sense to fastrack the first part of the series ?
>
> The other three patches (4,5,6) are quite independent and can go faster.
> >
> >>>
> >>>>>
> >>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>> +
> >>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >>>>>> + isc->pads);
> >>>>>> + if (ret < 0) {
> >>>>>> + dev_err(isc->dev, "media entity init failed\n");
> >>>>>> + return ret;
> >>>>>> + }
> >>>>>> +
> >>>>>> + isc->mdev.dev = isc->dev;
> >>>>>> +
> >>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
> >>>>>> + isc->dev->of_node);
> >>>>>> +
> >>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> >>>>>> + sizeof(isc->mdev.driver_name));
> >>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> >>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> >>>>>> + isc->v4l2_dev.name);
> >>>>>> + isc->mdev.hw_revision = ver;
> >>>>>> +
> >>>>>> + media_device_init(&isc->mdev);
> >>>>>> +
> >>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
> >>>>>> +
> >>>>>> + return isc_scaler_init(isc);
> >>>>>> +}
> >>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
> >>>>>> +
> >>>>>> +void isc_mc_cleanup(struct isc_device *isc)
> >>>>>> +{
> >>>>>> + media_entity_cleanup(&isc->video_dev.entity);
> >>>>>> +}
> >>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> >>>>>> +
> >>>>>> /* regmap configuration */
> >>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
> >>>>>> const struct regmap_config isc_regmap_config = {
> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>> new file mode 100644
> >>>>>> index 000000000000..ec95c9665883
> >>>>>> --- /dev/null
> >>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>> @@ -0,0 +1,245 @@
> >>>>>> +// SPDX-License-Identifier: GPL-2.0-only
> >>>>>> +/*
> >>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> >>>>>> + *
> >>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
> >>>>>
> >>>>> Time flies! It's 2022 already :)
> >>>>>
> >>>>>> + *
> >>>>>> + * Author: Eugen Hristev <[email protected]>
> >>>>>> + *
> >>>>>> + */
> >>>>>> +
> >>>>>> +#include <media/media-device.h>
> >>>>>> +#include <media/media-entity.h>
> >>>>>> +#include <media/v4l2-device.h>
> >>>>>> +#include <media/v4l2-subdev.h>
> >>>>>> +
> >>>>>> +#include "atmel-isc-regs.h"
> >>>>>> +#include "atmel-isc.h"
> >>>>>> +
> >>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> >>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>> + struct v4l2_subdev_format *format)
> >>>>>> +{
> >>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>> +
> >>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>> + format->pad);
> >>>>>> + format->format = *v4l2_try_fmt;
> >>>>>> +
> >>>>>> + return 0;
> >>>>>> + }
> >>>>>> +
> >>>>>> + format->format = isc->scaler_format;
> >>>>>
> >>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
> >>>>> I wonder why it lives in the isc_device struct.
> >>>>
> >>>> isc_device is a placeholder for all isc things.
> >>>>
> >>>> I would not create a separate struct in this file that would have to be
> >>>> allocated etc... just for one two things. So I preferred to have it in
> >>>> the same place as all the other things.
> >>>>
> >>>>>
> >>>>>> +
> >>>>>> + return 0;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> >>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>> + struct v4l2_subdev_format *req_fmt)
> >>>>>> +{
> >>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>> + struct isc_format *fmt;
> >>>>>> + unsigned int i;
> >>>>>> +
> >>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>> + v4l_bound_align_image
> >>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
> >>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> >>>>>> + else
> >>>>>> + v4l_bound_align_image
> >>>>>> + (&req_fmt->format.width, 16, 10000, 0,
> >>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
> >>>>>
> >>>>> Where does 10000 come from ?
> >>>>
> >>>> It's a random number. Do you have any suggestion for a better one ?
> >>>
> >>> An actual HW limit ? :)
> >>
> >> There is no hardware limit. The ISC just stops sampling pixels once the
> >> crop limit is reached. The element that generates pixels could go on
> >> forever , or until the next HBLANK/VBLANK. So what limit could I place
> >> here ?
> >> The cropping is mandatory, otherwise there could be an overflow w.r.t.
> >> the buffer size if the ISC would sample more pixels than the software
> >> expects it to.
> >>
> >>>
> >>>> Maybe this would be much more clear with my comments below (where I will
> >>>> explain what the isc scaler does )
> >>>> In short, it allows the other entity in the link to 'sink' a huge format
> >>>> into this 'scaler' . Because the scaler can crop anything basically down
> >>>> to the biggest format size the ISC could handle.
> >>>>
> >>>
> >>> doesn't it have any documented input size limit ?
> >>
> >> As stated above, ISC handles a pixel stream and stores it in an internal
> >> SRAM. ISC stops sampling when this SRAM is full or when the cropping
> >> limit is reached.
> >> After storing the pixels in the SRAM (which is limited to the ISC
> >> maximum frame size), the ISC will work on the pixels and then DMA the
> >> frame to another RAM.
> >>
> >
> > I understand...
> >
> > So whatever the input size is, only the first n-th bytes are actually
> > dumped to memory, the following ones are discarded.
>
> Yes ! That's the behavior.
> >
> > So the crop is always applied on the top-leftmost corner as a
> > consequence, right ?
>
> Right. Pixels come in order. The ISC does not really know how many
> pixels are coming until its memory is full (or counting until the crop
> limit is reached). After that it stops sampling pixels, and waits for
> vblank to finish the frame.
>
>
> >
> > I understand why you had difficulties picking a default :)
> >
> > This is not easy to handle, as the scaler doesn't actually applies a
> > crop but rather stops capturing after the destination buffer has been
> > saturated. So the limiting factor is the full image size, not one of
> > its dimensions, like in example the line length. This makes it hard
> > to adjust cropping to something meaningful for most use cases.
>
> I didn't know how to name this except crop. It 'crops out' what does not
> fit in the internal memory of the ISC
>
> >
> > In your below example you suggest that an image size of (800000 x 10)
> > would theoretically be captured correctly. What if the input is of
> > size (800000 x 10 + 1). What size would you set the crop rectangle to ?
> >
> > I wouldn't however be too much concerned for the moment, but I would
> > record this in a comment ?
>
> I discussed a bit more with designer and looked in datasheet.
> It looks like the horizontal line length is not a hard limit, but it
> makes the ISC internal ISP stop working, so I would keep the horizontal
> limit as well for now, unless an important use case (I am coming back to
> that). The vertical limit looks to be a hard requirement.
> So in short, we have to impose a vertical limit of 2464 , we cannot have
> more lines than these. So if the horizontal is 100 pixels e.g., we can
> capture only 100x2464 .
> In practice we could capture [16,3264] on the horizontal, and [16,2464]
> on the vertical, and in theory horizontal could go more like [16, 4000],
> but we have to limit the vertical, and without the internal ISP.
> So I would not haphazardly go in this direction. Let's stick to the
> datasheet maximum frame 3264x2464 .
> According to this , I changed the frame size of the ISC to be a
> continuous size from [16,16] to [3264,2464]
> About that use case that I talked about earlier, we have one sensor ,
> imx274, which is more wide, and outputs something like 3800x2100 .
> With the cropping and limitations we have in place now, this will be
> limited to 3264x2100 , which is in line with datasheet and what the ISC
> supports now.
> As in theory we could capture 3800x2100 as long as the SRAM doesn't get
> full, for now I would like to avoid capturing this size (and disable the
> internal ISP), until we have someone really asking for such kind of
> capture and with some good reasons.
> In short I would like to have the crop rectangle to 3264x2464 and the
> bounds would be limited to the incoming frame but no H>3264 and no
> V>2464. Does that make sense ?

Just to make sure I got it: is the ISC actually able to crop on the
line length ? If you apply a 3264 limit, will the last 536 pixels of
each line will be dropped if the input frame is 3800x2100 as your
above example ?

Or rather, is the SRAM size limit the issue and the lines gets
captured in full lenght no matter what, until the buffer gets filled up
and all the successive pixels dropped.

In other words, do your cropped image look like:

3264
+--------------------------------X-----------+ 3800
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
+--------------------------------------------+ 2100

Or rather


3264
+--------------------------------X-----------+ 3800
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
|xxxxxxxxxxxxxxxxx| |
| |
+--------------------------------------------+ 2100

I assume the first, as otherwise you wouldn't be able to interpret the
image as 3264x2100.

I'm asking because you had the un-cropped 3280x2464 size as the ISC
source pad size in the paste of the media graph you provided below,
which might mean the output image is actually full size in lines
lenght and it's maybe missing pixels at the end, in case the ISC
operates as in the above second case.

>
> >
> >>>
> >>>>>
> >>>>>> +
> >>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
> >>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>> +
> >>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
> >>>>>
> >>>>> So you rely on the isc format list for the scaler as well ?
> >>>>> I think it's fine as long as they are identical
> >>>>
> >>>> Yes, the scaler is kind of a dummy entity , but it's required by the
> >>>> media controller validation of the pipeline.
> >>>>
> >>>
> >>> More on this later
> >>>
> >>>>>
> >>>>>> +
> >>>>>> + if (!fmt)
> >>>>>> + fmt = &isc->formats_list[0];
> >>>>>> +
> >>>>>> + req_fmt->format.code = fmt->mbus_code;
> >>>>>> +
> >>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>> + req_fmt->pad);
> >>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>> + /* Trying on the pad sink makes the source sink change too */
> >>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> >>>>>> + v4l2_try_fmt =
> >>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>> +
> >>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
> >>>>>> + 16, isc->max_width, 0,
> >>>>>> + &v4l2_try_fmt->height,
> >>>>>> + 16, isc->max_height, 0, 0);
> >>>>>> + }
> >>>>>> + /* if we are just trying, we are done */
> >>>>>> + return 0;
> >>>>>> + }
> >>>>>> +
> >>>>>> + isc->scaler_format = req_fmt->format;
> >>>>>
> >>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
> >>>>>
> >>>>> Actually, I would like to know more about the scaler device
> >>>>> capabilities. What functions can this IP perform ? Does it do
> >>>>> cropping, can it also do (down)scaling or even composition ?
> >>>>>
> >>>>> I think it is worth to read
> >>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> >>>>> where it is reported how cropping and scaling are implemented by using
> >>>>> the selection API.
> >>>>>
> >>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
> >>>>> for which you need to implement support to by adding s_selection on
> >>>>> the sink pad TGT_CROP target.
> >>>>
> >>>> The scaler is kind of a dummy entity.
> >>>> The problem is that different subdevices cannot be connected directly to
> >>>> a v4l entity, because the ISC does cropping on the incoming frame.
> >>>> The ISC has a limit on the total number of pixels per frame, thus it
> >>>> will do a crop.
> >>>
> >>> What is this limit related to ? I don't know the ISC internals, but is
> >>> the limitation due to the overall image size, or maybe it's due to
> >>> some line buffers being limited in the size they can transfer to
> >>> memory ? Is it a limitation on the size of the image in pixels, or is
> >>> it due to a limitation of the processing bandwidth on the input
> >>> interface and thus depends on the actual size of the image in bytes ?
> >>
> >> Limitation is both on line size and full image size. From my
> >> understanding, is that the internal SRAM is limited (8 Mpixels with a
> >> small safe buffer on top in the case of sama7g5 )
> >> This SRAM could be used as 800000x10 for example, or 3200x2464 for
> >> another example. However, I am not sure if a real line of 800,000 pixels
> >> make sense, or if there is another internal mechanism in the ISC that
> >> stops it.
> >> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
> >> the maximum frame that it can handle
> >>
> >>>
> >>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
> >>>> only handle 3264x2464 , then we have two choices:
> >>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
> >>>> 3280x2464, and the whole pipeline fails to configure, there is a
> >>>> mismatch between /dev/video output and what the whole pipeline produces
> >>>
> >>> I'm confused in first place. If the ISC cannot output anything larger
> >>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
> >>> this should be adjusted to 3264x2464.
> >>
> >> yes, that is true. However, without an entity that says 'I am doing
> >> cropping', the media pipeline will fail, because previous entity will
> >> output 3280x2464 .
> >> The conclusion to create a scaler entitity was done during my last year
> >> discussions on the IRC. I could not find a way to perform this cropping
> >> at the v4l entity side, and one reason is that the v4l entity does not
> >> support what I needed , or at least that was my understanding.
> >> You claim that the v4l entity could perform the cropping in the same
> >> entity ?
> >
> > I thought so, but I haven't tried to be honest
> >
> >>
> >> Let's assume for a minute that it could do this. Even so, I would prefer
> >> to have a separate scaler entity to do this , for several reasons:
> >> -> I would like to be able to crop the frame arbitrarily if I want to,
> >> and the scaler would allow me to do this easier
> >> -> it would be more obvious that the frame is cropped by having this
> >> entity there
> >> -> in fact some of the ISC versions really have a down scaler, that I
> >> have not implemented yet. So it would be useful to already have this in
> >> place to be able to scale down with it at a later time.
> >
> > I think this last reason is enough to have an entity plumbed in
> > already
> >
> >>
> >>>
> >>> Or is the ISC limitation on the -input- side ?
> >>>
> >>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
> >>>> the ISC can handle at most -> cannot , because the sensor let's say can
> >>>> *only* produce 3280x2464 and *any* other frame size returns an error.
> >>>
> >>> buggy sensor I would say :)
> >>
> >> I cannot tell this to the customers, and I cannot fail the whole
> >> pipeline because the sensor does not crop 16 pixels, when in fact I can
> >> do this very easily on the ISC side.
> >>
> >
> > Fair enough :)
> >
> >>>
> >>>>
> >>>> The only solution to make both worlds happy, is to have a dummy entity
> >>>> called 'scaler' which in fact now it only performs a simple crop.
> >>>> It accepts any frame size at input (hence the 10000x10000 which you saw
> >>>> earlier), and outputs at most 3264x2464 the isc_max_height and
> >>>> isc_max_width.
> >>>
> >>> I have a maybe dumb question: can the cropping operation be modeled
> >>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
> >>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
> >>
> >> I am not sure. This is what I wanted initially. But then I thought about
> >> the three reasons stated above for the use of the scaler entity, and
> >> discussed with folks on the IRC, and come up with the solution for the
> >> scaler entity
> >>
> >>>
> >>>>
> >>>> So to answer your question, the isc scaler is a software model for a
> >>>> simple cropping, that would make media controller happy, and capture
> >>>> software happy.
> >>>>
> >>>> Here it how it looks :
> >>>>
> >>>> Device topology
> >>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
> >>>> type V4L2 subdev subtype Unknown flags 0
> >>>> device node name /dev/v4l-subdev0
> >>>> pad0: Sink
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>> crop.bounds:(0,0)/3264x2464
> >>>> crop:(0,0)/3264x2464]
> >>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>> pad1: Source
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>
> >>>> - entity 4: csi2dc (2 pads, 2 links)
> >>>> type V4L2 subdev subtype Unknown flags 0
> >>>> device node name /dev/v4l-subdev1
> >>>> pad0: Sink
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>> <- "dw-csi.0":1 [ENABLED]
> >>>> pad1: Source
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
> >>>>
> >>>> - entity 7: dw-csi.0 (2 pads, 2 links)
> >>>> type V4L2 subdev subtype Unknown flags 0
> >>>> device node name /dev/v4l-subdev2
> >>>> pad0: Sink
> >>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>> <- "imx219 1-0010":0 [ENABLED]
> >>>> pad1: Source
> >>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>> -> "csi2dc":0 [ENABLED]
> >>>>
> >>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
> >>>> type V4L2 subdev subtype Sensor flags 0
> >>>> device node name /dev/v4l-subdev3
> >>>> pad0: Source
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>> xfer:srgb ycbcr:601 quantization:full-range
> >>>> crop.bounds:(8,8)/3280x2464
> >>>> crop:(8,8)/3280x2464]
> >>>> -> "dw-csi.0":0 [ENABLED]
> >>>>
> >>>> - entity 24: atmel_isc_common (1 pad, 1 link)
> >>>> type Node subtype V4L flags 1
> >>>> device node name /dev/video0
> >>>> pad0: Sink
> >>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
> >>>>
> >>>>
> >>>> Scaler does this one cute little thing :
> >>>>
> >>>> pad0: Sink
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>> crop.bounds:(0,0)/3264x2464
> >>>> crop:(0,0)/3264x2464]
> >>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>> pad1: Source
> >>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>
> >>> Shouldn't this be 3264x2464 as that's what the entity outputs after
> >>> the cropping ? And shouldn't then be 3264x2464 the size output from
> >>> the video device too ?
> >>
> >> That's right.
> >> I don't know why the source format for the scaler is still 3280x2464.
> >> Maybe there is a bug on g_fmt for it.. have to check it.
> >
> > Thanks, I think this should be fixed ten
> >
> >
> >>
> >> Anyway, the video format is like this :
> >>
> >> # v4l2-ctl --get-fmt-video
> >> Format Video Capture:
> >> Width/Height : 3264/2464
> >> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
> >> Field : None
> >> Bytes per Line : 6528
> >> Size Image : 16084992
> >> Colorspace : sRGB
> >> Transfer Function : Default (maps to sRGB)
> >> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
> >> Quantization : Default (maps to Full Range)
> >> Flags :
> >> #
> >>
> >> and initializing the pipeline looks fine from user perspective.
> >> The scaler is in fact the solution to make this pipeline work with
> >> libcamera.
> >> I found this cropping to be an issue in media controller when trying it
> >> with libcamera. Otherwise, the other user space apps which I was using
> >> never complained that anything was wrong
> >> Libcamera simply refuses to acknowledge the pipeline if the video output
> >> is 3264x2464 but there is no entity that changes the format from
> >> 3280x2464 down
> >
> > Not sure why libcamera should play a role there. Isn't it the media
> > pipeline validation that complains and returns an -EPIPE ?
>
> I tried this before the media_pipeline_start .
> Perhaps libcamera was doing some similar checks, anyway, it was
> complaining and not adding the stream, and reported no usable streams/camera
>
> Once I will rework some of the things according to your review I will
> run libcamera again to see what it complains about
>
> >
> >>
> >>>
> >>>
> >>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>
> >>>>
> >>>> Which is what we needed.
> >>>>
> >>>>>
> >>>>>> +
> >>>>>> + return 0;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> >>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>> + struct v4l2_subdev_mbus_code_enum *code)
> >>>>>> +{
> >>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>> + int supported_index = 0;
> >>>>>> + int i;
> >>>>>> +
> >>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
> >>>>>> + if (!isc->formats_list[i].sd_support)
> >>>>>> + continue;
> >>>>>
> >>>>> The sd_support flag still doesn't click in my head.
> >>>>>
> >>>>> Shouldn't the enumeration of available formats on the scaler do not
> >>>>> depend on the sensor supproted formats ?
> >>>>
> >>>> You're right. I will have to check it again.
> >>>>
> >>>>>
> >>>>>> + if (supported_index == code->index) {
> >>>>>> + code->code = isc->formats_list[i].mbus_code;
> >>>>>> + return 0;
> >>>>>> + }
> >>>>>> + supported_index++;
> >>>>>> + }
> >>>>>> +
> >>>>>> + return -EINVAL;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> >>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>> + struct v4l2_subdev_selection *sel)
> >>>>>> +{
> >>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>> +
> >>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>> + return -EINVAL;
> >>>>>> +
> >>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> >>>>>> + sel->target != V4L2_SEL_TGT_CROP)
> >>>>>> + return -EINVAL;
> >>>>>> +
> >>>>>> + sel->r.height = isc->max_height;
> >>>>>> + sel->r.width = isc->max_width;
> >>>>>
> >>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
> >>>>> as it represents the maximum valid crop rectangle.
> >>>>>
> >>>>> TGT_CROP should report the configured crop rectangle which can be
> >>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
> >>>>> the spec is correct
> >>>>
> >>>> So you would like to have this differentiated, and report the
> >>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
> >>>> reported now, the maximum size of the ISC frame .
> >>>> My understanding is correct ?
> >>>>
> >>>
> >>> I didn't know you have an HW limitation, so your _BOUNDS is not the
> >>> input image size but rather 3264x2464 ( == max_width x max_height).
> >>>
> >>> What I meant is that _BOUNDS should report the maximum rectangle size
> >>> that can be applied to the _CROP target. In you case you have an HW
> >>> limitation 3264x2464 and that's the largest rectangle you can apply.
> >> So the CROP should be at 3264x2464
> >>> TGT_CROP can be initialized to the same as _BOUND, but if you
> >>> implement s_selection it should report what has been there applied.
> >> and BOUND to actual frame size ?
> >>> But as you don't implement s_selection yet, I think this is fine for
> >>> now. Maybe a little comment ?
> >>
> >> It could also be the case where e.g. the sensor is outputting 1920x1080,
> >> in this case the scaler would do nothing.
> >> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
> >> If the sensor is outputting 1920x1080, this format comes directly from
> >> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
> >> sensor's problem)
> >>>
> >>> Also, is set->r zeroed by the framework before getting here ?
> >>> Otherwise you should set r.left and r.top to 0
> >>
> >> If these are redundant, no problem to remove them
> >
> > I was actually confused. I was suggesting you to add...
> >
> >>>
> >>>>>
> >>>>>> +
> >>>>>> + sel->r.left = 0;
> >>>>>> + sel->r.top = 0;
> >
> > These ^
> >
> > So nothing to change. Sorry I've missed them
> >
> > Thanks
> > j
> >
> >
> >>>>>> +
> >>>>>> + return 0;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> >>>>>> + struct v4l2_subdev_state *sd_state)
> >>>>>> +{
> >>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> >>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> >>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>> +
> >>>>>> + *v4l2_try_fmt = isc->scaler_format;
> >>>>>> +
> >>>>>> + return 0;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> >>>>>> + .set_fmt = isc_scaler_set_fmt,
> >>>>>> + .get_fmt = isc_scaler_get_fmt,
> >>>>>> + .get_selection = isc_scaler_g_sel,
> >>>>>> + .init_cfg = isc_scaler_init_cfg,
> >>>>>
> >>>>> .link_validate = v4l2_subdev_link_validate_default,
> >>>>>
> >>>>> To have the formats at the end of links that point to this entity
> >>>>> validated (I think the framework already calls it if not set though,
> >>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
> >>>>>
> >>>>>> +};
> >>>>>> +
> >>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >>>>>> + .pad = &isc_scaler_pad_ops,
> >>>>>> +};
> >>>>>> +
> >>>>>> +int isc_scaler_init(struct isc_device *isc)
> >>>>>> +{
> >>>>>> + int ret;
> >>>>>> +
> >>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> >>>>>> +
> >>>>>> + isc->scaler_sd.owner = THIS_MODULE;
> >>>>>> + isc->scaler_sd.dev = isc->dev;
> >>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> >>>>>> + "atmel_isc_scaler");
> >>>>>
> >>>>> I would drop 'atmel' for brevity, unless other entities have this
> >>>>> prefix set already
> >>>>
> >>>> The v4l entity takes it's name from the module name, which is
> >>>> atmel_isc_common, so I thought to keep the prefix
> >>>>
> >>>
> >>> Ack!
> >>>
> >>>>>
> >>>>>> +
> >>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >>>>>> +
> >>>>>> + isc->scaler_format.height = isc->max_height;
> >>>>>> + isc->scaler_format.width = isc->max_width;
> >>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> >>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
> >>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>> +
> >>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> >>>>>> + ISC_SCALER_PADS_NUM,
> >>>>>> + isc->scaler_pads);
> >>>>>> + if (ret < 0) {
> >>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> >>>>>> + return ret;
> >>>>>> + }
> >>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> >>>>>> + if (ret < 0) {
> >>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> >>>>>> + return ret;
> >>>>>> + }
> >>>>>> +
> >>>>>> + return ret;
> >>>>>> +}
> >>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> >>>>>> +
> >>>>>> +int isc_scaler_link(struct isc_device *isc)
> >>>>>> +{
> >>>>>> + int ret;
> >>>>>> +
> >>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> >>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
> >>>>>> + ISC_SCALER_PAD_SINK,
> >>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>> +
> >>>>>> + if (ret < 0) {
> >>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>> + isc->current_subdev->sd->entity.name,
> >>>>>> + isc->scaler_sd.entity.name);
> >>>>>> + return ret;
> >>>>>> + }
> >>>>>> +
> >>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> >>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
> >>>>>> +
> >>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> >>>>>> + ISC_SCALER_PAD_SOURCE,
> >>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
> >>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>> +
> >>>>>> + if (ret < 0) {
> >>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>> + isc->scaler_sd.entity.name,
> >>>>>> + isc->video_dev.entity.name);
> >>>>>> + return ret;
> >>>>>> + }
> >>>>>> +
> >>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> >>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>> +
> >>>>>> + return ret;
> >>>>>> +}
> >>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
> >>>>>
> >>>>> From the DT graph point of view, the ISC appears as a single block
> >>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
> >>>>> add to its notifier the image sensor remote subdevice.
> >>>>
> >>>> Actually it's only a parallel port.
> >>>> And it can be connected either to a sensor directly or to the csi2dc
> >>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
> >>>>
> >>>>>
> >>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
> >>>>> entity which was initialized at isc_probe() time is linked in between
> >>>>> the ISC and the image sensor immutably.
> >>>>
> >>>> yes, because this cannot change at runtime, and usually, it can't change
> >>>> without altering the board hardware. (at least this is what my
> >>>> understanding of immutability is )
> >>>>
> >>>>>
> >>>>> I think it is fine for now, but I wonder if you plan to plumb more
> >>>>> components between the ISC video node and the sensor, if it's not
> >>>>> worth changing the DT bindings and their parsing logic to separate the
> >>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
> >>>>> time and have the CSI-2 receiver entity as downstream remote. I think
> >>>>> I need to get to know the ISC better to really have an idea. For now,
> >>>>> this seems ok to me, but please check with maintainers if this is fine
> >>>>> with them.
> >>>>
> >>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
> >>>> the remote subdev).
> >>>> The ISC will register a scaler, and connect the subdev to this scaler
> >>>> first, and then, the scaler to the isc itself (the v4l entity).
> >>>>
> >>>>>
> >>>>>> +
> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>> index 5fbf52a9080b..c9234c90ae58 100644
> >>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
> >>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> >>>>>> u32 his_entry;
> >>>>>> };
> >>>>>>
> >>>>>> +enum isc_mc_pads {
> >>>>>> + ISC_PAD_SINK = 0,
> >>>>>> + ISC_PADS_NUM = 1,
> >>>>>> +};
> >>>>>> +
> >>>>>> +enum isc_scaler_pads {
> >>>>>> + ISC_SCALER_PAD_SINK = 0,
> >>>>>> + ISC_SCALER_PAD_SOURCE = 1,
> >>>>>> + ISC_SCALER_PADS_NUM = 2,
> >>>>>> +};
> >>>>>> +
> >>>>>> /*
> >>>>>> * struct isc_device - ISC device driver data/config struct
> >>>>>> * @regmap: Register map
> >>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> >>>>>> * be used as an input to the controller
> >>>>>> * @controller_formats_size: size of controller_formats array
> >>>>>> * @formats_list_size: size of formats_list array
> >>>>>> + * @pads: media controller pads for isc video entity
> >>>>>> + * @mdev: media device that is registered by the isc
> >>>>>> + * @remote_pad: remote pad on the connected subdevice
> >>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
> >>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
> >>>>>> + * @scaler_format: current format for the scaler subdevice
> >>>>>> */
> >>>>>> struct isc_device {
> >>>>>> struct regmap *regmap;
> >>>>>> @@ -344,6 +361,19 @@ struct isc_device {
> >>>>>> struct isc_format *formats_list;
> >>>>>> u32 controller_formats_size;
> >>>>>> u32 formats_list_size;
> >>>>>> +
> >>>>>> + struct {
> >>>>>> + struct media_pad pads[ISC_PADS_NUM];
> >>>>>> + struct media_device mdev;
> >>>>>> +
> >>>>>> + u32 remote_pad;
> >>>>>> + };
> >>>>>> +
> >>>>>> + struct {
> >>>>>> + struct v4l2_subdev scaler_sd;
> >>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> >>>>>> + struct v4l2_mbus_framefmt scaler_format;
> >>>>
> >>>> Here are the scaler stuff which I added, in the same bulk struct for the
> >>>> whole isc device
> >>>>
> >>>>
> >>>>>> + };
> >>>>>> };
> >>>>>>
> >>>>>> extern const struct regmap_config isc_regmap_config;
> >>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> >>>>>> void isc_subdev_cleanup(struct isc_device *isc);
> >>>>>> void isc_clk_cleanup(struct isc_device *isc);
> >>>>>>
> >>>>>> +int isc_scaler_link(struct isc_device *isc);
> >>>>>> +int isc_scaler_init(struct isc_device *isc);
> >>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
> >>>>>> +void isc_mc_cleanup(struct isc_device *isc);
> >>>>>> +
> >>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>> + unsigned int code, int *index);
> >>>>>> #endif
> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>> index c5b9563e36cb..c244682ea22f 100644
> >>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>> break;
> >>>>>> }
> >>>>>>
> >>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>
> >>>>> I am surprised you can read a register before runtime_pm is
> >>>>> intialized!
> >>>>>
> >>>>
> >>>> Actually I can read any register after the moment of starting the clock
> >>>> on the hardware block.
> >>>
> >>> Ah right then
> >>>
> >>>> Maybe the starting and stopping of the clock needs to be moved to the
> >>>> runtime_pm calls, but this is another story, not related to the media
> >>>> controller.
> >>>
> >>> It's fine, but maybe worth recording with a todo ?
> >>>
> >>> Thanks
> >>> j
> >>>
> >>>> I moved this line because I had to pass the version to the isc_mc_init call
> >>>>
> >>>>> Thanks
> >>>>> j
> >>>>
> >>>> Thanks for reviewing !
> >>>> Eugen
> >>>>>
> >>>>>
> >>>>>> +
> >>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>> + if (ret < 0)
> >>>>>> + goto isc_probe_mc_init_err;
> >>>>>> +
> >>>>>> pm_runtime_set_active(dev);
> >>>>>> pm_runtime_enable(dev);
> >>>>>> pm_request_idle(dev);
> >>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>> ret = clk_prepare_enable(isc->ispck);
> >>>>>> if (ret) {
> >>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
> >>>>>> - goto cleanup_subdev;
> >>>>>> + goto isc_probe_mc_init_err;
> >>>>>> }
> >>>>>>
> >>>>>> /* ispck should be greater or equal to hclock */
> >>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>> goto unprepare_clk;
> >>>>>> }
> >>>>>>
> >>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
> >>>>>>
> >>>>>> return 0;
> >>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>> unprepare_clk:
> >>>>>> clk_disable_unprepare(isc->ispck);
> >>>>>>
> >>>>>> +isc_probe_mc_init_err:
> >>>>>> + isc_mc_cleanup(isc);
> >>>>>> +
> >>>>>> cleanup_subdev:
> >>>>>> isc_subdev_cleanup(isc);
> >>>>>>
> >>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
> >>>>>>
> >>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>
> >>>>>> + isc_mc_cleanup(isc);
> >>>>>> +
> >>>>>> isc_subdev_cleanup(isc);
> >>>>>>
> >>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>> index 07a80b08bc54..9dc75eed0098 100644
> >>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> >>>>>> break;
> >>>>>> }
> >>>>>>
> >>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>> +
> >>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>> + if (ret < 0)
> >>>>>> + goto isc_probe_mc_init_err;
> >>>>>> +
> >>>>>> pm_runtime_set_active(dev);
> >>>>>> pm_runtime_enable(dev);
> >>>>>> pm_request_idle(dev);
> >>>>>>
> >>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
> >>>>>>
> >>>>>> return 0;
> >>>>>>
> >>>>>> +isc_probe_mc_init_err:
> >>>>>> + isc_mc_cleanup(isc);
> >>>>>> +
> >>>>>> cleanup_subdev:
> >>>>>> isc_subdev_cleanup(isc);
> >>>>>>
> >>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
> >>>>>>
> >>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>
> >>>>>> + isc_mc_cleanup(isc);
> >>>>>> +
> >>>>>> isc_subdev_cleanup(isc);
> >>>>>>
> >>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>> --
> >>>>>> 2.25.1
> >>>>>>
> >>>>
> >>
>

2022-02-11 12:13:27

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 07/11] media: atmel: atmel-isc: change format propagation to subdev into only verification

On 2/10/22 8:43 PM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Fri, Jan 21, 2022 at 03:14:12PM +0200, Eugen Hristev wrote:
>> As a top MC video driver, the atmel-isc should not propagate the format to the
>> subdevice.
>> It should rather check at start_streaming() time if the subdev is properly
>> configured with a compatible format.
>> Removed the whole format finding logic, and reworked the format verification
>> at start_streaming time, such that the ISC will return an error if the subdevice
>> is not properly configured. To achieve this, media_pipeline_start
>> is called and a link_validate callback is created to check the formats.
>> With this being done, the module parameter 'sensor_preferred' makes no sense
>> anymore. The ISC should not decide which format the sensor is using. The
>> ISC should only cope with the situation and inform userspace if the streaming
>> is possible in the current configuration.
>
> This sounds right!
>
>> The redesign of the format propagation has also risen the question of the
>> enumfmt callback. If enumfmt is called with an mbus_code, the enumfmt handler
>> should only return the formats that are supported for this mbus_code.
>> To make it more easy to understand the formats, changed the report order
>> to report first the native formats, and after that the formats that the ISC
>> can convert to.
>>
>> Signed-off-by: Eugen Hristev <[email protected]>
>> ---
>> Changes in v4:
>> - moved validation code into link_validate and used media_pipeline_start
>> - merged this patch with the enum_fmt patch which was previously in v3 of
>> the series
>>
>> Changes in v3:
>> - clamp to maximum resolution once the frame size from the subdev is found
>>
>> drivers/media/platform/atmel/atmel-isc-base.c | 335 ++++++++++--------
>> .../media/platform/atmel/atmel-isc-scaler.c | 5 +
>> drivers/media/platform/atmel/atmel-isc.h | 3 +
>> 3 files changed, 191 insertions(+), 152 deletions(-)
>>
>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>> index f7a88399bd54..31c79313aadc 100644
>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>> @@ -36,11 +36,6 @@ static unsigned int debug;
>> module_param(debug, int, 0644);
>> MODULE_PARM_DESC(debug, "debug level (0-2)");
>>
>> -static unsigned int sensor_preferred = 1;
>> -module_param(sensor_preferred, uint, 0644);
>> -MODULE_PARM_DESC(sensor_preferred,
>> - "Sensor is preferred to output the specified format (1-on 0-off), default 1");
>> -
>> #define ISC_IS_FORMAT_RAW(mbus_code) \
>> (((mbus_code) & 0xf000) == 0x3000)
>>
>> @@ -337,6 +332,10 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
>> unsigned long flags;
>> int ret;
>>
>> + ret = media_pipeline_start(&isc->video_dev.entity, &isc->mpipe);
>> + if (ret)
>> + goto err_pipeline_start;
>> +
>> /* Enable stream on the sub device */
>> ret = v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 1);
>> if (ret && ret != -ENOIOCTLCMD) {
>> @@ -384,6 +383,9 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
>> v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0);
>>
>> err_start_stream:
>> + media_pipeline_stop(&isc->video_dev.entity);
>> +
>> +err_pipeline_start:
>> spin_lock_irqsave(&isc->dma_queue_lock, flags);
>> list_for_each_entry(buf, &isc->dma_queue, list)
>> vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_QUEUED);
>> @@ -420,6 +422,9 @@ static void isc_stop_streaming(struct vb2_queue *vq)
>> if (ret && ret != -ENOIOCTLCMD)
>> v4l2_err(&isc->v4l2_dev, "stream off failed in subdev\n");
>>
>> + /* Stop media pipeline */
>> + media_pipeline_stop(&isc->video_dev.entity);
>> +
>> /* Release all active buffers */
>> spin_lock_irqsave(&isc->dma_queue_lock, flags);
>> if (unlikely(isc->cur_frm)) {
>> @@ -496,21 +501,56 @@ static int isc_enum_fmt_vid_cap(struct file *file, void *priv,
>> u32 index = f->index;
>> u32 i, supported_index;
>>
>> - if (index < isc->controller_formats_size) {
>> - f->pixelformat = isc->controller_formats[index].fourcc;
>> - return 0;
>> + supported_index = 0;
>> +
>> + for (i = 0; i < isc->formats_list_size; i++) {
>> + if (!isc->formats_list[i].sd_support)
>> + continue;
>
> Sorry to ask again about sd_support, but if the ISC formats
> enumeration is not constrained to a specific mbus_code, shouldn't all
> the formats be enumerated regardless if they're supported from the
> sensor or not ?

