Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <03b551925d079fcc151239afa735562332cfd557.camel@pengutronix.de>
Subject: Re: [PATCH v3 2/4] drm/imx: Add initial support for DCSS on iMX8MQ
From:   Lucas Stach <l.stach@pengutronix.de>
To:     Laurentiu Palcu <laurentiu.palcu@nxp.com>,
        Philipp Zabel <p.zabel@pengutronix.de>,
        Shawn Guo <shawnguo@kernel.org>,
        Sascha Hauer <s.hauer@pengutronix.de>,
        Pengutronix Kernel Team <kernel@pengutronix.de>,
        Fabio Estevam <festevam@gmail.com>,
        NXP Linux Team <linux-imx@nxp.com>
Cc:     agx@sigxcpu.org, lukas@mntmn.com, linux-kernel@vger.kernel.org,
        dri-devel@lists.freedesktop.org,
        linux-arm-kernel@lists.infradead.org
Date:   Wed, 26 Feb 2020 14:19:11 +0100
In-Reply-To: <1575625964-27102-3-git-send-email-laurentiu.palcu@nxp.com>
References: <1575625964-27102-1-git-send-email-laurentiu.palcu@nxp.com>
         <1575625964-27102-3-git-send-email-laurentiu.palcu@nxp.com>
Content-Type: text/plain; charset="UTF-8"
User-Agent: Evolution 3.30.5-1.1 
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Laurentiu,

again a day later than promised, but here we go with some more
in-depth comments. Apologies if I missed something, my metal bandwidth
was pretty exhausted by the time I made my was to the scaler code. :)

On Fr, 2019-12-06 at 11:52 +0200, Laurentiu Palcu wrote:
> This adds initial support for iMX8MQ's Display Controller Subsystem (DCSS).
> Some of its capabilities include:
>  * 4K@60fps;
>  * HDR10;
>  * one graphics and 2 video pipelines;
>  * on-the-fly decompression of compressed video and graphics;
> 
> The reference manual can be found here:
> https://www.nxp.com/webapp/Download?colCode=IMX8MDQLQRM
> 
> The current patch adds only basic functionality: one primary plane for
> graphics, linear, tiled and super-tiled buffers support (no graphics
> decompression yet), no HDR10 and no video planes.
> 
> Video planes support and HDR10 will be added in subsequent patches once
> per-plane de-gamma/CSC/gamma support is in.
> 
> Signed-off-by: Laurentiu Palcu <laurentiu.palcu@nxp.com>
> ---

[...]

> diff --git a/drivers/gpu/drm/imx/dcss/dcss-blkctl.c b/drivers/gpu/drm/imx/dcss/dcss-blkctl.c
> new file mode 100644
> index 00000000..ee7ffa1
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-blkctl.c
> @@ -0,0 +1,75 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/device.h>
> +#include <linux/of.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_BLKCTL_RESET_CTRL		0x00
> +#define   B_CLK_RESETN			BIT(0)
> +#define   APB_CLK_RESETN		BIT(1)
> +#define   P_CLK_RESETN			BIT(2)
> +#define   RTR_CLK_RESETN		BIT(3)
> +#define DCSS_BLKCTL_CONTROL0		0x10
> +#define   HDMI_MIPI_CLK_SEL		BIT(0)
> +#define   DISPMIX_REFCLK_SEL_POS	4
> +#define   DISPMIX_REFCLK_SEL_MASK	GENMASK(5, 4)
> +#define   DISPMIX_PIXCLK_SEL		BIT(8)
> +#define   HDMI_SRC_SECURE_EN		BIT(16)
> +
> +struct dcss_blkctl {
> +	struct device *dev;
> +	void __iomem *base_reg;
> +
> +	bool hdmi_output;

Do we need this state in this structure? It's a duplication of state in
struct dcss_dev, so maybe better just keep a pointer to the parent
dcss_dev here?

> +};
> +
> +void dcss_blkctl_cfg(struct dcss_blkctl *blkctl)
> +{
> +	if (blkctl->hdmi_output)
> +		dcss_writel(0, blkctl->base_reg + DCSS_BLKCTL_CONTROL0);
> +	else
> +		dcss_writel(DISPMIX_PIXCLK_SEL,
> +			    blkctl->base_reg + DCSS_BLKCTL_CONTROL0);

The documentation is a bit sparse on why this needs to be configured
differently for HDMI output. Is HDMI the only case where this clock mux
needs to be configured to source from PLL2?

> +
> +	dcss_set(B_CLK_RESETN | APB_CLK_RESETN | P_CLK_RESETN | RTR_CLK_RESETN,
> +		 blkctl->base_reg + DCSS_BLKCTL_RESET_CTRL);
> +}
> +
> +int dcss_blkctl_init(struct dcss_dev *dcss, unsigned long blkctl_base)
> +{
> +	struct dcss_blkctl *blkctl;
> +
> +	blkctl = devm_kzalloc(dcss->dev, sizeof(*blkctl), GFP_KERNEL);

A general remark that also applies to all the other DCSS subblock
drivers: I don't think devm is the right choice for the resource
management here. All those drivers also have explicit resource freeing
in the _exit functions and there is an explicit rollback in
dcss_submodules_init, so I think we should not pretend to use devm
here, but just do explicit resource management without any devm
involvement.

> +	if (!blkctl)
> +		return -ENOMEM;
> +
> +	blkctl->base_reg = devm_ioremap(dcss->dev, blkctl_base, SZ_4K);
> +	if (!blkctl->base_reg) {
> +		dev_err(dcss->dev, "unable to remap BLK CTRL base\n");
> +		devm_kfree(dcss->dev, blkctl);
> +		return -ENOMEM;
> +	}
> +
> +	dcss->blkctl = blkctl;
> +	blkctl->dev = dcss->dev;
> +	blkctl->hdmi_output = dcss->hdmi_output;
> +
> +	dcss_blkctl_cfg(blkctl);
> +
> +	return 0;
> +}
> +
> +void dcss_blkctl_exit(struct dcss_blkctl *blkctl)
> +{
> +	dcss_clr(P_CLK_RESETN | RTR_CLK_RESETN,
> +		 blkctl->base_reg + DCSS_BLKCTL_RESET_CTRL);
> +
> +	if (blkctl->base_reg)
> +		devm_iounmap(blkctl->dev, blkctl->base_reg);
> +
> +	devm_kfree(blkctl->dev, blkctl);
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-crtc.c b/drivers/gpu/drm/imx/dcss/dcss-crtc.c
> new file mode 100644
> index 00000000..567bf07
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-crtc.c
> @@ -0,0 +1,224 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <drm/drm_atomic_helper.h>
> +#include <drm/drm_vblank.h>
> +#include <linux/platform_device.h>
> +#include <linux/pm_runtime.h>
> +
> +#include "dcss-dev.h"
> +#include "dcss-kms.h"
> +
> +static int dcss_enable_vblank(struct drm_crtc *crtc)
> +{
> +	struct dcss_crtc *dcss_crtc = container_of(crtc, struct dcss_crtc,
> +						   base);
> +	struct dcss_dev *dcss = crtc->dev->dev_private;
> +
> +	if (dcss_crtc->irq_enabled)
> +		return 0;
> +
> +	dcss_crtc->irq_enabled = true;

This state should not be necessary. Unless there is a reference
counting bug somewhere in the driver, the DRM core should never call
enable_vblank and disable_vblank unbalanced.

> +
> +	dcss_dtg_vblank_irq_enable(dcss->dtg, true);
> +
> +	dcss_dtg_ctxld_kick_irq_enable(dcss->dtg, true);
> +
> +	enable_irq(dcss_crtc->irq);
> +
> +	return 0;
> +}
> +
> +static void dcss_disable_vblank(struct drm_crtc *crtc)
> +{
> +	struct dcss_crtc *dcss_crtc = container_of(crtc, struct dcss_crtc,
> +						   base);
> +	struct dcss_dev *dcss = dcss_crtc->base.dev->dev_private;
> +
> +	disable_irq_nosync(dcss_crtc->irq);
> +
> +	dcss_dtg_vblank_irq_enable(dcss->dtg, false);
> +
> +	dcss_dtg_ctxld_kick_irq_enable(dcss->dtg, false);
> +
> +	dcss_crtc->irq_enabled = false;
> +}
> +
> +static const struct drm_crtc_funcs dcss_crtc_funcs = {
> +	.set_config = drm_atomic_helper_set_config,
> +	.destroy = drm_crtc_cleanup,
> +	.page_flip = drm_atomic_helper_page_flip,
> +	.reset = drm_atomic_helper_crtc_reset,
> +	.atomic_duplicate_state = drm_atomic_helper_crtc_duplicate_state,
> +	.atomic_destroy_state = drm_atomic_helper_crtc_destroy_state,
> +	.enable_vblank = dcss_enable_vblank,
> +	.disable_vblank = dcss_disable_vblank,
> +};
> +
> +static void dcss_crtc_atomic_begin(struct drm_crtc *crtc,
> +				   struct drm_crtc_state *old_crtc_state)
> +{
> +	drm_crtc_vblank_on(crtc);
> +
> +	spin_lock_irq(&crtc->dev->event_lock);
> +	if (crtc->state->event) {
> +		WARN_ON(drm_crtc_vblank_get(crtc));
> +		drm_crtc_arm_vblank_event(crtc, crtc->state->event);

Arming of the vblank event should move into atomic flush. Otherwise the
state update is racing with the vblank IRQ. In practice you probably
don't hit this issue on a moderately loaded system very often as the
CTXLD allows for almost a full frame duration to prepare the state
update, but the following race is present in the current code:

userspace                  commit worker                    HW

atomic-commit
                           dcss_crtc_atomic_begin
                            -> drm_crtc_arm_vblank_event
                           [prepare plane state updates]

                           ...
                                                            vblank
                           preempted
[reuses buffers from       ...
 last commit, while
 they are still scanned
 out, as state update
 didn't finish in time]
                           dcss_crtc_atomic_flush
                           -> CTXLD arm
                                                            vblank
                                                     (freeing buffers
                                                      from last commit)

> +		crtc->state->event = NULL;
> +	}
> +	spin_unlock_irq(&crtc->dev->event_lock);
> +}
> +
> +static void dcss_crtc_atomic_flush(struct drm_crtc *crtc,
> +				   struct drm_crtc_state *old_crtc_state)
> +{
> +	struct dcss_crtc *dcss_crtc = container_of(crtc, struct dcss_crtc,
> +						   base);
> +	struct dcss_dev *dcss = dcss_crtc->base.dev->dev_private;
> +
> +	if (dcss_dtg_is_enabled(dcss->dtg))
> +		dcss_ctxld_enable(dcss->ctxld);
> +}
> +
> +static void dcss_crtc_atomic_enable(struct drm_crtc *crtc,
> +				    struct drm_crtc_state *old_crtc_state)
> +{
> +	struct dcss_crtc *dcss_crtc = container_of(crtc, struct dcss_crtc,
> +						   base);
> +	struct dcss_dev *dcss = dcss_crtc->base.dev->dev_private;
> +	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
> +	struct videomode vm;
> +
> +	drm_display_mode_to_videomode(mode, &vm);
> +
> +	pm_runtime_get_sync(dcss->dev);
> +
> +	vm.pixelclock = mode->crtc_clock * 1000;
> +
> +	dcss_dtg_sync_set(dcss->dtg, &vm);
> +
> +	dcss_ss_subsam_set(dcss->ss);
> +	dcss_ss_sync_set(dcss->ss, &vm, mode->flags & DRM_MODE_FLAG_PHSYNC,
> +			 mode->flags & DRM_MODE_FLAG_PVSYNC);
> +
> +	dcss_dtg_css_set(dcss->dtg);
> +
> +	dcss_ss_enable(dcss->ss);
> +	dcss_dtg_enable(dcss->dtg, true, NULL);
> +	dcss_ctxld_enable(dcss->ctxld);
> +
> +	dcss_enable_vblank(crtc);

Why is this needed? The drm_crtc_vblank_on in dcss_crtc_atomic_begin
should keep vblank IRQs enabled as long as needed.

> +
> +	reinit_completion(&dcss_crtc->en_completion);

This is racing with the IRQ. The completion should be initialized
before enabling the hardware, as otherwise the IRQ might fire before we
get to the point where the completion is initialized, causing the wait
below to time out.

> +	wait_for_completion_timeout(&dcss_crtc->en_completion,
> +				    msecs_to_jiffies(500));
> +}
> +
> +static void dcss_crtc_atomic_disable(struct drm_crtc *crtc,
> +				     struct drm_crtc_state *old_crtc_state)
> +{
> +	struct dcss_crtc *dcss_crtc = container_of(crtc, struct dcss_crtc,
> +						   base);
> +	struct dcss_dev *dcss = dcss_crtc->base.dev->dev_private;
> +
> +	drm_atomic_helper_disable_planes_on_crtc(old_crtc_state, false);
> +
> +	spin_lock_irq(&crtc->dev->event_lock);
> +	if (crtc->state->event) {
> +		drm_crtc_send_vblank_event(crtc, crtc->state->event);
> +		crtc->state->event = NULL;
> +	}
> +	spin_unlock_irq(&crtc->dev->event_lock);
> +
> +	dcss_dtg_ctxld_kick_irq_enable(dcss->dtg, true);
> +
> +	dcss_ss_disable(dcss->ss);
> +	dcss_dtg_enable(dcss->dtg, false, &dcss_crtc->dis_completion);
> +	dcss_ctxld_enable(dcss->ctxld);
> +
> +	reinit_completion(&dcss_crtc->dis_completion);

Same as above. Init completion before calling the function, which might
potentially complete it.

> +	wait_for_completion_timeout(&dcss_crtc->dis_completion,
> +				    msecs_to_jiffies(100));
> +
> +	drm_crtc_vblank_off(crtc);

Those should be replaced by drm_crtc_vblank_get and drm_crtc_vblank_put
where actually needed.

> +
> +	dcss_dtg_ctxld_kick_irq_enable(dcss->dtg, false);
> +
> +	pm_runtime_put_sync(dcss->dev);
> +}
> +
> +static const struct drm_crtc_helper_funcs dcss_helper_funcs = {
> +	.atomic_begin = dcss_crtc_atomic_begin,
> +	.atomic_flush = dcss_crtc_atomic_flush,
> +	.atomic_enable = dcss_crtc_atomic_enable,
> +	.atomic_disable = dcss_crtc_atomic_disable,
> +};
> +
> +static irqreturn_t dcss_crtc_irq_handler(int irq, void *dev_id)
> +{
> +	struct dcss_crtc *dcss_crtc = dev_id;
> +	struct dcss_dev *dcss = dcss_crtc->base.dev->dev_private;
> +
> +	if (!dcss_dtg_vblank_irq_valid(dcss->dtg))
> +		return IRQ_HANDLED;

This should be IRQ_NONE, to give the IRQ core a chance to detect and
disable a runaway spurious IRQ.

> +
> +	complete(&dcss_crtc->en_completion);
> +
> +	if (dcss_ctxld_is_flushed(dcss->ctxld))
> +		drm_crtc_handle_vblank(&dcss_crtc->base);
> +
> +	dcss_dtg_vblank_irq_clear(dcss->dtg);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +int dcss_crtc_init(struct dcss_crtc *crtc, struct drm_device *drm)
> +{
> +	struct dcss_dev *dcss = drm->dev_private;
> +	struct platform_device *pdev = to_platform_device(dcss->dev);
> +	int ret;
> +
> +	crtc->plane[0] = dcss_plane_init(drm, drm_crtc_mask(&crtc->base),
> +					 DRM_PLANE_TYPE_PRIMARY, 0);
> +	if (IS_ERR(crtc->plane[0]))
> +		return PTR_ERR(crtc->plane[0]);
> +
> +	crtc->base.port = dcss->of_port;
> +
> +	drm_crtc_helper_add(&crtc->base, &dcss_helper_funcs);
> +	ret = drm_crtc_init_with_planes(drm, &crtc->base, &crtc->plane[0]->base,
> +					NULL, &dcss_crtc_funcs, NULL);
> +	if (ret) {
> +		dev_err(dcss->dev, "failed to init crtc\n");
> +		return ret;
> +	}
> +
> +	crtc->irq = platform_get_irq_byname(pdev, "vblank");
> +	if (crtc->irq < 0) {
> +		dev_err(dcss->dev, "unable to get vblank interrupt\n");

platform_get_irq_byname already prints an error message if the IRQ
can't be found, so no point in doing the same here.

> +		return crtc->irq;
> +	}
> +
> +	init_completion(&crtc->en_completion);
> +	init_completion(&crtc->dis_completion);
> +
> +	ret = devm_request_irq(dcss->dev, crtc->irq, dcss_crtc_irq_handler,
> +			       IRQF_TRIGGER_RISING, "dcss_drm", crtc);

AFAIK the irqsteer IRQs are level triggered. While irqsteer will just
ignore the wrong trigger flags, better just remove the the flags and
rely on the DT to provide the correct flags.