Yes. this slipped a little. I have to rework a bit the format reporting
to user space.

>
>> + /*
>> + * If specific mbus_code is requested, provide only
>> + * supported formats with this mbus code
>> + */
>> + if (f->mbus_code && f->mbus_code !=
>> + isc->formats_list[i].mbus_code)
>
> nit: for clarity
> if (f->mbus_code &&
> f->mbus_code != isc->formats_list[i].mbus_code)
>
>> + continue;
>> + if (supported_index == index) {
>> + f->pixelformat = isc->formats_list[i].fourcc;
>> + return 0;
>> + }
>> + supported_index++;
>> }
>>
>> - index -= isc->controller_formats_size;
>> + /*
>> + * If the sensor does not support this mbus_code whatsoever,
>> + * there is no reason to advertise any of our output formats
>> + */
>> + if (supported_index == 0)
>> + return -EINVAL;
>>
>> + /*
>> + * If the sensor uses a format that is not raw, then we cannot
>> + * convert it to any of the formats that we usually can with a
>> + * RAW sensor. Thus, do not advertise them.
>> + */
>> + if (isc->config.sd_format &&
>> + !ISC_IS_FORMAT_RAW(isc->config.sd_format->mbus_code))
>> + return -EINVAL;
>> +
>> + /*
>> + * Iterate again through the formats that we can convert to.
>> + * However, to avoid duplicates, skip the formats that
>> + * the sensor already supports directly
>> + */
>> + index -= supported_index;
>> supported_index = 0;
>>
>> - for (i = 0; i < isc->formats_list_size; i++) {
>> - if (!ISC_IS_FORMAT_RAW(isc->formats_list[i].mbus_code) ||
>> - !isc->formats_list[i].sd_support)
>> + for (i = 0; i < isc->controller_formats_size; i++) {
>> + /* if this format is already supported by sensor, skip it */
>> + if (find_format_by_fourcc(isc, isc->controller_formats[i].fourcc))
>> continue;
>> if (supported_index == index) {
>> - f->pixelformat = isc->formats_list[i].fourcc;
>> + f->pixelformat =
>> + isc->controller_formats[i].fourcc;
>
> I trust this last part to work correctly, as it's not 100% clear to me :)
>

I think I will remove this and just report all the formats to userspace.
Let libcamera handle this ;)

>> return 0;
>> }
>> supported_index++;
>> @@ -581,20 +621,30 @@ static int isc_try_validate_formats(struct isc_device *isc)
>> break;
>> default:
>> /* any other different formats are not supported */
>> + v4l2_err(&isc->v4l2_dev, "Requested unsupported format.\n");
>> ret = -EINVAL;
>> }
>> v4l2_dbg(1, debug, &isc->v4l2_dev,
>> "Format validation, requested rgb=%u, yuv=%u, grey=%u, bayer=%u\n",
>> rgb, yuv, grey, bayer);
>>
>> - /* we cannot output RAW if we do not receive RAW */
>> - if ((bayer) && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
>> + if ((bayer) &&
>
> just 'bayer' ?

not sure what you mean ?
Have to check if we can output raw bayer , and we can only do that if we
receive raw bayer from the sensor.

>
>> + !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code)) {
>> + v4l2_err(&isc->v4l2_dev, "Cannot output RAW if we do not receive RAW.\n");
>> return -EINVAL;
>> + }
>>
>> - /* we cannot output GREY if we do not receive RAW/GREY */
>> if (grey && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code) &&
>> - !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code))
>> + !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
>> + v4l2_err(&isc->v4l2_dev, "Cannot output GREY if we do not receive RAW/GREY.\n");
>> + return -EINVAL;
>> + }
>> +
>> + if ((rgb || bayer || yuv) &&
>> + ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
>> + v4l2_err(&isc->v4l2_dev, "Cannot convert GREY to another format.\n");
>> return -EINVAL;
>> + }
>>
>> return ret;
>> }
>> @@ -822,7 +872,7 @@ static void isc_try_fse(struct isc_device *isc,
>> * If we do not know yet which format the subdev is using, we cannot
>> * do anything.
>> */
>> - if (!isc->try_config.sd_format)
>> + if (!isc->config.sd_format)
>> return;
>>
>> fse.code = isc->try_config.sd_format->mbus_code;
>> @@ -843,180 +893,141 @@ static void isc_try_fse(struct isc_device *isc,
>> }
>> }
>>
>> -static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f,
>> - u32 *code)
>> +static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f)
>> {
>> - int i;
>> - struct isc_format *sd_fmt = NULL, *direct_fmt = NULL;
>> struct v4l2_pix_format *pixfmt = &f->fmt.pix;
>> - struct v4l2_subdev_pad_config pad_cfg = {};
>> - struct v4l2_subdev_state pad_state = {
>> - .pads = &pad_cfg
>> - };
>> - struct v4l2_subdev_format format = {
>> - .which = V4L2_SUBDEV_FORMAT_TRY,
>> - };
>> - u32 mbus_code;
>> - int ret;
>> - bool rlp_dma_direct_dump = false;
>> + unsigned int i;
>>
>> if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
>> return -EINVAL;
>>
>> - /* Step 1: find a RAW format that is supported */
>> - for (i = 0; i < isc->num_user_formats; i++) {
>> - if (ISC_IS_FORMAT_RAW(isc->user_formats[i]->mbus_code)) {
>> - sd_fmt = isc->user_formats[i];
>> + isc->try_config.fourcc = isc->user_formats[0]->fourcc;
>> +
>> + /* find if the format requested is supported */
>> + for (i = 0; i < isc->controller_formats_size; i++)
>> + if (isc->controller_formats[i].fourcc == pixfmt->pixelformat) {
>> + isc->try_config.fourcc = pixfmt->pixelformat;
>> break;
>> }
>> - }
>> - /* Step 2: We can continue with this RAW format, or we can look
>> - * for better: maybe sensor supports directly what we need.
>> - */
>> - direct_fmt = find_format_by_fourcc(isc, pixfmt->pixelformat);
>> -
>> - /* Step 3: We have both. We decide given the module parameter which
>> - * one to use.
>> - */
>> - if (direct_fmt && sd_fmt && sensor_preferred)
>> - sd_fmt = direct_fmt;
>> -
>> - /* Step 4: we do not have RAW but we have a direct format. Use it. */
>> - if (direct_fmt && !sd_fmt)
>> - sd_fmt = direct_fmt;
>> -
>> - /* Step 5: if we are using a direct format, we need to package
>> - * everything as 8 bit data and just dump it
>> - */
>> - if (sd_fmt == direct_fmt)
>> - rlp_dma_direct_dump = true;
>> -
>> - /* Step 6: We have no format. This can happen if the userspace
>> - * requests some weird/invalid format.
>> - * In this case, default to whatever we have
>> - */
>> - if (!sd_fmt && !direct_fmt) {
>> - sd_fmt = isc->user_formats[isc->num_user_formats - 1];
>> - v4l2_dbg(1, debug, &isc->v4l2_dev,
>> - "Sensor not supporting %.4s, using %.4s\n",
>> - (char *)&pixfmt->pixelformat, (char *)&sd_fmt->fourcc);
>> - }
>> -
>> - if (!sd_fmt) {
>> - ret = -EINVAL;
>> - goto isc_try_fmt_err;
>> - }
>> -
>> - /* Step 7: Print out what we decided for debugging */
>> - v4l2_dbg(1, debug, &isc->v4l2_dev,
>> - "Preferring to have sensor using format %.4s\n",
>> - (char *)&sd_fmt->fourcc);
>> -
>> - /* Step 8: at this moment we decided which format the subdev will use */
>> - isc->try_config.sd_format = sd_fmt;
>> -
>> - /* Limit to Atmel ISC hardware capabilities */
>> - if (pixfmt->width > isc->max_width)
>> - pixfmt->width = isc->max_width;
>> - if (pixfmt->height > isc->max_height)
>> - pixfmt->height = isc->max_height;
>> -
>> - /*
>> - * The mbus format is the one the subdev outputs.
>> - * The pixels will be transferred in this format Sensor -> ISC
>> - */
>> - mbus_code = sd_fmt->mbus_code;
>> -
>> - /*
>> - * Validate formats. If the required format is not OK, default to raw.
>> - */
>> -
>> - isc->try_config.fourcc = pixfmt->pixelformat;
>> -
>> - if (isc_try_validate_formats(isc)) {
>> - pixfmt->pixelformat = isc->try_config.fourcc = sd_fmt->fourcc;
>> - /* Re-try to validate the new format */
>> - ret = isc_try_validate_formats(isc);
>> - if (ret)
>> - goto isc_try_fmt_err;
>> - }
>> -
>> - ret = isc_try_configure_rlp_dma(isc, rlp_dma_direct_dump);
>> - if (ret)
>> - goto isc_try_fmt_err;
>> -
>> - ret = isc_try_configure_pipeline(isc);
>> - if (ret)
>> - goto isc_try_fmt_err;
>>
>> - /* Obtain frame sizes if possible to have crop requirements ready */
>> - isc_try_fse(isc, &pad_state);
>> -
>> - v4l2_fill_mbus_format(&format.format, pixfmt, mbus_code);
>> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad, set_fmt,
>> - &pad_state, &format);
>> - if (ret < 0)
>> - goto isc_try_fmt_subdev_err;
>> + /* If we did not find the requested format, we will fallback here */
>> + pixfmt->pixelformat = isc->try_config.fourcc;
>> + pixfmt->colorspace = V4L2_COLORSPACE_SRGB;
>> + pixfmt->field = V4L2_FIELD_NONE;
>>
>> - v4l2_fill_pix_format(pixfmt, &format.format);
>> + isc_try_configure_rlp_dma(isc, false);
>>
>> /* Limit to Atmel ISC hardware capabilities */
>> - if (pixfmt->width > isc->max_width)
>> - pixfmt->width = isc->max_width;
>> - if (pixfmt->height > isc->max_height)
>> - pixfmt->height = isc->max_height;
>> + v4l_bound_align_image(&pixfmt->width, 16, isc->max_width, 0,
>> + &pixfmt->height, 16, isc->max_height, 0, 0);
>>
>> pixfmt->field = V4L2_FIELD_NONE;
>> pixfmt->bytesperline = (pixfmt->width * isc->try_config.bpp_v4l2) >> 3;
>> pixfmt->sizeimage = ((pixfmt->width * isc->try_config.bpp) >> 3) *
>> pixfmt->height;
>
> Seems pixfmt->field is set twice.
>
> Could you move the assignment of all pixfmt to a single block ?

okay
>
>>
>> - if (code)
>> - *code = mbus_code;
>> + isc->try_fmt = *f;
>>
>> return 0;
>> +}
>>
>> -isc_try_fmt_err:
>> - v4l2_err(&isc->v4l2_dev, "Could not find any possible format for a working pipeline\n");
>> -isc_try_fmt_subdev_err:
>> - memset(&isc->try_config, 0, sizeof(isc->try_config));
>> +static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
>> +{
>> + isc_try_fmt(isc, f);
>>
>> - return ret;
>> + /* make the try configuration active */
>> + isc->config = isc->try_config;
>> + isc->fmt = isc->try_fmt;
>> +
>> + v4l2_dbg(1, debug, &isc->v4l2_dev, "ISC set_fmt to %.4s @%dx%d\n",
>> + (char *)&f->fmt.pix.pixelformat,
>> + f->fmt.pix.width, f->fmt.pix.height);
>> +
>> + return 0;
>> }
>>
>> -static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
>> +static int isc_validate(struct isc_device *isc)
>> {
>> + int ret;
>> + int i;
>> + struct isc_format *sd_fmt = NULL;
>> + struct v4l2_pix_format *pixfmt = &isc->fmt.fmt.pix;
>> struct v4l2_subdev_format format = {
>> .which = V4L2_SUBDEV_FORMAT_ACTIVE,
>> + .pad = isc->remote_pad,
>> + };
>> + struct v4l2_subdev_pad_config pad_cfg = {};
>> + struct v4l2_subdev_state pad_state = {
>> + .pads = &pad_cfg,
>> };
>> - u32 mbus_code = 0;
>> - int ret;
>>
>> - ret = isc_try_fmt(isc, f, &mbus_code);
>> + /* Get current format from subdev */
>> + ret = v4l2_subdev_call(isc->current_subdev->sd, pad, get_fmt, NULL,
>> + &format);
>> if (ret)
>> return ret;
>>
>> - v4l2_fill_mbus_format(&format.format, &f->fmt.pix, mbus_code);
>> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad,
>> - set_fmt, NULL, &format);
>> - if (ret < 0)
>> - return ret;
>> + /* Identify the subdev's format configuration */
>> + for (i = 0; i < isc->num_user_formats; i++)
>> + if (isc->user_formats[i]->mbus_code == format.format.code) {
>> + sd_fmt = isc->user_formats[i];
>> + break;
>> + }
>> +
>> + /* Check if the format is not supported */
>> + if (!sd_fmt) {
>> + v4l2_err(&isc->v4l2_dev,
>> + "Current subdevice is streaming a media bus code that is not supported 0x%x\n",
>> + format.format.code);
>> + return -EPIPE;
>> + }
>> +
>> + /* At this moment we know which format the subdev will use */
>> + isc->try_config.sd_format = sd_fmt;
>> +
>> + /* If the sensor is not RAW, we can only do a direct dump */
>> + if (!ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
>> + isc_try_configure_rlp_dma(isc, true);
>>
>> /* Limit to Atmel ISC hardware capabilities */
>> - if (f->fmt.pix.width > isc->max_width)
>> - f->fmt.pix.width = isc->max_width;
>> - if (f->fmt.pix.height > isc->max_height)
>> - f->fmt.pix.height = isc->max_height;
>> + v4l_bound_align_image(&format.format.width, 16, isc->max_width, 0,
>> + &format.format.height, 16, isc->max_height, 0, 0);
>> +
>> + /* Check if the frame size is the same. Otherwise we may overflow */
>> + if (pixfmt->height != format.format.height ||
>> + pixfmt->width != format.format.width) {
>> + v4l2_err(&isc->v4l2_dev,
>> + "ISC not configured with the proper frame size: %dx%d\n",
>> + format.format.width, format.format.height);
>> + return -EPIPE;
>> + }
>>
>> - isc->fmt = *f;
>> + v4l2_dbg(1, debug, &isc->v4l2_dev,
>> + "Identified subdev using format %.4s with %dx%d %d bpp\n",
>> + (char *)&sd_fmt->fourcc, pixfmt->width, pixfmt->height,
>> + isc->try_config.bpp);
>>
>> + /* Reset and restart AWB if the subdevice changed the format */
>> if (isc->try_config.sd_format && isc->config.sd_format &&
>> isc->try_config.sd_format != isc->config.sd_format) {
>> isc->ctrls.hist_stat = HIST_INIT;
>> isc_reset_awb_ctrls(isc);
>> isc_update_v4l2_ctrls(isc);
>> }
>> - /* make the try configuration active */
>> +
>> + /* Validate formats */
>> + ret = isc_try_validate_formats(isc);
>> + if (ret)
>> + return ret;
>> +
>> + /* Obtain frame sizes if possible to have crop requirements ready */
>> + isc_try_fse(isc, &pad_state);
>> +
>> + /* Configure ISC pipeline for the config */
>> + ret = isc_try_configure_pipeline(isc);
>> + if (ret)
>> + return ret;
>> +
>> isc->config = isc->try_config;
>>
>> v4l2_dbg(1, debug, &isc->v4l2_dev, "New ISC configuration in place\n");
>> @@ -1040,7 +1051,7 @@ static int isc_try_fmt_vid_cap(struct file *file, void *priv,
>> {
>> struct isc_device *isc = video_drvdata(file);
>>
>> - return isc_try_fmt(isc, f, NULL);
>> + return isc_try_fmt(isc, f);
>> }
>>
>> static int isc_enum_input(struct file *file, void *priv,
>> @@ -1841,7 +1852,7 @@ static int isc_set_default_fmt(struct isc_device *isc)
>> };
>> int ret;
>>
>> - ret = isc_try_fmt(isc, &f, NULL);
>> + ret = isc_try_fmt(isc, &f);
>> if (ret)
>> return ret;
>>
>> @@ -2015,6 +2026,24 @@ int isc_pipeline_init(struct isc_device *isc)
>> }
>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>
>> +int isc_link_validate(struct media_link *link)
>> +{
>> + struct video_device *vdev =
>> + media_entity_to_video_device(link->sink->entity);
>> + struct isc_device *isc = video_get_drvdata(vdev);
>> + int ret;
>> +
>> + ret = v4l2_subdev_link_validate(link);
>> + if (ret)
>> + return ret;
>
> Doesn't v4l2_subdev_link_validate() call
> v4l2_subdev_link_validate_default() if you don't define a .link_validate()
> pad operation ? v4l2_subdev_link_validate_default() compares format
> and sizes of both ends to the links and want them to be identical,
> something that defeat the purpose of your custom isc_validate() ?

I think so. However if I replace the validate call with my custom one, I
thought that the old default should be called as well.

Meaning, that maybe my custom validation does certain isc-related
checks, while the v4l2_subdev_link_validate does exactly what you said.
So, by defining a custom link_validate, do I lose the checks that are
done by the v4l2_subdev_link_validate ?
That's what I wanted to avoid. Chip in my checks, but keep the v4l2
standard validation in place.

Does that make sense ?

> Thanks
> j
>
>> +
>> + return isc_validate(isc);
>> +}
>> +
>> +static const struct media_entity_operations isc_entity_operations = {
>> + .link_validate = isc_link_validate,
>> +};
>> +
> };
>> int isc_mc_init(struct isc_device *isc, u32 ver)
>> {
>> const struct of_device_id *match;
>> @@ -2022,6 +2051,8 @@ int isc_mc_init(struct isc_device *isc, u32 ver)
>>
>> isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>> isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>> + isc->video_dev.entity.ops = &isc_entity_operations;
>> +
>> isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>
>> ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>> index ec95c9665883..93375a61aea6 100644
>> --- a/drivers/media/platform/atmel/atmel-isc-scaler.c
>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>> @@ -155,6 +155,10 @@ static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>> .init_cfg = isc_scaler_init_cfg,
>> };
>>
>> +static const struct media_entity_operations isc_scaler_entity_ops = {
>> + .link_validate = v4l2_subdev_link_validate,
>> +};
>> +
>> static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>> .pad = &isc_scaler_pad_ops,
>> };
>> @@ -172,6 +176,7 @@ int isc_scaler_init(struct isc_device *isc)
>>
>> isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>> isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>> + isc->scaler_sd.entity.ops = &isc_scaler_entity_ops;
>> isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>> isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>>
>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>> index 04ea5e340808..e36cf9d3942a 100644
>> --- a/drivers/media/platform/atmel/atmel-isc.h
>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>> @@ -270,6 +270,7 @@ enum isc_scaler_pads {
>> * @formats_list_size: size of formats_list array
>> * @pads: media controller pads for isc video entity
>> * @mdev: media device that is registered by the isc
>> + * @mpipe: media device pipeline used by the isc
>> * @remote_pad: remote pad on the connected subdevice
>> * @scaler_sd: subdevice for the scaler that isc registers
>> * @scaler_pads: media controller pads for the scaler subdevice
>> @@ -295,6 +296,7 @@ struct isc_device {
>> struct completion comp;
>>
>> struct v4l2_format fmt;
>> + struct v4l2_format try_fmt;
>> struct isc_format **user_formats;
>> unsigned int num_user_formats;
>>
>> @@ -366,6 +368,7 @@ struct isc_device {
>> struct {
>> struct media_pad pads[ISC_PADS_NUM];
>> struct media_device mdev;
>> + struct media_pipeline mpipe;
>>
>> u32 remote_pad;
>> };
>> --
>> 2.25.1
>>

2022-02-11 13:07:03

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Fri, Feb 11, 2022 at 10:32:43AM +0000, [email protected] wrote:
> On 2/10/22 7:49 PM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Wed, Feb 09, 2022 at 06:59:37AM +0000, [email protected] wrote:
> >> On 2/8/22 9:30 PM, Jacopo Mondi wrote:
> >>> Hi Eugen
> >>>
> >>> On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
> >>>> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
> >>>>> Hi Eugen
> >>>>>
> >>>>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
> >>>>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> >>>>>>> Hi Eugen
> >>>>>>>
> >>>>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> >>>>>>>> Implement the support for media-controller.
> >>>>>>>> This means that the capabilities of the driver have changed and now
> >>>>>>>> it also advertises the IO_MC .
> >>>>>>>> The driver will register it's media device, and add the video entity to this
> >>>>>>>> media device. The subdevices are registered to the same media device.
> >>>>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> >>>>>>>> It will also register a subdevice that allows cropping of the incoming frame
> >>>>>>>> to the maximum frame size supported by the ISC.
> >>>>>>>> The ISC will create a link between the subdevice that is asynchronously
> >>>>>>>> registered and the atmel_isc_scaler entity.
> >>>>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> >>>>>>>> link.
> >>>>>>>>
> >>>>>>>> Signed-off-by: Eugen Hristev <[email protected]>
> >>>>>>>> ---
> >>>>>>>> Changes in v4:
> >>>>>>>> As suggested by Jacopo:
> >>>>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
> >>>>>>>> - moved init_mc/clean_mc to isc_base file
> >>>>>>>>
> >>>>>>>> Changes in v2:
> >>>>>>>> - implement try formats
> >>>>>>>>
> >>>>>>>> drivers/media/platform/atmel/Makefile | 2 +-
> >>>>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> >>>>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> >>>>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> >>>>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> >>>>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> >>>>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
> >>>>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>>
> >>>>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> >>>>>>>> index 794e8f739287..f02d03df89d6 100644
> >>>>>>>> --- a/drivers/media/platform/atmel/Makefile
> >>>>>>>> +++ b/drivers/media/platform/atmel/Makefile
> >>>>>>>> @@ -1,7 +1,7 @@
> >>>>>>>> # SPDX-License-Identifier: GPL-2.0-only
> >>>>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
> >>>>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
> >>>>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> >>>>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
> >>>>>>>>
> >>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> >>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> >>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>> index 6b0005987a17..6b482270eb93 100644
> >>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>>>> struct isc_device, v4l2_dev);
> >>>>>>>> struct isc_subdev_entity *subdev_entity =
> >>>>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
> >>>>>>>> + int pad;
> >>>>>>>>
> >>>>>>>> if (video_is_registered(&isc->video_dev)) {
> >>>>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> >>>>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>>>>
> >>>>>>>> subdev_entity->sd = subdev;
> >>>>>>>>
> >>>>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> >>>>>>>> + MEDIA_PAD_FL_SOURCE);
> >>>>>>>> + if (pad < 0) {
> >>>>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> >>>>>>>> + subdev->name);
> >>>>>>>> + return pad;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + isc->remote_pad = pad;
> >>>>>>>> +
> >>>>>>>> return 0;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> >>>>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>>>> - unsigned int code, int *index)
> >>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>>>> + unsigned int code, int *index)
> >>>>>>>> {
> >>>>>>>> struct isc_format *fmt = &isc->formats_list[0];
> >>>>>>>> unsigned int i;
> >>>>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>>>>
> >>>>>>>> return NULL;
> >>>>>>>> }
> >>>>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
> >>>>>>>>
> >>>>>>>> static int isc_formats_init(struct isc_device *isc)
> >>>>>>>> {
> >>>>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> >>>>>>>> NULL, &mbus_code)) {
> >>>>>>>> mbus_code.index++;
> >>>>>>>>
> >>>>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> >>>>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> >>>>>>>> if (!fmt) {
> >>>>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> >>>>>>>> mbus_code.code);
> >>>>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>>>> vdev->queue = q;
> >>>>>>>> vdev->lock = &isc->lock;
> >>>>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
> >>>>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> >>>>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> >>>>>>>> + V4L2_CAP_IO_MC;
> >>>>>>>> video_set_drvdata(vdev, isc);
> >>>>>>>>
> >>>>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> >>>>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>>>> goto isc_async_complete_err;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> + ret = isc_scaler_link(isc);
> >>>>>>>> + if (ret < 0)
> >>>>>>>> + goto isc_async_complete_unregister_device;
> >>>>>>>> +
> >>>>>>>> + ret = media_device_register(&isc->mdev);
> >>>>>>>> + if (ret < 0)
> >>>>>>>> + goto isc_async_complete_unregister_device;
> >>>>>>>> return 0;
> >>>>>>>>
> >>>>>>>> +isc_async_complete_unregister_device:
> >>>>>>>> + video_unregister_device(vdev);
> >>>>>>>> +
> >>>>>>>> isc_async_complete_err:
> >>>>>>>> mutex_destroy(&isc->lock);
> >>>>>>>> return ret;
> >>>>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> >>>>>>>> }
> >>>>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>>>>>>>
> >>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
> >>>>>>>> +{
> >>>>>>>> + const struct of_device_id *match;
> >>>>>>>> + int ret;
> >>>>>>>> +
> >>>>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >>>>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >>>>>>>
> >>>>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> >>>>>>> to be able to have the media pipeline validated at
> >>>>>>> media_pipeline_start() time ?
> >>>>>>
> >>>>>> Hi,
> >>>>>>
> >>>>>> I am doing that in a subsequent patch. Things are not completely ready
> >>>>>> at this moment, because ISC still relies on the old mechanism to call
> >>>>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
> >>>>>> removes and adds checks to the link validate ).
> >>>>>>
> >>>>>
> >>>>> I see.. the subsequent patches are not part of this series, right ?
> >>>>
> >>>> No. Patch 7 does this. In that patch, the link validation is created and
> >>>> all the logic of format propagation is removed and reworked, and creates
> >>>> the need for the link_validate call
> >>>> Could you have also a look at that patch ?
> >>>
> >>> Ah ups, 7 was not in my inbox with the rest of the series. Just
> >>> noticed. I had even reviewed the previous version, I should have
> >>> remembered link_validate was added later
> >>>
> >>>>>
> >>>>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
> >>>>> the introduction of media controller or the scaler, right ? (I'm not
> >>>>> sure if the DTS patches do).
> >>>
> >>> Do DTS patches depend on media-controller plumbing ?
> >>
> >> I think not, but, I have not tested them without the media controller.
> >>
> >>>
> >>>>>
> >>>>> If that's the case, would it make sense to to fast-track them (as some
> >>>>> of them are fixes) and then on top plumb the media controller
> >>>>> infrastructure with this patch and the ones you have been mentioning ?
> >>>>
> >>>> It would make sense.
> >>>>
> >>>
> >>> Once the other discussion with Hans about the stop state it might make
> >>> sense to fastrack the first part of the series ?
> >>
> >> The other three patches (4,5,6) are quite independent and can go faster.
> >>>
> >>>>>
> >>>>>>>
> >>>>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>>>> +
> >>>>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >>>>>>>> + isc->pads);
> >>>>>>>> + if (ret < 0) {
> >>>>>>>> + dev_err(isc->dev, "media entity init failed\n");
> >>>>>>>> + return ret;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + isc->mdev.dev = isc->dev;
> >>>>>>>> +
> >>>>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
> >>>>>>>> + isc->dev->of_node);
> >>>>>>>> +
> >>>>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> >>>>>>>> + sizeof(isc->mdev.driver_name));
> >>>>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> >>>>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> >>>>>>>> + isc->v4l2_dev.name);
> >>>>>>>> + isc->mdev.hw_revision = ver;
> >>>>>>>> +
> >>>>>>>> + media_device_init(&isc->mdev);
> >>>>>>>> +
> >>>>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
> >>>>>>>> +
> >>>>>>>> + return isc_scaler_init(isc);
> >>>>>>>> +}
> >>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
> >>>>>>>> +
> >>>>>>>> +void isc_mc_cleanup(struct isc_device *isc)
> >>>>>>>> +{
> >>>>>>>> + media_entity_cleanup(&isc->video_dev.entity);
> >>>>>>>> +}
> >>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> >>>>>>>> +
> >>>>>>>> /* regmap configuration */
> >>>>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
> >>>>>>>> const struct regmap_config isc_regmap_config = {
> >>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>> new file mode 100644
> >>>>>>>> index 000000000000..ec95c9665883
> >>>>>>>> --- /dev/null
> >>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>> @@ -0,0 +1,245 @@
> >>>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
> >>>>>>>> +/*
> >>>>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> >>>>>>>> + *
> >>>>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
> >>>>>>>
> >>>>>>> Time flies! It's 2022 already :)
> >>>>>>>
> >>>>>>>> + *
> >>>>>>>> + * Author: Eugen Hristev <[email protected]>
> >>>>>>>> + *
> >>>>>>>> + */
> >>>>>>>> +
> >>>>>>>> +#include <media/media-device.h>
> >>>>>>>> +#include <media/media-entity.h>
> >>>>>>>> +#include <media/v4l2-device.h>
> >>>>>>>> +#include <media/v4l2-subdev.h>
> >>>>>>>> +
> >>>>>>>> +#include "atmel-isc-regs.h"
> >>>>>>>> +#include "atmel-isc.h"
> >>>>>>>> +
> >>>>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> >>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>> + struct v4l2_subdev_format *format)
> >>>>>>>> +{
> >>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>>>> +
> >>>>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>> + format->pad);
> >>>>>>>> + format->format = *v4l2_try_fmt;
> >>>>>>>> +
> >>>>>>>> + return 0;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + format->format = isc->scaler_format;
> >>>>>>>
> >>>>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
> >>>>>>> I wonder why it lives in the isc_device struct.
> >>>>>>
> >>>>>> isc_device is a placeholder for all isc things.
> >>>>>>
> >>>>>> I would not create a separate struct in this file that would have to be
> >>>>>> allocated etc... just for one two things. So I preferred to have it in
> >>>>>> the same place as all the other things.
> >>>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + return 0;
> >>>>>>>> +}
> >>>>>>>> +
> >>>>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> >>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>> + struct v4l2_subdev_format *req_fmt)
> >>>>>>>> +{
> >>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>>>> + struct isc_format *fmt;
> >>>>>>>> + unsigned int i;
> >>>>>>>> +
> >>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>>>> + v4l_bound_align_image
> >>>>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
> >>>>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> >>>>>>>> + else
> >>>>>>>> + v4l_bound_align_image
> >>>>>>>> + (&req_fmt->format.width, 16, 10000, 0,
> >>>>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
> >>>>>>>
> >>>>>>> Where does 10000 come from ?
> >>>>>>
> >>>>>> It's a random number. Do you have any suggestion for a better one ?
> >>>>>
> >>>>> An actual HW limit ? :)
> >>>>
> >>>> There is no hardware limit. The ISC just stops sampling pixels once the
> >>>> crop limit is reached. The element that generates pixels could go on
> >>>> forever , or until the next HBLANK/VBLANK. So what limit could I place
> >>>> here ?
> >>>> The cropping is mandatory, otherwise there could be an overflow w.r.t.
> >>>> the buffer size if the ISC would sample more pixels than the software
> >>>> expects it to.
> >>>>
> >>>>>
> >>>>>> Maybe this would be much more clear with my comments below (where I will
> >>>>>> explain what the isc scaler does )
> >>>>>> In short, it allows the other entity in the link to 'sink' a huge format
> >>>>>> into this 'scaler' . Because the scaler can crop anything basically down
> >>>>>> to the biggest format size the ISC could handle.
> >>>>>>
> >>>>>
> >>>>> doesn't it have any documented input size limit ?
> >>>>
> >>>> As stated above, ISC handles a pixel stream and stores it in an internal
> >>>> SRAM. ISC stops sampling when this SRAM is full or when the cropping
> >>>> limit is reached.
> >>>> After storing the pixels in the SRAM (which is limited to the ISC
> >>>> maximum frame size), the ISC will work on the pixels and then DMA the
> >>>> frame to another RAM.
> >>>>
> >>>
> >>> I understand...
> >>>
> >>> So whatever the input size is, only the first n-th bytes are actually
> >>> dumped to memory, the following ones are discarded.
> >>
> >> Yes ! That's the behavior.
> >>>
> >>> So the crop is always applied on the top-leftmost corner as a
> >>> consequence, right ?
> >>
> >> Right. Pixels come in order. The ISC does not really know how many
> >> pixels are coming until its memory is full (or counting until the crop
> >> limit is reached). After that it stops sampling pixels, and waits for
> >> vblank to finish the frame.
> >>
> >>
> >>>
> >>> I understand why you had difficulties picking a default :)
> >>>
> >>> This is not easy to handle, as the scaler doesn't actually applies a
> >>> crop but rather stops capturing after the destination buffer has been
> >>> saturated. So the limiting factor is the full image size, not one of
> >>> its dimensions, like in example the line length. This makes it hard
> >>> to adjust cropping to something meaningful for most use cases.
> >>
> >> I didn't know how to name this except crop. It 'crops out' what does not
> >> fit in the internal memory of the ISC
> >>
> >>>
> >>> In your below example you suggest that an image size of (800000 x 10)
> >>> would theoretically be captured correctly. What if the input is of
> >>> size (800000 x 10 + 1). What size would you set the crop rectangle to ?
> >>>
> >>> I wouldn't however be too much concerned for the moment, but I would
> >>> record this in a comment ?
> >>
> >> I discussed a bit more with designer and looked in datasheet.
> >> It looks like the horizontal line length is not a hard limit, but it
> >> makes the ISC internal ISP stop working, so I would keep the horizontal
> >> limit as well for now, unless an important use case (I am coming back to
> >> that). The vertical limit looks to be a hard requirement.
> >> So in short, we have to impose a vertical limit of 2464 , we cannot have
> >> more lines than these. So if the horizontal is 100 pixels e.g., we can
> >> capture only 100x2464 .
> >> In practice we could capture [16,3264] on the horizontal, and [16,2464]
> >> on the vertical, and in theory horizontal could go more like [16, 4000],
> >> but we have to limit the vertical, and without the internal ISP.
> >> So I would not haphazardly go in this direction. Let's stick to the
> >> datasheet maximum frame 3264x2464 .
> >> According to this , I changed the frame size of the ISC to be a
> >> continuous size from [16,16] to [3264,2464]
> >> About that use case that I talked about earlier, we have one sensor ,
> >> imx274, which is more wide, and outputs something like 3800x2100 .
> >> With the cropping and limitations we have in place now, this will be
> >> limited to 3264x2100 , which is in line with datasheet and what the ISC
> >> supports now.
> >> As in theory we could capture 3800x2100 as long as the SRAM doesn't get
> >> full, for now I would like to avoid capturing this size (and disable the
> >> internal ISP), until we have someone really asking for such kind of
> >> capture and with some good reasons.
> >> In short I would like to have the crop rectangle to 3264x2464 and the
> >> bounds would be limited to the incoming frame but no H>3264 and no
> >> V>2464. Does that make sense ?
> >
> > Just to make sure I got it: is the ISC actually able to crop on the
> > line length ? If you apply a 3264 limit, will the last 536 pixels of
> > each line will be dropped if the input frame is 3800x2100 as your
> > above example ?
>
> Yes. Last 536 pixels will be dropped.
> After the 3264 limit, no more pixels are being sampled until HBLANK.
> >
> > Or rather, is the SRAM size limit the issue and the lines gets
> > captured in full lenght no matter what, until the buffer gets filled up
> > and all the successive pixels dropped.
> >
> > In other words, do your cropped image look like:
> >
> > 3264
> > +--------------------------------X-----------+ 3800
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> > +--------------------------------------------+ 2100
> >
> > Or rather
> >
> >
> > 3264
> > +--------------------------------X-----------+ 3800
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> > |xxxxxxxxxxxxxxxxx| |
> > | |
> > +--------------------------------------------+ 2100
> >
> > I assume the first, as otherwise you wouldn't be able to interpret the
> > image as 3264x2100.
>
> Yes, the first.
>
> >
> > I'm asking because you had the un-cropped 3280x2464 size as the ISC
> > source pad size in the paste of the media graph you provided below,
> > which might mean the output image is actually full size in lines
> > lenght and it's maybe missing pixels at the end, in case the ISC
> > operates as in the above second case.
>
> No, the image won't be padded with blanks on the right side. Actually
> this cannot happen due to a very big reason: the frame is being written
> to the main memory by a DMA controller. This DMA controller will write
> all the lines at consecutive addresses, and it will write exactly how
> many pixels are being captured by the module from the ISC named
> 'parallel front end' which does the cropping. Thus, the DMA will never
> output empty padding bytes at the end of each line, and it will not add
> a certain stride to each line if not instructed to. In fact the DMA can
> add a stride offset for the address of each line, which can be used for
> composing or copying the frame into another frame, but this is a
> different story. ( I tried this for fun, but not implemented in the
> driver at the moment)

Oh that will really be interesting and it's indeed an argument for
hainvg a scaler subdevice indeed

>
> The cropping is also mandatory even if the captured frame is less than
> the maximum size, because the ISC has no configuration related to frame
> size in fact. The parallel front end captures pixels and the dma engine
> will write them to memory. There is no control over this except the
> cropping.
> So for example if the software expects 640x480 but somehow someone sends
> more pixels, it can lead to buffer overflow if the parallel front end is
> not configured to ignore(crop) anything beyond 640x480 .
>

I see and I think you have correctly plumbed that with the scaler sink
pad crop rectangle. I don't seem to find where the crop rectangle
sizes are actually applied to the ISC in this series ? Could you point
me to it for sake of my better understanding ?