> +	if (ret) {
> +		dev_err(dcss->dev, "irq request failed with %d.\n", ret);
> +		return ret;
> +	}
> +
> +	disable_irq(crtc->irq);
> +
> +	return 0;
> +}
> +
> +void dcss_crtc_deinit(struct dcss_crtc *crtc, struct drm_device *drm)
> +{
> +	struct dcss_dev *dcss = drm->dev_private;
> +
> +	devm_free_irq(dcss->dev, crtc->irq, crtc);
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-ctxld.c b/drivers/gpu/drm/imx/dcss/dcss-ctxld.c
> new file mode 100644
> index 00000000..4fe35b2b
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-ctxld.c
> @@ -0,0 +1,447 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/delay.h>
> +#include <linux/dma-mapping.h>
> +#include <linux/interrupt.h>
> +#include <linux/platform_device.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_CTXLD_DEVNAME		"dcss_ctxld"
> +
> +#define DCSS_CTXLD_CONTROL_STATUS	0x0
> +#define   CTXLD_ENABLE			BIT(0)
> +#define   ARB_SEL			BIT(1)
> +#define   RD_ERR_EN			BIT(2)
> +#define   DB_COMP_EN			BIT(3)
> +#define   SB_HP_COMP_EN			BIT(4)
> +#define   SB_LP_COMP_EN			BIT(5)
> +#define   DB_PEND_SB_REC_EN		BIT(6)
> +#define   SB_PEND_DISP_ACTIVE_EN	BIT(7)
> +#define   AHB_ERR_EN			BIT(8)
> +#define   RD_ERR			BIT(16)
> +#define   DB_COMP			BIT(17)
> +#define   SB_HP_COMP			BIT(18)
> +#define   SB_LP_COMP			BIT(19)
> +#define   DB_PEND_SB_REC		BIT(20)
> +#define   SB_PEND_DISP_ACTIVE		BIT(21)
> +#define   AHB_ERR			BIT(22)
> +#define DCSS_CTXLD_DB_BASE_ADDR		0x10
> +#define DCSS_CTXLD_DB_COUNT		0x14
> +#define DCSS_CTXLD_SB_BASE_ADDR		0x18
> +#define DCSS_CTXLD_SB_COUNT		0x1C
> +#define   SB_HP_COUNT_POS		0
> +#define   SB_HP_COUNT_MASK		0xffff
> +#define   SB_LP_COUNT_POS		16
> +#define   SB_LP_COUNT_MASK		0xffff0000
> +#define DCSS_AHB_ERR_ADDR		0x20
> +
> +#define CTXLD_IRQ_NAME			"ctx_ld"
> +#define CTXLD_IRQ_COMPLETION		(DB_COMP | SB_HP_COMP | SB_LP_COMP)
> +#define CTXLD_IRQ_ERROR			(RD_ERR | DB_PEND_SB_REC | AHB_ERR)
> +
> +/* The following sizes are in context loader entries, 8 bytes each. */
> +#define CTXLD_DB_CTX_ENTRIES		1024	/* max 65536 */
> +#define CTXLD_SB_LP_CTX_ENTRIES		10240	/* max 65536 */
> +#define CTXLD_SB_HP_CTX_ENTRIES		20000	/* max 65536 */
> +#define CTXLD_SB_CTX_ENTRIES		(CTXLD_SB_LP_CTX_ENTRIES + \
> +					 CTXLD_SB_HP_CTX_ENTRIES)
> +
> +/* Sizes, in entries, of the DB, SB_HP and SB_LP context regions. */
> +static u16 dcss_ctxld_ctx_size[3] = {
> +	CTXLD_DB_CTX_ENTRIES,
> +	CTXLD_SB_HP_CTX_ENTRIES,
> +	CTXLD_SB_LP_CTX_ENTRIES
> +};
> +
> +/* this represents an entry in the context loader map */
> +struct dcss_ctxld_item {
> +	u32 val;
> +	u32 ofs;
> +};
> +
> +#define CTX_ITEM_SIZE			sizeof(struct dcss_ctxld_item)
> +
> +struct dcss_ctxld {
> +	struct device *dev;
> +	void __iomem *ctxld_reg;
> +	int irq;
> +	bool irq_en;
> +
> +	struct dcss_ctxld_item *db[2];
> +	struct dcss_ctxld_item *sb_hp[2];
> +	struct dcss_ctxld_item *sb_lp[2];
> +
> +	dma_addr_t db_paddr[2];
> +	dma_addr_t sb_paddr[2];
> +
> +	u16 ctx_size[2][3]; /* holds the sizes of DB, SB_HP and SB_LP ctx */
> +	u8 current_ctx;
> +
> +	bool in_use;
> +	bool armed;
> +
> +	spinlock_t lock; /* protects concurent access to private data */
> +
> +	void (*dtg_disable_cb)(void *data);
> +	void *dtg_disable_data;

The only use of this callback is to signal a completion. As this can be
done from IRQ context, couldn't we just remember the completion and
remove this callback indirection?

> +};
> +
> +static int __dcss_ctxld_enable(struct dcss_ctxld *ctxld);

This forward declaration isn't needed AFAICS.

> +static irqreturn_t dcss_ctxld_irq_handler(int irq, void *data)
> +{
> +	struct dcss_ctxld *ctxld = data;
> +	u32 irq_status;
> +
> +	irq_status = dcss_readl(ctxld->ctxld_reg + DCSS_CTXLD_CONTROL_STATUS);
> +
> +	if (irq_status & CTXLD_IRQ_COMPLETION &&
> +	    !(irq_status & CTXLD_ENABLE) && ctxld->in_use) {
> +		ctxld->in_use = false;
> +
> +		if (ctxld->dtg_disable_cb) {
> +			ctxld->dtg_disable_cb(ctxld->dtg_disable_data);
> +			ctxld->dtg_disable_cb = NULL;
> +			ctxld->dtg_disable_data = NULL;
> +		}
> +	} else if (irq_status & CTXLD_IRQ_ERROR) {
> +		/*
> +		 * Except for throwing an error message and clearing the status
> +		 * register, there's not much we can do here.
> +		 */
> +		dev_err(ctxld->dev, "ctxld: error encountered: %08x\n",
> +			irq_status);
> +		dev_err(ctxld->dev, "ctxld: db=%d, sb_hp=%d, sb_lp=%d\n",
> +			ctxld->ctx_size[ctxld->current_ctx ^ 1][CTX_DB],
> +			ctxld->ctx_size[ctxld->current_ctx ^ 1][CTX_SB_HP],
> +			ctxld->ctx_size[ctxld->current_ctx ^ 1][CTX_SB_LP]);
> +	}
> +
> +	dcss_clr(irq_status & (CTXLD_IRQ_ERROR | CTXLD_IRQ_COMPLETION),
> +		 ctxld->ctxld_reg + DCSS_CTXLD_CONTROL_STATUS);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static int dcss_ctxld_irq_config(struct dcss_ctxld *ctxld,
> +				 struct platform_device *pdev)
> +{
> +	int ret;
> +
> +	ctxld->irq = platform_get_irq_byname(pdev, CTXLD_IRQ_NAME);

Why have a define for the IRQ name? It's only used a single time in
this call, so I think it would be clearer to just have the string here.

> +	if (ctxld->irq < 0) {
> +		dev_err(ctxld->dev, "ctxld: can't get irq number\n");

platform_get_irq_byname already prints an error message if the IRQ
can't be found, so no point in doing the same here.

> +		return ctxld->irq;
> +	}
> +
> +	ret = devm_request_irq(ctxld->dev, ctxld->irq,
> +			       dcss_ctxld_irq_handler,
> +			       IRQF_ONESHOT | IRQF_TRIGGER_HIGH,

Same as with the vblank IRQ, remove the trigger flags and rely on DT to
provide the correct trigger. Also IRQF_ONESHOT doesn't really makes
sense for a non-threaded interrupt handler.

> +			       DCSS_CTXLD_DEVNAME, ctxld);
> +	if (ret) {
> +		dev_err(ctxld->dev, "ctxld: irq request failed.\n");
> +		return ret;
> +	}
> +
> +	ctxld->irq_en = true;
> +
> +	return 0;
> +}
> +
> +void dcss_ctxld_hw_cfg(struct dcss_ctxld *ctxld)

Missing static annotation for local function.

> +{
> +	dcss_writel(RD_ERR_EN | SB_HP_COMP_EN |
> +		    DB_PEND_SB_REC_EN | AHB_ERR_EN | RD_ERR | AHB_ERR,
> +		    ctxld->ctxld_reg + DCSS_CTXLD_CONTROL_STATUS);
> +}
> +
> +static void dcss_ctxld_free_ctx(struct dcss_ctxld *ctxld)
> +{
> +	struct dcss_ctxld_item *ctx;
> +	int i;
> +
> +	for (i = 0; i < 2; i++) {
> +		if (ctxld->db[i]) {
> +			dmam_free_coherent(ctxld->dev,
> +					   CTXLD_DB_CTX_ENTRIES * sizeof(*ctx),
> +					   ctxld->db[i], ctxld->db_paddr[i]);

Same comment as with the other devm_ functions: just use the raw
dma_*_coherent functions when doing explicit resource management.

> +			ctxld->db[i] = NULL;
> +			ctxld->db_paddr[i] = 0;
> +		}
> +
> +		if (ctxld->sb_hp[i]) {
> +			dmam_free_coherent(ctxld->dev,
> +					   CTXLD_SB_CTX_ENTRIES * sizeof(*ctx),
> +					   ctxld->sb_hp[i], ctxld->sb_paddr[i]);
> +			ctxld->sb_hp[i] = NULL;
> +			ctxld->sb_paddr[i] = 0;
> +		}
> +	}
> +}
> +
> +static int dcss_ctxld_alloc_ctx(struct dcss_ctxld *ctxld)
> +{
> +	struct dcss_ctxld_item *ctx;
> +	int i;
> +	dma_addr_t dma_handle;
> +
> +	for (i = 0; i < 2; i++) {
> +		ctx = dmam_alloc_coherent(ctxld->dev,
> +					  CTXLD_DB_CTX_ENTRIES * sizeof(*ctx),
> +					  &dma_handle, GFP_KERNEL);

What's the point of the local dma_handle variable? Just use
ctxld->db_paddr[i] directly.