> With this being said, or having an understanding about the sink pad of
> the ISC, how should I configure the sink frame size ?

I don't think you should, the video device format is set through the
usual S_FMT as it happens already.

What I think it's missing in your implementation is only to propagate the
crop rectangle size of the scaler sink pad to the scaler source pad
format (as you don't have any composition rectangles yet) [1]

With that done, you should be able to capture from your video device
in the cropped dimensions.

The (simplified) media pipeline will look something like

sensor
type: Subdev
pad 0 (sink) (3280, 2462)
-> scaler:0

scaler
type: Subdev
pad 0 (sink) (3280, 2464)
crop: (3262, 2464)
<- sensor:0
pad 1 (source) (3264, 2464)
-> isc_common

isc_common
type V4L2
pad 0 (sink)
<- scaler:1

And the video device format set to (3264, 2464)

Then later you might want to implement s_selection on the scaler to have
the crop size configurable.

One additional question: can the crop be panned ? What I mean is that
you explained you can instruct the ISC to drop pixels after a certain
line lenght. This will cause the crop not to be centered on the input
image. Can the ISC be configured to ignore data -before- and -after-
a certain number of pixels or lines ? In that case you might want to
handle the top,left corner of the crop rectangle too. Anyway,
something for later..

[1] https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#order-of-configuration-and-format-propagation

> Is there any possibility to avoid limits on the frame size and have it
> arbitrary ?
> I am 99% sure that v4l2-compliance will not be happy if there is no
> limit on the frame on entity pads (hence, I bound-aligned the frame to
> 16x16 up to 10k x 10k )
>

That I'm not sure, but I think having size limits is required ?

Anyway, from my side, the next version should be good to go. Thanks
for sticking to all my comments and questions so far!

Thanks
j


> >
> >>
> >>>
> >>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
> >>>>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>>>> +
> >>>>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
> >>>>>>>
> >>>>>>> So you rely on the isc format list for the scaler as well ?
> >>>>>>> I think it's fine as long as they are identical
> >>>>>>
> >>>>>> Yes, the scaler is kind of a dummy entity , but it's required by the
> >>>>>> media controller validation of the pipeline.
> >>>>>>
> >>>>>
> >>>>> More on this later
> >>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + if (!fmt)
> >>>>>>>> + fmt = &isc->formats_list[0];
> >>>>>>>> +
> >>>>>>>> + req_fmt->format.code = fmt->mbus_code;
> >>>>>>>> +
> >>>>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>> + req_fmt->pad);
> >>>>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>>>> + /* Trying on the pad sink makes the source sink change too */
> >>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> >>>>>>>> + v4l2_try_fmt =
> >>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>>>> +
> >>>>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
> >>>>>>>> + 16, isc->max_width, 0,
> >>>>>>>> + &v4l2_try_fmt->height,
> >>>>>>>> + 16, isc->max_height, 0, 0);
> >>>>>>>> + }
> >>>>>>>> + /* if we are just trying, we are done */
> >>>>>>>> + return 0;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + isc->scaler_format = req_fmt->format;
> >>>>>>>
> >>>>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
> >>>>>>>
> >>>>>>> Actually, I would like to know more about the scaler device
> >>>>>>> capabilities. What functions can this IP perform ? Does it do
> >>>>>>> cropping, can it also do (down)scaling or even composition ?
> >>>>>>>
> >>>>>>> I think it is worth to read
> >>>>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> >>>>>>> where it is reported how cropping and scaling are implemented by using
> >>>>>>> the selection API.
> >>>>>>>
> >>>>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
> >>>>>>> for which you need to implement support to by adding s_selection on
> >>>>>>> the sink pad TGT_CROP target.
> >>>>>>
> >>>>>> The scaler is kind of a dummy entity.
> >>>>>> The problem is that different subdevices cannot be connected directly to
> >>>>>> a v4l entity, because the ISC does cropping on the incoming frame.
> >>>>>> The ISC has a limit on the total number of pixels per frame, thus it
> >>>>>> will do a crop.
> >>>>>
> >>>>> What is this limit related to ? I don't know the ISC internals, but is
> >>>>> the limitation due to the overall image size, or maybe it's due to
> >>>>> some line buffers being limited in the size they can transfer to
> >>>>> memory ? Is it a limitation on the size of the image in pixels, or is
> >>>>> it due to a limitation of the processing bandwidth on the input
> >>>>> interface and thus depends on the actual size of the image in bytes ?
> >>>>
> >>>> Limitation is both on line size and full image size. From my
> >>>> understanding, is that the internal SRAM is limited (8 Mpixels with a
> >>>> small safe buffer on top in the case of sama7g5 )
> >>>> This SRAM could be used as 800000x10 for example, or 3200x2464 for
> >>>> another example. However, I am not sure if a real line of 800,000 pixels
> >>>> make sense, or if there is another internal mechanism in the ISC that
> >>>> stops it.
> >>>> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
> >>>> the maximum frame that it can handle
> >>>>
> >>>>>
> >>>>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
> >>>>>> only handle 3264x2464 , then we have two choices:
> >>>>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
> >>>>>> 3280x2464, and the whole pipeline fails to configure, there is a
> >>>>>> mismatch between /dev/video output and what the whole pipeline produces
> >>>>>
> >>>>> I'm confused in first place. If the ISC cannot output anything larger
> >>>>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
> >>>>> this should be adjusted to 3264x2464.
> >>>>
> >>>> yes, that is true. However, without an entity that says 'I am doing
> >>>> cropping', the media pipeline will fail, because previous entity will
> >>>> output 3280x2464 .
> >>>> The conclusion to create a scaler entitity was done during my last year
> >>>> discussions on the IRC. I could not find a way to perform this cropping
> >>>> at the v4l entity side, and one reason is that the v4l entity does not
> >>>> support what I needed , or at least that was my understanding.
> >>>> You claim that the v4l entity could perform the cropping in the same
> >>>> entity ?
> >>>
> >>> I thought so, but I haven't tried to be honest
> >>>
> >>>>
> >>>> Let's assume for a minute that it could do this. Even so, I would prefer
> >>>> to have a separate scaler entity to do this , for several reasons:
> >>>> -> I would like to be able to crop the frame arbitrarily if I want to,
> >>>> and the scaler would allow me to do this easier
> >>>> -> it would be more obvious that the frame is cropped by having this
> >>>> entity there
> >>>> -> in fact some of the ISC versions really have a down scaler, that I
> >>>> have not implemented yet. So it would be useful to already have this in
> >>>> place to be able to scale down with it at a later time.
> >>>
> >>> I think this last reason is enough to have an entity plumbed in
> >>> already
> >>>
> >>>>
> >>>>>
> >>>>> Or is the ISC limitation on the -input- side ?
> >>>>>
> >>>>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
> >>>>>> the ISC can handle at most -> cannot , because the sensor let's say can
> >>>>>> *only* produce 3280x2464 and *any* other frame size returns an error.
> >>>>>
> >>>>> buggy sensor I would say :)
> >>>>
> >>>> I cannot tell this to the customers, and I cannot fail the whole
> >>>> pipeline because the sensor does not crop 16 pixels, when in fact I can
> >>>> do this very easily on the ISC side.
> >>>>
> >>>
> >>> Fair enough :)
> >>>
> >>>>>
> >>>>>>
> >>>>>> The only solution to make both worlds happy, is to have a dummy entity
> >>>>>> called 'scaler' which in fact now it only performs a simple crop.
> >>>>>> It accepts any frame size at input (hence the 10000x10000 which you saw
> >>>>>> earlier), and outputs at most 3264x2464 the isc_max_height and
> >>>>>> isc_max_width.
> >>>>>
> >>>>> I have a maybe dumb question: can the cropping operation be modeled
> >>>>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
> >>>>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
> >>>>
> >>>> I am not sure. This is what I wanted initially. But then I thought about
> >>>> the three reasons stated above for the use of the scaler entity, and
> >>>> discussed with folks on the IRC, and come up with the solution for the
> >>>> scaler entity
> >>>>
> >>>>>
> >>>>>>
> >>>>>> So to answer your question, the isc scaler is a software model for a
> >>>>>> simple cropping, that would make media controller happy, and capture
> >>>>>> software happy.
> >>>>>>
> >>>>>> Here it how it looks :
> >>>>>>
> >>>>>> Device topology
> >>>>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
> >>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>> device node name /dev/v4l-subdev0
> >>>>>> pad0: Sink
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>> crop.bounds:(0,0)/3264x2464
> >>>>>> crop:(0,0)/3264x2464]
> >>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>>>> pad1: Source
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>>>
> >>>>>> - entity 4: csi2dc (2 pads, 2 links)
> >>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>> device node name /dev/v4l-subdev1
> >>>>>> pad0: Sink
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>> <- "dw-csi.0":1 [ENABLED]
> >>>>>> pad1: Source
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
> >>>>>>
> >>>>>> - entity 7: dw-csi.0 (2 pads, 2 links)
> >>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>> device node name /dev/v4l-subdev2
> >>>>>> pad0: Sink
> >>>>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>>>> <- "imx219 1-0010":0 [ENABLED]
> >>>>>> pad1: Source
> >>>>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>>>> -> "csi2dc":0 [ENABLED]
> >>>>>>
> >>>>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
> >>>>>> type V4L2 subdev subtype Sensor flags 0
> >>>>>> device node name /dev/v4l-subdev3
> >>>>>> pad0: Source
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>> xfer:srgb ycbcr:601 quantization:full-range
> >>>>>> crop.bounds:(8,8)/3280x2464
> >>>>>> crop:(8,8)/3280x2464]
> >>>>>> -> "dw-csi.0":0 [ENABLED]
> >>>>>>
> >>>>>> - entity 24: atmel_isc_common (1 pad, 1 link)
> >>>>>> type Node subtype V4L flags 1
> >>>>>> device node name /dev/video0
> >>>>>> pad0: Sink
> >>>>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
> >>>>>>
> >>>>>>
> >>>>>> Scaler does this one cute little thing :
> >>>>>>
> >>>>>> pad0: Sink
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>> crop.bounds:(0,0)/3264x2464
> >>>>>> crop:(0,0)/3264x2464]
> >>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>>>> pad1: Source
> >>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>
> >>>>> Shouldn't this be 3264x2464 as that's what the entity outputs after
> >>>>> the cropping ? And shouldn't then be 3264x2464 the size output from
> >>>>> the video device too ?
> >>>>
> >>>> That's right.
> >>>> I don't know why the source format for the scaler is still 3280x2464.
> >>>> Maybe there is a bug on g_fmt for it.. have to check it.
> >>>
> >>> Thanks, I think this should be fixed ten
> >>>
> >>>
> >>>>
> >>>> Anyway, the video format is like this :
> >>>>
> >>>> # v4l2-ctl --get-fmt-video
> >>>> Format Video Capture:
> >>>> Width/Height : 3264/2464
> >>>> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
> >>>> Field : None
> >>>> Bytes per Line : 6528
> >>>> Size Image : 16084992
> >>>> Colorspace : sRGB
> >>>> Transfer Function : Default (maps to sRGB)
> >>>> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
> >>>> Quantization : Default (maps to Full Range)
> >>>> Flags :
> >>>> #
> >>>>
> >>>> and initializing the pipeline looks fine from user perspective.
> >>>> The scaler is in fact the solution to make this pipeline work with
> >>>> libcamera.
> >>>> I found this cropping to be an issue in media controller when trying it
> >>>> with libcamera. Otherwise, the other user space apps which I was using
> >>>> never complained that anything was wrong
> >>>> Libcamera simply refuses to acknowledge the pipeline if the video output
> >>>> is 3264x2464 but there is no entity that changes the format from
> >>>> 3280x2464 down
> >>>
> >>> Not sure why libcamera should play a role there. Isn't it the media
> >>> pipeline validation that complains and returns an -EPIPE ?
> >>
> >> I tried this before the media_pipeline_start .
> >> Perhaps libcamera was doing some similar checks, anyway, it was
> >> complaining and not adding the stream, and reported no usable streams/camera
> >>
> >> Once I will rework some of the things according to your review I will
> >> run libcamera again to see what it complains about
> >>
> >>>
> >>>>
> >>>>>
> >>>>>
> >>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>>>
> >>>>>>
> >>>>>> Which is what we needed.
> >>>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + return 0;
> >>>>>>>> +}
> >>>>>>>> +
> >>>>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> >>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>> + struct v4l2_subdev_mbus_code_enum *code)
> >>>>>>>> +{
> >>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>> + int supported_index = 0;
> >>>>>>>> + int i;
> >>>>>>>> +
> >>>>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
> >>>>>>>> + if (!isc->formats_list[i].sd_support)
> >>>>>>>> + continue;
> >>>>>>>
> >>>>>>> The sd_support flag still doesn't click in my head.
> >>>>>>>
> >>>>>>> Shouldn't the enumeration of available formats on the scaler do not
> >>>>>>> depend on the sensor supproted formats ?
> >>>>>>
> >>>>>> You're right. I will have to check it again.
> >>>>>>
> >>>>>>>
> >>>>>>>> + if (supported_index == code->index) {
> >>>>>>>> + code->code = isc->formats_list[i].mbus_code;
> >>>>>>>> + return 0;
> >>>>>>>> + }
> >>>>>>>> + supported_index++;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + return -EINVAL;
> >>>>>>>> +}
> >>>>>>>> +
> >>>>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> >>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>> + struct v4l2_subdev_selection *sel)
> >>>>>>>> +{
> >>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>> +
> >>>>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>>>> + return -EINVAL;
> >>>>>>>> +
> >>>>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> >>>>>>>> + sel->target != V4L2_SEL_TGT_CROP)
> >>>>>>>> + return -EINVAL;
> >>>>>>>> +
> >>>>>>>> + sel->r.height = isc->max_height;
> >>>>>>>> + sel->r.width = isc->max_width;
> >>>>>>>
> >>>>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
> >>>>>>> as it represents the maximum valid crop rectangle.
> >>>>>>>
> >>>>>>> TGT_CROP should report the configured crop rectangle which can be
> >>>>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
> >>>>>>> the spec is correct
> >>>>>>
> >>>>>> So you would like to have this differentiated, and report the
> >>>>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
> >>>>>> reported now, the maximum size of the ISC frame .
> >>>>>> My understanding is correct ?
> >>>>>>
> >>>>>
> >>>>> I didn't know you have an HW limitation, so your _BOUNDS is not the
> >>>>> input image size but rather 3264x2464 ( == max_width x max_height).
> >>>>>
> >>>>> What I meant is that _BOUNDS should report the maximum rectangle size
> >>>>> that can be applied to the _CROP target. In you case you have an HW
> >>>>> limitation 3264x2464 and that's the largest rectangle you can apply.
> >>>> So the CROP should be at 3264x2464
> >>>>> TGT_CROP can be initialized to the same as _BOUND, but if you
> >>>>> implement s_selection it should report what has been there applied.
> >>>> and BOUND to actual frame size ?
> >>>>> But as you don't implement s_selection yet, I think this is fine for
> >>>>> now. Maybe a little comment ?
> >>>>
> >>>> It could also be the case where e.g. the sensor is outputting 1920x1080,
> >>>> in this case the scaler would do nothing.
> >>>> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
> >>>> If the sensor is outputting 1920x1080, this format comes directly from
> >>>> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
> >>>> sensor's problem)
> >>>>>
> >>>>> Also, is set->r zeroed by the framework before getting here ?
> >>>>> Otherwise you should set r.left and r.top to 0
> >>>>
> >>>> If these are redundant, no problem to remove them
> >>>
> >>> I was actually confused. I was suggesting you to add...
> >>>
> >>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + sel->r.left = 0;
> >>>>>>>> + sel->r.top = 0;
> >>>
> >>> These ^
> >>>
> >>> So nothing to change. Sorry I've missed them
> >>>
> >>> Thanks
> >>> j
> >>>
> >>>
> >>>>>>>> +
> >>>>>>>> + return 0;
> >>>>>>>> +}
> >>>>>>>> +
> >>>>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> >>>>>>>> + struct v4l2_subdev_state *sd_state)
> >>>>>>>> +{
> >>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> >>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> >>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>> +
> >>>>>>>> + *v4l2_try_fmt = isc->scaler_format;
> >>>>>>>> +
> >>>>>>>> + return 0;
> >>>>>>>> +}
> >>>>>>>> +
> >>>>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >>>>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> >>>>>>>> + .set_fmt = isc_scaler_set_fmt,
> >>>>>>>> + .get_fmt = isc_scaler_get_fmt,
> >>>>>>>> + .get_selection = isc_scaler_g_sel,
> >>>>>>>> + .init_cfg = isc_scaler_init_cfg,
> >>>>>>>
> >>>>>>> .link_validate = v4l2_subdev_link_validate_default,
> >>>>>>>
> >>>>>>> To have the formats at the end of links that point to this entity
> >>>>>>> validated (I think the framework already calls it if not set though,
> >>>>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
> >>>>>>>
> >>>>>>>> +};
> >>>>>>>> +
> >>>>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >>>>>>>> + .pad = &isc_scaler_pad_ops,
> >>>>>>>> +};
> >>>>>>>> +
> >>>>>>>> +int isc_scaler_init(struct isc_device *isc)
> >>>>>>>> +{
> >>>>>>>> + int ret;
> >>>>>>>> +
> >>>>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> >>>>>>>> +
> >>>>>>>> + isc->scaler_sd.owner = THIS_MODULE;
> >>>>>>>> + isc->scaler_sd.dev = isc->dev;
> >>>>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> >>>>>>>> + "atmel_isc_scaler");
> >>>>>>>
> >>>>>>> I would drop 'atmel' for brevity, unless other entities have this
> >>>>>>> prefix set already
> >>>>>>
> >>>>>> The v4l entity takes it's name from the module name, which is
> >>>>>> atmel_isc_common, so I thought to keep the prefix
> >>>>>>
> >>>>>
> >>>>> Ack!
> >>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >>>>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >>>>>>>> +
> >>>>>>>> + isc->scaler_format.height = isc->max_height;
> >>>>>>>> + isc->scaler_format.width = isc->max_width;
> >>>>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> >>>>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
> >>>>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>>>> +
> >>>>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> >>>>>>>> + ISC_SCALER_PADS_NUM,
> >>>>>>>> + isc->scaler_pads);
> >>>>>>>> + if (ret < 0) {
> >>>>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> >>>>>>>> + return ret;
> >>>>>>>> + }
> >>>>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> >>>>>>>> + if (ret < 0) {
> >>>>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> >>>>>>>> + return ret;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + return ret;
> >>>>>>>> +}
> >>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> >>>>>>>> +
> >>>>>>>> +int isc_scaler_link(struct isc_device *isc)
> >>>>>>>> +{
> >>>>>>>> + int ret;
> >>>>>>>> +
> >>>>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> >>>>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
> >>>>>>>> + ISC_SCALER_PAD_SINK,
> >>>>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>>>> +
> >>>>>>>> + if (ret < 0) {
> >>>>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>>>> + isc->current_subdev->sd->entity.name,
> >>>>>>>> + isc->scaler_sd.entity.name);
> >>>>>>>> + return ret;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> >>>>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
> >>>>>>>> +
> >>>>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> >>>>>>>> + ISC_SCALER_PAD_SOURCE,
> >>>>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
> >>>>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>>>> +
> >>>>>>>> + if (ret < 0) {
> >>>>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>>>> + isc->scaler_sd.entity.name,
> >>>>>>>> + isc->video_dev.entity.name);
> >>>>>>>> + return ret;
> >>>>>>>> + }
> >>>>>>>> +
> >>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> >>>>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>>>> +
> >>>>>>>> + return ret;
> >>>>>>>> +}
> >>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
> >>>>>>>
> >>>>>>> From the DT graph point of view, the ISC appears as a single block
> >>>>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
> >>>>>>> add to its notifier the image sensor remote subdevice.
> >>>>>>
> >>>>>> Actually it's only a parallel port.
> >>>>>> And it can be connected either to a sensor directly or to the csi2dc
> >>>>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
> >>>>>>
> >>>>>>>
> >>>>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
> >>>>>>> entity which was initialized at isc_probe() time is linked in between
> >>>>>>> the ISC and the image sensor immutably.
> >>>>>>
> >>>>>> yes, because this cannot change at runtime, and usually, it can't change
> >>>>>> without altering the board hardware. (at least this is what my
> >>>>>> understanding of immutability is )
> >>>>>>
> >>>>>>>
> >>>>>>> I think it is fine for now, but I wonder if you plan to plumb more
> >>>>>>> components between the ISC video node and the sensor, if it's not
> >>>>>>> worth changing the DT bindings and their parsing logic to separate the
> >>>>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
> >>>>>>> time and have the CSI-2 receiver entity as downstream remote. I think
> >>>>>>> I need to get to know the ISC better to really have an idea. For now,
> >>>>>>> this seems ok to me, but please check with maintainers if this is fine
> >>>>>>> with them.
> >>>>>>
> >>>>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
> >>>>>> the remote subdev).
> >>>>>> The ISC will register a scaler, and connect the subdev to this scaler
> >>>>>> first, and then, the scaler to the isc itself (the v4l entity).
> >>>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>> index 5fbf52a9080b..c9234c90ae58 100644
> >>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> >>>>>>>> u32 his_entry;
> >>>>>>>> };
> >>>>>>>>
> >>>>>>>> +enum isc_mc_pads {
> >>>>>>>> + ISC_PAD_SINK = 0,
> >>>>>>>> + ISC_PADS_NUM = 1,
> >>>>>>>> +};
> >>>>>>>> +
> >>>>>>>> +enum isc_scaler_pads {
> >>>>>>>> + ISC_SCALER_PAD_SINK = 0,
> >>>>>>>> + ISC_SCALER_PAD_SOURCE = 1,
> >>>>>>>> + ISC_SCALER_PADS_NUM = 2,
> >>>>>>>> +};
> >>>>>>>> +
> >>>>>>>> /*
> >>>>>>>> * struct isc_device - ISC device driver data/config struct
> >>>>>>>> * @regmap: Register map
> >>>>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> >>>>>>>> * be used as an input to the controller
> >>>>>>>> * @controller_formats_size: size of controller_formats array
> >>>>>>>> * @formats_list_size: size of formats_list array
> >>>>>>>> + * @pads: media controller pads for isc video entity
> >>>>>>>> + * @mdev: media device that is registered by the isc
> >>>>>>>> + * @remote_pad: remote pad on the connected subdevice
> >>>>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
> >>>>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
> >>>>>>>> + * @scaler_format: current format for the scaler subdevice
> >>>>>>>> */
> >>>>>>>> struct isc_device {
> >>>>>>>> struct regmap *regmap;
> >>>>>>>> @@ -344,6 +361,19 @@ struct isc_device {
> >>>>>>>> struct isc_format *formats_list;
> >>>>>>>> u32 controller_formats_size;
> >>>>>>>> u32 formats_list_size;
> >>>>>>>> +
> >>>>>>>> + struct {
> >>>>>>>> + struct media_pad pads[ISC_PADS_NUM];
> >>>>>>>> + struct media_device mdev;
> >>>>>>>> +
> >>>>>>>> + u32 remote_pad;
> >>>>>>>> + };
> >>>>>>>> +
> >>>>>>>> + struct {
> >>>>>>>> + struct v4l2_subdev scaler_sd;
> >>>>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> >>>>>>>> + struct v4l2_mbus_framefmt scaler_format;
> >>>>>>
> >>>>>> Here are the scaler stuff which I added, in the same bulk struct for the
> >>>>>> whole isc device
> >>>>>>
> >>>>>>
> >>>>>>>> + };
> >>>>>>>> };
> >>>>>>>>
> >>>>>>>> extern const struct regmap_config isc_regmap_config;
> >>>>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> >>>>>>>> void isc_subdev_cleanup(struct isc_device *isc);
> >>>>>>>> void isc_clk_cleanup(struct isc_device *isc);
> >>>>>>>>
> >>>>>>>> +int isc_scaler_link(struct isc_device *isc);
> >>>>>>>> +int isc_scaler_init(struct isc_device *isc);
> >>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
> >>>>>>>> +void isc_mc_cleanup(struct isc_device *isc);
> >>>>>>>> +
> >>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>>>> + unsigned int code, int *index);
> >>>>>>>> #endif
> >>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>> index c5b9563e36cb..c244682ea22f 100644
> >>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>> break;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>
> >>>>>>> I am surprised you can read a register before runtime_pm is
> >>>>>>> intialized!
> >>>>>>>
> >>>>>>
> >>>>>> Actually I can read any register after the moment of starting the clock
> >>>>>> on the hardware block.
> >>>>>
> >>>>> Ah right then
> >>>>>
> >>>>>> Maybe the starting and stopping of the clock needs to be moved to the
> >>>>>> runtime_pm calls, but this is another story, not related to the media
> >>>>>> controller.
> >>>>>
> >>>>> It's fine, but maybe worth recording with a todo ?
> >>>>>
> >>>>> Thanks
> >>>>> j
> >>>>>
> >>>>>> I moved this line because I had to pass the version to the isc_mc_init call
> >>>>>>
> >>>>>>> Thanks
> >>>>>>> j
> >>>>>>
> >>>>>> Thanks for reviewing !
> >>>>>> Eugen
> >>>>>>>
> >>>>>>>
> >>>>>>>> +
> >>>>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>>>> + if (ret < 0)
> >>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>> +
> >>>>>>>> pm_runtime_set_active(dev);
> >>>>>>>> pm_runtime_enable(dev);
> >>>>>>>> pm_request_idle(dev);
> >>>>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>> ret = clk_prepare_enable(isc->ispck);
> >>>>>>>> if (ret) {
> >>>>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
> >>>>>>>> - goto cleanup_subdev;
> >>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> /* ispck should be greater or equal to hclock */
> >>>>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>> goto unprepare_clk;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
> >>>>>>>>
> >>>>>>>> return 0;
> >>>>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>> unprepare_clk:
> >>>>>>>> clk_disable_unprepare(isc->ispck);
> >>>>>>>>
> >>>>>>>> +isc_probe_mc_init_err:
> >>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>> +
> >>>>>>>> cleanup_subdev:
> >>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>
> >>>>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
> >>>>>>>>
> >>>>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>>>
> >>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>> +
> >>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>
> >>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>> index 07a80b08bc54..9dc75eed0098 100644
> >>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> >>>>>>>> break;
> >>>>>>>> }
> >>>>>>>>
> >>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>> +
> >>>>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>>>> + if (ret < 0)
> >>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>> +
> >>>>>>>> pm_runtime_set_active(dev);
> >>>>>>>> pm_runtime_enable(dev);
> >>>>>>>> pm_request_idle(dev);
> >>>>>>>>
> >>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
> >>>>>>>>
> >>>>>>>> return 0;
> >>>>>>>>
> >>>>>>>> +isc_probe_mc_init_err:
> >>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>> +
> >>>>>>>> cleanup_subdev:
> >>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>
> >>>>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
> >>>>>>>>
> >>>>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>>>
> >>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>> +
> >>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>
> >>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>>>> --
> >>>>>>>> 2.25.1
> >>>>>>>>
> >>>>>>
> >>>>
> >>
>

2022-02-11 14:58:15

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

On 2/11/22 1:55 PM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Fri, Feb 11, 2022 at 10:32:43AM +0000, [email protected] wrote:
>> On 2/10/22 7:49 PM, Jacopo Mondi wrote:
>>> Hi Eugen
>>>
>>> On Wed, Feb 09, 2022 at 06:59:37AM +0000, [email protected] wrote:
>>>> On 2/8/22 9:30 PM, Jacopo Mondi wrote:
>>>>> Hi Eugen
>>>>>
>>>>> On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
>>>>>> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
>>>>>>> Hi Eugen
>>>>>>>
>>>>>>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
>>>>>>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
>>>>>>>>> Hi Eugen
>>>>>>>>>
>>>>>>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
>>>>>>>>>> Implement the support for media-controller.
>>>>>>>>>> This means that the capabilities of the driver have changed and now
>>>>>>>>>> it also advertises the IO_MC .
>>>>>>>>>> The driver will register it's media device, and add the video entity to this
>>>>>>>>>> media device. The subdevices are registered to the same media device.
>>>>>>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
>>>>>>>>>> It will also register a subdevice that allows cropping of the incoming frame
>>>>>>>>>> to the maximum frame size supported by the ISC.
>>>>>>>>>> The ISC will create a link between the subdevice that is asynchronously
>>>>>>>>>> registered and the atmel_isc_scaler entity.
>>>>>>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
>>>>>>>>>> link.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Eugen Hristev <[email protected]>
>>>>>>>>>> ---
>>>>>>>>>> Changes in v4:
>>>>>>>>>> As suggested by Jacopo:
>>>>>>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
>>>>>>>>>> - moved init_mc/clean_mc to isc_base file
>>>>>>>>>>
>>>>>>>>>> Changes in v2:
>>>>>>>>>> - implement try formats
>>>>>>>>>>
>>>>>>>>>> drivers/media/platform/atmel/Makefile | 2 +-
>>>>>>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
>>>>>>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
>>>>>>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
>>>>>>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
>>>>>>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
>>>>>>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
>>>>>>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
>>>>>>>>>> index 794e8f739287..f02d03df89d6 100644
>>>>>>>>>> --- a/drivers/media/platform/atmel/Makefile
>>>>>>>>>> +++ b/drivers/media/platform/atmel/Makefile
>>>>>>>>>> @@ -1,7 +1,7 @@
>>>>>>>>>> # SPDX-License-Identifier: GPL-2.0-only
>>>>>>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
>>>>>>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
>>>>>>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
>>>>>>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>>>>>>>>>>
>>>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
>>>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>>>> index 6b0005987a17..6b482270eb93 100644
>>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>>>>>> struct isc_device, v4l2_dev);
>>>>>>>>>> struct isc_subdev_entity *subdev_entity =
>>>>>>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
>>>>>>>>>> + int pad;
>>>>>>>>>>
>>>>>>>>>> if (video_is_registered(&isc->video_dev)) {
>>>>>>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
>>>>>>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>>>>>>
>>>>>>>>>> subdev_entity->sd = subdev;
>>>>>>>>>>
>>>>>>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
>>>>>>>>>> + MEDIA_PAD_FL_SOURCE);
>>>>>>>>>> + if (pad < 0) {
>>>>>>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
>>>>>>>>>> + subdev->name);
>>>>>>>>>> + return pad;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + isc->remote_pad = pad;
>>>>>>>>>> +
>>>>>>>>>> return 0;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
>>>>>>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>>>>>> - unsigned int code, int *index)
>>>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>>>>>> + unsigned int code, int *index)
>>>>>>>>>> {
>>>>>>>>>> struct isc_format *fmt = &isc->formats_list[0];
>>>>>>>>>> unsigned int i;
>>>>>>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>>>>>>
>>>>>>>>>> return NULL;
>>>>>>>>>> }
>>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>>>>>>>>>>
>>>>>>>>>> static int isc_formats_init(struct isc_device *isc)
>>>>>>>>>> {
>>>>>>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
>>>>>>>>>> NULL, &mbus_code)) {
>>>>>>>>>> mbus_code.index++;
>>>>>>>>>>
>>>>>>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
>>>>>>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
>>>>>>>>>> if (!fmt) {
>>>>>>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
>>>>>>>>>> mbus_code.code);
>>>>>>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>>>>>> vdev->queue = q;
>>>>>>>>>> vdev->lock = &isc->lock;
>>>>>>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
>>>>>>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
>>>>>>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
>>>>>>>>>> + V4L2_CAP_IO_MC;
>>>>>>>>>> video_set_drvdata(vdev, isc);
>>>>>>>>>>
>>>>>>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
>>>>>>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>>>>>> goto isc_async_complete_err;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> + ret = isc_scaler_link(isc);
>>>>>>>>>> + if (ret < 0)
>>>>>>>>>> + goto isc_async_complete_unregister_device;
>>>>>>>>>> +
>>>>>>>>>> + ret = media_device_register(&isc->mdev);
>>>>>>>>>> + if (ret < 0)
>>>>>>>>>> + goto isc_async_complete_unregister_device;
>>>>>>>>>> return 0;
>>>>>>>>>>
>>>>>>>>>> +isc_async_complete_unregister_device:
>>>>>>>>>> + video_unregister_device(vdev);
>>>>>>>>>> +
>>>>>>>>>> isc_async_complete_err:
>>>>>>>>>> mutex_destroy(&isc->lock);
>>>>>>>>>> return ret;
>>>>>>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
>>>>>>>>>> }
>>>>>>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>>>>>>>>>
>>>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
>>>>>>>>>> +{
>>>>>>>>>> + const struct of_device_id *match;
>>>>>>>>>> + int ret;
>>>>>>>>>> +
>>>>>>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>>>>>>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>>>>>>>>>
>>>>>>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
>>>>>>>>> to be able to have the media pipeline validated at
>>>>>>>>> media_pipeline_start() time ?
>>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> I am doing that in a subsequent patch. Things are not completely ready
>>>>>>>> at this moment, because ISC still relies on the old mechanism to call
>>>>>>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
>>>>>>>> removes and adds checks to the link validate ).
>>>>>>>>
>>>>>>>
>>>>>>> I see.. the subsequent patches are not part of this series, right ?
>>>>>>
>>>>>> No. Patch 7 does this. In that patch, the link validation is created and
>>>>>> all the logic of format propagation is removed and reworked, and creates
>>>>>> the need for the link_validate call
>>>>>> Could you have also a look at that patch ?
>>>>>
>>>>> Ah ups, 7 was not in my inbox with the rest of the series. Just
>>>>> noticed. I had even reviewed the previous version, I should have
>>>>> remembered link_validate was added later
>>>>>
>>>>>>>
>>>>>>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
>>>>>>> the introduction of media controller or the scaler, right ? (I'm not
>>>>>>> sure if the DTS patches do).
>>>>>
>>>>> Do DTS patches depend on media-controller plumbing ?
>>>>
>>>> I think not, but, I have not tested them without the media controller.
>>>>
>>>>>
>>>>>>>
>>>>>>> If that's the case, would it make sense to to fast-track them (as some
>>>>>>> of them are fixes) and then on top plumb the media controller
>>>>>>> infrastructure with this patch and the ones you have been mentioning ?
>>>>>>
>>>>>> It would make sense.
>>>>>>
>>>>>
>>>>> Once the other discussion with Hans about the stop state it might make
>>>>> sense to fastrack the first part of the series ?
>>>>
>>>> The other three patches (4,5,6) are quite independent and can go faster.
>>>>>
>>>>>>>
>>>>>>>>>
>>>>>>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>>>>>> +
>>>>>>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>>>>>>>>>> + isc->pads);
>>>>>>>>>> + if (ret < 0) {
>>>>>>>>>> + dev_err(isc->dev, "media entity init failed\n");
>>>>>>>>>> + return ret;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + isc->mdev.dev = isc->dev;
>>>>>>>>>> +
>>>>>>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
>>>>>>>>>> + isc->dev->of_node);
>>>>>>>>>> +
>>>>>>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
>>>>>>>>>> + sizeof(isc->mdev.driver_name));
>>>>>>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
>>>>>>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
>>>>>>>>>> + isc->v4l2_dev.name);
>>>>>>>>>> + isc->mdev.hw_revision = ver;
>>>>>>>>>> +
>>>>>>>>>> + media_device_init(&isc->mdev);
>>>>>>>>>> +
>>>>>>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
>>>>>>>>>> +
>>>>>>>>>> + return isc_scaler_init(isc);
>>>>>>>>>> +}
>>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
>>>>>>>>>> +
>>>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc)
>>>>>>>>>> +{
>>>>>>>>>> + media_entity_cleanup(&isc->video_dev.entity);
>>>>>>>>>> +}
>>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
>>>>>>>>>> +
>>>>>>>>>> /* regmap configuration */
>>>>>>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
>>>>>>>>>> const struct regmap_config isc_regmap_config = {
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>>>> new file mode 100644
>>>>>>>>>> index 000000000000..ec95c9665883
>>>>>>>>>> --- /dev/null
>>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>>>> @@ -0,0 +1,245 @@
>>>>>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
>>>>>>>>>> +/*
>>>>>>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
>>>>>>>>>> + *
>>>>>>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
>>>>>>>>>
>>>>>>>>> Time flies! It's 2022 already :)
>>>>>>>>>
>>>>>>>>>> + *
>>>>>>>>>> + * Author: Eugen Hristev <[email protected]>
>>>>>>>>>> + *
>>>>>>>>>> + */
>>>>>>>>>> +
>>>>>>>>>> +#include <media/media-device.h>
>>>>>>>>>> +#include <media/media-entity.h>
>>>>>>>>>> +#include <media/v4l2-device.h>
>>>>>>>>>> +#include <media/v4l2-subdev.h>
>>>>>>>>>> +
>>>>>>>>>> +#include "atmel-isc-regs.h"
>>>>>>>>>> +#include "atmel-isc.h"
>>>>>>>>>> +
>>>>>>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
>>>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>>>> + struct v4l2_subdev_format *format)
>>>>>>>>>> +{
>>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>>>>>> +
>>>>>>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>>>> + format->pad);
>>>>>>>>>> + format->format = *v4l2_try_fmt;
>>>>>>>>>> +
>>>>>>>>>> + return 0;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + format->format = isc->scaler_format;
>>>>>>>>>
>>>>>>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
>>>>>>>>> I wonder why it lives in the isc_device struct.
>>>>>>>>
>>>>>>>> isc_device is a placeholder for all isc things.
>>>>>>>>
>>>>>>>> I would not create a separate struct in this file that would have to be
>>>>>>>> allocated etc... just for one two things. So I preferred to have it in
>>>>>>>> the same place as all the other things.
>>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + return 0;
>>>>>>>>>> +}
>>>>>>>>>> +
>>>>>>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
>>>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>>>> + struct v4l2_subdev_format *req_fmt)
>>>>>>>>>> +{
>>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>>>>>> + struct isc_format *fmt;
>>>>>>>>>> + unsigned int i;
>>>>>>>>>> +
>>>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
>>>>>>>>>> + v4l_bound_align_image
>>>>>>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
>>>>>>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
>>>>>>>>>> + else
>>>>>>>>>> + v4l_bound_align_image
>>>>>>>>>> + (&req_fmt->format.width, 16, 10000, 0,
>>>>>>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
>>>>>>>>>
>>>>>>>>> Where does 10000 come from ?
>>>>>>>>
>>>>>>>> It's a random number. Do you have any suggestion for a better one ?
>>>>>>>
>>>>>>> An actual HW limit ? :)
>>>>>>
>>>>>> There is no hardware limit. The ISC just stops sampling pixels once the
>>>>>> crop limit is reached. The element that generates pixels could go on
>>>>>> forever , or until the next HBLANK/VBLANK. So what limit could I place
>>>>>> here ?
>>>>>> The cropping is mandatory, otherwise there could be an overflow w.r.t.
>>>>>> the buffer size if the ISC would sample more pixels than the software
>>>>>> expects it to.
>>>>>>
>>>>>>>
>>>>>>>> Maybe this would be much more clear with my comments below (where I will
>>>>>>>> explain what the isc scaler does )
>>>>>>>> In short, it allows the other entity in the link to 'sink' a huge format
>>>>>>>> into this 'scaler' . Because the scaler can crop anything basically down
>>>>>>>> to the biggest format size the ISC could handle.
>>>>>>>>
>>>>>>>
>>>>>>> doesn't it have any documented input size limit ?
>>>>>>
>>>>>> As stated above, ISC handles a pixel stream and stores it in an internal
>>>>>> SRAM. ISC stops sampling when this SRAM is full or when the cropping
>>>>>> limit is reached.
>>>>>> After storing the pixels in the SRAM (which is limited to the ISC
>>>>>> maximum frame size), the ISC will work on the pixels and then DMA the
>>>>>> frame to another RAM.
>>>>>>
>>>>>
>>>>> I understand...
>>>>>
>>>>> So whatever the input size is, only the first n-th bytes are actually
>>>>> dumped to memory, the following ones are discarded.
>>>>
>>>> Yes ! That's the behavior.
>>>>>
>>>>> So the crop is always applied on the top-leftmost corner as a
>>>>> consequence, right ?
>>>>
>>>> Right. Pixels come in order. The ISC does not really know how many
>>>> pixels are coming until its memory is full (or counting until the crop
>>>> limit is reached). After that it stops sampling pixels, and waits for
>>>> vblank to finish the frame.
>>>>
>>>>
>>>>>
>>>>> I understand why you had difficulties picking a default :)
>>>>>
>>>>> This is not easy to handle, as the scaler doesn't actually applies a
>>>>> crop but rather stops capturing after the destination buffer has been
>>>>> saturated. So the limiting factor is the full image size, not one of
>>>>> its dimensions, like in example the line length. This makes it hard
>>>>> to adjust cropping to something meaningful for most use cases.
>>>>
>>>> I didn't know how to name this except crop. It 'crops out' what does not
>>>> fit in the internal memory of the ISC
>>>>
>>>>>
>>>>> In your below example you suggest that an image size of (800000 x 10)
>>>>> would theoretically be captured correctly. What if the input is of
>>>>> size (800000 x 10 + 1). What size would you set the crop rectangle to ?
>>>>>
>>>>> I wouldn't however be too much concerned for the moment, but I would
>>>>> record this in a comment ?
>>>>
>>>> I discussed a bit more with designer and looked in datasheet.
>>>> It looks like the horizontal line length is not a hard limit, but it
>>>> makes the ISC internal ISP stop working, so I would keep the horizontal
>>>> limit as well for now, unless an important use case (I am coming back to
>>>> that). The vertical limit looks to be a hard requirement.
>>>> So in short, we have to impose a vertical limit of 2464 , we cannot have
>>>> more lines than these. So if the horizontal is 100 pixels e.g., we can
>>>> capture only 100x2464 .
>>>> In practice we could capture [16,3264] on the horizontal, and [16,2464]
>>>> on the vertical, and in theory horizontal could go more like [16, 4000],
>>>> but we have to limit the vertical, and without the internal ISP.
>>>> So I would not haphazardly go in this direction. Let's stick to the
>>>> datasheet maximum frame 3264x2464 .
>>>> According to this , I changed the frame size of the ISC to be a
>>>> continuous size from [16,16] to [3264,2464]
>>>> About that use case that I talked about earlier, we have one sensor ,
>>>> imx274, which is more wide, and outputs something like 3800x2100 .
>>>> With the cropping and limitations we have in place now, this will be
>>>> limited to 3264x2100 , which is in line with datasheet and what the ISC
>>>> supports now.
>>>> As in theory we could capture 3800x2100 as long as the SRAM doesn't get
>>>> full, for now I would like to avoid capturing this size (and disable the
>>>> internal ISP), until we have someone really asking for such kind of
>>>> capture and with some good reasons.
>>>> In short I would like to have the crop rectangle to 3264x2464 and the
>>>> bounds would be limited to the incoming frame but no H>3264 and no
>>>> V>2464. Does that make sense ?
>>>
>>> Just to make sure I got it: is the ISC actually able to crop on the
>>> line length ? If you apply a 3264 limit, will the last 536 pixels of
>>> each line will be dropped if the input frame is 3800x2100 as your
>>> above example ?
>>
>> Yes. Last 536 pixels will be dropped.
>> After the 3264 limit, no more pixels are being sampled until HBLANK.
>>>
>>> Or rather, is the SRAM size limit the issue and the lines gets
>>> captured in full lenght no matter what, until the buffer gets filled up
>>> and all the successive pixels dropped.
>>>
>>> In other words, do your cropped image look like:
>>>
>>> 3264
>>> +--------------------------------X-----------+ 3800
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
>>> +--------------------------------------------+ 2100
>>>
>>> Or rather
>>>
>>>
>>> 3264
>>> +--------------------------------X-----------+ 3800
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
>>> |xxxxxxxxxxxxxxxxx| |
>>> | |
>>> +--------------------------------------------+ 2100
>>>
>>> I assume the first, as otherwise you wouldn't be able to interpret the
>>> image as 3264x2100.
>>
>> Yes, the first.
>>
>>>
>>> I'm asking because you had the un-cropped 3280x2464 size as the ISC
>>> source pad size in the paste of the media graph you provided below,
>>> which might mean the output image is actually full size in lines
>>> lenght and it's maybe missing pixels at the end, in case the ISC
>>> operates as in the above second case.
>>
>> No, the image won't be padded with blanks on the right side. Actually
>> this cannot happen due to a very big reason: the frame is being written
>> to the main memory by a DMA controller. This DMA controller will write
>> all the lines at consecutive addresses, and it will write exactly how
>> many pixels are being captured by the module from the ISC named
>> 'parallel front end' which does the cropping. Thus, the DMA will never
>> output empty padding bytes at the end of each line, and it will not add
>> a certain stride to each line if not instructed to. In fact the DMA can
>> add a stride offset for the address of each line, which can be used for
>> composing or copying the frame into another frame, but this is a
>> different story. ( I tried this for fun, but not implemented in the
>> driver at the moment)
>
> Oh that will really be interesting and it's indeed an argument for
> hainvg a scaler subdevice indeed
>
>>
>> The cropping is also mandatory even if the captured frame is less than
>> the maximum size, because the ISC has no configuration related to frame
>> size in fact. The parallel front end captures pixels and the dma engine
>> will write them to memory. There is no control over this except the
>> cropping.
>> So for example if the software expects 640x480 but somehow someone sends
>> more pixels, it can lead to buffer overflow if the parallel front end is
>> not configured to ignore(crop) anything beyond 640x480 .
>>
>
> I see and I think you have correctly plumbed that with the scaler sink
> pad crop rectangle. I don't seem to find where the crop rectangle
> sizes are actually applied to the ISC in this series ? Could you point
> me to it for sake of my better understanding ?

The ISC is already doing that for some time now:

https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/atmel/atmel-isc-base.c#L459

Initially it was more like a safety feature, because I discovered that
there was no way of limiting the incoming pixels, and if some malevolent
entity would just keep sending pixels, it could lead to exploits.

Now, the scaler is just exposing this to the user.

>
>> With this being said, or having an understanding about the sink pad of
>> the ISC, how should I configure the sink frame size ?
>
> I don't think you should, the video device format is set through the
> usual S_FMT as it happens already.
>
> What I think it's missing in your implementation is only to propagate the
> crop rectangle size of the scaler sink pad to the scaler source pad
> format (as you don't have any composition rectangles yet) [1]
>
> With that done, you should be able to capture from your video device
> in the cropped dimensions.
>
> The (simplified) media pipeline will look something like
>
> sensor
> type: Subdev
> pad 0 (sink) (3280, 2462)
> -> scaler:0
>
> scaler
> type: Subdev
> pad 0 (sink) (3280, 2464)
> crop: (3262, 2464)
> <- sensor:0
> pad 1 (source) (3264, 2464)
> -> isc_common
>
> isc_common
> type V4L2
> pad 0 (sink)
> <- scaler:1
>
> And the video device format set to (3264, 2464)
>
> Then later you might want to implement s_selection on the scaler to have
> the crop size configurable.
>
> One additional question: can the crop be panned ? What I mean is that
> you explained you can instruct the ISC to drop pixels after a certain
> line lenght. This will cause the crop not to be centered on the input
> image. Can the ISC be configured to ignore data -before- and -after-
> a certain number of pixels or lines ? In that case you might want to
> handle the top,left corner of the crop rectangle too. Anyway,
> something for later..

Looking over this, it looks possible. I have not tried it.
It looks like configuring a minimum row /col number would make the ISC
drop the first n pixels/ first n lines.

I will have to try this, but if it works, user could configure the
cropping area, that would be interesting ! Anyway I can't implement this
right now, but it's a nice thing to work on later.

>
> [1] https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#order-of-configuration-and-format-propagation
>
>> Is there any possibility to avoid limits on the frame size and have it
>> arbitrary ?
>> I am 99% sure that v4l2-compliance will not be happy if there is no
>> limit on the frame on entity pads (hence, I bound-aligned the frame to
>> 16x16 up to 10k x 10k )
>>
>
> That I'm not sure, but I think having size limits is required ?
>
> Anyway, from my side, the next version should be good to go. Thanks
> for sticking to all my comments and questions so far!

What should I do about the 10000 x 10000 pixels ?

Thanks for reviewing !

Eugen
>
> Thanks
> j
>
>
>>>
>>>>
>>>>>
>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
>>>>>>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>>>>>> +
>>>>>>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
>>>>>>>>>
>>>>>>>>> So you rely on the isc format list for the scaler as well ?
>>>>>>>>> I think it's fine as long as they are identical
>>>>>>>>
>>>>>>>> Yes, the scaler is kind of a dummy entity , but it's required by the
>>>>>>>> media controller validation of the pipeline.
>>>>>>>>
>>>>>>>
>>>>>>> More on this later
>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + if (!fmt)
>>>>>>>>>> + fmt = &isc->formats_list[0];
>>>>>>>>>> +
>>>>>>>>>> + req_fmt->format.code = fmt->mbus_code;
>>>>>>>>>> +
>>>>>>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>>>> + req_fmt->pad);
>>>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>>>>>> + /* Trying on the pad sink makes the source sink change too */
>>>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
>>>>>>>>>> + v4l2_try_fmt =
>>>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>>>>>> +
>>>>>>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
>>>>>>>>>> + 16, isc->max_width, 0,
>>>>>>>>>> + &v4l2_try_fmt->height,
>>>>>>>>>> + 16, isc->max_height, 0, 0);
>>>>>>>>>> + }
>>>>>>>>>> + /* if we are just trying, we are done */
>>>>>>>>>> + return 0;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + isc->scaler_format = req_fmt->format;
>>>>>>>>>
>>>>>>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
>>>>>>>>>
>>>>>>>>> Actually, I would like to know more about the scaler device
>>>>>>>>> capabilities. What functions can this IP perform ? Does it do
>>>>>>>>> cropping, can it also do (down)scaling or even composition ?
>>>>>>>>>
>>>>>>>>> I think it is worth to read
>>>>>>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
>>>>>>>>> where it is reported how cropping and scaling are implemented by using
>>>>>>>>> the selection API.
>>>>>>>>>
>>>>>>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
>>>>>>>>> for which you need to implement support to by adding s_selection on
>>>>>>>>> the sink pad TGT_CROP target.
>>>>>>>>
>>>>>>>> The scaler is kind of a dummy entity.
>>>>>>>> The problem is that different subdevices cannot be connected directly to
>>>>>>>> a v4l entity, because the ISC does cropping on the incoming frame.
>>>>>>>> The ISC has a limit on the total number of pixels per frame, thus it
>>>>>>>> will do a crop.
>>>>>>>
>>>>>>> What is this limit related to ? I don't know the ISC internals, but is
>>>>>>> the limitation due to the overall image size, or maybe it's due to
>>>>>>> some line buffers being limited in the size they can transfer to
>>>>>>> memory ? Is it a limitation on the size of the image in pixels, or is
>>>>>>> it due to a limitation of the processing bandwidth on the input
>>>>>>> interface and thus depends on the actual size of the image in bytes ?
>>>>>>
>>>>>> Limitation is both on line size and full image size. From my
>>>>>> understanding, is that the internal SRAM is limited (8 Mpixels with a
>>>>>> small safe buffer on top in the case of sama7g5 )
>>>>>> This SRAM could be used as 800000x10 for example, or 3200x2464 for
>>>>>> another example. However, I am not sure if a real line of 800,000 pixels
>>>>>> make sense, or if there is another internal mechanism in the ISC that
>>>>>> stops it.
>>>>>> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
>>>>>> the maximum frame that it can handle
>>>>>>
>>>>>>>
>>>>>>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
>>>>>>>> only handle 3264x2464 , then we have two choices:
>>>>>>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
>>>>>>>> 3280x2464, and the whole pipeline fails to configure, there is a
>>>>>>>> mismatch between /dev/video output and what the whole pipeline produces
>>>>>>>
>>>>>>> I'm confused in first place. If the ISC cannot output anything larger
>>>>>>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
>>>>>>> this should be adjusted to 3264x2464.
>>>>>>
>>>>>> yes, that is true. However, without an entity that says 'I am doing
>>>>>> cropping', the media pipeline will fail, because previous entity will
>>>>>> output 3280x2464 .
>>>>>> The conclusion to create a scaler entitity was done during my last year
>>>>>> discussions on the IRC. I could not find a way to perform this cropping
>>>>>> at the v4l entity side, and one reason is that the v4l entity does not
>>>>>> support what I needed , or at least that was my understanding.
>>>>>> You claim that the v4l entity could perform the cropping in the same
>>>>>> entity ?
>>>>>
>>>>> I thought so, but I haven't tried to be honest
>>>>>
>>>>>>
>>>>>> Let's assume for a minute that it could do this. Even so, I would prefer
>>>>>> to have a separate scaler entity to do this , for several reasons:
>>>>>> -> I would like to be able to crop the frame arbitrarily if I want to,
>>>>>> and the scaler would allow me to do this easier
>>>>>> -> it would be more obvious that the frame is cropped by having this
>>>>>> entity there
>>>>>> -> in fact some of the ISC versions really have a down scaler, that I
>>>>>> have not implemented yet. So it would be useful to already have this in
>>>>>> place to be able to scale down with it at a later time.
>>>>>
>>>>> I think this last reason is enough to have an entity plumbed in
>>>>> already
>>>>>
>>>>>>
>>>>>>>
>>>>>>> Or is the ISC limitation on the -input- side ?
>>>>>>>
>>>>>>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
>>>>>>>> the ISC can handle at most -> cannot , because the sensor let's say can
>>>>>>>> *only* produce 3280x2464 and *any* other frame size returns an error.
>>>>>>>
>>>>>>> buggy sensor I would say :)
>>>>>>
>>>>>> I cannot tell this to the customers, and I cannot fail the whole
>>>>>> pipeline because the sensor does not crop 16 pixels, when in fact I can
>>>>>> do this very easily on the ISC side.
>>>>>>
>>>>>
>>>>> Fair enough :)
>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> The only solution to make both worlds happy, is to have a dummy entity
>>>>>>>> called 'scaler' which in fact now it only performs a simple crop.
>>>>>>>> It accepts any frame size at input (hence the 10000x10000 which you saw
>>>>>>>> earlier), and outputs at most 3264x2464 the isc_max_height and
>>>>>>>> isc_max_width.
>>>>>>>
>>>>>>> I have a maybe dumb question: can the cropping operation be modeled
>>>>>>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
>>>>>>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
>>>>>>
>>>>>> I am not sure. This is what I wanted initially. But then I thought about
>>>>>> the three reasons stated above for the use of the scaler entity, and
>>>>>> discussed with folks on the IRC, and come up with the solution for the
>>>>>> scaler entity
>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>> So to answer your question, the isc scaler is a software model for a
>>>>>>>> simple cropping, that would make media controller happy, and capture
>>>>>>>> software happy.
>>>>>>>>
>>>>>>>> Here it how it looks :
>>>>>>>>
>>>>>>>> Device topology
>>>>>>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
>>>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>>>> device node name /dev/v4l-subdev0
>>>>>>>> pad0: Sink
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>>>> crop.bounds:(0,0)/3264x2464
>>>>>>>> crop:(0,0)/3264x2464]
>>>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>>>>>> pad1: Source
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>>>>>
>>>>>>>> - entity 4: csi2dc (2 pads, 2 links)
>>>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>>>> device node name /dev/v4l-subdev1
>>>>>>>> pad0: Sink
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>>>> <- "dw-csi.0":1 [ENABLED]
>>>>>>>> pad1: Source
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
>>>>>>>>
>>>>>>>> - entity 7: dw-csi.0 (2 pads, 2 links)
>>>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>>>> device node name /dev/v4l-subdev2
>>>>>>>> pad0: Sink
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>>>>>> <- "imx219 1-0010":0 [ENABLED]
>>>>>>>> pad1: Source
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>>>>>> -> "csi2dc":0 [ENABLED]
>>>>>>>>
>>>>>>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
>>>>>>>> type V4L2 subdev subtype Sensor flags 0
>>>>>>>> device node name /dev/v4l-subdev3
>>>>>>>> pad0: Source
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>>>> xfer:srgb ycbcr:601 quantization:full-range
>>>>>>>> crop.bounds:(8,8)/3280x2464
>>>>>>>> crop:(8,8)/3280x2464]
>>>>>>>> -> "dw-csi.0":0 [ENABLED]
>>>>>>>>
>>>>>>>> - entity 24: atmel_isc_common (1 pad, 1 link)
>>>>>>>> type Node subtype V4L flags 1
>>>>>>>> device node name /dev/video0
>>>>>>>> pad0: Sink
>>>>>>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
>>>>>>>>
>>>>>>>>
>>>>>>>> Scaler does this one cute little thing :
>>>>>>>>
>>>>>>>> pad0: Sink
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>>>> crop.bounds:(0,0)/3264x2464
>>>>>>>> crop:(0,0)/3264x2464]
>>>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>>>>>> pad1: Source
>>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>>>
>>>>>>> Shouldn't this be 3264x2464 as that's what the entity outputs after
>>>>>>> the cropping ? And shouldn't then be 3264x2464 the size output from
>>>>>>> the video device too ?
>>>>>>
>>>>>> That's right.
>>>>>> I don't know why the source format for the scaler is still 3280x2464.
>>>>>> Maybe there is a bug on g_fmt for it.. have to check it.
>>>>>
>>>>> Thanks, I think this should be fixed ten
>>>>>
>>>>>
>>>>>>
>>>>>> Anyway, the video format is like this :
>>>>>>
>>>>>> # v4l2-ctl --get-fmt-video
>>>>>> Format Video Capture:
>>>>>> Width/Height : 3264/2464
>>>>>> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
>>>>>> Field : None
>>>>>> Bytes per Line : 6528
>>>>>> Size Image : 16084992
>>>>>> Colorspace : sRGB
>>>>>> Transfer Function : Default (maps to sRGB)
>>>>>> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
>>>>>> Quantization : Default (maps to Full Range)
>>>>>> Flags :
>>>>>> #
>>>>>>
>>>>>> and initializing the pipeline looks fine from user perspective.
>>>>>> The scaler is in fact the solution to make this pipeline work with
>>>>>> libcamera.
>>>>>> I found this cropping to be an issue in media controller when trying it
>>>>>> with libcamera. Otherwise, the other user space apps which I was using
>>>>>> never complained that anything was wrong
>>>>>> Libcamera simply refuses to acknowledge the pipeline if the video output
>>>>>> is 3264x2464 but there is no entity that changes the format from
>>>>>> 3280x2464 down
>>>>>
>>>>> Not sure why libcamera should play a role there. Isn't it the media
>>>>> pipeline validation that complains and returns an -EPIPE ?
>>>>
>>>> I tried this before the media_pipeline_start .
>>>> Perhaps libcamera was doing some similar checks, anyway, it was
>>>> complaining and not adding the stream, and reported no usable streams/camera
>>>>
>>>> Once I will rework some of the things according to your review I will
>>>> run libcamera again to see what it complains about
>>>>
>>>>>
>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>>>>>
>>>>>>>>
>>>>>>>> Which is what we needed.
>>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + return 0;
>>>>>>>>>> +}
>>>>>>>>>> +
>>>>>>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
>>>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>>>> + struct v4l2_subdev_mbus_code_enum *code)
>>>>>>>>>> +{
>>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>>>> + int supported_index = 0;
>>>>>>>>>> + int i;
>>>>>>>>>> +
>>>>>>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
>>>>>>>>>> + if (!isc->formats_list[i].sd_support)
>>>>>>>>>> + continue;
>>>>>>>>>
>>>>>>>>> The sd_support flag still doesn't click in my head.
>>>>>>>>>
>>>>>>>>> Shouldn't the enumeration of available formats on the scaler do not
>>>>>>>>> depend on the sensor supproted formats ?
>>>>>>>>
>>>>>>>> You're right. I will have to check it again.
>>>>>>>>
>>>>>>>>>
>>>>>>>>>> + if (supported_index == code->index) {
>>>>>>>>>> + code->code = isc->formats_list[i].mbus_code;
>>>>>>>>>> + return 0;
>>>>>>>>>> + }
>>>>>>>>>> + supported_index++;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + return -EINVAL;
>>>>>>>>>> +}
>>>>>>>>>> +
>>>>>>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
>>>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>>>> + struct v4l2_subdev_selection *sel)
>>>>>>>>>> +{
>>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>>>> +
>>>>>>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
>>>>>>>>>> + return -EINVAL;
>>>>>>>>>> +
>>>>>>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
>>>>>>>>>> + sel->target != V4L2_SEL_TGT_CROP)
>>>>>>>>>> + return -EINVAL;
>>>>>>>>>> +
>>>>>>>>>> + sel->r.height = isc->max_height;
>>>>>>>>>> + sel->r.width = isc->max_width;
>>>>>>>>>
>>>>>>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
>>>>>>>>> as it represents the maximum valid crop rectangle.
>>>>>>>>>
>>>>>>>>> TGT_CROP should report the configured crop rectangle which can be
>>>>>>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
>>>>>>>>> the spec is correct
>>>>>>>>
>>>>>>>> So you would like to have this differentiated, and report the
>>>>>>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
>>>>>>>> reported now, the maximum size of the ISC frame .
>>>>>>>> My understanding is correct ?
>>>>>>>>
>>>>>>>
>>>>>>> I didn't know you have an HW limitation, so your _BOUNDS is not the
>>>>>>> input image size but rather 3264x2464 ( == max_width x max_height).
>>>>>>>
>>>>>>> What I meant is that _BOUNDS should report the maximum rectangle size
>>>>>>> that can be applied to the _CROP target. In you case you have an HW
>>>>>>> limitation 3264x2464 and that's the largest rectangle you can apply.
>>>>>> So the CROP should be at 3264x2464
>>>>>>> TGT_CROP can be initialized to the same as _BOUND, but if you
>>>>>>> implement s_selection it should report what has been there applied.
>>>>>> and BOUND to actual frame size ?
>>>>>>> But as you don't implement s_selection yet, I think this is fine for
>>>>>>> now. Maybe a little comment ?
>>>>>>
>>>>>> It could also be the case where e.g. the sensor is outputting 1920x1080,
>>>>>> in this case the scaler would do nothing.
>>>>>> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
>>>>>> If the sensor is outputting 1920x1080, this format comes directly from
>>>>>> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
>>>>>> sensor's problem)
>>>>>>>
>>>>>>> Also, is set->r zeroed by the framework before getting here ?
>>>>>>> Otherwise you should set r.left and r.top to 0
>>>>>>
>>>>>> If these are redundant, no problem to remove them
>>>>>
>>>>> I was actually confused. I was suggesting you to add...
>>>>>
>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + sel->r.left = 0;
>>>>>>>>>> + sel->r.top = 0;
>>>>>
>>>>> These ^
>>>>>
>>>>> So nothing to change. Sorry I've missed them
>>>>>
>>>>> Thanks
>>>>> j
>>>>>
>>>>>
>>>>>>>>>> +
>>>>>>>>>> + return 0;
>>>>>>>>>> +}
>>>>>>>>>> +
>>>>>>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
>>>>>>>>>> + struct v4l2_subdev_state *sd_state)
>>>>>>>>>> +{
>>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
>>>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
>>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>>>> +
>>>>>>>>>> + *v4l2_try_fmt = isc->scaler_format;
>>>>>>>>>> +
>>>>>>>>>> + return 0;
>>>>>>>>>> +}
>>>>>>>>>> +
>>>>>>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>>>>>>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
>>>>>>>>>> + .set_fmt = isc_scaler_set_fmt,
>>>>>>>>>> + .get_fmt = isc_scaler_get_fmt,
>>>>>>>>>> + .get_selection = isc_scaler_g_sel,
>>>>>>>>>> + .init_cfg = isc_scaler_init_cfg,
>>>>>>>>>
>>>>>>>>> .link_validate = v4l2_subdev_link_validate_default,
>>>>>>>>>
>>>>>>>>> To have the formats at the end of links that point to this entity
>>>>>>>>> validated (I think the framework already calls it if not set though,
>>>>>>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
>>>>>>>>>
>>>>>>>>>> +};
>>>>>>>>>> +
>>>>>>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>>>>>>>>>> + .pad = &isc_scaler_pad_ops,
>>>>>>>>>> +};
>>>>>>>>>> +
>>>>>>>>>> +int isc_scaler_init(struct isc_device *isc)
>>>>>>>>>> +{
>>>>>>>>>> + int ret;
>>>>>>>>>> +
>>>>>>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
>>>>>>>>>> +
>>>>>>>>>> + isc->scaler_sd.owner = THIS_MODULE;
>>>>>>>>>> + isc->scaler_sd.dev = isc->dev;
>>>>>>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
>>>>>>>>>> + "atmel_isc_scaler");
>>>>>>>>>
>>>>>>>>> I would drop 'atmel' for brevity, unless other entities have this
>>>>>>>>> prefix set already
>>>>>>>>
>>>>>>>> The v4l entity takes it's name from the module name, which is
>>>>>>>> atmel_isc_common, so I thought to keep the prefix
>>>>>>>>
>>>>>>>
>>>>>>> Ack!
>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>>>>>>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>>>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>>>>>>>>>> +
>>>>>>>>>> + isc->scaler_format.height = isc->max_height;
>>>>>>>>>> + isc->scaler_format.width = isc->max_width;
>>>>>>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
>>>>>>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
>>>>>>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>>>>>> +
>>>>>>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
>>>>>>>>>> + ISC_SCALER_PADS_NUM,
>>>>>>>>>> + isc->scaler_pads);
>>>>>>>>>> + if (ret < 0) {
>>>>>>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
>>>>>>>>>> + return ret;
>>>>>>>>>> + }
>>>>>>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
>>>>>>>>>> + if (ret < 0) {
>>>>>>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
>>>>>>>>>> + return ret;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + return ret;
>>>>>>>>>> +}
>>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
>>>>>>>>>> +
>>>>>>>>>> +int isc_scaler_link(struct isc_device *isc)
>>>>>>>>>> +{
>>>>>>>>>> + int ret;
>>>>>>>>>> +
>>>>>>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
>>>>>>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
>>>>>>>>>> + ISC_SCALER_PAD_SINK,
>>>>>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>>>>>> +
>>>>>>>>>> + if (ret < 0) {
>>>>>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>>>>>> + isc->current_subdev->sd->entity.name,
>>>>>>>>>> + isc->scaler_sd.entity.name);
>>>>>>>>>> + return ret;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
>>>>>>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
>>>>>>>>>> +
>>>>>>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
>>>>>>>>>> + ISC_SCALER_PAD_SOURCE,
>>>>>>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
>>>>>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>>>>>> +
>>>>>>>>>> + if (ret < 0) {
>>>>>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>>>>>> + isc->scaler_sd.entity.name,
>>>>>>>>>> + isc->video_dev.entity.name);
>>>>>>>>>> + return ret;
>>>>>>>>>> + }
>>>>>>>>>> +
>>>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
>>>>>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>>>>>> +
>>>>>>>>>> + return ret;
>>>>>>>>>> +}
>>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
>>>>>>>>>
>>>>>>>>> From the DT graph point of view, the ISC appears as a single block
>>>>>>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
>>>>>>>>> add to its notifier the image sensor remote subdevice.
>>>>>>>>
>>>>>>>> Actually it's only a parallel port.
>>>>>>>> And it can be connected either to a sensor directly or to the csi2dc
>>>>>>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
>>>>>>>>
>>>>>>>>>
>>>>>>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
>>>>>>>>> entity which was initialized at isc_probe() time is linked in between
>>>>>>>>> the ISC and the image sensor immutably.
>>>>>>>>
>>>>>>>> yes, because this cannot change at runtime, and usually, it can't change
>>>>>>>> without altering the board hardware. (at least this is what my
>>>>>>>> understanding of immutability is )
>>>>>>>>
>>>>>>>>>
>>>>>>>>> I think it is fine for now, but I wonder if you plan to plumb more
>>>>>>>>> components between the ISC video node and the sensor, if it's not
>>>>>>>>> worth changing the DT bindings and their parsing logic to separate the
>>>>>>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
>>>>>>>>> time and have the CSI-2 receiver entity as downstream remote. I think
>>>>>>>>> I need to get to know the ISC better to really have an idea. For now,
>>>>>>>>> this seems ok to me, but please check with maintainers if this is fine
>>>>>>>>> with them.
>>>>>>>>
>>>>>>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
>>>>>>>> the remote subdev).
>>>>>>>> The ISC will register a scaler, and connect the subdev to this scaler
>>>>>>>> first, and then, the scaler to the isc itself (the v4l entity).
>>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>>>> index 5fbf52a9080b..c9234c90ae58 100644
>>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
>>>>>>>>>> u32 his_entry;
>>>>>>>>>> };
>>>>>>>>>>
>>>>>>>>>> +enum isc_mc_pads {
>>>>>>>>>> + ISC_PAD_SINK = 0,
>>>>>>>>>> + ISC_PADS_NUM = 1,
>>>>>>>>>> +};
>>>>>>>>>> +
>>>>>>>>>> +enum isc_scaler_pads {
>>>>>>>>>> + ISC_SCALER_PAD_SINK = 0,
>>>>>>>>>> + ISC_SCALER_PAD_SOURCE = 1,
>>>>>>>>>> + ISC_SCALER_PADS_NUM = 2,
>>>>>>>>>> +};
>>>>>>>>>> +
>>>>>>>>>> /*
>>>>>>>>>> * struct isc_device - ISC device driver data/config struct
>>>>>>>>>> * @regmap: Register map
>>>>>>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
>>>>>>>>>> * be used as an input to the controller
>>>>>>>>>> * @controller_formats_size: size of controller_formats array
>>>>>>>>>> * @formats_list_size: size of formats_list array
>>>>>>>>>> + * @pads: media controller pads for isc video entity
>>>>>>>>>> + * @mdev: media device that is registered by the isc
>>>>>>>>>> + * @remote_pad: remote pad on the connected subdevice
>>>>>>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
>>>>>>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
>>>>>>>>>> + * @scaler_format: current format for the scaler subdevice
>>>>>>>>>> */
>>>>>>>>>> struct isc_device {
>>>>>>>>>> struct regmap *regmap;
>>>>>>>>>> @@ -344,6 +361,19 @@ struct isc_device {
>>>>>>>>>> struct isc_format *formats_list;
>>>>>>>>>> u32 controller_formats_size;
>>>>>>>>>> u32 formats_list_size;
>>>>>>>>>> +
>>>>>>>>>> + struct {
>>>>>>>>>> + struct media_pad pads[ISC_PADS_NUM];
>>>>>>>>>> + struct media_device mdev;
>>>>>>>>>> +
>>>>>>>>>> + u32 remote_pad;
>>>>>>>>>> + };
>>>>>>>>>> +
>>>>>>>>>> + struct {
>>>>>>>>>> + struct v4l2_subdev scaler_sd;
>>>>>>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
>>>>>>>>>> + struct v4l2_mbus_framefmt scaler_format;
>>>>>>>>
>>>>>>>> Here are the scaler stuff which I added, in the same bulk struct for the
>>>>>>>> whole isc device
>>>>>>>>
>>>>>>>>
>>>>>>>>>> + };
>>>>>>>>>> };
>>>>>>>>>>
>>>>>>>>>> extern const struct regmap_config isc_regmap_config;
>>>>>>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
>>>>>>>>>> void isc_subdev_cleanup(struct isc_device *isc);
>>>>>>>>>> void isc_clk_cleanup(struct isc_device *isc);
>>>>>>>>>>
>>>>>>>>>> +int isc_scaler_link(struct isc_device *isc);
>>>>>>>>>> +int isc_scaler_init(struct isc_device *isc);
>>>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
>>>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc);
>>>>>>>>>> +
>>>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>>>>>> + unsigned int code, int *index);
>>>>>>>>>> #endif
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>>>> index c5b9563e36cb..c244682ea22f 100644
>>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>>>> break;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>>>
>>>>>>>>> I am surprised you can read a register before runtime_pm is
>>>>>>>>> intialized!
>>>>>>>>>
>>>>>>>>
>>>>>>>> Actually I can read any register after the moment of starting the clock
>>>>>>>> on the hardware block.
>>>>>>>
>>>>>>> Ah right then
>>>>>>>
>>>>>>>> Maybe the starting and stopping of the clock needs to be moved to the
>>>>>>>> runtime_pm calls, but this is another story, not related to the media
>>>>>>>> controller.
>>>>>>>
>>>>>>> It's fine, but maybe worth recording with a todo ?
>>>>>>>
>>>>>>> Thanks
>>>>>>> j
>>>>>>>
>>>>>>>> I moved this line because I had to pass the version to the isc_mc_init call
>>>>>>>>
>>>>>>>>> Thanks
>>>>>>>>> j
>>>>>>>>
>>>>>>>> Thanks for reviewing !
>>>>>>>> Eugen
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> +
>>>>>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>>>>>> + if (ret < 0)
>>>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>>>> +
>>>>>>>>>> pm_runtime_set_active(dev);
>>>>>>>>>> pm_runtime_enable(dev);
>>>>>>>>>> pm_request_idle(dev);
>>>>>>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>>>> ret = clk_prepare_enable(isc->ispck);
>>>>>>>>>> if (ret) {
>>>>>>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
>>>>>>>>>> - goto cleanup_subdev;
>>>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> /* ispck should be greater or equal to hclock */
>>>>>>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>>>> goto unprepare_clk;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
>>>>>>>>>>
>>>>>>>>>> return 0;
>>>>>>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>>>> unprepare_clk:
>>>>>>>>>> clk_disable_unprepare(isc->ispck);
>>>>>>>>>>
>>>>>>>>>> +isc_probe_mc_init_err:
>>>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>>>> +
>>>>>>>>>> cleanup_subdev:
>>>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>>>
>>>>>>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>>>>>>>>>>
>>>>>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>>>>>
>>>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>>>> +
>>>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>>>
>>>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>>>> index 07a80b08bc54..9dc75eed0098 100644
>>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
>>>>>>>>>> break;
>>>>>>>>>> }
>>>>>>>>>>
>>>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>>>> +
>>>>>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>>>>>> + if (ret < 0)
>>>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>>>> +
>>>>>>>>>> pm_runtime_set_active(dev);
>>>>>>>>>> pm_runtime_enable(dev);
>>>>>>>>>> pm_request_idle(dev);
>>>>>>>>>>
>>>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
>>>>>>>>>>
>>>>>>>>>> return 0;
>>>>>>>>>>
>>>>>>>>>> +isc_probe_mc_init_err:
>>>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>>>> +
>>>>>>>>>> cleanup_subdev:
>>>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>>>
>>>>>>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>>>>>>>>>>
>>>>>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>>>>>
>>>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>>>> +
>>>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>>>
>>>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>>>>>> --
>>>>>>>>>> 2.25.1
>>>>>>>>>>
>>>>>>>>
>>>>>>
>>>>
>>