> +		if (!ctx)
> +			return -ENOMEM;
> +
> +		ctxld->db[i] = ctx;
> +		ctxld->db_paddr[i] = dma_handle;
> +
> +		ctx = dmam_alloc_coherent(ctxld->dev,
> +					  CTXLD_SB_CTX_ENTRIES * sizeof(*ctx),
> +					  &dma_handle, GFP_KERNEL);
> +		if (!ctx)
> +			return -ENOMEM;
> +
> +		ctxld->sb_hp[i] = ctx;
> +		ctxld->sb_lp[i] = ctx + CTXLD_SB_HP_CTX_ENTRIES;
> +
> +		ctxld->sb_paddr[i] = dma_handle;
> +	}
> +
> +	return 0;
> +}
> +
> +int dcss_ctxld_init(struct dcss_dev *dcss, unsigned long ctxld_base)
> +{
> +	struct dcss_ctxld *ctxld;
> +	int ret;
> +
> +	ctxld = devm_kzalloc(dcss->dev, sizeof(struct dcss_ctxld),
> +			     GFP_KERNEL);
> +	if (!ctxld)
> +		return -ENOMEM;
> +
> +	dcss->ctxld = ctxld;
> +	ctxld->dev = dcss->dev;
> +
> +	spin_lock_init(&ctxld->lock);
> +
> +	ret = dcss_ctxld_alloc_ctx(ctxld);
> +	if (ret) {
> +		dev_err(dcss->dev, "ctxld: cannot allocate context memory.\n");
> +		goto err;
> +	}
> +
> +	ctxld->ctxld_reg = devm_ioremap(dcss->dev, ctxld_base, SZ_4K);
> +	if (!ctxld->ctxld_reg) {
> +		dev_err(dcss->dev, "ctxld: unable to remap ctxld base\n");
> +		ret = -ENOMEM;
> +		goto err;
> +	}
> +
> +	ret = dcss_ctxld_irq_config(ctxld, to_platform_device(dcss->dev));
> +	if (ret)
> +		goto err_irq;
> +
> +	dcss_ctxld_hw_cfg(ctxld);
> +
> +	return 0;
> +
> +err_irq:
> +	devm_iounmap(ctxld->dev, ctxld->ctxld_reg);
> +
> +err:
> +	dcss_ctxld_free_ctx(ctxld);
> +	devm_kfree(ctxld->dev, ctxld);
> +
> +	return ret;
> +}
> +
> +void dcss_ctxld_exit(struct dcss_ctxld *ctxld)
> +{
> +	devm_free_irq(ctxld->dev, ctxld->irq, ctxld);
> +
> +	if (ctxld->ctxld_reg)
> +		devm_iounmap(ctxld->dev, ctxld->ctxld_reg);
> +
> +	dcss_ctxld_free_ctx(ctxld);
> +	devm_kfree(ctxld->dev, ctxld);
> +}
> +
> +static int __dcss_ctxld_enable(struct dcss_ctxld *ctxld)

I don't like those __something local functions. Maybe call this
dcss_ctxld_enable_locked instead?

> +{
> +	int curr_ctx = ctxld->current_ctx;
> +	u32 db_base, sb_base, sb_count;
> +	u32 sb_hp_cnt, sb_lp_cnt, db_cnt;
> +	struct dcss_dev *dcss = dcss_drv_dev_to_dcss(ctxld->dev);
> +
> +	dcss_dpr_write_sysctrl(dcss->dpr);
> +
> +	dcss_scaler_write_sclctrl(dcss->scaler);
> +
> +	sb_hp_cnt = ctxld->ctx_size[curr_ctx][CTX_SB_HP];
> +	sb_lp_cnt = ctxld->ctx_size[curr_ctx][CTX_SB_LP];
> +	db_cnt = ctxld->ctx_size[curr_ctx][CTX_DB];
> +
> +	/* make sure SB_LP context area comes after SB_HP */
> +	if (sb_lp_cnt &&
> +	    ctxld->sb_lp[curr_ctx] != ctxld->sb_hp[curr_ctx] + sb_hp_cnt) {
> +		struct dcss_ctxld_item *sb_lp_adjusted;
> +
> +		sb_lp_adjusted = ctxld->sb_hp[curr_ctx] + sb_hp_cnt;
> +
> +		memcpy(sb_lp_adjusted, ctxld->sb_lp[curr_ctx],
> +		       sb_lp_cnt * CTX_ITEM_SIZE);
> +	}
> +
> +	db_base = db_cnt ? ctxld->db_paddr[curr_ctx] : 0;
> +
> +	dcss_writel(db_base, ctxld->ctxld_reg + DCSS_CTXLD_DB_BASE_ADDR);
> +	dcss_writel(db_cnt, ctxld->ctxld_reg + DCSS_CTXLD_DB_COUNT);
> +
> +	if (sb_hp_cnt)
> +		sb_count = ((sb_hp_cnt << SB_HP_COUNT_POS) & SB_HP_COUNT_MASK) |
> +			   ((sb_lp_cnt << SB_LP_COUNT_POS) & SB_LP_COUNT_MASK);
> +	else
> +		sb_count = (sb_lp_cnt << SB_HP_COUNT_POS) & SB_HP_COUNT_MASK;
> +
> +	sb_base = sb_count ? ctxld->sb_paddr[curr_ctx] : 0;
> +
> +	dcss_writel(sb_base, ctxld->ctxld_reg + DCSS_CTXLD_SB_BASE_ADDR);
> +	dcss_writel(sb_count, ctxld->ctxld_reg + DCSS_CTXLD_SB_COUNT);
> +
> +	/* enable the context loader */
> +	dcss_set(CTXLD_ENABLE, ctxld->ctxld_reg + DCSS_CTXLD_CONTROL_STATUS);
> +
> +	ctxld->in_use = true;
> +
> +	/*
> +	 * Toggle the current context to the alternate one so that any updates
> +	 * in the modules' settings take place there.
> +	 */
> +	ctxld->current_ctx ^= 1;
> +
> +	ctxld->ctx_size[ctxld->current_ctx][CTX_DB] = 0;
> +	ctxld->ctx_size[ctxld->current_ctx][CTX_SB_HP] = 0;
> +	ctxld->ctx_size[ctxld->current_ctx][CTX_SB_LP] = 0;
> +
> +	return 0;
> +}
> +
> +int dcss_ctxld_enable(struct dcss_ctxld *ctxld)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ctxld->lock, flags);

This function is never called from IRQ context, so you can use the
lower overhead spin_lock_irq here.

> +	ctxld->armed = true;
> +	spin_unlock_irqrestore(&ctxld->lock, flags);
> +
> +	return 0;
> +}
> +
> +void dcss_ctxld_kick(struct dcss_ctxld *ctxld)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ctxld->lock, flags);
> +	if (ctxld->armed && !ctxld->in_use) {
> +		ctxld->armed = false;
> +		__dcss_ctxld_enable(ctxld);
> +	}
> +	spin_unlock_irqrestore(&ctxld->lock, flags);
> +}
> +
> +void dcss_ctxld_write_irqsafe(struct dcss_ctxld *ctxld, u32 ctx_id, u32 val,
> +			      u32 reg_ofs)
> +{
> +	int curr_ctx = ctxld->current_ctx;
> +	struct dcss_ctxld_item *ctx[] = {
> +		[CTX_DB] = ctxld->db[curr_ctx],
> +		[CTX_SB_HP] = ctxld->sb_hp[curr_ctx],
> +		[CTX_SB_LP] = ctxld->sb_lp[curr_ctx]
> +	};
> +	int item_idx = ctxld->ctx_size[curr_ctx][ctx_id];
> +
> +	if (item_idx + 1 > dcss_ctxld_ctx_size[ctx_id]) {
> +		WARN_ON(1);
> +		return;
> +	}
> +
> +	ctx[ctx_id][item_idx].val = val;
> +	ctx[ctx_id][item_idx].ofs = reg_ofs;
> +	ctxld->ctx_size[curr_ctx][ctx_id] += 1;
> +}
> +
> +void dcss_ctxld_write(struct dcss_ctxld *ctxld, u32 ctx_id,
> +		      u32 val, u32 reg_ofs)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ctxld->lock, flags);

I have not actually validated all callers, but I suspect that this
function also isn't called from IRQ context, so can use spin_lock_irq.

> +	dcss_ctxld_write_irqsafe(ctxld, ctx_id, val, reg_ofs);
> +	spin_unlock_irqrestore(&ctxld->lock, flags);
> +}
> +
> +bool dcss_ctxld_is_flushed(struct dcss_ctxld *ctxld)
> +{
> +	return ctxld->ctx_size[ctxld->current_ctx][CTX_DB] == 0 &&
> +		ctxld->ctx_size[ctxld->current_ctx][CTX_SB_HP] == 0 &&
> +		ctxld->ctx_size[ctxld->current_ctx][CTX_SB_LP] == 0;
> +}
> +
> +int dcss_ctxld_resume(struct dcss_ctxld *ctxld)
> +{
> +	dcss_ctxld_hw_cfg(ctxld);
> +
> +	if (!ctxld->irq_en) {
> +		enable_irq(ctxld->irq);
> +		ctxld->irq_en = true;
> +	}
> +
> +	return 0;
> +}
> +
> +int dcss_ctxld_suspend(struct dcss_ctxld *ctxld)
> +{
> +	int ret = 0;
> +	int wait_time_ms = 0;
> +	unsigned long flags;
> +
> +	dcss_ctxld_kick(ctxld);
> +
> +	while (ctxld->in_use && wait_time_ms < 500) {
> +		msleep(20);
> +		wait_time_ms += 20;
> +	}

msleep might sleep a lot longer than the specified time, so if you care
about the timeout being somewhat accurate you should use a timeout
based on jiffies:

unsigned long timeout = jiffies + msecs_to_jiffies(500);

while (!time_after(jiffies, timout) && ctxld->in_use)
	msleep(20);


> +	if (wait_time_ms > 500)
> +		return -ETIMEDOUT;
> +
> +	spin_lock_irqsave(&ctxld->lock, flags);

spin_lock_irq

> +
> +	if (ctxld->irq_en) {
> +		disable_irq_nosync(ctxld->irq);
> +		ctxld->irq_en = false;
> +	}
> +
> +	/* reset context region and sizes */
> +	ctxld->current_ctx = 0;
> +	ctxld->ctx_size[0][CTX_DB] = 0;
> +	ctxld->ctx_size[0][CTX_SB_HP] = 0;
> +	ctxld->ctx_size[0][CTX_SB_LP] = 0;
> +
> +	spin_unlock_irqrestore(&ctxld->lock, flags);
> +
> +	return ret;
> +}
> +
> +void dcss_ctxld_register_dtg_disable_cb(struct dcss_ctxld *ctxld,
> +					void (*cb)(void *),
> +					void *data)
> +{
> +	ctxld->dtg_disable_cb = cb;
> +	ctxld->dtg_disable_data = data;
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-dev.c b/drivers/gpu/drm/imx/dcss/dcss-dev.c
> new file mode 100644
> index 00000000..265bf3c
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-dev.c
> @@ -0,0 +1,286 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/of_device.h>
> +#include <linux/of_graph.h>
> +#include <linux/pm_runtime.h>
> +#include <drm/drm_modeset_helper.h>
> +
> +#include "dcss-dev.h"
> +
> +static void dcss_clocks_enable(struct dcss_dev *dcss)
> +{
> +	if (dcss->clks_on)
> +		return;

AFAICS dcss_clocks_enable and dcss_clocks_disable are always called
balanced, so there is no point in having this clks_on state, right?

> +
> +	clk_prepare_enable(dcss->axi_clk);
> +	clk_prepare_enable(dcss->apb_clk);
> +	clk_prepare_enable(dcss->rtrm_clk);
> +	clk_prepare_enable(dcss->dtrc_clk);
> +	clk_prepare_enable(dcss->pix_clk);
> +
> +	dcss->clks_on = true;
> +}
> +
> +static void dcss_clocks_disable(struct dcss_dev *dcss)
> +{
> +	if (!dcss->clks_on)
> +		return;
> +
> +	clk_disable_unprepare(dcss->pix_clk);
> +	clk_disable_unprepare(dcss->dtrc_clk);
> +	clk_disable_unprepare(dcss->rtrm_clk);
> +	clk_disable_unprepare(dcss->apb_clk);
> +	clk_disable_unprepare(dcss->axi_clk);
> +
> +	dcss->clks_on = false;
> +}
> +
> +static int dcss_submodules_init(struct dcss_dev *dcss)
> +{
> +	int ret = 0;
> +	u32 base_addr = dcss->start_addr;
> +	const struct dcss_type_data *devtype = dcss->devtype;
> +
> +	dcss_clocks_enable(dcss);
> +
> +	ret = dcss_blkctl_init(dcss, base_addr + devtype->blkctl_ofs);
> +	if (ret)
> +		return ret;
> +
> +	ret = dcss_ctxld_init(dcss, base_addr + devtype->ctxld_ofs);
> +	if (ret)
> +		goto ctxld_err;
> +
> +	ret = dcss_dtg_init(dcss, base_addr + devtype->dtg_ofs);
> +	if (ret)
> +		goto dtg_err;
> +
> +	ret = dcss_ss_init(dcss, base_addr + devtype->ss_ofs);
> +	if (ret)
> +		goto ss_err;
> +
> +	ret = dcss_dpr_init(dcss, base_addr + devtype->dpr_ofs);
> +	if (ret)
> +		goto dpr_err;
> +
> +	ret = dcss_scaler_init(dcss, base_addr + devtype->scaler_ofs);
> +	if (ret)
> +		goto scaler_err;
> +
> +	return 0;
> +
> +scaler_err:
> +	dcss_dpr_exit(dcss->dpr);
> +
> +dpr_err:
> +	dcss_ss_exit(dcss->ss);
> +
> +ss_err:
> +	dcss_dtg_exit(dcss->dtg);
> +
> +dtg_err:
> +	dcss_ctxld_exit(dcss->ctxld);
> +
> +ctxld_err:
> +	dcss_blkctl_exit(dcss->blkctl);
> +
> +	dcss_clocks_disable(dcss);
> +
> +	return ret;
> +}
> +
> +static void dcss_submodules_stop(struct dcss_dev *dcss)
> +{
> +	dcss_clocks_enable(dcss);
> +	dcss_scaler_exit(dcss->scaler);
> +	dcss_dpr_exit(dcss->dpr);
> +	dcss_ss_exit(dcss->ss);
> +	dcss_dtg_exit(dcss->dtg);
> +	dcss_ctxld_exit(dcss->ctxld);
> +	dcss_blkctl_exit(dcss->blkctl);
> +	dcss_clocks_disable(dcss);
> +}
> +
> +static int dcss_clks_init(struct dcss_dev *dcss)
> +{
> +	int i;
> +	struct {
> +		const char *id;
> +		struct clk **clk;
> +	} clks[] = {
> +		{"apb",   &dcss->apb_clk},
> +		{"axi",   &dcss->axi_clk},
> +		{"pix",   &dcss->pix_clk},
> +		{"rtrm",  &dcss->rtrm_clk},
> +		{"dtrc",  &dcss->dtrc_clk},
> +	};
> +
> +	for (i = 0; i < ARRAY_SIZE(clks); i++) {
> +		*clks[i].clk = devm_clk_get(dcss->dev, clks[i].id);
> +		if (IS_ERR(*clks[i].clk)) {
> +			dev_err(dcss->dev, "failed to get %s clock\n",
> +				clks[i].id);
> +			return PTR_ERR(*clks[i].clk);
> +		}
> +	}
> +
> +	dcss->clks_on = false;
> +
> +	return 0;
> +}
> +
> +static void dcss_clks_release(struct dcss_dev *dcss)
> +{
> +	devm_clk_put(dcss->dev, dcss->dtrc_clk);
> +	devm_clk_put(dcss->dev, dcss->rtrm_clk);
> +	devm_clk_put(dcss->dev, dcss->pix_clk);
> +	devm_clk_put(dcss->dev, dcss->axi_clk);
> +	devm_clk_put(dcss->dev, dcss->apb_clk);
> +}
> +
> +struct dcss_dev *dcss_dev_create(struct device *dev, bool hdmi_output)
> +{
> +	struct platform_device *pdev = to_platform_device(dev);
> +	int ret;
> +	struct resource *res;
> +	struct dcss_dev *dcss;
> +	const struct dcss_type_data *devtype;
> +
> +	devtype = of_device_get_match_data(dev);
> +	if (!devtype) {
> +		dev_err(dev, "no device match found\n");
> +		return ERR_PTR(-ENODEV);
> +	}
> +
> +	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (!res) {
> +		dev_err(dev, "cannot get memory resource\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	dcss = devm_kzalloc(dev, sizeof(struct dcss_dev), GFP_KERNEL);
> +	if (!dcss)
> +		return ERR_PTR(-ENOMEM);
> +
> +	dcss->dev = dev;
> +	dcss->devtype = devtype;
> +	dcss->hdmi_output = hdmi_output;
> +
> +	ret = dcss_clks_init(dcss);
> +	if (ret) {
> +		dev_err(dev, "clocks initialization failed\n");
> +		goto err;
> +	}
> +
> +	dcss->of_port = of_graph_get_port_by_id(dev->of_node, 0);
> +	if (!dcss->of_port) {
> +		dev_err(dev, "no port@0 node in %s\n", dev->of_node->full_name);
> +		ret = -ENODEV;
> +		goto clks_err;
> +	}
> +
> +	dcss->start_addr = res->start;
> +
> +	ret = dcss_submodules_init(dcss);
> +	if (ret) {
> +		dev_err(dev, "submodules initialization failed\n");
> +		goto clks_err;
> +	}
> +
> +	pm_runtime_enable(dev);
> +
> +	return dcss;
> +
> +clks_err:
> +	dcss_clks_release(dcss);
> +
> +err:
> +	devm_kfree(dcss->dev, dcss);
> +
> +	return ERR_PTR(ret);
> +}
> +
> +void dcss_dev_destroy(struct dcss_dev *dcss)
> +{
> +	pm_runtime_disable(dcss->dev);
> +
> +	dcss_submodules_stop(dcss);
> +
> +	dcss_clks_release(dcss);
> +
> +	devm_kfree(dcss->dev, dcss);
> +}
> +
> +#ifdef CONFIG_PM_SLEEP
> +int dcss_dev_suspend(struct device *dev)
> +{
> +	struct dcss_dev *dcss = dcss_drv_dev_to_dcss(dev);
> +	int ret;
> +
> +	drm_mode_config_helper_suspend(dcss_drv_dev_to_drm(dev));
> +
> +	if (pm_runtime_suspended(dev))
> +		return 0;
> +
> +	ret = dcss_ctxld_suspend(dcss->ctxld);
> +	if (ret)
> +		return ret;
> +
> +	dcss_clocks_disable(dcss);
> +
> +	return 0;
> +}
> +
> +int dcss_dev_resume(struct device *dev)
> +{
> +	struct dcss_dev *dcss = dcss_drv_dev_to_dcss(dev);
> +
> +	if (pm_runtime_suspended(dev)) {
> +		drm_mode_config_helper_resume(dcss_drv_dev_to_drm(dev));
> +		return 0;
> +	}
> +
> +	dcss_clocks_enable(dcss);
> +
> +	dcss_blkctl_cfg(dcss->blkctl);
> +
> +	dcss_ctxld_resume(dcss->ctxld);
> +
> +	drm_mode_config_helper_resume(dcss_drv_dev_to_drm(dev));
> +
> +	return 0;
> +}
> +#endif /* CONFIG_PM_SLEEP */
> +
> +#ifdef CONFIG_PM
> +int dcss_dev_runtime_suspend(struct device *dev)
> +{
> +	struct dcss_dev *dcss = dcss_drv_dev_to_dcss(dev);
> +	int ret;
> +
> +	ret = dcss_ctxld_suspend(dcss->ctxld);
> +	if (ret)
> +		return ret;
> +
> +	dcss_clocks_disable(dcss);
> +
> +	return 0;
> +}
> +
> +int dcss_dev_runtime_resume(struct device *dev)
> +{
> +	struct dcss_dev *dcss = dcss_drv_dev_to_dcss(dev);
> +
> +	dcss_clocks_enable(dcss);
> +
> +	dcss_blkctl_cfg(dcss->blkctl);
> +
> +	dcss_ctxld_resume(dcss->ctxld);
> +
> +	return 0;
> +}
> +#endif /* CONFIG_PM */
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-dev.h b/drivers/gpu/drm/imx/dcss/dcss-dev.h
> new file mode 100644
> index 00000000..15c5de3
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-dev.h
> @@ -0,0 +1,195 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#ifndef __DCSS_PRV_H__
> +#define __DCSS_PRV_H__
> +
> +#include <drm/drm_fourcc.h>
> +#include <linux/io.h>
> +#include <video/videomode.h>
> +
> +#define SET			0x04
> +#define CLR			0x08
> +#define TGL			0x0C
> +
> +#define dcss_writel(v, c)	writel((v), (c))
> +#define dcss_readl(c)		readl(c)
> +#define dcss_set(v, c)		writel((v), (c) + SET)
> +#define dcss_clr(v, c)		writel((v), (c) + CLR)
> +#define dcss_toggle(v, c)	writel((v), (c) + TGL)
> +
> +static inline void dcss_update(u32 v, u32 m, void __iomem *c)
> +{
> +	writel((readl(c) & ~(m)) | (v), (c));
> +}
> +
> +#define DCSS_DBG_REG(reg)	{.name = #reg, .ofs = reg}
> +
> +enum {
> +	DCSS_IMX8MQ = 0,
> +};
> +
> +struct dcss_type_data {
> +	const char *name;
> +	u32 blkctl_ofs;
> +	u32 ctxld_ofs;
> +	u32 rdsrc_ofs;
> +	u32 wrscl_ofs;
> +	u32 dtg_ofs;
> +	u32 scaler_ofs;
> +	u32 ss_ofs;
> +	u32 dpr_ofs;
> +	u32 dtrc_ofs;
> +	u32 dec400d_ofs;
> +	u32 hdr10_ofs;
> +};
> +
> +struct dcss_debug_reg {
> +	char *name;
> +	u32 ofs;
> +};
> +
> +enum dcss_ctxld_ctx_type {
> +	CTX_DB,
> +	CTX_SB_HP, /* high-priority */
> +	CTX_SB_LP, /* low-priority  */
> +};
> +
> +struct dcss_dev {
> +	struct device *dev;
> +	const struct dcss_type_data *devtype;
> +	struct device_node *of_port;
> +
> +	u32 start_addr;
> +
> +	struct dcss_blkctl *blkctl;
> +	struct dcss_ctxld *ctxld;
> +	struct dcss_dpr *dpr;
> +	struct dcss_dtg *dtg;
> +	struct dcss_ss *ss;
> +	struct dcss_hdr10 *hdr10;
> +	struct dcss_scaler *scaler;
> +	struct dcss_dtrc *dtrc;
> +	struct dcss_dec400d *dec400d;
> +	struct dcss_wrscl *wrscl;
> +	struct dcss_rdsrc *rdsrc;
> +
> +	struct clk *apb_clk;
> +	struct clk *axi_clk;
> +	struct clk *pix_clk;
> +	struct clk *rtrm_clk;
> +	struct clk *dtrc_clk;
> +	struct clk *pll_src_clk;
> +	struct clk *pll_phy_ref_clk;
> +
> +	void (*dcss_disable_callback)(void *data);
> +
> +	bool clks_on;
> +	bool hdmi_output;
> +};
> +
> +enum dcss_color_space {
> +	DCSS_COLORSPACE_RGB,
> +	DCSS_COLORSPACE_YUV,
> +	DCSS_COLORSPACE_UNKNOWN,
> +};