2022-02-11 22:37:19

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 07/11] media: atmel: atmel-isc: change format propagation to subdev into only verification

Hi Eugen,

On Fri, Feb 11, 2022 at 10:38:56AM +0000, [email protected] wrote:
> On 2/10/22 8:43 PM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Fri, Jan 21, 2022 at 03:14:12PM +0200, Eugen Hristev wrote:
> >> As a top MC video driver, the atmel-isc should not propagate the format to the
> >> subdevice.
> >> It should rather check at start_streaming() time if the subdev is properly
> >> configured with a compatible format.
> >> Removed the whole format finding logic, and reworked the format verification
> >> at start_streaming time, such that the ISC will return an error if the subdevice
> >> is not properly configured. To achieve this, media_pipeline_start
> >> is called and a link_validate callback is created to check the formats.
> >> With this being done, the module parameter 'sensor_preferred' makes no sense
> >> anymore. The ISC should not decide which format the sensor is using. The
> >> ISC should only cope with the situation and inform userspace if the streaming
> >> is possible in the current configuration.
> >
> > This sounds right!
> >
> >> The redesign of the format propagation has also risen the question of the
> >> enumfmt callback. If enumfmt is called with an mbus_code, the enumfmt handler
> >> should only return the formats that are supported for this mbus_code.
> >> To make it more easy to understand the formats, changed the report order
> >> to report first the native formats, and after that the formats that the ISC
> >> can convert to.
> >>
> >> Signed-off-by: Eugen Hristev <[email protected]>
> >> ---
> >> Changes in v4:
> >> - moved validation code into link_validate and used media_pipeline_start
> >> - merged this patch with the enum_fmt patch which was previously in v3 of
> >> the series
> >>
> >> Changes in v3:
> >> - clamp to maximum resolution once the frame size from the subdev is found
> >>
> >> drivers/media/platform/atmel/atmel-isc-base.c | 335 ++++++++++--------
> >> .../media/platform/atmel/atmel-isc-scaler.c | 5 +
> >> drivers/media/platform/atmel/atmel-isc.h | 3 +
> >> 3 files changed, 191 insertions(+), 152 deletions(-)
> >>
> >> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >> index f7a88399bd54..31c79313aadc 100644
> >> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >> @@ -36,11 +36,6 @@ static unsigned int debug;
> >> module_param(debug, int, 0644);
> >> MODULE_PARM_DESC(debug, "debug level (0-2)");
> >>
> >> -static unsigned int sensor_preferred = 1;
> >> -module_param(sensor_preferred, uint, 0644);
> >> -MODULE_PARM_DESC(sensor_preferred,
> >> - "Sensor is preferred to output the specified format (1-on 0-off), default 1");
> >> -
> >> #define ISC_IS_FORMAT_RAW(mbus_code) \
> >> (((mbus_code) & 0xf000) == 0x3000)
> >>
> >> @@ -337,6 +332,10 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
> >> unsigned long flags;
> >> int ret;
> >>
> >> + ret = media_pipeline_start(&isc->video_dev.entity, &isc->mpipe);
> >> + if (ret)
> >> + goto err_pipeline_start;
> >> +
> >> /* Enable stream on the sub device */
> >> ret = v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 1);
> >> if (ret && ret != -ENOIOCTLCMD) {
> >> @@ -384,6 +383,9 @@ static int isc_start_streaming(struct vb2_queue *vq, unsigned int count)
> >> v4l2_subdev_call(isc->current_subdev->sd, video, s_stream, 0);
> >>
> >> err_start_stream:
> >> + media_pipeline_stop(&isc->video_dev.entity);
> >> +
> >> +err_pipeline_start:
> >> spin_lock_irqsave(&isc->dma_queue_lock, flags);
> >> list_for_each_entry(buf, &isc->dma_queue, list)
> >> vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_QUEUED);
> >> @@ -420,6 +422,9 @@ static void isc_stop_streaming(struct vb2_queue *vq)
> >> if (ret && ret != -ENOIOCTLCMD)
> >> v4l2_err(&isc->v4l2_dev, "stream off failed in subdev\n");
> >>
> >> + /* Stop media pipeline */
> >> + media_pipeline_stop(&isc->video_dev.entity);
> >> +
> >> /* Release all active buffers */
> >> spin_lock_irqsave(&isc->dma_queue_lock, flags);
> >> if (unlikely(isc->cur_frm)) {
> >> @@ -496,21 +501,56 @@ static int isc_enum_fmt_vid_cap(struct file *file, void *priv,
> >> u32 index = f->index;
> >> u32 i, supported_index;
> >>
> >> - if (index < isc->controller_formats_size) {
> >> - f->pixelformat = isc->controller_formats[index].fourcc;
> >> - return 0;
> >> + supported_index = 0;
> >> +
> >> + for (i = 0; i < isc->formats_list_size; i++) {
> >> + if (!isc->formats_list[i].sd_support)
> >> + continue;
> >
> > Sorry to ask again about sd_support, but if the ISC formats
> > enumeration is not constrained to a specific mbus_code, shouldn't all
> > the formats be enumerated regardless if they're supported from the
> > sensor or not ?
>
> Yes. this slipped a little. I have to rework a bit the format reporting
> to user space.
>
> >
> >> + /*
> >> + * If specific mbus_code is requested, provide only
> >> + * supported formats with this mbus code
> >> + */
> >> + if (f->mbus_code && f->mbus_code !=
> >> + isc->formats_list[i].mbus_code)
> >
> > nit: for clarity
> > if (f->mbus_code &&
> > f->mbus_code != isc->formats_list[i].mbus_code)
> >
> >> + continue;
> >> + if (supported_index == index) {
> >> + f->pixelformat = isc->formats_list[i].fourcc;
> >> + return 0;
> >> + }
> >> + supported_index++;
> >> }
> >>
> >> - index -= isc->controller_formats_size;
> >> + /*
> >> + * If the sensor does not support this mbus_code whatsoever,
> >> + * there is no reason to advertise any of our output formats
> >> + */
> >> + if (supported_index == 0)
> >> + return -EINVAL;
> >>
> >> + /*
> >> + * If the sensor uses a format that is not raw, then we cannot
> >> + * convert it to any of the formats that we usually can with a
> >> + * RAW sensor. Thus, do not advertise them.
> >> + */
> >> + if (isc->config.sd_format &&
> >> + !ISC_IS_FORMAT_RAW(isc->config.sd_format->mbus_code))
> >> + return -EINVAL;
> >> +
> >> + /*
> >> + * Iterate again through the formats that we can convert to.
> >> + * However, to avoid duplicates, skip the formats that
> >> + * the sensor already supports directly
> >> + */
> >> + index -= supported_index;
> >> supported_index = 0;
> >>
> >> - for (i = 0; i < isc->formats_list_size; i++) {
> >> - if (!ISC_IS_FORMAT_RAW(isc->formats_list[i].mbus_code) ||
> >> - !isc->formats_list[i].sd_support)
> >> + for (i = 0; i < isc->controller_formats_size; i++) {
> >> + /* if this format is already supported by sensor, skip it */
> >> + if (find_format_by_fourcc(isc, isc->controller_formats[i].fourcc))
> >> continue;
> >> if (supported_index == index) {
> >> - f->pixelformat = isc->formats_list[i].fourcc;
> >> + f->pixelformat =
> >> + isc->controller_formats[i].fourcc;
> >
> > I trust this last part to work correctly, as it's not 100% clear to me :)
> >
>
> I think I will remove this and just report all the formats to userspace.
> Let libcamera handle this ;)
>

I think so, the driver will be greatly simplified with this change.

> >> return 0;
> >> }
> >> supported_index++;
> >> @@ -581,20 +621,30 @@ static int isc_try_validate_formats(struct isc_device *isc)
> >> break;
> >> default:
> >> /* any other different formats are not supported */
> >> + v4l2_err(&isc->v4l2_dev, "Requested unsupported format.\n");
> >> ret = -EINVAL;
> >> }
> >> v4l2_dbg(1, debug, &isc->v4l2_dev,
> >> "Format validation, requested rgb=%u, yuv=%u, grey=%u, bayer=%u\n",
> >> rgb, yuv, grey, bayer);
> >>
> >> - /* we cannot output RAW if we do not receive RAW */
> >> - if ((bayer) && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
> >> + if ((bayer) &&
> >
> > just 'bayer' ?
>
> not sure what you mean ?
> Have to check if we can output raw bayer , and we can only do that if we
> receive raw bayer from the sensor.
>

I just meant that you could remove () from (bayer) ?

Sorry for not being clear :)

> >
> >> + !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code)) {
> >> + v4l2_err(&isc->v4l2_dev, "Cannot output RAW if we do not receive RAW.\n");
> >> return -EINVAL;
> >> + }
> >>
> >> - /* we cannot output GREY if we do not receive RAW/GREY */
> >> if (grey && !ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code) &&
> >> - !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code))
> >> + !ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
> >> + v4l2_err(&isc->v4l2_dev, "Cannot output GREY if we do not receive RAW/GREY.\n");
> >> + return -EINVAL;
> >> + }
> >> +
> >> + if ((rgb || bayer || yuv) &&
> >> + ISC_IS_FORMAT_GREY(isc->try_config.sd_format->mbus_code)) {
> >> + v4l2_err(&isc->v4l2_dev, "Cannot convert GREY to another format.\n");
> >> return -EINVAL;
> >> + }
> >>
> >> return ret;
> >> }
> >> @@ -822,7 +872,7 @@ static void isc_try_fse(struct isc_device *isc,
> >> * If we do not know yet which format the subdev is using, we cannot
> >> * do anything.
> >> */
> >> - if (!isc->try_config.sd_format)
> >> + if (!isc->config.sd_format)
> >> return;
> >>
> >> fse.code = isc->try_config.sd_format->mbus_code;
> >> @@ -843,180 +893,141 @@ static void isc_try_fse(struct isc_device *isc,
> >> }
> >> }
> >>
> >> -static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f,
> >> - u32 *code)
> >> +static int isc_try_fmt(struct isc_device *isc, struct v4l2_format *f)
> >> {
> >> - int i;
> >> - struct isc_format *sd_fmt = NULL, *direct_fmt = NULL;
> >> struct v4l2_pix_format *pixfmt = &f->fmt.pix;
> >> - struct v4l2_subdev_pad_config pad_cfg = {};
> >> - struct v4l2_subdev_state pad_state = {
> >> - .pads = &pad_cfg
> >> - };
> >> - struct v4l2_subdev_format format = {
> >> - .which = V4L2_SUBDEV_FORMAT_TRY,
> >> - };
> >> - u32 mbus_code;
> >> - int ret;
> >> - bool rlp_dma_direct_dump = false;
> >> + unsigned int i;
> >>
> >> if (f->type != V4L2_BUF_TYPE_VIDEO_CAPTURE)
> >> return -EINVAL;
> >>
> >> - /* Step 1: find a RAW format that is supported */
> >> - for (i = 0; i < isc->num_user_formats; i++) {
> >> - if (ISC_IS_FORMAT_RAW(isc->user_formats[i]->mbus_code)) {
> >> - sd_fmt = isc->user_formats[i];
> >> + isc->try_config.fourcc = isc->user_formats[0]->fourcc;
> >> +
> >> + /* find if the format requested is supported */
> >> + for (i = 0; i < isc->controller_formats_size; i++)
> >> + if (isc->controller_formats[i].fourcc == pixfmt->pixelformat) {
> >> + isc->try_config.fourcc = pixfmt->pixelformat;
> >> break;
> >> }
> >> - }
> >> - /* Step 2: We can continue with this RAW format, or we can look
> >> - * for better: maybe sensor supports directly what we need.
> >> - */
> >> - direct_fmt = find_format_by_fourcc(isc, pixfmt->pixelformat);
> >> -
> >> - /* Step 3: We have both. We decide given the module parameter which
> >> - * one to use.
> >> - */
> >> - if (direct_fmt && sd_fmt && sensor_preferred)
> >> - sd_fmt = direct_fmt;
> >> -
> >> - /* Step 4: we do not have RAW but we have a direct format. Use it. */
> >> - if (direct_fmt && !sd_fmt)
> >> - sd_fmt = direct_fmt;
> >> -
> >> - /* Step 5: if we are using a direct format, we need to package
> >> - * everything as 8 bit data and just dump it
> >> - */
> >> - if (sd_fmt == direct_fmt)
> >> - rlp_dma_direct_dump = true;
> >> -
> >> - /* Step 6: We have no format. This can happen if the userspace
> >> - * requests some weird/invalid format.
> >> - * In this case, default to whatever we have
> >> - */
> >> - if (!sd_fmt && !direct_fmt) {
> >> - sd_fmt = isc->user_formats[isc->num_user_formats - 1];
> >> - v4l2_dbg(1, debug, &isc->v4l2_dev,
> >> - "Sensor not supporting %.4s, using %.4s\n",
> >> - (char *)&pixfmt->pixelformat, (char *)&sd_fmt->fourcc);
> >> - }
> >> -
> >> - if (!sd_fmt) {
> >> - ret = -EINVAL;
> >> - goto isc_try_fmt_err;
> >> - }
> >> -
> >> - /* Step 7: Print out what we decided for debugging */
> >> - v4l2_dbg(1, debug, &isc->v4l2_dev,
> >> - "Preferring to have sensor using format %.4s\n",
> >> - (char *)&sd_fmt->fourcc);
> >> -
> >> - /* Step 8: at this moment we decided which format the subdev will use */
> >> - isc->try_config.sd_format = sd_fmt;
> >> -
> >> - /* Limit to Atmel ISC hardware capabilities */
> >> - if (pixfmt->width > isc->max_width)
> >> - pixfmt->width = isc->max_width;
> >> - if (pixfmt->height > isc->max_height)
> >> - pixfmt->height = isc->max_height;
> >> -
> >> - /*
> >> - * The mbus format is the one the subdev outputs.
> >> - * The pixels will be transferred in this format Sensor -> ISC
> >> - */
> >> - mbus_code = sd_fmt->mbus_code;
> >> -
> >> - /*
> >> - * Validate formats. If the required format is not OK, default to raw.
> >> - */
> >> -
> >> - isc->try_config.fourcc = pixfmt->pixelformat;
> >> -
> >> - if (isc_try_validate_formats(isc)) {
> >> - pixfmt->pixelformat = isc->try_config.fourcc = sd_fmt->fourcc;
> >> - /* Re-try to validate the new format */
> >> - ret = isc_try_validate_formats(isc);
> >> - if (ret)
> >> - goto isc_try_fmt_err;
> >> - }
> >> -
> >> - ret = isc_try_configure_rlp_dma(isc, rlp_dma_direct_dump);
> >> - if (ret)
> >> - goto isc_try_fmt_err;
> >> -
> >> - ret = isc_try_configure_pipeline(isc);
> >> - if (ret)
> >> - goto isc_try_fmt_err;
> >>
> >> - /* Obtain frame sizes if possible to have crop requirements ready */
> >> - isc_try_fse(isc, &pad_state);
> >> -
> >> - v4l2_fill_mbus_format(&format.format, pixfmt, mbus_code);
> >> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad, set_fmt,
> >> - &pad_state, &format);
> >> - if (ret < 0)
> >> - goto isc_try_fmt_subdev_err;
> >> + /* If we did not find the requested format, we will fallback here */
> >> + pixfmt->pixelformat = isc->try_config.fourcc;
> >> + pixfmt->colorspace = V4L2_COLORSPACE_SRGB;
> >> + pixfmt->field = V4L2_FIELD_NONE;
> >>
> >> - v4l2_fill_pix_format(pixfmt, &format.format);
> >> + isc_try_configure_rlp_dma(isc, false);
> >>
> >> /* Limit to Atmel ISC hardware capabilities */
> >> - if (pixfmt->width > isc->max_width)
> >> - pixfmt->width = isc->max_width;
> >> - if (pixfmt->height > isc->max_height)
> >> - pixfmt->height = isc->max_height;
> >> + v4l_bound_align_image(&pixfmt->width, 16, isc->max_width, 0,
> >> + &pixfmt->height, 16, isc->max_height, 0, 0);
> >>
> >> pixfmt->field = V4L2_FIELD_NONE;
> >> pixfmt->bytesperline = (pixfmt->width * isc->try_config.bpp_v4l2) >> 3;
> >> pixfmt->sizeimage = ((pixfmt->width * isc->try_config.bpp) >> 3) *
> >> pixfmt->height;
> >
> > Seems pixfmt->field is set twice.
> >
> > Could you move the assignment of all pixfmt to a single block ?
>
> okay
> >
> >>
> >> - if (code)
> >> - *code = mbus_code;
> >> + isc->try_fmt = *f;
> >>
> >> return 0;
> >> +}
> >>
> >> -isc_try_fmt_err:
> >> - v4l2_err(&isc->v4l2_dev, "Could not find any possible format for a working pipeline\n");
> >> -isc_try_fmt_subdev_err:
> >> - memset(&isc->try_config, 0, sizeof(isc->try_config));
> >> +static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
> >> +{
> >> + isc_try_fmt(isc, f);
> >>
> >> - return ret;
> >> + /* make the try configuration active */
> >> + isc->config = isc->try_config;
> >> + isc->fmt = isc->try_fmt;
> >> +
> >> + v4l2_dbg(1, debug, &isc->v4l2_dev, "ISC set_fmt to %.4s @%dx%d\n",
> >> + (char *)&f->fmt.pix.pixelformat,
> >> + f->fmt.pix.width, f->fmt.pix.height);
> >> +
> >> + return 0;
> >> }
> >>
> >> -static int isc_set_fmt(struct isc_device *isc, struct v4l2_format *f)
> >> +static int isc_validate(struct isc_device *isc)
> >> {
> >> + int ret;
> >> + int i;
> >> + struct isc_format *sd_fmt = NULL;
> >> + struct v4l2_pix_format *pixfmt = &isc->fmt.fmt.pix;
> >> struct v4l2_subdev_format format = {
> >> .which = V4L2_SUBDEV_FORMAT_ACTIVE,
> >> + .pad = isc->remote_pad,
> >> + };
> >> + struct v4l2_subdev_pad_config pad_cfg = {};
> >> + struct v4l2_subdev_state pad_state = {
> >> + .pads = &pad_cfg,
> >> };
> >> - u32 mbus_code = 0;
> >> - int ret;
> >>
> >> - ret = isc_try_fmt(isc, f, &mbus_code);
> >> + /* Get current format from subdev */
> >> + ret = v4l2_subdev_call(isc->current_subdev->sd, pad, get_fmt, NULL,
> >> + &format);
> >> if (ret)
> >> return ret;
> >>
> >> - v4l2_fill_mbus_format(&format.format, &f->fmt.pix, mbus_code);
> >> - ret = v4l2_subdev_call(isc->current_subdev->sd, pad,
> >> - set_fmt, NULL, &format);
> >> - if (ret < 0)
> >> - return ret;
> >> + /* Identify the subdev's format configuration */
> >> + for (i = 0; i < isc->num_user_formats; i++)
> >> + if (isc->user_formats[i]->mbus_code == format.format.code) {
> >> + sd_fmt = isc->user_formats[i];
> >> + break;
> >> + }
> >> +
> >> + /* Check if the format is not supported */
> >> + if (!sd_fmt) {
> >> + v4l2_err(&isc->v4l2_dev,
> >> + "Current subdevice is streaming a media bus code that is not supported 0x%x\n",
> >> + format.format.code);
> >> + return -EPIPE;
> >> + }
> >> +
> >> + /* At this moment we know which format the subdev will use */
> >> + isc->try_config.sd_format = sd_fmt;
> >> +
> >> + /* If the sensor is not RAW, we can only do a direct dump */
> >> + if (!ISC_IS_FORMAT_RAW(isc->try_config.sd_format->mbus_code))
> >> + isc_try_configure_rlp_dma(isc, true);
> >>
> >> /* Limit to Atmel ISC hardware capabilities */
> >> - if (f->fmt.pix.width > isc->max_width)
> >> - f->fmt.pix.width = isc->max_width;
> >> - if (f->fmt.pix.height > isc->max_height)
> >> - f->fmt.pix.height = isc->max_height;
> >> + v4l_bound_align_image(&format.format.width, 16, isc->max_width, 0,
> >> + &format.format.height, 16, isc->max_height, 0, 0);
> >> +
> >> + /* Check if the frame size is the same. Otherwise we may overflow */
> >> + if (pixfmt->height != format.format.height ||
> >> + pixfmt->width != format.format.width) {
> >> + v4l2_err(&isc->v4l2_dev,
> >> + "ISC not configured with the proper frame size: %dx%d\n",
> >> + format.format.width, format.format.height);
> >> + return -EPIPE;
> >> + }
> >>
> >> - isc->fmt = *f;
> >> + v4l2_dbg(1, debug, &isc->v4l2_dev,
> >> + "Identified subdev using format %.4s with %dx%d %d bpp\n",
> >> + (char *)&sd_fmt->fourcc, pixfmt->width, pixfmt->height,
> >> + isc->try_config.bpp);
> >>
> >> + /* Reset and restart AWB if the subdevice changed the format */
> >> if (isc->try_config.sd_format && isc->config.sd_format &&
> >> isc->try_config.sd_format != isc->config.sd_format) {
> >> isc->ctrls.hist_stat = HIST_INIT;
> >> isc_reset_awb_ctrls(isc);
> >> isc_update_v4l2_ctrls(isc);
> >> }
> >> - /* make the try configuration active */
> >> +
> >> + /* Validate formats */
> >> + ret = isc_try_validate_formats(isc);
> >> + if (ret)
> >> + return ret;
> >> +
> >> + /* Obtain frame sizes if possible to have crop requirements ready */
> >> + isc_try_fse(isc, &pad_state);
> >> +
> >> + /* Configure ISC pipeline for the config */
> >> + ret = isc_try_configure_pipeline(isc);
> >> + if (ret)
> >> + return ret;
> >> +
> >> isc->config = isc->try_config;
> >>
> >> v4l2_dbg(1, debug, &isc->v4l2_dev, "New ISC configuration in place\n");
> >> @@ -1040,7 +1051,7 @@ static int isc_try_fmt_vid_cap(struct file *file, void *priv,
> >> {
> >> struct isc_device *isc = video_drvdata(file);
> >>
> >> - return isc_try_fmt(isc, f, NULL);
> >> + return isc_try_fmt(isc, f);
> >> }
> >>
> >> static int isc_enum_input(struct file *file, void *priv,
> >> @@ -1841,7 +1852,7 @@ static int isc_set_default_fmt(struct isc_device *isc)
> >> };
> >> int ret;
> >>
> >> - ret = isc_try_fmt(isc, &f, NULL);
> >> + ret = isc_try_fmt(isc, &f);
> >> if (ret)
> >> return ret;
> >>
> >> @@ -2015,6 +2026,24 @@ int isc_pipeline_init(struct isc_device *isc)
> >> }
> >> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>
> >> +int isc_link_validate(struct media_link *link)
> >> +{
> >> + struct video_device *vdev =
> >> + media_entity_to_video_device(link->sink->entity);
> >> + struct isc_device *isc = video_get_drvdata(vdev);
> >> + int ret;
> >> +
> >> + ret = v4l2_subdev_link_validate(link);
> >> + if (ret)
> >> + return ret;
> >
> > Doesn't v4l2_subdev_link_validate() call
> > v4l2_subdev_link_validate_default() if you don't define a .link_validate()
> > pad operation ? v4l2_subdev_link_validate_default() compares format
> > and sizes of both ends to the links and want them to be identical,
> > something that defeat the purpose of your custom isc_validate() ?
>
> I think so. However if I replace the validate call with my custom one, I
> thought that the old default should be called as well.
>

I was actually suggesting to remove the call to
v4l2_subdev_link_validate() completely as I thought you were
performing the same checks in your custom validate. If you're not
please ignore this comment, and even if you are my comment is wrong,
as I should have suggested you to remove the duplicated code and let
the framework do that :)


> Meaning, that maybe my custom validation does certain isc-related
> checks, while the v4l2_subdev_link_validate does exactly what you said.
> So, by defining a custom link_validate, do I lose the checks that are
> done by the v4l2_subdev_link_validate ?
> That's what I wanted to avoid. Chip in my checks, but keep the v4l2
> standard validation in place.
>
> Does that make sense ?

Absolutely, sorry for bothering and if the pipeline validation works
please go ahead with what you have!