This enum isn't used anywhere, please remove it and add it back in the
patch adding support for other color spaces.

> +
> +struct dcss_dev *dcss_drv_dev_to_dcss(struct device *dev);
> +struct drm_device *dcss_drv_dev_to_drm(struct device *dev);
> +struct dcss_dev *dcss_dev_create(struct device *dev, bool mipi_output);
> +void dcss_dev_destroy(struct dcss_dev *dcss);
> +int dcss_dev_runtime_suspend(struct device *dev);
> +int dcss_dev_runtime_resume(struct device *dev);
> +int dcss_dev_suspend(struct device *dev);
> +int dcss_dev_resume(struct device *dev);
> +
> +/* BLKCTL */
> +int dcss_blkctl_init(struct dcss_dev *dcss, unsigned long blkctl_base);
> +void dcss_blkctl_cfg(struct dcss_blkctl *blkctl);
> +void dcss_blkctl_exit(struct dcss_blkctl *blkctl);
> +
> +/* CTXLD */
> +int dcss_ctxld_init(struct dcss_dev *dcss, unsigned long ctxld_base);
> +void dcss_ctxld_exit(struct dcss_ctxld *ctxld);
> +void dcss_ctxld_write(struct dcss_ctxld *ctxld, u32 ctx_id,
> +		      u32 val, u32 reg_idx);
> +int dcss_ctxld_resume(struct dcss_ctxld *dcss_ctxld);
> +int dcss_ctxld_suspend(struct dcss_ctxld *dcss_ctxld);
> +void dcss_ctxld_write_irqsafe(struct dcss_ctxld *ctlxd, u32 ctx_id, u32 val,
> +			      u32 reg_ofs);
> +void dcss_ctxld_kick(struct dcss_ctxld *ctxld);
> +bool dcss_ctxld_is_flushed(struct dcss_ctxld *ctxld);
> +int dcss_ctxld_enable(struct dcss_ctxld *ctxld);
> +void dcss_ctxld_register_dtg_disable_cb(struct dcss_ctxld *ctxld,
> +					void (*cb)(void *),
> +					void *data);
> +
> +/* DPR */
> +enum dcss_tile_type {
> +	TILE_LINEAR = 0,
> +	TILE_GPU_STANDARD,
> +	TILE_GPU_SUPER,
> +	TILE_VPU_YUV420,
> +	TILE_VPU_VP9,
> +};

This is only used in dcss-dpr.c, so doesn't need to be in a global
header.

> +
> +enum dcss_pix_size {
> +	PIX_SIZE_8,
> +	PIX_SIZE_16,
> +	PIX_SIZE_32,
> +};

Unused enum, but should be used?

> +
> +int dcss_dpr_init(struct dcss_dev *dcss, unsigned long dpr_base);
> +void dcss_dpr_exit(struct dcss_dpr *dpr);
> +void dcss_dpr_write_sysctrl(struct dcss_dpr *dpr);
> +void dcss_dpr_set_res(struct dcss_dpr *dpr, int ch_num, u32 xres, u32 yres);
> +void dcss_dpr_addr_set(struct dcss_dpr *dpr, int ch_num, u32 luma_base_addr,
> +		       u32 chroma_base_addr, u16 pitch);
> +void dcss_dpr_enable(struct dcss_dpr *dpr, int ch_num, bool en);
> +void dcss_dpr_format_set(struct dcss_dpr *dpr, int ch_num,
> +			 const struct drm_format_info *format, u64 modifier);
> +void dcss_dpr_set_rotation(struct dcss_dpr *dpr, int ch_num, u32 rotation);
> +
> +/* DTG */
> +int dcss_dtg_init(struct dcss_dev *dcss, unsigned long dtg_base);
> +void dcss_dtg_exit(struct dcss_dtg *dtg);
> +bool dcss_dtg_vblank_irq_valid(struct dcss_dtg *dtg);
> +void dcss_dtg_vblank_irq_enable(struct dcss_dtg *dtg, bool en);
> +void dcss_dtg_vblank_irq_clear(struct dcss_dtg *dtg);
> +void dcss_dtg_sync_set(struct dcss_dtg *dtg, struct videomode *vm);
> +void dcss_dtg_css_set(struct dcss_dtg *dtg);
> +void dcss_dtg_enable(struct dcss_dtg *dtg, bool en,
> +		     struct completion *dis_completion);
> +bool dcss_dtg_is_enabled(struct dcss_dtg *dtg);
> +void dcss_dtg_ctxld_kick_irq_enable(struct dcss_dtg *dtg, bool en);
> +bool dcss_dtg_global_alpha_changed(struct dcss_dtg *dtg, int ch_num, int alpha);
> +void dcss_dtg_plane_alpha_set(struct dcss_dtg *dtg, int ch_num,
> +			      const struct drm_format_info *format, int alpha);
> +void dcss_dtg_plane_pos_set(struct dcss_dtg *dtg, int ch_num,
> +			    int px, int py, int pw, int ph);
> +void dcss_dtg_ch_enable(struct dcss_dtg *dtg, int ch_num, bool en);
> +
> +/* SUBSAM */
> +int dcss_ss_init(struct dcss_dev *dcss, unsigned long subsam_base);
> +void dcss_ss_exit(struct dcss_ss *ss);
> +void dcss_ss_enable(struct dcss_ss *ss);
> +void dcss_ss_disable(struct dcss_ss *ss);
> +void dcss_ss_subsam_set(struct dcss_ss *ss);
> +void dcss_ss_sync_set(struct dcss_ss *ss, struct videomode *vm,
> +		      bool phsync, bool pvsync);
> +
> +/* SCALER */
> +int dcss_scaler_init(struct dcss_dev *dcss, unsigned long scaler_base);
> +void dcss_scaler_exit(struct dcss_scaler *scl);
> +void dcss_scaler_setup(struct dcss_scaler *scl, int ch_num,
> +		       const struct drm_format_info *format,
> +		       int src_xres, int src_yres, int dst_xres, int dst_yres,
> +		       u32 vrefresh_hz);
> +void dcss_scaler_ch_enable(struct dcss_scaler *scl, int ch_num, bool en);
> +int dcss_scaler_get_min_max_ratios(struct dcss_scaler *scl, int ch_num,
> +				   int *min, int *max);
> +void dcss_scaler_write_sclctrl(struct dcss_scaler *scl);
> +
> +#endif /* __DCSS_PRV_H__ */
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-dpr.c b/drivers/gpu/drm/imx/dcss/dcss-dpr.c
> new file mode 100644
> index 00000000..15aa6d8
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-dpr.c
> @@ -0,0 +1,550 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/device.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_DPR_SYSTEM_CTRL0			0x000
> +#define   RUN_EN				BIT(0)
> +#define   SOFT_RESET				BIT(1)
> +#define   REPEAT_EN				BIT(2)
> +#define   SHADOW_LOAD_EN			BIT(3)
> +#define   SW_SHADOW_LOAD_SEL			BIT(4)
> +#define   BCMD2AXI_MSTR_ID_CTRL			BIT(16)
> +#define DCSS_DPR_IRQ_MASK			0x020
> +#define DCSS_DPR_IRQ_MASK_STATUS		0x030
> +#define DCSS_DPR_IRQ_NONMASK_STATUS		0x040
> +#define   IRQ_DPR_CTRL_DONE			BIT(0)
> +#define   IRQ_DPR_RUN				BIT(1)
> +#define   IRQ_DPR_SHADOW_LOADED			BIT(2)
> +#define   IRQ_AXI_READ_ERR			BIT(3)
> +#define   DPR2RTR_YRGB_FIFO_OVFL		BIT(4)
> +#define   DPR2RTR_UV_FIFO_OVFL			BIT(5)
> +#define   DPR2RTR_FIFO_LD_BUF_RDY_YRGB_ERR	BIT(6)
> +#define   DPR2RTR_FIFO_LD_BUF_RDY_UV_ERR	BIT(7)
> +#define DCSS_DPR_MODE_CTRL0			0x050
> +#define   RTR_3BUF_EN				BIT(0)
> +#define   RTR_4LINE_BUF_EN			BIT(1)
> +#define   TILE_TYPE_POS				2
> +#define   TILE_TYPE_MASK			GENMASK(4, 2)
> +#define   YUV_EN				BIT(6)
> +#define   COMP_2PLANE_EN			BIT(7)
> +#define   PIX_SIZE_POS				8
> +#define   PIX_SIZE_MASK				GENMASK(9, 8)
> +#define   PIX_LUMA_UV_SWAP			BIT(10)
> +#define   PIX_UV_SWAP				BIT(11)
> +#define   B_COMP_SEL_POS			12
> +#define   B_COMP_SEL_MASK			GENMASK(13, 12)
> +#define   G_COMP_SEL_POS			14
> +#define   G_COMP_SEL_MASK			GENMASK(15, 14)
> +#define   R_COMP_SEL_POS			16
> +#define   R_COMP_SEL_MASK			GENMASK(17, 16)
> +#define   A_COMP_SEL_POS			18
> +#define   A_COMP_SEL_MASK			GENMASK(19, 18)
> +#define DCSS_DPR_FRAME_CTRL0			0x070
> +#define   HFLIP_EN				BIT(0)
> +#define   VFLIP_EN				BIT(1)
> +#define   ROT_ENC_POS				2
> +#define   ROT_ENC_MASK				GENMASK(3, 2)
> +#define   ROT_FLIP_ORDER_EN			BIT(4)
> +#define   PITCH_POS				16
> +#define   PITCH_MASK				GENMASK(31, 16)
> +#define DCSS_DPR_FRAME_1P_CTRL0			0x090
> +#define DCSS_DPR_FRAME_1P_PIX_X_CTRL		0x0A0
> +#define DCSS_DPR_FRAME_1P_PIX_Y_CTRL		0x0B0
> +#define DCSS_DPR_FRAME_1P_BASE_ADDR		0x0C0
> +#define DCSS_DPR_FRAME_2P_CTRL0			0x0E0
> +#define DCSS_DPR_FRAME_2P_PIX_X_CTRL		0x0F0
> +#define DCSS_DPR_FRAME_2P_PIX_Y_CTRL		0x100
> +#define DCSS_DPR_FRAME_2P_BASE_ADDR		0x110
> +#define DCSS_DPR_STATUS_CTRL0			0x130
> +#define   STATUS_MUX_SEL_MASK			GENMASK(2, 0)
> +#define   STATUS_SRC_SEL_POS			16
> +#define   STATUS_SRC_SEL_MASK			GENMASK(18, 16)
> +#define DCSS_DPR_STATUS_CTRL1			0x140
> +#define DCSS_DPR_RTRAM_CTRL0			0x200
> +#define   NUM_ROWS_ACTIVE			BIT(0)
> +#define   THRES_HIGH_POS			1
> +#define   THRES_HIGH_MASK			GENMASK(3, 1)
> +#define   THRES_LOW_POS				4
> +#define   THRES_LOW_MASK			GENMASK(6, 4)
> +#define   ABORT_SEL				BIT(7)
> +
> +struct dcss_dpr_ch {
> +	struct dcss_dpr *dpr;
> +	void __iomem *base_reg;
> +	u32 base_ofs;
> +
> +	struct drm_format_info format;
> +	enum dcss_pix_size pix_size;
> +	enum dcss_tile_type tile;
> +	bool rtram_4line_en;
> +	bool rtram_3buf_en;
> +
> +	u32 frame_ctrl;
> +	u32 mode_ctrl;
> +	u32 sys_ctrl;
> +	u32 rtram_ctrl;
> +
> +	bool sys_ctrl_chgd;
> +
> +	u32 pitch;

Written, but not read anywhere.