Thanks
j

>
> > Thanks
> > j
> >
> >> +
> >> + return isc_validate(isc);
> >> +}
> >> +
> >> +static const struct media_entity_operations isc_entity_operations = {
> >> + .link_validate = isc_link_validate,
> >> +};
> >> +
> > };
> >> int isc_mc_init(struct isc_device *isc, u32 ver)
> >> {
> >> const struct of_device_id *match;
> >> @@ -2022,6 +2051,8 @@ int isc_mc_init(struct isc_device *isc, u32 ver)
> >>
> >> isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >> isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >> + isc->video_dev.entity.ops = &isc_entity_operations;
> >> +
> >> isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>
> >> ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >> index ec95c9665883..93375a61aea6 100644
> >> --- a/drivers/media/platform/atmel/atmel-isc-scaler.c
> >> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >> @@ -155,6 +155,10 @@ static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >> .init_cfg = isc_scaler_init_cfg,
> >> };
> >>
> >> +static const struct media_entity_operations isc_scaler_entity_ops = {
> >> + .link_validate = v4l2_subdev_link_validate,
> >> +};
> >> +
> >> static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >> .pad = &isc_scaler_pad_ops,
> >> };
> >> @@ -172,6 +176,7 @@ int isc_scaler_init(struct isc_device *isc)
> >>
> >> isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >> isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >> + isc->scaler_sd.entity.ops = &isc_scaler_entity_ops;
> >> isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >> isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >>
> >> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >> index 04ea5e340808..e36cf9d3942a 100644
> >> --- a/drivers/media/platform/atmel/atmel-isc.h
> >> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >> @@ -270,6 +270,7 @@ enum isc_scaler_pads {
> >> * @formats_list_size: size of formats_list array
> >> * @pads: media controller pads for isc video entity
> >> * @mdev: media device that is registered by the isc
> >> + * @mpipe: media device pipeline used by the isc
> >> * @remote_pad: remote pad on the connected subdevice
> >> * @scaler_sd: subdevice for the scaler that isc registers
> >> * @scaler_pads: media controller pads for the scaler subdevice
> >> @@ -295,6 +296,7 @@ struct isc_device {
> >> struct completion comp;
> >>
> >> struct v4l2_format fmt;
> >> + struct v4l2_format try_fmt;
> >> struct isc_format **user_formats;
> >> unsigned int num_user_formats;
> >>
> >> @@ -366,6 +368,7 @@ struct isc_device {
> >> struct {
> >> struct media_pad pads[ISC_PADS_NUM];
> >> struct media_device mdev;
> >> + struct media_pipeline mpipe;
> >>
> >> u32 remote_pad;
> >> };
> >> --
> >> 2.25.1
> >>
>

2022-02-12 00:20:38

by jacopo mondi

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

Hi Eugen

On Fri, Feb 11, 2022 at 12:32:14PM +0000, [email protected] wrote:
> On 2/11/22 1:55 PM, Jacopo Mondi wrote:
> > Hi Eugen
> >
> > On Fri, Feb 11, 2022 at 10:32:43AM +0000, [email protected] wrote:
> >> On 2/10/22 7:49 PM, Jacopo Mondi wrote:
> >>> Hi Eugen
> >>>
> >>> On Wed, Feb 09, 2022 at 06:59:37AM +0000, [email protected] wrote:
> >>>> On 2/8/22 9:30 PM, Jacopo Mondi wrote:
> >>>>> Hi Eugen
> >>>>>
> >>>>> On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
> >>>>>> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
> >>>>>>> Hi Eugen
> >>>>>>>
> >>>>>>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
> >>>>>>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
> >>>>>>>>> Hi Eugen
> >>>>>>>>>
> >>>>>>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
> >>>>>>>>>> Implement the support for media-controller.
> >>>>>>>>>> This means that the capabilities of the driver have changed and now
> >>>>>>>>>> it also advertises the IO_MC .
> >>>>>>>>>> The driver will register it's media device, and add the video entity to this
> >>>>>>>>>> media device. The subdevices are registered to the same media device.
> >>>>>>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
> >>>>>>>>>> It will also register a subdevice that allows cropping of the incoming frame
> >>>>>>>>>> to the maximum frame size supported by the ISC.
> >>>>>>>>>> The ISC will create a link between the subdevice that is asynchronously
> >>>>>>>>>> registered and the atmel_isc_scaler entity.
> >>>>>>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
> >>>>>>>>>> link.
> >>>>>>>>>>
> >>>>>>>>>> Signed-off-by: Eugen Hristev <[email protected]>
> >>>>>>>>>> ---
> >>>>>>>>>> Changes in v4:
> >>>>>>>>>> As suggested by Jacopo:
> >>>>>>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
> >>>>>>>>>> - moved init_mc/clean_mc to isc_base file
> >>>>>>>>>>
> >>>>>>>>>> Changes in v2:
> >>>>>>>>>> - implement try formats
> >>>>>>>>>>
> >>>>>>>>>> drivers/media/platform/atmel/Makefile | 2 +-
> >>>>>>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
> >>>>>>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
> >>>>>>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
> >>>>>>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
> >>>>>>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
> >>>>>>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
> >>>>>>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>>>>
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
> >>>>>>>>>> index 794e8f739287..f02d03df89d6 100644
> >>>>>>>>>> --- a/drivers/media/platform/atmel/Makefile
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/Makefile
> >>>>>>>>>> @@ -1,7 +1,7 @@
> >>>>>>>>>> # SPDX-License-Identifier: GPL-2.0-only
> >>>>>>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
> >>>>>>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
> >>>>>>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
> >>>>>>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
> >>>>>>>>>>
> >>>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
> >>>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>>>> index 6b0005987a17..6b482270eb93 100644
> >>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
> >>>>>>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>>>>>> struct isc_device, v4l2_dev);
> >>>>>>>>>> struct isc_subdev_entity *subdev_entity =
> >>>>>>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
> >>>>>>>>>> + int pad;
> >>>>>>>>>>
> >>>>>>>>>> if (video_is_registered(&isc->video_dev)) {
> >>>>>>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
> >>>>>>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
> >>>>>>>>>>
> >>>>>>>>>> subdev_entity->sd = subdev;
> >>>>>>>>>>
> >>>>>>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
> >>>>>>>>>> + MEDIA_PAD_FL_SOURCE);
> >>>>>>>>>> + if (pad < 0) {
> >>>>>>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
> >>>>>>>>>> + subdev->name);
> >>>>>>>>>> + return pad;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + isc->remote_pad = pad;
> >>>>>>>>>> +
> >>>>>>>>>> return 0;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
> >>>>>>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>>>>>> - unsigned int code, int *index)
> >>>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>>>>>> + unsigned int code, int *index)
> >>>>>>>>>> {
> >>>>>>>>>> struct isc_format *fmt = &isc->formats_list[0];
> >>>>>>>>>> unsigned int i;
> >>>>>>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
> >>>>>>>>>>
> >>>>>>>>>> return NULL;
> >>>>>>>>>> }
> >>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
> >>>>>>>>>>
> >>>>>>>>>> static int isc_formats_init(struct isc_device *isc)
> >>>>>>>>>> {
> >>>>>>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
> >>>>>>>>>> NULL, &mbus_code)) {
> >>>>>>>>>> mbus_code.index++;
> >>>>>>>>>>
> >>>>>>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
> >>>>>>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
> >>>>>>>>>> if (!fmt) {
> >>>>>>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
> >>>>>>>>>> mbus_code.code);
> >>>>>>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>>>>>> vdev->queue = q;
> >>>>>>>>>> vdev->lock = &isc->lock;
> >>>>>>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
> >>>>>>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
> >>>>>>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
> >>>>>>>>>> + V4L2_CAP_IO_MC;
> >>>>>>>>>> video_set_drvdata(vdev, isc);
> >>>>>>>>>>
> >>>>>>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
> >>>>>>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
> >>>>>>>>>> goto isc_async_complete_err;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> + ret = isc_scaler_link(isc);
> >>>>>>>>>> + if (ret < 0)
> >>>>>>>>>> + goto isc_async_complete_unregister_device;
> >>>>>>>>>> +
> >>>>>>>>>> + ret = media_device_register(&isc->mdev);
> >>>>>>>>>> + if (ret < 0)
> >>>>>>>>>> + goto isc_async_complete_unregister_device;
> >>>>>>>>>> return 0;
> >>>>>>>>>>
> >>>>>>>>>> +isc_async_complete_unregister_device:
> >>>>>>>>>> + video_unregister_device(vdev);
> >>>>>>>>>> +
> >>>>>>>>>> isc_async_complete_err:
> >>>>>>>>>> mutex_destroy(&isc->lock);
> >>>>>>>>>> return ret;
> >>>>>>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
> >>>>>>>>>> }
> >>>>>>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
> >>>>>>>>>>
> >>>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
> >>>>>>>>>> +{
> >>>>>>>>>> + const struct of_device_id *match;
> >>>>>>>>>> + int ret;
> >>>>>>>>>> +
> >>>>>>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
> >>>>>>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
> >>>>>>>>>
> >>>>>>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
> >>>>>>>>> to be able to have the media pipeline validated at
> >>>>>>>>> media_pipeline_start() time ?
> >>>>>>>>
> >>>>>>>> Hi,
> >>>>>>>>
> >>>>>>>> I am doing that in a subsequent patch. Things are not completely ready
> >>>>>>>> at this moment, because ISC still relies on the old mechanism to call
> >>>>>>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
> >>>>>>>> removes and adds checks to the link validate ).
> >>>>>>>>
> >>>>>>>
> >>>>>>> I see.. the subsequent patches are not part of this series, right ?
> >>>>>>
> >>>>>> No. Patch 7 does this. In that patch, the link validation is created and
> >>>>>> all the logic of format propagation is removed and reworked, and creates
> >>>>>> the need for the link_validate call
> >>>>>> Could you have also a look at that patch ?
> >>>>>
> >>>>> Ah ups, 7 was not in my inbox with the rest of the series. Just
> >>>>> noticed. I had even reviewed the previous version, I should have
> >>>>> remembered link_validate was added later
> >>>>>
> >>>>>>>
> >>>>>>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
> >>>>>>> the introduction of media controller or the scaler, right ? (I'm not
> >>>>>>> sure if the DTS patches do).
> >>>>>
> >>>>> Do DTS patches depend on media-controller plumbing ?
> >>>>
> >>>> I think not, but, I have not tested them without the media controller.
> >>>>
> >>>>>
> >>>>>>>
> >>>>>>> If that's the case, would it make sense to to fast-track them (as some
> >>>>>>> of them are fixes) and then on top plumb the media controller
> >>>>>>> infrastructure with this patch and the ones you have been mentioning ?
> >>>>>>
> >>>>>> It would make sense.
> >>>>>>
> >>>>>
> >>>>> Once the other discussion with Hans about the stop state it might make
> >>>>> sense to fastrack the first part of the series ?
> >>>>
> >>>> The other three patches (4,5,6) are quite independent and can go faster.
> >>>>>
> >>>>>>>
> >>>>>>>>>
> >>>>>>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>>>>>> +
> >>>>>>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
> >>>>>>>>>> + isc->pads);
> >>>>>>>>>> + if (ret < 0) {
> >>>>>>>>>> + dev_err(isc->dev, "media entity init failed\n");
> >>>>>>>>>> + return ret;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + isc->mdev.dev = isc->dev;
> >>>>>>>>>> +
> >>>>>>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
> >>>>>>>>>> + isc->dev->of_node);
> >>>>>>>>>> +
> >>>>>>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
> >>>>>>>>>> + sizeof(isc->mdev.driver_name));
> >>>>>>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
> >>>>>>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
> >>>>>>>>>> + isc->v4l2_dev.name);
> >>>>>>>>>> + isc->mdev.hw_revision = ver;
> >>>>>>>>>> +
> >>>>>>>>>> + media_device_init(&isc->mdev);
> >>>>>>>>>> +
> >>>>>>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
> >>>>>>>>>> +
> >>>>>>>>>> + return isc_scaler_init(isc);
> >>>>>>>>>> +}
> >>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
> >>>>>>>>>> +
> >>>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc)
> >>>>>>>>>> +{
> >>>>>>>>>> + media_entity_cleanup(&isc->video_dev.entity);
> >>>>>>>>>> +}
> >>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
> >>>>>>>>>> +
> >>>>>>>>>> /* regmap configuration */
> >>>>>>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
> >>>>>>>>>> const struct regmap_config isc_regmap_config = {
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>>>> new file mode 100644
> >>>>>>>>>> index 000000000000..ec95c9665883
> >>>>>>>>>> --- /dev/null
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
> >>>>>>>>>> @@ -0,0 +1,245 @@
> >>>>>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
> >>>>>>>>>> +/*
> >>>>>>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
> >>>>>>>>>> + *
> >>>>>>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
> >>>>>>>>>
> >>>>>>>>> Time flies! It's 2022 already :)
> >>>>>>>>>
> >>>>>>>>>> + *
> >>>>>>>>>> + * Author: Eugen Hristev <[email protected]>
> >>>>>>>>>> + *
> >>>>>>>>>> + */
> >>>>>>>>>> +
> >>>>>>>>>> +#include <media/media-device.h>
> >>>>>>>>>> +#include <media/media-entity.h>
> >>>>>>>>>> +#include <media/v4l2-device.h>
> >>>>>>>>>> +#include <media/v4l2-subdev.h>
> >>>>>>>>>> +
> >>>>>>>>>> +#include "atmel-isc-regs.h"
> >>>>>>>>>> +#include "atmel-isc.h"
> >>>>>>>>>> +
> >>>>>>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
> >>>>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>>>> + struct v4l2_subdev_format *format)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>>>>>> +
> >>>>>>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>>>> + format->pad);
> >>>>>>>>>> + format->format = *v4l2_try_fmt;
> >>>>>>>>>> +
> >>>>>>>>>> + return 0;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + format->format = isc->scaler_format;
> >>>>>>>>>
> >>>>>>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
> >>>>>>>>> I wonder why it lives in the isc_device struct.
> >>>>>>>>
> >>>>>>>> isc_device is a placeholder for all isc things.
> >>>>>>>>
> >>>>>>>> I would not create a separate struct in this file that would have to be
> >>>>>>>> allocated etc... just for one two things. So I preferred to have it in
> >>>>>>>> the same place as all the other things.
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + return 0;
> >>>>>>>>>> +}
> >>>>>>>>>> +
> >>>>>>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
> >>>>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>>>> + struct v4l2_subdev_format *req_fmt)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
> >>>>>>>>>> + struct isc_format *fmt;
> >>>>>>>>>> + unsigned int i;
> >>>>>>>>>> +
> >>>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>>>>>> + v4l_bound_align_image
> >>>>>>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
> >>>>>>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
> >>>>>>>>>> + else
> >>>>>>>>>> + v4l_bound_align_image
> >>>>>>>>>> + (&req_fmt->format.width, 16, 10000, 0,
> >>>>>>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
> >>>>>>>>>
> >>>>>>>>> Where does 10000 come from ?
> >>>>>>>>
> >>>>>>>> It's a random number. Do you have any suggestion for a better one ?
> >>>>>>>
> >>>>>>> An actual HW limit ? :)
> >>>>>>
> >>>>>> There is no hardware limit. The ISC just stops sampling pixels once the
> >>>>>> crop limit is reached. The element that generates pixels could go on
> >>>>>> forever , or until the next HBLANK/VBLANK. So what limit could I place
> >>>>>> here ?
> >>>>>> The cropping is mandatory, otherwise there could be an overflow w.r.t.
> >>>>>> the buffer size if the ISC would sample more pixels than the software
> >>>>>> expects it to.
> >>>>>>
> >>>>>>>
> >>>>>>>> Maybe this would be much more clear with my comments below (where I will
> >>>>>>>> explain what the isc scaler does )
> >>>>>>>> In short, it allows the other entity in the link to 'sink' a huge format
> >>>>>>>> into this 'scaler' . Because the scaler can crop anything basically down
> >>>>>>>> to the biggest format size the ISC could handle.
> >>>>>>>>
> >>>>>>>
> >>>>>>> doesn't it have any documented input size limit ?
> >>>>>>
> >>>>>> As stated above, ISC handles a pixel stream and stores it in an internal
> >>>>>> SRAM. ISC stops sampling when this SRAM is full or when the cropping
> >>>>>> limit is reached.
> >>>>>> After storing the pixels in the SRAM (which is limited to the ISC
> >>>>>> maximum frame size), the ISC will work on the pixels and then DMA the
> >>>>>> frame to another RAM.
> >>>>>>
> >>>>>
> >>>>> I understand...
> >>>>>
> >>>>> So whatever the input size is, only the first n-th bytes are actually
> >>>>> dumped to memory, the following ones are discarded.
> >>>>
> >>>> Yes ! That's the behavior.
> >>>>>
> >>>>> So the crop is always applied on the top-leftmost corner as a
> >>>>> consequence, right ?
> >>>>
> >>>> Right. Pixels come in order. The ISC does not really know how many
> >>>> pixels are coming until its memory is full (or counting until the crop
> >>>> limit is reached). After that it stops sampling pixels, and waits for
> >>>> vblank to finish the frame.
> >>>>
> >>>>
> >>>>>
> >>>>> I understand why you had difficulties picking a default :)
> >>>>>
> >>>>> This is not easy to handle, as the scaler doesn't actually applies a
> >>>>> crop but rather stops capturing after the destination buffer has been
> >>>>> saturated. So the limiting factor is the full image size, not one of
> >>>>> its dimensions, like in example the line length. This makes it hard
> >>>>> to adjust cropping to something meaningful for most use cases.
> >>>>
> >>>> I didn't know how to name this except crop. It 'crops out' what does not
> >>>> fit in the internal memory of the ISC
> >>>>
> >>>>>
> >>>>> In your below example you suggest that an image size of (800000 x 10)
> >>>>> would theoretically be captured correctly. What if the input is of
> >>>>> size (800000 x 10 + 1). What size would you set the crop rectangle to ?
> >>>>>
> >>>>> I wouldn't however be too much concerned for the moment, but I would
> >>>>> record this in a comment ?
> >>>>
> >>>> I discussed a bit more with designer and looked in datasheet.
> >>>> It looks like the horizontal line length is not a hard limit, but it
> >>>> makes the ISC internal ISP stop working, so I would keep the horizontal
> >>>> limit as well for now, unless an important use case (I am coming back to
> >>>> that). The vertical limit looks to be a hard requirement.
> >>>> So in short, we have to impose a vertical limit of 2464 , we cannot have
> >>>> more lines than these. So if the horizontal is 100 pixels e.g., we can
> >>>> capture only 100x2464 .
> >>>> In practice we could capture [16,3264] on the horizontal, and [16,2464]
> >>>> on the vertical, and in theory horizontal could go more like [16, 4000],
> >>>> but we have to limit the vertical, and without the internal ISP.
> >>>> So I would not haphazardly go in this direction. Let's stick to the
> >>>> datasheet maximum frame 3264x2464 .
> >>>> According to this , I changed the frame size of the ISC to be a
> >>>> continuous size from [16,16] to [3264,2464]
> >>>> About that use case that I talked about earlier, we have one sensor ,
> >>>> imx274, which is more wide, and outputs something like 3800x2100 .
> >>>> With the cropping and limitations we have in place now, this will be
> >>>> limited to 3264x2100 , which is in line with datasheet and what the ISC
> >>>> supports now.
> >>>> As in theory we could capture 3800x2100 as long as the SRAM doesn't get
> >>>> full, for now I would like to avoid capturing this size (and disable the
> >>>> internal ISP), until we have someone really asking for such kind of
> >>>> capture and with some good reasons.
> >>>> In short I would like to have the crop rectangle to 3264x2464 and the
> >>>> bounds would be limited to the incoming frame but no H>3264 and no
> >>>> V>2464. Does that make sense ?
> >>>
> >>> Just to make sure I got it: is the ISC actually able to crop on the
> >>> line length ? If you apply a 3264 limit, will the last 536 pixels of
> >>> each line will be dropped if the input frame is 3800x2100 as your
> >>> above example ?
> >>
> >> Yes. Last 536 pixels will be dropped.
> >> After the 3264 limit, no more pixels are being sampled until HBLANK.
> >>>
> >>> Or rather, is the SRAM size limit the issue and the lines gets
> >>> captured in full lenght no matter what, until the buffer gets filled up
> >>> and all the successive pixels dropped.
> >>>
> >>> In other words, do your cropped image look like:
> >>>
> >>> 3264
> >>> +--------------------------------X-----------+ 3800
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> >>> +--------------------------------------------+ 2100
> >>>
> >>> Or rather
> >>>
> >>>
> >>> 3264
> >>> +--------------------------------X-----------+ 3800
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> >>> |xxxxxxxxxxxxxxxxx| |
> >>> | |
> >>> +--------------------------------------------+ 2100
> >>>
> >>> I assume the first, as otherwise you wouldn't be able to interpret the
> >>> image as 3264x2100.
> >>
> >> Yes, the first.
> >>
> >>>
> >>> I'm asking because you had the un-cropped 3280x2464 size as the ISC
> >>> source pad size in the paste of the media graph you provided below,
> >>> which might mean the output image is actually full size in lines
> >>> lenght and it's maybe missing pixels at the end, in case the ISC
> >>> operates as in the above second case.
> >>
> >> No, the image won't be padded with blanks on the right side. Actually
> >> this cannot happen due to a very big reason: the frame is being written
> >> to the main memory by a DMA controller. This DMA controller will write
> >> all the lines at consecutive addresses, and it will write exactly how
> >> many pixels are being captured by the module from the ISC named
> >> 'parallel front end' which does the cropping. Thus, the DMA will never
> >> output empty padding bytes at the end of each line, and it will not add
> >> a certain stride to each line if not instructed to. In fact the DMA can
> >> add a stride offset for the address of each line, which can be used for
> >> composing or copying the frame into another frame, but this is a
> >> different story. ( I tried this for fun, but not implemented in the
> >> driver at the moment)
> >
> > Oh that will really be interesting and it's indeed an argument for
> > hainvg a scaler subdevice indeed
> >
> >>
> >> The cropping is also mandatory even if the captured frame is less than
> >> the maximum size, because the ISC has no configuration related to frame
> >> size in fact. The parallel front end captures pixels and the dma engine
> >> will write them to memory. There is no control over this except the
> >> cropping.
> >> So for example if the software expects 640x480 but somehow someone sends
> >> more pixels, it can lead to buffer overflow if the parallel front end is
> >> not configured to ignore(crop) anything beyond 640x480 .
> >>
> >
> > I see and I think you have correctly plumbed that with the scaler sink
> > pad crop rectangle. I don't seem to find where the crop rectangle
> > sizes are actually applied to the ISC in this series ? Could you point
> > me to it for sake of my better understanding ?
>
> The ISC is already doing that for some time now:
>
> https://elixir.bootlin.com/linux/latest/source/drivers/media/platform/atmel/atmel-isc-base.c#L459
>
> Initially it was more like a safety feature, because I discovered that
> there was no way of limiting the incoming pixels, and if some malevolent
> entity would just keep sending pixels, it could lead to exploits.
>
> Now, the scaler is just exposing this to the user.
>
> >
> >> With this being said, or having an understanding about the sink pad of
> >> the ISC, how should I configure the sink frame size ?
> >
> > I don't think you should, the video device format is set through the
> > usual S_FMT as it happens already.
> >
> > What I think it's missing in your implementation is only to propagate the
> > crop rectangle size of the scaler sink pad to the scaler source pad
> > format (as you don't have any composition rectangles yet) [1]
> >
> > With that done, you should be able to capture from your video device
> > in the cropped dimensions.
> >
> > The (simplified) media pipeline will look something like
> >
> > sensor
> > type: Subdev
> > pad 0 (sink) (3280, 2462)
> > -> scaler:0
> >
> > scaler
> > type: Subdev
> > pad 0 (sink) (3280, 2464)
> > crop: (3262, 2464)
> > <- sensor:0
> > pad 1 (source) (3264, 2464)
> > -> isc_common
> >
> > isc_common
> > type V4L2
> > pad 0 (sink)
> > <- scaler:1
> >
> > And the video device format set to (3264, 2464)
> >
> > Then later you might want to implement s_selection on the scaler to have
> > the crop size configurable.
> >
> > One additional question: can the crop be panned ? What I mean is that
> > you explained you can instruct the ISC to drop pixels after a certain
> > line lenght. This will cause the crop not to be centered on the input
> > image. Can the ISC be configured to ignore data -before- and -after-
> > a certain number of pixels or lines ? In that case you might want to
> > handle the top,left corner of the crop rectangle too. Anyway,
> > something for later..
>
> Looking over this, it looks possible. I have not tried it.
> It looks like configuring a minimum row /col number would make the ISC
> drop the first n pixels/ first n lines.
>
> I will have to try this, but if it works, user could configure the
> cropping area, that would be interesting ! Anyway I can't implement this
> right now, but it's a nice thing to work on later.
>
> >
> > [1] https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#order-of-configuration-and-format-propagation
> >
> >> Is there any possibility to avoid limits on the frame size and have it
> >> arbitrary ?
> >> I am 99% sure that v4l2-compliance will not be happy if there is no
> >> limit on the frame on entity pads (hence, I bound-aligned the frame to
> >> 16x16 up to 10k x 10k )
> >>
> >
> > That I'm not sure, but I think having size limits is required ?
> >
> > Anyway, from my side, the next version should be good to go. Thanks
> > for sticking to all my comments and questions so far!
>
> What should I do about the 10000 x 10000 pixels ?
>

If only used for aligning in

v4l_bound_align_image
(&req_fmt->format.width, 16, 10000, 0,
&req_fmt->format.height, 16, 10000, 0, 0);

I think you can use UINT_MAX to make it more explicit ?


> Thanks for reviewing !
>
> Eugen
> >
> > Thanks
> > j
> >
> >
> >>>
> >>>>
> >>>>>
> >>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
> >>>>>>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>>>>>> +
> >>>>>>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
> >>>>>>>>>
> >>>>>>>>> So you rely on the isc format list for the scaler as well ?
> >>>>>>>>> I think it's fine as long as they are identical
> >>>>>>>>
> >>>>>>>> Yes, the scaler is kind of a dummy entity , but it's required by the
> >>>>>>>> media controller validation of the pipeline.
> >>>>>>>>
> >>>>>>>
> >>>>>>> More on this later
> >>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + if (!fmt)
> >>>>>>>>>> + fmt = &isc->formats_list[0];
> >>>>>>>>>> +
> >>>>>>>>>> + req_fmt->format.code = fmt->mbus_code;
> >>>>>>>>>> +
> >>>>>>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
> >>>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>>>> + req_fmt->pad);
> >>>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>>>>>> + /* Trying on the pad sink makes the source sink change too */
> >>>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
> >>>>>>>>>> + v4l2_try_fmt =
> >>>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
> >>>>>>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
> >>>>>>>>>> +
> >>>>>>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
> >>>>>>>>>> + 16, isc->max_width, 0,
> >>>>>>>>>> + &v4l2_try_fmt->height,
> >>>>>>>>>> + 16, isc->max_height, 0, 0);
> >>>>>>>>>> + }
> >>>>>>>>>> + /* if we are just trying, we are done */
> >>>>>>>>>> + return 0;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + isc->scaler_format = req_fmt->format;
> >>>>>>>>>
> >>>>>>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
> >>>>>>>>>
> >>>>>>>>> Actually, I would like to know more about the scaler device
> >>>>>>>>> capabilities. What functions can this IP perform ? Does it do
> >>>>>>>>> cropping, can it also do (down)scaling or even composition ?
> >>>>>>>>>
> >>>>>>>>> I think it is worth to read
> >>>>>>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
> >>>>>>>>> where it is reported how cropping and scaling are implemented by using
> >>>>>>>>> the selection API.
> >>>>>>>>>
> >>>>>>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
> >>>>>>>>> for which you need to implement support to by adding s_selection on
> >>>>>>>>> the sink pad TGT_CROP target.
> >>>>>>>>
> >>>>>>>> The scaler is kind of a dummy entity.
> >>>>>>>> The problem is that different subdevices cannot be connected directly to
> >>>>>>>> a v4l entity, because the ISC does cropping on the incoming frame.
> >>>>>>>> The ISC has a limit on the total number of pixels per frame, thus it
> >>>>>>>> will do a crop.
> >>>>>>>
> >>>>>>> What is this limit related to ? I don't know the ISC internals, but is
> >>>>>>> the limitation due to the overall image size, or maybe it's due to
> >>>>>>> some line buffers being limited in the size they can transfer to
> >>>>>>> memory ? Is it a limitation on the size of the image in pixels, or is
> >>>>>>> it due to a limitation of the processing bandwidth on the input
> >>>>>>> interface and thus depends on the actual size of the image in bytes ?
> >>>>>>
> >>>>>> Limitation is both on line size and full image size. From my
> >>>>>> understanding, is that the internal SRAM is limited (8 Mpixels with a
> >>>>>> small safe buffer on top in the case of sama7g5 )
> >>>>>> This SRAM could be used as 800000x10 for example, or 3200x2464 for
> >>>>>> another example. However, I am not sure if a real line of 800,000 pixels
> >>>>>> make sense, or if there is another internal mechanism in the ISC that
> >>>>>> stops it.
> >>>>>> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
> >>>>>> the maximum frame that it can handle
> >>>>>>
> >>>>>>>
> >>>>>>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
> >>>>>>>> only handle 3264x2464 , then we have two choices:
> >>>>>>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
> >>>>>>>> 3280x2464, and the whole pipeline fails to configure, there is a
> >>>>>>>> mismatch between /dev/video output and what the whole pipeline produces
> >>>>>>>
> >>>>>>> I'm confused in first place. If the ISC cannot output anything larger
> >>>>>>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
> >>>>>>> this should be adjusted to 3264x2464.
> >>>>>>
> >>>>>> yes, that is true. However, without an entity that says 'I am doing
> >>>>>> cropping', the media pipeline will fail, because previous entity will
> >>>>>> output 3280x2464 .
> >>>>>> The conclusion to create a scaler entitity was done during my last year
> >>>>>> discussions on the IRC. I could not find a way to perform this cropping
> >>>>>> at the v4l entity side, and one reason is that the v4l entity does not
> >>>>>> support what I needed , or at least that was my understanding.
> >>>>>> You claim that the v4l entity could perform the cropping in the same
> >>>>>> entity ?
> >>>>>
> >>>>> I thought so, but I haven't tried to be honest
> >>>>>
> >>>>>>
> >>>>>> Let's assume for a minute that it could do this. Even so, I would prefer
> >>>>>> to have a separate scaler entity to do this , for several reasons:
> >>>>>> -> I would like to be able to crop the frame arbitrarily if I want to,
> >>>>>> and the scaler would allow me to do this easier
> >>>>>> -> it would be more obvious that the frame is cropped by having this
> >>>>>> entity there
> >>>>>> -> in fact some of the ISC versions really have a down scaler, that I
> >>>>>> have not implemented yet. So it would be useful to already have this in
> >>>>>> place to be able to scale down with it at a later time.
> >>>>>
> >>>>> I think this last reason is enough to have an entity plumbed in
> >>>>> already
> >>>>>
> >>>>>>
> >>>>>>>
> >>>>>>> Or is the ISC limitation on the -input- side ?
> >>>>>>>
> >>>>>>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
> >>>>>>>> the ISC can handle at most -> cannot , because the sensor let's say can
> >>>>>>>> *only* produce 3280x2464 and *any* other frame size returns an error.
> >>>>>>>
> >>>>>>> buggy sensor I would say :)
> >>>>>>
> >>>>>> I cannot tell this to the customers, and I cannot fail the whole
> >>>>>> pipeline because the sensor does not crop 16 pixels, when in fact I can
> >>>>>> do this very easily on the ISC side.
> >>>>>>
> >>>>>
> >>>>> Fair enough :)
> >>>>>
> >>>>>>>
> >>>>>>>>
> >>>>>>>> The only solution to make both worlds happy, is to have a dummy entity
> >>>>>>>> called 'scaler' which in fact now it only performs a simple crop.
> >>>>>>>> It accepts any frame size at input (hence the 10000x10000 which you saw
> >>>>>>>> earlier), and outputs at most 3264x2464 the isc_max_height and
> >>>>>>>> isc_max_width.
> >>>>>>>
> >>>>>>> I have a maybe dumb question: can the cropping operation be modeled
> >>>>>>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
> >>>>>>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
> >>>>>>
> >>>>>> I am not sure. This is what I wanted initially. But then I thought about
> >>>>>> the three reasons stated above for the use of the scaler entity, and
> >>>>>> discussed with folks on the IRC, and come up with the solution for the
> >>>>>> scaler entity
> >>>>>>
> >>>>>>>
> >>>>>>>>
> >>>>>>>> So to answer your question, the isc scaler is a software model for a
> >>>>>>>> simple cropping, that would make media controller happy, and capture
> >>>>>>>> software happy.
> >>>>>>>>
> >>>>>>>> Here it how it looks :
> >>>>>>>>
> >>>>>>>> Device topology
> >>>>>>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
> >>>>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>>>> device node name /dev/v4l-subdev0
> >>>>>>>> pad0: Sink
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>>>> crop.bounds:(0,0)/3264x2464
> >>>>>>>> crop:(0,0)/3264x2464]
> >>>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>>>>>> pad1: Source
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>>>>>
> >>>>>>>> - entity 4: csi2dc (2 pads, 2 links)
> >>>>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>>>> device node name /dev/v4l-subdev1
> >>>>>>>> pad0: Sink
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>>>> <- "dw-csi.0":1 [ENABLED]
> >>>>>>>> pad1: Source
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
> >>>>>>>>
> >>>>>>>> - entity 7: dw-csi.0 (2 pads, 2 links)
> >>>>>>>> type V4L2 subdev subtype Unknown flags 0
> >>>>>>>> device node name /dev/v4l-subdev2
> >>>>>>>> pad0: Sink
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>>>>>> <- "imx219 1-0010":0 [ENABLED]
> >>>>>>>> pad1: Source
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464]
> >>>>>>>> -> "csi2dc":0 [ENABLED]
> >>>>>>>>
> >>>>>>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
> >>>>>>>> type V4L2 subdev subtype Sensor flags 0
> >>>>>>>> device node name /dev/v4l-subdev3
> >>>>>>>> pad0: Source
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>>>> xfer:srgb ycbcr:601 quantization:full-range
> >>>>>>>> crop.bounds:(8,8)/3280x2464
> >>>>>>>> crop:(8,8)/3280x2464]
> >>>>>>>> -> "dw-csi.0":0 [ENABLED]
> >>>>>>>>
> >>>>>>>> - entity 24: atmel_isc_common (1 pad, 1 link)
> >>>>>>>> type Node subtype V4L flags 1
> >>>>>>>> device node name /dev/video0
> >>>>>>>> pad0: Sink
> >>>>>>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> Scaler does this one cute little thing :
> >>>>>>>>
> >>>>>>>> pad0: Sink
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
> >>>>>>>> crop.bounds:(0,0)/3264x2464
> >>>>>>>> crop:(0,0)/3264x2464]
> >>>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
> >>>>>>>> pad1: Source
> >>>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
> >>>>>>>
> >>>>>>> Shouldn't this be 3264x2464 as that's what the entity outputs after
> >>>>>>> the cropping ? And shouldn't then be 3264x2464 the size output from
> >>>>>>> the video device too ?
> >>>>>>
> >>>>>> That's right.
> >>>>>> I don't know why the source format for the scaler is still 3280x2464.
> >>>>>> Maybe there is a bug on g_fmt for it.. have to check it.
> >>>>>
> >>>>> Thanks, I think this should be fixed ten
> >>>>>
> >>>>>
> >>>>>>
> >>>>>> Anyway, the video format is like this :
> >>>>>>
> >>>>>> # v4l2-ctl --get-fmt-video
> >>>>>> Format Video Capture:
> >>>>>> Width/Height : 3264/2464
> >>>>>> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
> >>>>>> Field : None
> >>>>>> Bytes per Line : 6528
> >>>>>> Size Image : 16084992
> >>>>>> Colorspace : sRGB
> >>>>>> Transfer Function : Default (maps to sRGB)
> >>>>>> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
> >>>>>> Quantization : Default (maps to Full Range)
> >>>>>> Flags :
> >>>>>> #
> >>>>>>
> >>>>>> and initializing the pipeline looks fine from user perspective.
> >>>>>> The scaler is in fact the solution to make this pipeline work with
> >>>>>> libcamera.
> >>>>>> I found this cropping to be an issue in media controller when trying it
> >>>>>> with libcamera. Otherwise, the other user space apps which I was using
> >>>>>> never complained that anything was wrong
> >>>>>> Libcamera simply refuses to acknowledge the pipeline if the video output
> >>>>>> is 3264x2464 but there is no entity that changes the format from
> >>>>>> 3280x2464 down
> >>>>>
> >>>>> Not sure why libcamera should play a role there. Isn't it the media
> >>>>> pipeline validation that complains and returns an -EPIPE ?
> >>>>
> >>>> I tried this before the media_pipeline_start .
> >>>> Perhaps libcamera was doing some similar checks, anyway, it was
> >>>> complaining and not adding the stream, and reported no usable streams/camera
> >>>>
> >>>> Once I will rework some of the things according to your review I will
> >>>> run libcamera again to see what it complains about
> >>>>
> >>>>>
> >>>>>>
> >>>>>>>
> >>>>>>>
> >>>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> Which is what we needed.
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + return 0;
> >>>>>>>>>> +}
> >>>>>>>>>> +
> >>>>>>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
> >>>>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>>>> + struct v4l2_subdev_mbus_code_enum *code)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>>>> + int supported_index = 0;
> >>>>>>>>>> + int i;
> >>>>>>>>>> +
> >>>>>>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
> >>>>>>>>>> + if (!isc->formats_list[i].sd_support)
> >>>>>>>>>> + continue;
> >>>>>>>>>
> >>>>>>>>> The sd_support flag still doesn't click in my head.
> >>>>>>>>>
> >>>>>>>>> Shouldn't the enumeration of available formats on the scaler do not
> >>>>>>>>> depend on the sensor supproted formats ?
> >>>>>>>>
> >>>>>>>> You're right. I will have to check it again.
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> + if (supported_index == code->index) {
> >>>>>>>>>> + code->code = isc->formats_list[i].mbus_code;
> >>>>>>>>>> + return 0;
> >>>>>>>>>> + }
> >>>>>>>>>> + supported_index++;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + return -EINVAL;
> >>>>>>>>>> +}
> >>>>>>>>>> +
> >>>>>>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
> >>>>>>>>>> + struct v4l2_subdev_state *sd_state,
> >>>>>>>>>> + struct v4l2_subdev_selection *sel)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>>>> +
> >>>>>>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
> >>>>>>>>>> + return -EINVAL;
> >>>>>>>>>> +
> >>>>>>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
> >>>>>>>>>> + sel->target != V4L2_SEL_TGT_CROP)
> >>>>>>>>>> + return -EINVAL;
> >>>>>>>>>> +
> >>>>>>>>>> + sel->r.height = isc->max_height;
> >>>>>>>>>> + sel->r.width = isc->max_width;
> >>>>>>>>>
> >>>>>>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
> >>>>>>>>> as it represents the maximum valid crop rectangle.
> >>>>>>>>>
> >>>>>>>>> TGT_CROP should report the configured crop rectangle which can be
> >>>>>>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
> >>>>>>>>> the spec is correct
> >>>>>>>>
> >>>>>>>> So you would like to have this differentiated, and report the
> >>>>>>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
> >>>>>>>> reported now, the maximum size of the ISC frame .
> >>>>>>>> My understanding is correct ?
> >>>>>>>>
> >>>>>>>
> >>>>>>> I didn't know you have an HW limitation, so your _BOUNDS is not the
> >>>>>>> input image size but rather 3264x2464 ( == max_width x max_height).
> >>>>>>>
> >>>>>>> What I meant is that _BOUNDS should report the maximum rectangle size
> >>>>>>> that can be applied to the _CROP target. In you case you have an HW
> >>>>>>> limitation 3264x2464 and that's the largest rectangle you can apply.
> >>>>>> So the CROP should be at 3264x2464
> >>>>>>> TGT_CROP can be initialized to the same as _BOUND, but if you
> >>>>>>> implement s_selection it should report what has been there applied.
> >>>>>> and BOUND to actual frame size ?
> >>>>>>> But as you don't implement s_selection yet, I think this is fine for
> >>>>>>> now. Maybe a little comment ?
> >>>>>>
> >>>>>> It could also be the case where e.g. the sensor is outputting 1920x1080,
> >>>>>> in this case the scaler would do nothing.
> >>>>>> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
> >>>>>> If the sensor is outputting 1920x1080, this format comes directly from
> >>>>>> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
> >>>>>> sensor's problem)
> >>>>>>>
> >>>>>>> Also, is set->r zeroed by the framework before getting here ?
> >>>>>>> Otherwise you should set r.left and r.top to 0
> >>>>>>
> >>>>>> If these are redundant, no problem to remove them
> >>>>>
> >>>>> I was actually confused. I was suggesting you to add...
> >>>>>
> >>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + sel->r.left = 0;
> >>>>>>>>>> + sel->r.top = 0;
> >>>>>
> >>>>> These ^
> >>>>>
> >>>>> So nothing to change. Sorry I've missed them
> >>>>>
> >>>>> Thanks
> >>>>> j
> >>>>>
> >>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + return 0;
> >>>>>>>>>> +}
> >>>>>>>>>> +
> >>>>>>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
> >>>>>>>>>> + struct v4l2_subdev_state *sd_state)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
> >>>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
> >>>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
> >>>>>>>>>> +
> >>>>>>>>>> + *v4l2_try_fmt = isc->scaler_format;
> >>>>>>>>>> +
> >>>>>>>>>> + return 0;
> >>>>>>>>>> +}
> >>>>>>>>>> +
> >>>>>>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
> >>>>>>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
> >>>>>>>>>> + .set_fmt = isc_scaler_set_fmt,
> >>>>>>>>>> + .get_fmt = isc_scaler_get_fmt,
> >>>>>>>>>> + .get_selection = isc_scaler_g_sel,
> >>>>>>>>>> + .init_cfg = isc_scaler_init_cfg,
> >>>>>>>>>
> >>>>>>>>> .link_validate = v4l2_subdev_link_validate_default,
> >>>>>>>>>
> >>>>>>>>> To have the formats at the end of links that point to this entity
> >>>>>>>>> validated (I think the framework already calls it if not set though,
> >>>>>>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
> >>>>>>>>>
> >>>>>>>>>> +};
> >>>>>>>>>> +
> >>>>>>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
> >>>>>>>>>> + .pad = &isc_scaler_pad_ops,
> >>>>>>>>>> +};
> >>>>>>>>>> +
> >>>>>>>>>> +int isc_scaler_init(struct isc_device *isc)
> >>>>>>>>>> +{
> >>>>>>>>>> + int ret;
> >>>>>>>>>> +
> >>>>>>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
> >>>>>>>>>> +
> >>>>>>>>>> + isc->scaler_sd.owner = THIS_MODULE;
> >>>>>>>>>> + isc->scaler_sd.dev = isc->dev;
> >>>>>>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
> >>>>>>>>>> + "atmel_isc_scaler");
> >>>>>>>>>
> >>>>>>>>> I would drop 'atmel' for brevity, unless other entities have this
> >>>>>>>>> prefix set already
> >>>>>>>>
> >>>>>>>> The v4l entity takes it's name from the module name, which is
> >>>>>>>> atmel_isc_common, so I thought to keep the prefix
> >>>>>>>>
> >>>>>>>
> >>>>>>> Ack!
> >>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> >>>>>>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
> >>>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
> >>>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
> >>>>>>>>>> +
> >>>>>>>>>> + isc->scaler_format.height = isc->max_height;
> >>>>>>>>>> + isc->scaler_format.width = isc->max_width;
> >>>>>>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
> >>>>>>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
> >>>>>>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
> >>>>>>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
> >>>>>>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
> >>>>>>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
> >>>>>>>>>> +
> >>>>>>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
> >>>>>>>>>> + ISC_SCALER_PADS_NUM,
> >>>>>>>>>> + isc->scaler_pads);
> >>>>>>>>>> + if (ret < 0) {
> >>>>>>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
> >>>>>>>>>> + return ret;
> >>>>>>>>>> + }
> >>>>>>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
> >>>>>>>>>> + if (ret < 0) {
> >>>>>>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
> >>>>>>>>>> + return ret;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + return ret;
> >>>>>>>>>> +}
> >>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
> >>>>>>>>>> +
> >>>>>>>>>> +int isc_scaler_link(struct isc_device *isc)
> >>>>>>>>>> +{
> >>>>>>>>>> + int ret;
> >>>>>>>>>> +
> >>>>>>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
> >>>>>>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
> >>>>>>>>>> + ISC_SCALER_PAD_SINK,
> >>>>>>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>>>>>> +
> >>>>>>>>>> + if (ret < 0) {
> >>>>>>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>>>>>> + isc->current_subdev->sd->entity.name,
> >>>>>>>>>> + isc->scaler_sd.entity.name);
> >>>>>>>>>> + return ret;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
> >>>>>>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
> >>>>>>>>>> +
> >>>>>>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
> >>>>>>>>>> + ISC_SCALER_PAD_SOURCE,
> >>>>>>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
> >>>>>>>>>> + MEDIA_LNK_FL_ENABLED |
> >>>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
> >>>>>>>>>> +
> >>>>>>>>>> + if (ret < 0) {
> >>>>>>>>>> + v4l2_err(&isc->v4l2_dev,
> >>>>>>>>>> + "Failed to create pad link: %s to %s\n",
> >>>>>>>>>> + isc->scaler_sd.entity.name,
> >>>>>>>>>> + isc->video_dev.entity.name);
> >>>>>>>>>> + return ret;
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
> >>>>>>>>>> + ISC_SCALER_PAD_SOURCE);
> >>>>>>>>>> +
> >>>>>>>>>> + return ret;
> >>>>>>>>>> +}
> >>>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
> >>>>>>>>>
> >>>>>>>>> From the DT graph point of view, the ISC appears as a single block
> >>>>>>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
> >>>>>>>>> add to its notifier the image sensor remote subdevice.
> >>>>>>>>
> >>>>>>>> Actually it's only a parallel port.
> >>>>>>>> And it can be connected either to a sensor directly or to the csi2dc
> >>>>>>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
> >>>>>>>>> entity which was initialized at isc_probe() time is linked in between
> >>>>>>>>> the ISC and the image sensor immutably.
> >>>>>>>>
> >>>>>>>> yes, because this cannot change at runtime, and usually, it can't change
> >>>>>>>> without altering the board hardware. (at least this is what my
> >>>>>>>> understanding of immutability is )
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>> I think it is fine for now, but I wonder if you plan to plumb more
> >>>>>>>>> components between the ISC video node and the sensor, if it's not
> >>>>>>>>> worth changing the DT bindings and their parsing logic to separate the
> >>>>>>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
> >>>>>>>>> time and have the CSI-2 receiver entity as downstream remote. I think
> >>>>>>>>> I need to get to know the ISC better to really have an idea. For now,
> >>>>>>>>> this seems ok to me, but please check with maintainers if this is fine
> >>>>>>>>> with them.
> >>>>>>>>
> >>>>>>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
> >>>>>>>> the remote subdev).
> >>>>>>>> The ISC will register a scaler, and connect the subdev to this scaler
> >>>>>>>> first, and then, the scaler to the isc itself (the v4l entity).
> >>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>>>> index 5fbf52a9080b..c9234c90ae58 100644
> >>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
> >>>>>>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
> >>>>>>>>>> u32 his_entry;
> >>>>>>>>>> };
> >>>>>>>>>>
> >>>>>>>>>> +enum isc_mc_pads {
> >>>>>>>>>> + ISC_PAD_SINK = 0,
> >>>>>>>>>> + ISC_PADS_NUM = 1,
> >>>>>>>>>> +};
> >>>>>>>>>> +
> >>>>>>>>>> +enum isc_scaler_pads {
> >>>>>>>>>> + ISC_SCALER_PAD_SINK = 0,
> >>>>>>>>>> + ISC_SCALER_PAD_SOURCE = 1,
> >>>>>>>>>> + ISC_SCALER_PADS_NUM = 2,
> >>>>>>>>>> +};
> >>>>>>>>>> +
> >>>>>>>>>> /*
> >>>>>>>>>> * struct isc_device - ISC device driver data/config struct
> >>>>>>>>>> * @regmap: Register map
> >>>>>>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
> >>>>>>>>>> * be used as an input to the controller
> >>>>>>>>>> * @controller_formats_size: size of controller_formats array
> >>>>>>>>>> * @formats_list_size: size of formats_list array
> >>>>>>>>>> + * @pads: media controller pads for isc video entity
> >>>>>>>>>> + * @mdev: media device that is registered by the isc
> >>>>>>>>>> + * @remote_pad: remote pad on the connected subdevice
> >>>>>>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
> >>>>>>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
> >>>>>>>>>> + * @scaler_format: current format for the scaler subdevice
> >>>>>>>>>> */
> >>>>>>>>>> struct isc_device {
> >>>>>>>>>> struct regmap *regmap;
> >>>>>>>>>> @@ -344,6 +361,19 @@ struct isc_device {
> >>>>>>>>>> struct isc_format *formats_list;
> >>>>>>>>>> u32 controller_formats_size;
> >>>>>>>>>> u32 formats_list_size;
> >>>>>>>>>> +
> >>>>>>>>>> + struct {
> >>>>>>>>>> + struct media_pad pads[ISC_PADS_NUM];
> >>>>>>>>>> + struct media_device mdev;
> >>>>>>>>>> +
> >>>>>>>>>> + u32 remote_pad;
> >>>>>>>>>> + };
> >>>>>>>>>> +
> >>>>>>>>>> + struct {
> >>>>>>>>>> + struct v4l2_subdev scaler_sd;
> >>>>>>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
> >>>>>>>>>> + struct v4l2_mbus_framefmt scaler_format;
> >>>>>>>>
> >>>>>>>> Here are the scaler stuff which I added, in the same bulk struct for the
> >>>>>>>> whole isc device
> >>>>>>>>
> >>>>>>>>
> >>>>>>>>>> + };
> >>>>>>>>>> };
> >>>>>>>>>>
> >>>>>>>>>> extern const struct regmap_config isc_regmap_config;
> >>>>>>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
> >>>>>>>>>> void isc_subdev_cleanup(struct isc_device *isc);
> >>>>>>>>>> void isc_clk_cleanup(struct isc_device *isc);
> >>>>>>>>>>
> >>>>>>>>>> +int isc_scaler_link(struct isc_device *isc);
> >>>>>>>>>> +int isc_scaler_init(struct isc_device *isc);
> >>>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
> >>>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc);
> >>>>>>>>>> +
> >>>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
> >>>>>>>>>> + unsigned int code, int *index);
> >>>>>>>>>> #endif
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>>>> index c5b9563e36cb..c244682ea22f 100644
> >>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
> >>>>>>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>>>> break;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>>>
> >>>>>>>>> I am surprised you can read a register before runtime_pm is
> >>>>>>>>> intialized!
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>> Actually I can read any register after the moment of starting the clock
> >>>>>>>> on the hardware block.
> >>>>>>>
> >>>>>>> Ah right then
> >>>>>>>
> >>>>>>>> Maybe the starting and stopping of the clock needs to be moved to the
> >>>>>>>> runtime_pm calls, but this is another story, not related to the media
> >>>>>>>> controller.
> >>>>>>>
> >>>>>>> It's fine, but maybe worth recording with a todo ?
> >>>>>>>
> >>>>>>> Thanks
> >>>>>>> j
> >>>>>>>
> >>>>>>>> I moved this line because I had to pass the version to the isc_mc_init call
> >>>>>>>>
> >>>>>>>>> Thanks
> >>>>>>>>> j
> >>>>>>>>
> >>>>>>>> Thanks for reviewing !
> >>>>>>>> Eugen
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> +
> >>>>>>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>>>>>> + if (ret < 0)
> >>>>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>>>> +
> >>>>>>>>>> pm_runtime_set_active(dev);
> >>>>>>>>>> pm_runtime_enable(dev);
> >>>>>>>>>> pm_request_idle(dev);
> >>>>>>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>>>> ret = clk_prepare_enable(isc->ispck);
> >>>>>>>>>> if (ret) {
> >>>>>>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
> >>>>>>>>>> - goto cleanup_subdev;
> >>>>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> /* ispck should be greater or equal to hclock */
> >>>>>>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>>>> goto unprepare_clk;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
> >>>>>>>>>>
> >>>>>>>>>> return 0;
> >>>>>>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
> >>>>>>>>>> unprepare_clk:
> >>>>>>>>>> clk_disable_unprepare(isc->ispck);
> >>>>>>>>>>
> >>>>>>>>>> +isc_probe_mc_init_err:
> >>>>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>>>> +
> >>>>>>>>>> cleanup_subdev:
> >>>>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>>>
> >>>>>>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
> >>>>>>>>>>
> >>>>>>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>>>>>
> >>>>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>>>> +
> >>>>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>>>
> >>>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>>>> index 07a80b08bc54..9dc75eed0098 100644
> >>>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
> >>>>>>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
> >>>>>>>>>> break;
> >>>>>>>>>> }
> >>>>>>>>>>
> >>>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>>>> +
> >>>>>>>>>> + ret = isc_mc_init(isc, ver);
> >>>>>>>>>> + if (ret < 0)
> >>>>>>>>>> + goto isc_probe_mc_init_err;
> >>>>>>>>>> +
> >>>>>>>>>> pm_runtime_set_active(dev);
> >>>>>>>>>> pm_runtime_enable(dev);
> >>>>>>>>>> pm_request_idle(dev);
> >>>>>>>>>>
> >>>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
> >>>>>>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
> >>>>>>>>>>
> >>>>>>>>>> return 0;
> >>>>>>>>>>
> >>>>>>>>>> +isc_probe_mc_init_err:
> >>>>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>>>> +
> >>>>>>>>>> cleanup_subdev:
> >>>>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>>>
> >>>>>>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
> >>>>>>>>>>
> >>>>>>>>>> pm_runtime_disable(&pdev->dev);
> >>>>>>>>>>
> >>>>>>>>>> + isc_mc_cleanup(isc);
> >>>>>>>>>> +
> >>>>>>>>>> isc_subdev_cleanup(isc);
> >>>>>>>>>>
> >>>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
> >>>>>>>>>> --
> >>>>>>>>>> 2.25.1
> >>>>>>>>>>
> >>>>>>>>
> >>>>>>
> >>>>
> >>
>

2022-02-14 02:33:00

by Eugen Hristev

[permalink] [raw]
Subject: Re: [PATCH v4 03/11] media: atmel: atmel-isc: implement media controller

On 2/10/22 7:49 PM, Jacopo Mondi wrote:
> Hi Eugen
>
> On Wed, Feb 09, 2022 at 06:59:37AM +0000, [email protected] wrote:
>> On 2/8/22 9:30 PM, Jacopo Mondi wrote:
>>> Hi Eugen
>>>
>>> On Tue, Feb 08, 2022 at 10:33:31AM +0000, [email protected] wrote:
>>>> On 2/8/22 10:59 AM, Jacopo Mondi wrote:
>>>>> Hi Eugen
>>>>>
>>>>> On Mon, Feb 07, 2022 at 01:40:31PM +0000, [email protected] wrote:
>>>>>> On 2/7/22 2:44 PM, Jacopo Mondi wrote:
>>>>>>> Hi Eugen
>>>>>>>
>>>>>>> On Fri, Jan 21, 2022 at 03:14:08PM +0200, Eugen Hristev wrote:
>>>>>>>> Implement the support for media-controller.
>>>>>>>> This means that the capabilities of the driver have changed and now
>>>>>>>> it also advertises the IO_MC .
>>>>>>>> The driver will register it's media device, and add the video entity to this
>>>>>>>> media device. The subdevices are registered to the same media device.
>>>>>>>> The ISC will have a base entity which is auto-detected as atmel_isc_base.
>>>>>>>> It will also register a subdevice that allows cropping of the incoming frame
>>>>>>>> to the maximum frame size supported by the ISC.
>>>>>>>> The ISC will create a link between the subdevice that is asynchronously
>>>>>>>> registered and the atmel_isc_scaler entity.
>>>>>>>> Then, the atmel_isc_scaler and atmel_isc_base are connected through another
>>>>>>>> link.
>>>>>>>>
>>>>>>>> Signed-off-by: Eugen Hristev <[email protected]>
>>>>>>>> ---
>>>>>>>> Changes in v4:
>>>>>>>> As suggested by Jacopo:
>>>>>>>> - renamed atmel_isc_mc to atmel_isc_scaler.c
>>>>>>>> - moved init_mc/clean_mc to isc_base file
>>>>>>>>
>>>>>>>> Changes in v2:
>>>>>>>> - implement try formats
>>>>>>>>
>>>>>>>> drivers/media/platform/atmel/Makefile | 2 +-
>>>>>>>> drivers/media/platform/atmel/atmel-isc-base.c | 73 +++++-
>>>>>>>> .../media/platform/atmel/atmel-isc-scaler.c | 245 ++++++++++++++++++
>>>>>>>> drivers/media/platform/atmel/atmel-isc.h | 37 +++
>>>>>>>> .../media/platform/atmel/atmel-sama5d2-isc.c | 14 +-
>>>>>>>> .../media/platform/atmel/atmel-sama7g5-isc.c | 12 +-
>>>>>>>> 6 files changed, 375 insertions(+), 8 deletions(-)
>>>>>>>> create mode 100644 drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>>
>>>>>>>> diff --git a/drivers/media/platform/atmel/Makefile b/drivers/media/platform/atmel/Makefile
>>>>>>>> index 794e8f739287..f02d03df89d6 100644
>>>>>>>> --- a/drivers/media/platform/atmel/Makefile
>>>>>>>> +++ b/drivers/media/platform/atmel/Makefile
>>>>>>>> @@ -1,7 +1,7 @@
>>>>>>>> # SPDX-License-Identifier: GPL-2.0-only
>>>>>>>> atmel-isc-objs = atmel-sama5d2-isc.o
>>>>>>>> atmel-xisc-objs = atmel-sama7g5-isc.o
>>>>>>>> -atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o
>>>>>>>> +atmel-isc-common-objs = atmel-isc-base.o atmel-isc-clk.o atmel-isc-scaler.o
>>>>>>>>
>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISI) += atmel-isi.o
>>>>>>>> obj-$(CONFIG_VIDEO_ATMEL_ISC_BASE) += atmel-isc-common.o
>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-base.c b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>> index 6b0005987a17..6b482270eb93 100644
>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-base.c
>>>>>>>> @@ -1710,6 +1710,7 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>>>> struct isc_device, v4l2_dev);
>>>>>>>> struct isc_subdev_entity *subdev_entity =
>>>>>>>> container_of(notifier, struct isc_subdev_entity, notifier);
>>>>>>>> + int pad;
>>>>>>>>
>>>>>>>> if (video_is_registered(&isc->video_dev)) {
>>>>>>>> v4l2_err(&isc->v4l2_dev, "only supports one sub-device.\n");
>>>>>>>> @@ -1718,6 +1719,16 @@ static int isc_async_bound(struct v4l2_async_notifier *notifier,
>>>>>>>>
>>>>>>>> subdev_entity->sd = subdev;
>>>>>>>>
>>>>>>>> + pad = media_entity_get_fwnode_pad(&subdev->entity, asd->match.fwnode,
>>>>>>>> + MEDIA_PAD_FL_SOURCE);
>>>>>>>> + if (pad < 0) {
>>>>>>>> + v4l2_err(&isc->v4l2_dev, "failed to find pad for %s\n",
>>>>>>>> + subdev->name);
>>>>>>>> + return pad;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + isc->remote_pad = pad;
>>>>>>>> +
>>>>>>>> return 0;
>>>>>>>> }
>>>>>>>>
>>>>>>>> @@ -1732,8 +1743,8 @@ static void isc_async_unbind(struct v4l2_async_notifier *notifier,
>>>>>>>> v4l2_ctrl_handler_free(&isc->ctrls.handler);
>>>>>>>> }
>>>>>>>>
>>>>>>>> -static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>>>> - unsigned int code, int *index)
>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>>>> + unsigned int code, int *index)
>>>>>>>> {
>>>>>>>> struct isc_format *fmt = &isc->formats_list[0];
>>>>>>>> unsigned int i;
>>>>>>>> @@ -1749,6 +1760,7 @@ static struct isc_format *find_format_by_code(struct isc_device *isc,
>>>>>>>>
>>>>>>>> return NULL;
>>>>>>>> }
>>>>>>>> +EXPORT_SYMBOL_GPL(isc_find_format_by_code);
>>>>>>>>
>>>>>>>> static int isc_formats_init(struct isc_device *isc)
>>>>>>>> {
>>>>>>>> @@ -1765,7 +1777,7 @@ static int isc_formats_init(struct isc_device *isc)
>>>>>>>> NULL, &mbus_code)) {
>>>>>>>> mbus_code.index++;
>>>>>>>>
>>>>>>>> - fmt = find_format_by_code(isc, mbus_code.code, &i);
>>>>>>>> + fmt = isc_find_format_by_code(isc, mbus_code.code, &i);
>>>>>>>> if (!fmt) {
>>>>>>>> v4l2_warn(&isc->v4l2_dev, "Mbus code %x not supported\n",
>>>>>>>> mbus_code.code);
>>>>>>>> @@ -1891,7 +1903,8 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>>>> vdev->queue = q;
>>>>>>>> vdev->lock = &isc->lock;
>>>>>>>> vdev->ctrl_handler = &isc->ctrls.handler;
>>>>>>>> - vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE;
>>>>>>>> + vdev->device_caps = V4L2_CAP_STREAMING | V4L2_CAP_VIDEO_CAPTURE |
>>>>>>>> + V4L2_CAP_IO_MC;
>>>>>>>> video_set_drvdata(vdev, isc);
>>>>>>>>
>>>>>>>> ret = video_register_device(vdev, VFL_TYPE_VIDEO, -1);
>>>>>>>> @@ -1901,8 +1914,18 @@ static int isc_async_complete(struct v4l2_async_notifier *notifier)
>>>>>>>> goto isc_async_complete_err;
>>>>>>>> }
>>>>>>>>
>>>>>>>> + ret = isc_scaler_link(isc);
>>>>>>>> + if (ret < 0)
>>>>>>>> + goto isc_async_complete_unregister_device;
>>>>>>>> +
>>>>>>>> + ret = media_device_register(&isc->mdev);
>>>>>>>> + if (ret < 0)
>>>>>>>> + goto isc_async_complete_unregister_device;
>>>>>>>> return 0;
>>>>>>>>
>>>>>>>> +isc_async_complete_unregister_device:
>>>>>>>> + video_unregister_device(vdev);
>>>>>>>> +
>>>>>>>> isc_async_complete_err:
>>>>>>>> mutex_destroy(&isc->lock);
>>>>>>>> return ret;
>>>>>>>> @@ -1969,6 +1992,48 @@ int isc_pipeline_init(struct isc_device *isc)
>>>>>>>> }
>>>>>>>> EXPORT_SYMBOL_GPL(isc_pipeline_init);
>>>>>>>>
>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver)
>>>>>>>> +{
>>>>>>>> + const struct of_device_id *match;
>>>>>>>> + int ret;
>>>>>>>> +
>>>>>>>> + isc->video_dev.entity.function = MEDIA_ENT_F_IO_V4L;
>>>>>>>> + isc->video_dev.entity.flags = MEDIA_ENT_FL_DEFAULT;
>>>>>>>
>>>>>>> Should you set entity.ops.link_validate = v4l2_subdev_link_validate
>>>>>>> to be able to have the media pipeline validated at
>>>>>>> media_pipeline_start() time ?
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I am doing that in a subsequent patch. Things are not completely ready
>>>>>> at this moment, because ISC still relies on the old mechanism to call
>>>>>> v4l2_subdev_call with set_fmt on the subdevice (which subsequent patch
>>>>>> removes and adds checks to the link validate ).
>>>>>>
>>>>>
>>>>> I see.. the subsequent patches are not part of this series, right ?
>>>>
>>>> No. Patch 7 does this. In that patch, the link validation is created and
>>>> all the logic of format propagation is removed and reworked, and creates
>>>> the need for the link_validate call
>>>> Could you have also a look at that patch ?
>>>
>>> Ah ups, 7 was not in my inbox with the rest of the series. Just
>>> noticed. I had even reviewed the previous version, I should have
>>> remembered link_validate was added later
>>>
>>>>>
>>>>> I think patches 1, 2, 4, 5 and 6 from this series do not depend on
>>>>> the introduction of media controller or the scaler, right ? (I'm not
>>>>> sure if the DTS patches do).
>>>
>>> Do DTS patches depend on media-controller plumbing ?
>>
>> I think not, but, I have not tested them without the media controller.
>>
>>>
>>>>>
>>>>> If that's the case, would it make sense to to fast-track them (as some
>>>>> of them are fixes) and then on top plumb the media controller
>>>>> infrastructure with this patch and the ones you have been mentioning ?
>>>>
>>>> It would make sense.
>>>>
>>>
>>> Once the other discussion with Hans about the stop state it might make
>>> sense to fastrack the first part of the series ?
>>
>> The other three patches (4,5,6) are quite independent and can go faster.
>>>
>>>>>
>>>>>>>
>>>>>>>> + isc->pads[ISC_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>>>> +
>>>>>>>> + ret = media_entity_pads_init(&isc->video_dev.entity, ISC_PADS_NUM,
>>>>>>>> + isc->pads);
>>>>>>>> + if (ret < 0) {
>>>>>>>> + dev_err(isc->dev, "media entity init failed\n");
>>>>>>>> + return ret;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + isc->mdev.dev = isc->dev;
>>>>>>>> +
>>>>>>>> + match = of_match_node(isc->dev->driver->of_match_table,
>>>>>>>> + isc->dev->of_node);
>>>>>>>> +
>>>>>>>> + strscpy(isc->mdev.driver_name, KBUILD_MODNAME,
>>>>>>>> + sizeof(isc->mdev.driver_name));
>>>>>>>> + strscpy(isc->mdev.model, match->compatible, sizeof(isc->mdev.model));
>>>>>>>> + snprintf(isc->mdev.bus_info, sizeof(isc->mdev.bus_info), "platform:%s",
>>>>>>>> + isc->v4l2_dev.name);
>>>>>>>> + isc->mdev.hw_revision = ver;
>>>>>>>> +
>>>>>>>> + media_device_init(&isc->mdev);
>>>>>>>> +
>>>>>>>> + isc->v4l2_dev.mdev = &isc->mdev;
>>>>>>>> +
>>>>>>>> + return isc_scaler_init(isc);
>>>>>>>> +}
>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_init);
>>>>>>>> +
>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc)
>>>>>>>> +{
>>>>>>>> + media_entity_cleanup(&isc->video_dev.entity);
>>>>>>>> +}
>>>>>>>> +EXPORT_SYMBOL_GPL(isc_mc_cleanup);
>>>>>>>> +
>>>>>>>> /* regmap configuration */
>>>>>>>> #define ATMEL_ISC_REG_MAX 0xd5c
>>>>>>>> const struct regmap_config isc_regmap_config = {
>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc-scaler.c b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..ec95c9665883
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc-scaler.c
>>>>>>>> @@ -0,0 +1,245 @@
>>>>>>>> +// SPDX-License-Identifier: GPL-2.0-only
>>>>>>>> +/*
>>>>>>>> + * Microchip Image Sensor Controller (ISC) Scaler entity support
>>>>>>>> + *
>>>>>>>> + * Copyright (C) 2021 Microchip Technology, Inc.
>>>>>>>
>>>>>>> Time flies! It's 2022 already :)
>>>>>>>
>>>>>>>> + *
>>>>>>>> + * Author: Eugen Hristev <[email protected]>
>>>>>>>> + *
>>>>>>>> + */
>>>>>>>> +
>>>>>>>> +#include <media/media-device.h>
>>>>>>>> +#include <media/media-entity.h>
>>>>>>>> +#include <media/v4l2-device.h>
>>>>>>>> +#include <media/v4l2-subdev.h>
>>>>>>>> +
>>>>>>>> +#include "atmel-isc-regs.h"
>>>>>>>> +#include "atmel-isc.h"
>>>>>>>> +
>>>>>>>> +static int isc_scaler_get_fmt(struct v4l2_subdev *sd,
>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>> + struct v4l2_subdev_format *format)
>>>>>>>> +{
>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>>>> +
>>>>>>>> + if (format->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>> + format->pad);
>>>>>>>> + format->format = *v4l2_try_fmt;
>>>>>>>> +
>>>>>>>> + return 0;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + format->format = isc->scaler_format;
>>>>>>>
>>>>>>> isc->scaler_format is only used inside this file if I'm not mistaken.
>>>>>>> I wonder why it lives in the isc_device struct.
>>>>>>
>>>>>> isc_device is a placeholder for all isc things.
>>>>>>
>>>>>> I would not create a separate struct in this file that would have to be
>>>>>> allocated etc... just for one two things. So I preferred to have it in
>>>>>> the same place as all the other things.
>>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + return 0;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +static int isc_scaler_set_fmt(struct v4l2_subdev *sd,
>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>> + struct v4l2_subdev_format *req_fmt)
>>>>>>>> +{
>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt;
>>>>>>>> + struct isc_format *fmt;
>>>>>>>> + unsigned int i;
>>>>>>>> +
>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SOURCE)
>>>>>>>> + v4l_bound_align_image
>>>>>>>> + (&req_fmt->format.width, 16, isc->max_width, 0,
>>>>>>>> + &req_fmt->format.height, 16, isc->max_height, 0, 0);
>>>>>>>> + else
>>>>>>>> + v4l_bound_align_image
>>>>>>>> + (&req_fmt->format.width, 16, 10000, 0,
>>>>>>>> + &req_fmt->format.height, 16, 10000, 0, 0);
>>>>>>>
>>>>>>> Where does 10000 come from ?
>>>>>>
>>>>>> It's a random number. Do you have any suggestion for a better one ?
>>>>>
>>>>> An actual HW limit ? :)
>>>>
>>>> There is no hardware limit. The ISC just stops sampling pixels once the
>>>> crop limit is reached. The element that generates pixels could go on
>>>> forever , or until the next HBLANK/VBLANK. So what limit could I place
>>>> here ?
>>>> The cropping is mandatory, otherwise there could be an overflow w.r.t.
>>>> the buffer size if the ISC would sample more pixels than the software
>>>> expects it to.
>>>>
>>>>>
>>>>>> Maybe this would be much more clear with my comments below (where I will
>>>>>> explain what the isc scaler does )
>>>>>> In short, it allows the other entity in the link to 'sink' a huge format
>>>>>> into this 'scaler' . Because the scaler can crop anything basically down
>>>>>> to the biggest format size the ISC could handle.
>>>>>>
>>>>>
>>>>> doesn't it have any documented input size limit ?
>>>>
>>>> As stated above, ISC handles a pixel stream and stores it in an internal
>>>> SRAM. ISC stops sampling when this SRAM is full or when the cropping
>>>> limit is reached.
>>>> After storing the pixels in the SRAM (which is limited to the ISC
>>>> maximum frame size), the ISC will work on the pixels and then DMA the
>>>> frame to another RAM.
>>>>
>>>
>>> I understand...
>>>
>>> So whatever the input size is, only the first n-th bytes are actually
>>> dumped to memory, the following ones are discarded.
>>
>> Yes ! That's the behavior.
>>>
>>> So the crop is always applied on the top-leftmost corner as a
>>> consequence, right ?
>>
>> Right. Pixels come in order. The ISC does not really know how many
>> pixels are coming until its memory is full (or counting until the crop
>> limit is reached). After that it stops sampling pixels, and waits for
>> vblank to finish the frame.
>>
>>
>>>
>>> I understand why you had difficulties picking a default :)
>>>
>>> This is not easy to handle, as the scaler doesn't actually applies a
>>> crop but rather stops capturing after the destination buffer has been
>>> saturated. So the limiting factor is the full image size, not one of
>>> its dimensions, like in example the line length. This makes it hard
>>> to adjust cropping to something meaningful for most use cases.
>>
>> I didn't know how to name this except crop. It 'crops out' what does not
>> fit in the internal memory of the ISC
>>
>>>
>>> In your below example you suggest that an image size of (800000 x 10)
>>> would theoretically be captured correctly. What if the input is of
>>> size (800000 x 10 + 1). What size would you set the crop rectangle to ?
>>>
>>> I wouldn't however be too much concerned for the moment, but I would
>>> record this in a comment ?
>>
>> I discussed a bit more with designer and looked in datasheet.
>> It looks like the horizontal line length is not a hard limit, but it
>> makes the ISC internal ISP stop working, so I would keep the horizontal
>> limit as well for now, unless an important use case (I am coming back to
>> that). The vertical limit looks to be a hard requirement.
>> So in short, we have to impose a vertical limit of 2464 , we cannot have
>> more lines than these. So if the horizontal is 100 pixels e.g., we can
>> capture only 100x2464 .
>> In practice we could capture [16,3264] on the horizontal, and [16,2464]
>> on the vertical, and in theory horizontal could go more like [16, 4000],
>> but we have to limit the vertical, and without the internal ISP.
>> So I would not haphazardly go in this direction. Let's stick to the
>> datasheet maximum frame 3264x2464 .
>> According to this , I changed the frame size of the ISC to be a
>> continuous size from [16,16] to [3264,2464]
>> About that use case that I talked about earlier, we have one sensor ,
>> imx274, which is more wide, and outputs something like 3800x2100 .
>> With the cropping and limitations we have in place now, this will be
>> limited to 3264x2100 , which is in line with datasheet and what the ISC
>> supports now.
>> As in theory we could capture 3800x2100 as long as the SRAM doesn't get
>> full, for now I would like to avoid capturing this size (and disable the
>> internal ISP), until we have someone really asking for such kind of
>> capture and with some good reasons.
>> In short I would like to have the crop rectangle to 3264x2464 and the
>> bounds would be limited to the incoming frame but no H>3264 and no
>> V>2464. Does that make sense ?
>
> Just to make sure I got it: is the ISC actually able to crop on the
> line length ? If you apply a 3264 limit, will the last 536 pixels of
> each line will be dropped if the input frame is 3800x2100 as your
> above example ?

Yes. Last 536 pixels will be dropped.
After the 3264 limit, no more pixels are being sampled until HBLANK.
>
> Or rather, is the SRAM size limit the issue and the lines gets
> captured in full lenght no matter what, until the buffer gets filled up
> and all the successive pixels dropped.
>
> In other words, do your cropped image look like:
>
> 3264
> +--------------------------------X-----------+ 3800
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx| |
> +--------------------------------------------+ 2100
>
> Or rather
>
>
> 3264
> +--------------------------------X-----------+ 3800
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx|
> |xxxxxxxxxxxxxxxxx| |
> | |
> +--------------------------------------------+ 2100
>
> I assume the first, as otherwise you wouldn't be able to interpret the
> image as 3264x2100.

Yes, the first.

>
> I'm asking because you had the un-cropped 3280x2464 size as the ISC
> source pad size in the paste of the media graph you provided below,
> which might mean the output image is actually full size in lines
> lenght and it's maybe missing pixels at the end, in case the ISC
> operates as in the above second case.

No, the image won't be padded with blanks on the right side. Actually
this cannot happen due to a very big reason: the frame is being written
to the main memory by a DMA controller. This DMA controller will write
all the lines at consecutive addresses, and it will write exactly how
many pixels are being captured by the module from the ISC named
'parallel front end' which does the cropping. Thus, the DMA will never
output empty padding bytes at the end of each line, and it will not add
a certain stride to each line if not instructed to. In fact the DMA can
add a stride offset for the address of each line, which can be used for
composing or copying the frame into another frame, but this is a
different story. ( I tried this for fun, but not implemented in the
driver at the moment)

The cropping is also mandatory even if the captured frame is less than
the maximum size, because the ISC has no configuration related to frame
size in fact. The parallel front end captures pixels and the dma engine
will write them to memory. There is no control over this except the
cropping.
So for example if the software expects 640x480 but somehow someone sends
more pixels, it can lead to buffer overflow if the parallel front end is
not configured to ignore(crop) anything beyond 640x480 .

With this being said, or having an understanding about the sink pad of
the ISC, how should I configure the sink frame size ?
Is there any possibility to avoid limits on the frame size and have it
arbitrary ?
I am 99% sure that v4l2-compliance will not be happy if there is no
limit on the frame on entity pads (hence, I bound-aligned the frame to
16x16 up to 10k x 10k )

>
>>
>>>
>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + req_fmt->format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>>>> + req_fmt->format.field = V4L2_FIELD_NONE;
>>>>>>>> + req_fmt->format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>>>> + req_fmt->format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>>>> + req_fmt->format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>>>> +
>>>>>>>> + fmt = isc_find_format_by_code(isc, req_fmt->format.code, &i);
>>>>>>>
>>>>>>> So you rely on the isc format list for the scaler as well ?
>>>>>>> I think it's fine as long as they are identical
>>>>>>
>>>>>> Yes, the scaler is kind of a dummy entity , but it's required by the
>>>>>> media controller validation of the pipeline.
>>>>>>
>>>>>
>>>>> More on this later
>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + if (!fmt)
>>>>>>>> + fmt = &isc->formats_list[0];
>>>>>>>> +
>>>>>>>> + req_fmt->format.code = fmt->mbus_code;
>>>>>>>> +
>>>>>>>> + if (req_fmt->which == V4L2_SUBDEV_FORMAT_TRY) {
>>>>>>>> + v4l2_try_fmt = v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>> + req_fmt->pad);
>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>>>> + /* Trying on the pad sink makes the source sink change too */
>>>>>>>> + if (req_fmt->pad == ISC_SCALER_PAD_SINK) {
>>>>>>>> + v4l2_try_fmt =
>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state,
>>>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>>>> + *v4l2_try_fmt = req_fmt->format;
>>>>>>>> +
>>>>>>>> + v4l_bound_align_image(&v4l2_try_fmt->width,
>>>>>>>> + 16, isc->max_width, 0,
>>>>>>>> + &v4l2_try_fmt->height,
>>>>>>>> + 16, isc->max_height, 0, 0);
>>>>>>>> + }
>>>>>>>> + /* if we are just trying, we are done */
>>>>>>>> + return 0;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + isc->scaler_format = req_fmt->format;
>>>>>>>
>>>>>>> No per-pad format but a global scaler_format ? How do you configure scaling ?
>>>>>>>
>>>>>>> Actually, I would like to know more about the scaler device
>>>>>>> capabilities. What functions can this IP perform ? Does it do
>>>>>>> cropping, can it also do (down)scaling or even composition ?
>>>>>>>
>>>>>>> I think it is worth to read
>>>>>>> https://www.kernel.org/doc/html/latest/userspace-api/media/v4l/dev-subdev.html#v4l2-subdev-selections
>>>>>>> where it is reported how cropping and scaling are implemented by using
>>>>>>> the selection API.
>>>>>>>
>>>>>>> Figure 4.5 is particularly helpful to explain the simple crop case,
>>>>>>> for which you need to implement support to by adding s_selection on
>>>>>>> the sink pad TGT_CROP target.
>>>>>>
>>>>>> The scaler is kind of a dummy entity.
>>>>>> The problem is that different subdevices cannot be connected directly to
>>>>>> a v4l entity, because the ISC does cropping on the incoming frame.
>>>>>> The ISC has a limit on the total number of pixels per frame, thus it
>>>>>> will do a crop.
>>>>>
>>>>> What is this limit related to ? I don't know the ISC internals, but is
>>>>> the limitation due to the overall image size, or maybe it's due to
>>>>> some line buffers being limited in the size they can transfer to
>>>>> memory ? Is it a limitation on the size of the image in pixels, or is
>>>>> it due to a limitation of the processing bandwidth on the input
>>>>> interface and thus depends on the actual size of the image in bytes ?
>>>>
>>>> Limitation is both on line size and full image size. From my
>>>> understanding, is that the internal SRAM is limited (8 Mpixels with a
>>>> small safe buffer on top in the case of sama7g5 )
>>>> This SRAM could be used as 800000x10 for example, or 3200x2464 for
>>>> another example. However, I am not sure if a real line of 800,000 pixels
>>>> make sense, or if there is another internal mechanism in the ISC that
>>>> stops it.
>>>> In any case, the ISC is cropping the incoming frame from aaaXbbb up to
>>>> the maximum frame that it can handle
>>>>
>>>>>
>>>>>> So, if I allow e.g. an entity that sends 3280x2464 , when the ISC can
>>>>>> only handle 3264x2464 , then we have two choices:
>>>>>> 1/ crop it and output the 3264x2464 -> cannot, because userspace expects
>>>>>> 3280x2464, and the whole pipeline fails to configure, there is a
>>>>>> mismatch between /dev/video output and what the whole pipeline produces
>>>>>
>>>>> I'm confused in first place. If the ISC cannot output anything larger
>>>>> than 3264x2464, then if userspace sets a 3280x2464 size on the ISC,
>>>>> this should be adjusted to 3264x2464.
>>>>
>>>> yes, that is true. However, without an entity that says 'I am doing
>>>> cropping', the media pipeline will fail, because previous entity will
>>>> output 3280x2464 .
>>>> The conclusion to create a scaler entitity was done during my last year
>>>> discussions on the IRC. I could not find a way to perform this cropping
>>>> at the v4l entity side, and one reason is that the v4l entity does not
>>>> support what I needed , or at least that was my understanding.
>>>> You claim that the v4l entity could perform the cropping in the same
>>>> entity ?
>>>
>>> I thought so, but I haven't tried to be honest
>>>
>>>>
>>>> Let's assume for a minute that it could do this. Even so, I would prefer
>>>> to have a separate scaler entity to do this , for several reasons:
>>>> -> I would like to be able to crop the frame arbitrarily if I want to,
>>>> and the scaler would allow me to do this easier
>>>> -> it would be more obvious that the frame is cropped by having this
>>>> entity there
>>>> -> in fact some of the ISC versions really have a down scaler, that I
>>>> have not implemented yet. So it would be useful to already have this in
>>>> place to be able to scale down with it at a later time.
>>>
>>> I think this last reason is enough to have an entity plumbed in
>>> already
>>>
>>>>
>>>>>
>>>>> Or is the ISC limitation on the -input- side ?
>>>>>
>>>>>> 2/ deny the link, and ask the subdevice to produce the 3264x2464 which
>>>>>> the ISC can handle at most -> cannot , because the sensor let's say can
>>>>>> *only* produce 3280x2464 and *any* other frame size returns an error.
>>>>>
>>>>> buggy sensor I would say :)
>>>>
>>>> I cannot tell this to the customers, and I cannot fail the whole
>>>> pipeline because the sensor does not crop 16 pixels, when in fact I can
>>>> do this very easily on the ISC side.
>>>>
>>>
>>> Fair enough :)
>>>
>>>>>
>>>>>>
>>>>>> The only solution to make both worlds happy, is to have a dummy entity
>>>>>> called 'scaler' which in fact now it only performs a simple crop.
>>>>>> It accepts any frame size at input (hence the 10000x10000 which you saw
>>>>>> earlier), and outputs at most 3264x2464 the isc_max_height and
>>>>>> isc_max_width.
>>>>>
>>>>> I have a maybe dumb question: can the cropping operation be modeled
>>>>> by applying a TGT_CROP rectangle (whose _BOUND size is 3280x2464) to
>>>>> the atmel_isc_common entity sink pad and avoid the scaler completely ?
>>>>
>>>> I am not sure. This is what I wanted initially. But then I thought about
>>>> the three reasons stated above for the use of the scaler entity, and
>>>> discussed with folks on the IRC, and come up with the solution for the
>>>> scaler entity
>>>>
>>>>>
>>>>>>
>>>>>> So to answer your question, the isc scaler is a software model for a
>>>>>> simple cropping, that would make media controller happy, and capture
>>>>>> software happy.
>>>>>>
>>>>>> Here it how it looks :
>>>>>>
>>>>>> Device topology
>>>>>> - entity 1: atmel_isc_scaler (2 pads, 2 links)
>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>> device node name /dev/v4l-subdev0
>>>>>> pad0: Sink
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>> crop.bounds:(0,0)/3264x2464
>>>>>> crop:(0,0)/3264x2464]
>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>>>> pad1: Source
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>>>
>>>>>> - entity 4: csi2dc (2 pads, 2 links)
>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>> device node name /dev/v4l-subdev1
>>>>>> pad0: Sink
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>> <- "dw-csi.0":1 [ENABLED]
>>>>>> pad1: Source
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>> -> "atmel_isc_scaler":0 [ENABLED,IMMUTABLE]
>>>>>>
>>>>>> - entity 7: dw-csi.0 (2 pads, 2 links)
>>>>>> type V4L2 subdev subtype Unknown flags 0
>>>>>> device node name /dev/v4l-subdev2
>>>>>> pad0: Sink
>>>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>>>> <- "imx219 1-0010":0 [ENABLED]
>>>>>> pad1: Source
>>>>>> [fmt:SRGGB10_1X10/3280x2464]
>>>>>> -> "csi2dc":0 [ENABLED]
>>>>>>
>>>>>> - entity 12: imx219 1-0010 (1 pad, 1 link)
>>>>>> type V4L2 subdev subtype Sensor flags 0
>>>>>> device node name /dev/v4l-subdev3
>>>>>> pad0: Source
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>> xfer:srgb ycbcr:601 quantization:full-range
>>>>>> crop.bounds:(8,8)/3280x2464
>>>>>> crop:(8,8)/3280x2464]
>>>>>> -> "dw-csi.0":0 [ENABLED]
>>>>>>
>>>>>> - entity 24: atmel_isc_common (1 pad, 1 link)
>>>>>> type Node subtype V4L flags 1
>>>>>> device node name /dev/video0
>>>>>> pad0: Sink
>>>>>> <- "atmel_isc_scaler":1 [ENABLED,IMMUTABLE]
>>>>>>
>>>>>>
>>>>>> Scaler does this one cute little thing :
>>>>>>
>>>>>> pad0: Sink
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb
>>>>>> crop.bounds:(0,0)/3264x2464
>>>>>> crop:(0,0)/3264x2464]
>>>>>> <- "csi2dc":1 [ENABLED,IMMUTABLE]
>>>>>> pad1: Source
>>>>>> [fmt:SRGGB10_1X10/3280x2464 field:none colorspace:srgb]
>>>>>
>>>>> Shouldn't this be 3264x2464 as that's what the entity outputs after
>>>>> the cropping ? And shouldn't then be 3264x2464 the size output from
>>>>> the video device too ?
>>>>
>>>> That's right.
>>>> I don't know why the source format for the scaler is still 3280x2464.
>>>> Maybe there is a bug on g_fmt for it.. have to check it.
>>>
>>> Thanks, I think this should be fixed ten
>>>
>>>
>>>>
>>>> Anyway, the video format is like this :
>>>>
>>>> # v4l2-ctl --get-fmt-video
>>>> Format Video Capture:
>>>> Width/Height : 3264/2464
>>>> Pixel Format : 'RGBP' (16-bit RGB 5-6-5)
>>>> Field : None
>>>> Bytes per Line : 6528
>>>> Size Image : 16084992
>>>> Colorspace : sRGB
>>>> Transfer Function : Default (maps to sRGB)
>>>> YCbCr/HSV Encoding: Default (maps to ITU-R 601)
>>>> Quantization : Default (maps to Full Range)
>>>> Flags :
>>>> #
>>>>
>>>> and initializing the pipeline looks fine from user perspective.
>>>> The scaler is in fact the solution to make this pipeline work with
>>>> libcamera.
>>>> I found this cropping to be an issue in media controller when trying it
>>>> with libcamera. Otherwise, the other user space apps which I was using
>>>> never complained that anything was wrong
>>>> Libcamera simply refuses to acknowledge the pipeline if the video output
>>>> is 3264x2464 but there is no entity that changes the format from
>>>> 3280x2464 down
>>>
>>> Not sure why libcamera should play a role there. Isn't it the media
>>> pipeline validation that complains and returns an -EPIPE ?
>>
>> I tried this before the media_pipeline_start .
>> Perhaps libcamera was doing some similar checks, anyway, it was
>> complaining and not adding the stream, and reported no usable streams/camera
>>
>> Once I will rework some of the things according to your review I will
>> run libcamera again to see what it complains about
>>
>>>
>>>>
>>>>>
>>>>>
>>>>>> -> "atmel_isc_common":0 [ENABLED,IMMUTABLE]
>>>>>>
>>>>>>
>>>>>> Which is what we needed.
>>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + return 0;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +static int isc_scaler_enum_mbus_code(struct v4l2_subdev *sd,
>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>> + struct v4l2_subdev_mbus_code_enum *code)
>>>>>>>> +{
>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>> + int supported_index = 0;
>>>>>>>> + int i;
>>>>>>>> +
>>>>>>>> + for (i = 0; i < isc->formats_list_size; i++) {
>>>>>>>> + if (!isc->formats_list[i].sd_support)
>>>>>>>> + continue;
>>>>>>>
>>>>>>> The sd_support flag still doesn't click in my head.
>>>>>>>
>>>>>>> Shouldn't the enumeration of available formats on the scaler do not
>>>>>>> depend on the sensor supproted formats ?
>>>>>>
>>>>>> You're right. I will have to check it again.
>>>>>>
>>>>>>>
>>>>>>>> + if (supported_index == code->index) {
>>>>>>>> + code->code = isc->formats_list[i].mbus_code;
>>>>>>>> + return 0;
>>>>>>>> + }
>>>>>>>> + supported_index++;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + return -EINVAL;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +static int isc_scaler_g_sel(struct v4l2_subdev *sd,
>>>>>>>> + struct v4l2_subdev_state *sd_state,
>>>>>>>> + struct v4l2_subdev_selection *sel)
>>>>>>>> +{
>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>> +
>>>>>>>> + if (sel->pad == ISC_SCALER_PAD_SOURCE)
>>>>>>>> + return -EINVAL;
>>>>>>>> +
>>>>>>>> + if (sel->target != V4L2_SEL_TGT_CROP_BOUNDS &&
>>>>>>>> + sel->target != V4L2_SEL_TGT_CROP)
>>>>>>>> + return -EINVAL;
>>>>>>>> +
>>>>>>>> + sel->r.height = isc->max_height;
>>>>>>>> + sel->r.width = isc->max_width;
>>>>>>>
>>>>>>> The CROP_BOUNDS should be set to the same size as the sink pad image format,
>>>>>>> as it represents the maximum valid crop rectangle.
>>>>>>>
>>>>>>> TGT_CROP should report the configured crop rectangle which can be
>>>>>>> intiialized to the same size as CROP_BOUNDS, if my understanding of
>>>>>>> the spec is correct
>>>>>>
>>>>>> So you would like to have this differentiated, and report the
>>>>>> CROP_BOUNDS to whatever is on the sink pad, and the TGT_CROP to what is
>>>>>> reported now, the maximum size of the ISC frame .
>>>>>> My understanding is correct ?
>>>>>>
>>>>>
>>>>> I didn't know you have an HW limitation, so your _BOUNDS is not the
>>>>> input image size but rather 3264x2464 ( == max_width x max_height).
>>>>>
>>>>> What I meant is that _BOUNDS should report the maximum rectangle size
>>>>> that can be applied to the _CROP target. In you case you have an HW
>>>>> limitation 3264x2464 and that's the largest rectangle you can apply.
>>>> So the CROP should be at 3264x2464
>>>>> TGT_CROP can be initialized to the same as _BOUND, but if you
>>>>> implement s_selection it should report what has been there applied.
>>>> and BOUND to actual frame size ?
>>>>> But as you don't implement s_selection yet, I think this is fine for
>>>>> now. Maybe a little comment ?
>>>>
>>>> It could also be the case where e.g. the sensor is outputting 1920x1080,
>>>> in this case the scaler would do nothing.
>>>> CROP is still 3264x2464 and BOUND in this case is 1920x1080 ?
>>>> If the sensor is outputting 1920x1080, this format comes directly from
>>>> the sensor (the sensor is cropping it from 3280x2464 or not... it's the
>>>> sensor's problem)
>>>>>
>>>>> Also, is set->r zeroed by the framework before getting here ?
>>>>> Otherwise you should set r.left and r.top to 0
>>>>
>>>> If these are redundant, no problem to remove them
>>>
>>> I was actually confused. I was suggesting you to add...
>>>
>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + sel->r.left = 0;
>>>>>>>> + sel->r.top = 0;
>>>
>>> These ^
>>>
>>> So nothing to change. Sorry I've missed them
>>>
>>> Thanks
>>> j
>>>
>>>
>>>>>>>> +
>>>>>>>> + return 0;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +static int isc_scaler_init_cfg(struct v4l2_subdev *sd,
>>>>>>>> + struct v4l2_subdev_state *sd_state)
>>>>>>>> +{
>>>>>>>> + struct v4l2_mbus_framefmt *v4l2_try_fmt =
>>>>>>>> + v4l2_subdev_get_try_format(sd, sd_state, 0);
>>>>>>>> + struct isc_device *isc = container_of(sd, struct isc_device, scaler_sd);
>>>>>>>> +
>>>>>>>> + *v4l2_try_fmt = isc->scaler_format;
>>>>>>>> +
>>>>>>>> + return 0;
>>>>>>>> +}
>>>>>>>> +
>>>>>>>> +static const struct v4l2_subdev_pad_ops isc_scaler_pad_ops = {
>>>>>>>> + .enum_mbus_code = isc_scaler_enum_mbus_code,
>>>>>>>> + .set_fmt = isc_scaler_set_fmt,
>>>>>>>> + .get_fmt = isc_scaler_get_fmt,
>>>>>>>> + .get_selection = isc_scaler_g_sel,
>>>>>>>> + .init_cfg = isc_scaler_init_cfg,
>>>>>>>
>>>>>>> .link_validate = v4l2_subdev_link_validate_default,
>>>>>>>
>>>>>>> To have the formats at the end of links that point to this entity
>>>>>>> validated (I think the framework already calls it if not set though,
>>>>>>> please check v4l2-subdev.c:v4l2_subdev_link_validate())
>>>>>>>
>>>>>>>> +};
>>>>>>>> +
>>>>>>>> +static const struct v4l2_subdev_ops xisc_scaler_subdev_ops = {
>>>>>>>> + .pad = &isc_scaler_pad_ops,
>>>>>>>> +};
>>>>>>>> +
>>>>>>>> +int isc_scaler_init(struct isc_device *isc)
>>>>>>>> +{
>>>>>>>> + int ret;
>>>>>>>> +
>>>>>>>> + v4l2_subdev_init(&isc->scaler_sd, &xisc_scaler_subdev_ops);
>>>>>>>> +
>>>>>>>> + isc->scaler_sd.owner = THIS_MODULE;
>>>>>>>> + isc->scaler_sd.dev = isc->dev;
>>>>>>>> + snprintf(isc->scaler_sd.name, sizeof(isc->scaler_sd.name),
>>>>>>>> + "atmel_isc_scaler");
>>>>>>>
>>>>>>> I would drop 'atmel' for brevity, unless other entities have this
>>>>>>> prefix set already
>>>>>>
>>>>>> The v4l entity takes it's name from the module name, which is
>>>>>> atmel_isc_common, so I thought to keep the prefix
>>>>>>
>>>>>
>>>>> Ack!
>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + isc->scaler_sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
>>>>>>>> + isc->scaler_sd.entity.function = MEDIA_ENT_F_PROC_VIDEO_SCALER;
>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SINK].flags = MEDIA_PAD_FL_SINK;
>>>>>>>> + isc->scaler_pads[ISC_SCALER_PAD_SOURCE].flags = MEDIA_PAD_FL_SOURCE;
>>>>>>>> +
>>>>>>>> + isc->scaler_format.height = isc->max_height;
>>>>>>>> + isc->scaler_format.width = isc->max_width;
>>>>>>>> + isc->scaler_format.code = isc->formats_list[0].mbus_code;
>>>>>>>> + isc->scaler_format.colorspace = V4L2_COLORSPACE_SRGB;
>>>>>>>> + isc->scaler_format.field = V4L2_FIELD_NONE;
>>>>>>>> + isc->scaler_format.ycbcr_enc = V4L2_YCBCR_ENC_DEFAULT;
>>>>>>>> + isc->scaler_format.quantization = V4L2_QUANTIZATION_DEFAULT;
>>>>>>>> + isc->scaler_format.xfer_func = V4L2_XFER_FUNC_DEFAULT;
>>>>>>>> +
>>>>>>>> + ret = media_entity_pads_init(&isc->scaler_sd.entity,
>>>>>>>> + ISC_SCALER_PADS_NUM,
>>>>>>>> + isc->scaler_pads);
>>>>>>>> + if (ret < 0) {
>>>>>>>> + dev_err(isc->dev, "scaler sd media entity init failed\n");
>>>>>>>> + return ret;
>>>>>>>> + }
>>>>>>>> + ret = v4l2_device_register_subdev(&isc->v4l2_dev, &isc->scaler_sd);
>>>>>>>> + if (ret < 0) {
>>>>>>>> + dev_err(isc->dev, "scaler sd failed to register subdev\n");
>>>>>>>> + return ret;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + return ret;
>>>>>>>> +}
>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_init);
>>>>>>>> +
>>>>>>>> +int isc_scaler_link(struct isc_device *isc)
>>>>>>>> +{
>>>>>>>> + int ret;
>>>>>>>> +
>>>>>>>> + ret = media_create_pad_link(&isc->current_subdev->sd->entity,
>>>>>>>> + isc->remote_pad, &isc->scaler_sd.entity,
>>>>>>>> + ISC_SCALER_PAD_SINK,
>>>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>>>> +
>>>>>>>> + if (ret < 0) {
>>>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>>>> + isc->current_subdev->sd->entity.name,
>>>>>>>> + isc->scaler_sd.entity.name);
>>>>>>>> + return ret;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n",
>>>>>>>> + isc->current_subdev->sd->name, isc->remote_pad);
>>>>>>>> +
>>>>>>>> + ret = media_create_pad_link(&isc->scaler_sd.entity,
>>>>>>>> + ISC_SCALER_PAD_SOURCE,
>>>>>>>> + &isc->video_dev.entity, ISC_PAD_SINK,
>>>>>>>> + MEDIA_LNK_FL_ENABLED |
>>>>>>>> + MEDIA_LNK_FL_IMMUTABLE);
>>>>>>>> +
>>>>>>>> + if (ret < 0) {
>>>>>>>> + v4l2_err(&isc->v4l2_dev,
>>>>>>>> + "Failed to create pad link: %s to %s\n",
>>>>>>>> + isc->scaler_sd.entity.name,
>>>>>>>> + isc->video_dev.entity.name);
>>>>>>>> + return ret;
>>>>>>>> + }
>>>>>>>> +
>>>>>>>> + dev_dbg(isc->dev, "link with %s pad: %d\n", isc->scaler_sd.name,
>>>>>>>> + ISC_SCALER_PAD_SOURCE);
>>>>>>>> +
>>>>>>>> + return ret;
>>>>>>>> +}
>>>>>>>> +EXPORT_SYMBOL_GPL(isc_scaler_link);
>>>>>>>
>>>>>>> From the DT graph point of view, the ISC appears as a single block
>>>>>>> with an CSI-2 input port. It is then in charge of parsing the .dts and
>>>>>>> add to its notifier the image sensor remote subdevice.
>>>>>>
>>>>>> Actually it's only a parallel port.
>>>>>> And it can be connected either to a sensor directly or to the csi2dc
>>>>>> bridge. (the csi2dc bridge can be connected to a CSI-2 receiver)
>>>>>>
>>>>>>>
>>>>>>> When the sensor subdev registers and the ISC notifier completes, the scaler
>>>>>>> entity which was initialized at isc_probe() time is linked in between
>>>>>>> the ISC and the image sensor immutably.
>>>>>>
>>>>>> yes, because this cannot change at runtime, and usually, it can't change
>>>>>> without altering the board hardware. (at least this is what my
>>>>>> understanding of immutability is )
>>>>>>
>>>>>>>
>>>>>>> I think it is fine for now, but I wonder if you plan to plumb more
>>>>>>> components between the ISC video node and the sensor, if it's not
>>>>>>> worth changing the DT bindings and their parsing logic to separate the
>>>>>>> CSI-2 receiver from the ISC, whcih can create its media graph at probe
>>>>>>> time and have the CSI-2 receiver entity as downstream remote. I think
>>>>>>> I need to get to know the ISC better to really have an idea. For now,
>>>>>>> this seems ok to me, but please check with maintainers if this is fine
>>>>>>> with them.
>>>>>>
>>>>>> In the XISC case (sama7g5-isc), the csi2dc receiver is the subdev (so,
>>>>>> the remote subdev).
>>>>>> The ISC will register a scaler, and connect the subdev to this scaler
>>>>>> first, and then, the scaler to the isc itself (the v4l entity).
>>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-isc.h b/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>> index 5fbf52a9080b..c9234c90ae58 100644
>>>>>>>> --- a/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-isc.h
>>>>>>>> @@ -183,6 +183,17 @@ struct isc_reg_offsets {
>>>>>>>> u32 his_entry;
>>>>>>>> };
>>>>>>>>
>>>>>>>> +enum isc_mc_pads {
>>>>>>>> + ISC_PAD_SINK = 0,
>>>>>>>> + ISC_PADS_NUM = 1,
>>>>>>>> +};
>>>>>>>> +
>>>>>>>> +enum isc_scaler_pads {
>>>>>>>> + ISC_SCALER_PAD_SINK = 0,
>>>>>>>> + ISC_SCALER_PAD_SOURCE = 1,
>>>>>>>> + ISC_SCALER_PADS_NUM = 2,
>>>>>>>> +};
>>>>>>>> +
>>>>>>>> /*
>>>>>>>> * struct isc_device - ISC device driver data/config struct
>>>>>>>> * @regmap: Register map
>>>>>>>> @@ -257,6 +268,12 @@ struct isc_reg_offsets {
>>>>>>>> * be used as an input to the controller
>>>>>>>> * @controller_formats_size: size of controller_formats array
>>>>>>>> * @formats_list_size: size of formats_list array
>>>>>>>> + * @pads: media controller pads for isc video entity
>>>>>>>> + * @mdev: media device that is registered by the isc
>>>>>>>> + * @remote_pad: remote pad on the connected subdevice
>>>>>>>> + * @scaler_sd: subdevice for the scaler that isc registers
>>>>>>>> + * @scaler_pads: media controller pads for the scaler subdevice
>>>>>>>> + * @scaler_format: current format for the scaler subdevice
>>>>>>>> */
>>>>>>>> struct isc_device {
>>>>>>>> struct regmap *regmap;
>>>>>>>> @@ -344,6 +361,19 @@ struct isc_device {
>>>>>>>> struct isc_format *formats_list;
>>>>>>>> u32 controller_formats_size;
>>>>>>>> u32 formats_list_size;
>>>>>>>> +
>>>>>>>> + struct {
>>>>>>>> + struct media_pad pads[ISC_PADS_NUM];
>>>>>>>> + struct media_device mdev;
>>>>>>>> +
>>>>>>>> + u32 remote_pad;
>>>>>>>> + };
>>>>>>>> +
>>>>>>>> + struct {
>>>>>>>> + struct v4l2_subdev scaler_sd;
>>>>>>>> + struct media_pad scaler_pads[ISC_SCALER_PADS_NUM];
>>>>>>>> + struct v4l2_mbus_framefmt scaler_format;
>>>>>>
>>>>>> Here are the scaler stuff which I added, in the same bulk struct for the
>>>>>> whole isc device
>>>>>>
>>>>>>
>>>>>>>> + };
>>>>>>>> };
>>>>>>>>
>>>>>>>> extern const struct regmap_config isc_regmap_config;
>>>>>>>> @@ -355,4 +385,11 @@ int isc_clk_init(struct isc_device *isc);
>>>>>>>> void isc_subdev_cleanup(struct isc_device *isc);
>>>>>>>> void isc_clk_cleanup(struct isc_device *isc);
>>>>>>>>
>>>>>>>> +int isc_scaler_link(struct isc_device *isc);
>>>>>>>> +int isc_scaler_init(struct isc_device *isc);
>>>>>>>> +int isc_mc_init(struct isc_device *isc, u32 ver);
>>>>>>>> +void isc_mc_cleanup(struct isc_device *isc);
>>>>>>>> +
>>>>>>>> +struct isc_format *isc_find_format_by_code(struct isc_device *isc,
>>>>>>>> + unsigned int code, int *index);
>>>>>>>> #endif
>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama5d2-isc.c b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>> index c5b9563e36cb..c244682ea22f 100644
>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama5d2-isc.c
>>>>>>>> @@ -553,6 +553,12 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>> break;
>>>>>>>> }
>>>>>>>>
>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>
>>>>>>> I am surprised you can read a register before runtime_pm is
>>>>>>> intialized!
>>>>>>>
>>>>>>
>>>>>> Actually I can read any register after the moment of starting the clock
>>>>>> on the hardware block.
>>>>>
>>>>> Ah right then
>>>>>
>>>>>> Maybe the starting and stopping of the clock needs to be moved to the
>>>>>> runtime_pm calls, but this is another story, not related to the media
>>>>>> controller.
>>>>>
>>>>> It's fine, but maybe worth recording with a todo ?
>>>>>
>>>>> Thanks
>>>>> j
>>>>>
>>>>>> I moved this line because I had to pass the version to the isc_mc_init call
>>>>>>
>>>>>>> Thanks
>>>>>>> j
>>>>>>
>>>>>> Thanks for reviewing !
>>>>>> Eugen
>>>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>>>> + if (ret < 0)
>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>> +
>>>>>>>> pm_runtime_set_active(dev);
>>>>>>>> pm_runtime_enable(dev);
>>>>>>>> pm_request_idle(dev);
>>>>>>>> @@ -562,7 +568,7 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>> ret = clk_prepare_enable(isc->ispck);
>>>>>>>> if (ret) {
>>>>>>>> dev_err(dev, "failed to enable ispck: %d\n", ret);
>>>>>>>> - goto cleanup_subdev;
>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>> }
>>>>>>>>
>>>>>>>> /* ispck should be greater or equal to hclock */
>>>>>>>> @@ -572,7 +578,6 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>> goto unprepare_clk;
>>>>>>>> }
>>>>>>>>
>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>> dev_info(dev, "Microchip ISC version %x\n", ver);
>>>>>>>>
>>>>>>>> return 0;
>>>>>>>> @@ -580,6 +585,9 @@ static int atmel_isc_probe(struct platform_device *pdev)
>>>>>>>> unprepare_clk:
>>>>>>>> clk_disable_unprepare(isc->ispck);
>>>>>>>>
>>>>>>>> +isc_probe_mc_init_err:
>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>> +
>>>>>>>> cleanup_subdev:
>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>
>>>>>>>> @@ -600,6 +608,8 @@ static int atmel_isc_remove(struct platform_device *pdev)
>>>>>>>>
>>>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>>>
>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>> +
>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>
>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>>>> diff --git a/drivers/media/platform/atmel/atmel-sama7g5-isc.c b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>> index 07a80b08bc54..9dc75eed0098 100644
>>>>>>>> --- a/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>> +++ b/drivers/media/platform/atmel/atmel-sama7g5-isc.c
>>>>>>>> @@ -547,15 +547,23 @@ static int microchip_xisc_probe(struct platform_device *pdev)
>>>>>>>> break;
>>>>>>>> }
>>>>>>>>
>>>>>>>> + regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>> +
>>>>>>>> + ret = isc_mc_init(isc, ver);
>>>>>>>> + if (ret < 0)
>>>>>>>> + goto isc_probe_mc_init_err;
>>>>>>>> +
>>>>>>>> pm_runtime_set_active(dev);
>>>>>>>> pm_runtime_enable(dev);
>>>>>>>> pm_request_idle(dev);
>>>>>>>>
>>>>>>>> - regmap_read(isc->regmap, ISC_VERSION + isc->offsets.version, &ver);
>>>>>>>> dev_info(dev, "Microchip XISC version %x\n", ver);
>>>>>>>>
>>>>>>>> return 0;
>>>>>>>>
>>>>>>>> +isc_probe_mc_init_err:
>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>> +
>>>>>>>> cleanup_subdev:
>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>
>>>>>>>> @@ -576,6 +584,8 @@ static int microchip_xisc_remove(struct platform_device *pdev)
>>>>>>>>
>>>>>>>> pm_runtime_disable(&pdev->dev);
>>>>>>>>
>>>>>>>> + isc_mc_cleanup(isc);
>>>>>>>> +
>>>>>>>> isc_subdev_cleanup(isc);
>>>>>>>>
>>>>>>>> v4l2_device_unregister(&isc->v4l2_dev);
>>>>>>>> --
>>>>>>>> 2.25.1
>>>>>>>>
>>>>>>
>>>>
>>