> +
> +	int ch_num;
> +	int irq;
> +};
> +
> +struct dcss_dpr {
> +	struct device *dev;
> +	struct dcss_ctxld *ctxld;
> +	u32  ctx_id;
> +
> +	struct dcss_dpr_ch ch[3];
> +};
> +
> +static void dcss_dpr_write(struct dcss_dpr_ch *ch, u32 val, u32 ofs)
> +{
> +	struct dcss_dpr *dpr = ch->dpr;
> +
> +	dcss_ctxld_write(dpr->ctxld, dpr->ctx_id, val, ch->base_ofs + ofs);
> +}
> +
> +static int dcss_dpr_ch_init_all(struct dcss_dpr *dpr, unsigned long dpr_base)
> +{
> +	struct dcss_dpr_ch *ch;
> +	int i;
> +
> +	for (i = 0; i < 3; i++) {
> +		ch = &dpr->ch[i];
> +
> +		ch->base_ofs = dpr_base + i * 0x1000;
> +
> +		ch->base_reg = devm_ioremap(dpr->dev, ch->base_ofs, SZ_4K);
> +		if (!ch->base_reg) {
> +			dev_err(dpr->dev, "dpr: unable to remap ch %d base\n",
> +				i);
> +			return -ENOMEM;
> +		}
> +
> +		ch->dpr = dpr;
> +		ch->ch_num = i;
> +
> +		dcss_writel(0xff, ch->base_reg + DCSS_DPR_IRQ_MASK);
> +	}
> +
> +	return 0;
> +}
> +
> +int dcss_dpr_init(struct dcss_dev *dcss, unsigned long dpr_base)
> +{
> +	struct dcss_dpr *dpr;
> +
> +	dpr = devm_kzalloc(dcss->dev, sizeof(struct dcss_dpr), GFP_KERNEL);
> +	if (!dpr)
> +		return -ENOMEM;
> +
> +	dcss->dpr = dpr;
> +	dpr->dev = dcss->dev;
> +	dpr->ctxld = dcss->ctxld;
> +	dpr->ctx_id = CTX_SB_HP;
> +
> +	if (dcss_dpr_ch_init_all(dpr, dpr_base)) {
> +		int i;
> +
> +		for (i = 0; i < 3; i++) {
> +			if (dpr->ch[i].base_reg)
> +				devm_iounmap(dpr->dev, dpr->ch[i].base_reg);
> +		}
> +
> +		devm_kfree(dpr->dev, dpr);
> +
> +		return -ENOMEM;
> +	}
> +
> +	return 0;
> +}
> +
> +void dcss_dpr_exit(struct dcss_dpr *dpr)
> +{
> +	int ch_no;
> +
> +	/* stop DPR on all channels */
> +	for (ch_no = 0; ch_no < 3; ch_no++) {
> +		struct dcss_dpr_ch *ch = &dpr->ch[ch_no];
> +
> +		dcss_writel(0, ch->base_reg + DCSS_DPR_SYSTEM_CTRL0);
> +
> +		if (ch->base_reg)
> +			devm_iounmap(dpr->dev, ch->base_reg);
> +	}
> +
> +	devm_kfree(dpr->dev, dpr);
> +}
> +
> +static u32 dcss_dpr_x_pix_wide_adjust(struct dcss_dpr_ch *ch, u32 pix_wide,
> +				      u32 pix_format)
> +{
> +	u8 pix_in_64byte_map[3][5] = {
> +		/* LIN, GPU_STD, GPU_SUP, VPU_YUV420, VPU_VP9 */
> +		{   64,       8,       8,          8,     16}, /* PIX_SIZE_8  */
> +		{   32,       8,       8,          8,      8}, /* PIX_SIZE_16 */
> +		{   16,       4,       4,          8,      8}, /* PIX_SIZE_32 */
> +	};
> +	u32 offset;
> +	u32 div_64byte_mod, pix_in_64byte;
> +
> +	pix_in_64byte = pix_in_64byte_map[ch->pix_size][ch->tile];
> +
> +	div_64byte_mod = pix_wide % pix_in_64byte;
> +	offset = (div_64byte_mod == 0) ? 0 : (pix_in_64byte - div_64byte_mod);
> +
> +	return pix_wide + offset;
> +}
> +
> +static u32 dcss_dpr_y_pix_high_adjust(struct dcss_dpr_ch *ch, u32 pix_high,
> +				      u32 pix_format)
> +{
> +	u8 num_rows_buf = ch->rtram_4line_en ? 4 : 8;
> +	u32 offset, pix_y_mod;
> +
> +	pix_y_mod = pix_high % num_rows_buf;
> +	offset = pix_y_mod ? (num_rows_buf - pix_y_mod) : 0;
> +
> +	return pix_high + offset;
> +}
> +
> +void dcss_dpr_set_res(struct dcss_dpr *dpr, int ch_num, u32 xres, u32 yres)
> +{
> +	struct dcss_dpr_ch *ch = &dpr->ch[ch_num];
> +	u32 pix_format = ch->format.format;
> +	u32 gap = DCSS_DPR_FRAME_2P_BASE_ADDR - DCSS_DPR_FRAME_1P_BASE_ADDR;
> +	int plane, max_planes = 1;
> +	u32 pix_x_wide, pix_y_high;
> +
> +	if (pix_format == DRM_FORMAT_NV12 ||
> +	    pix_format == DRM_FORMAT_NV21)
> +		max_planes = 2;
> +
> +	for (plane = 0; plane < max_planes; plane++) {
> +		yres = plane == 1 ? yres >> 1 : yres;
> +
> +		pix_x_wide = dcss_dpr_x_pix_wide_adjust(ch, xres, pix_format);
> +		pix_y_high = dcss_dpr_y_pix_high_adjust(ch, yres, pix_format);
> +
> +		if (plane == 0)
> +			ch->pitch = pix_x_wide;
> +
> +		dcss_dpr_write(ch, pix_x_wide,
> +			       DCSS_DPR_FRAME_1P_PIX_X_CTRL + plane * gap);
> +		dcss_dpr_write(ch, pix_y_high,
> +			       DCSS_DPR_FRAME_1P_PIX_Y_CTRL + plane * gap);
> +
> +		dcss_dpr_write(ch, 2, DCSS_DPR_FRAME_1P_CTRL0 + plane * gap);
> +	}
> +}
> +
> +void dcss_dpr_addr_set(struct dcss_dpr *dpr, int ch_num, u32 luma_base_addr,
> +		       u32 chroma_base_addr, u16 pitch)
> +{
> +	struct dcss_dpr_ch *ch = &dpr->ch[ch_num];
> +
> +	dcss_dpr_write(ch, luma_base_addr, DCSS_DPR_FRAME_1P_BASE_ADDR);
> +
> +	dcss_dpr_write(ch, chroma_base_addr, DCSS_DPR_FRAME_2P_BASE_ADDR);
> +
> +	ch->frame_ctrl &= ~PITCH_MASK;
> +	ch->frame_ctrl |= (((u32)pitch << PITCH_POS) & PITCH_MASK);
> +}
> +
> +static void dcss_dpr_argb_comp_sel(struct dcss_dpr_ch *ch, int a_sel, int r_sel,
> +				   int g_sel, int b_sel)
> +{
> +	u32 sel;
> +
> +	sel = ((a_sel << A_COMP_SEL_POS) & A_COMP_SEL_MASK) |
> +	      ((r_sel << R_COMP_SEL_POS) & R_COMP_SEL_MASK) |
> +	      ((g_sel << G_COMP_SEL_POS) & G_COMP_SEL_MASK) |
> +	      ((b_sel << B_COMP_SEL_POS) & B_COMP_SEL_MASK);
> +
> +	ch->mode_ctrl &= ~(A_COMP_SEL_MASK | R_COMP_SEL_MASK |
> +			   G_COMP_SEL_MASK | B_COMP_SEL_MASK);
> +	ch->mode_ctrl |= sel;
> +}
> +
> +static void dcss_dpr_pix_size_set(struct dcss_dpr_ch *ch,
> +				  const struct drm_format_info *format)
> +{
> +	u32 val;
> +
> +	switch (format->format) {
> +	case DRM_FORMAT_NV12:
> +	case DRM_FORMAT_NV21:
> +		val = 0;
> +		break;
> +
> +	case DRM_FORMAT_UYVY:
> +	case DRM_FORMAT_VYUY:
> +	case DRM_FORMAT_YUYV:
> +	case DRM_FORMAT_YVYU:
> +		val = 1;
> +		break;
> +
> +	default:
> +		val = 2;
> +		break;
> +	}
> +
> +	ch->pix_size = val;
> +
> +	ch->mode_ctrl &= ~PIX_SIZE_MASK;
> +	ch->mode_ctrl |= ((val << PIX_SIZE_POS) & PIX_SIZE_MASK);
> +}
> +
> +static void dcss_dpr_uv_swap(struct dcss_dpr_ch *ch, bool swap)
> +{
> +	ch->mode_ctrl &= ~PIX_UV_SWAP;
> +	ch->mode_ctrl |= (swap ? PIX_UV_SWAP : 0);
> +}
> +
> +static void dcss_dpr_y_uv_swap(struct dcss_dpr_ch *ch, bool swap)
> +{
> +	ch->mode_ctrl &= ~PIX_LUMA_UV_SWAP;
> +	ch->mode_ctrl |= (swap ? PIX_LUMA_UV_SWAP : 0);
> +}
> +
> +static void dcss_dpr_2plane_en(struct dcss_dpr_ch *ch, bool en)
> +{
> +	ch->mode_ctrl &= ~COMP_2PLANE_EN;
> +	ch->mode_ctrl |= (en ? COMP_2PLANE_EN : 0);
> +}
> +
> +static void dcss_dpr_yuv_en(struct dcss_dpr_ch *ch, bool en)
> +{
> +	ch->mode_ctrl &= ~YUV_EN;
> +	ch->mode_ctrl |= (en ? YUV_EN : 0);
> +}
> +
> +void dcss_dpr_enable(struct dcss_dpr *dpr, int ch_num, bool en)
> +{
> +	struct dcss_dpr_ch *ch = &dpr->ch[ch_num];
> +	u32 sys_ctrl;
> +
> +	sys_ctrl = (en ? REPEAT_EN | RUN_EN : 0);
> +
> +	if (en) {
> +		dcss_dpr_write(ch, ch->mode_ctrl, DCSS_DPR_MODE_CTRL0);
> +		dcss_dpr_write(ch, ch->frame_ctrl, DCSS_DPR_FRAME_CTRL0);
> +		dcss_dpr_write(ch, ch->rtram_ctrl, DCSS_DPR_RTRAM_CTRL0);
> +	}
> +
> +	if (ch->sys_ctrl != sys_ctrl)
> +		ch->sys_ctrl_chgd = true;
> +
> +	ch->sys_ctrl = sys_ctrl;
> +}
> +
> +struct rgb_comp_sel {
> +	u32 drm_format;
> +	int a_sel;
> +	int r_sel;
> +	int g_sel;
> +	int b_sel;
> +};
> +
> +static struct rgb_comp_sel comp_sel_map[] = {
> +	{DRM_FORMAT_ARGB8888, 3, 2, 1, 0},
> +	{DRM_FORMAT_XRGB8888, 3, 2, 1, 0},
> +	{DRM_FORMAT_ABGR8888, 3, 0, 1, 2},
> +	{DRM_FORMAT_XBGR8888, 3, 0, 1, 2},
> +	{DRM_FORMAT_RGBA8888, 0, 3, 2, 1},
> +	{DRM_FORMAT_RGBX8888, 0, 3, 2, 1},
> +	{DRM_FORMAT_BGRA8888, 0, 1, 2, 3},
> +	{DRM_FORMAT_BGRX8888, 0, 1, 2, 3},
> +};
> +
> +static int to_comp_sel(u32 pix_fmt, int *a_sel, int *r_sel, int *g_sel,
> +		       int *b_sel)
> +{
> +	int i;
> +
> +	for (i = 0; i < ARRAY_SIZE(comp_sel_map); i++) {
> +		if (comp_sel_map[i].drm_format == pix_fmt) {
> +			*a_sel = comp_sel_map[i].a_sel;
> +			*r_sel = comp_sel_map[i].r_sel;
> +			*g_sel = comp_sel_map[i].g_sel;
> +			*b_sel = comp_sel_map[i].b_sel;
> +
> +			return 0;
> +		}
> +	}
> +
> +	return -1;
> +}
> +
> +static void dcss_dpr_rtram_set(struct dcss_dpr_ch *ch, u32 pix_format)
> +{
> +	u32 val, mask;
> +
> +	switch (pix_format) {
> +	case DRM_FORMAT_NV21:
> +	case DRM_FORMAT_NV12:
> +		ch->rtram_3buf_en = true;
> +		ch->rtram_4line_en = false;
> +		break;
> +
> +	default:
> +		ch->rtram_3buf_en = true;
> +		ch->rtram_4line_en = true;
> +		break;
> +	}
> +
> +	val = (ch->rtram_4line_en ? RTR_4LINE_BUF_EN : 0);
> +	val |= (ch->rtram_3buf_en ? RTR_3BUF_EN : 0);
> +	mask = RTR_4LINE_BUF_EN | RTR_3BUF_EN;
> +
> +	ch->mode_ctrl &= ~mask;
> +	ch->mode_ctrl |= (val & mask);
> +
> +	val = (ch->rtram_4line_en ? 0 : NUM_ROWS_ACTIVE);
> +	val |= (3 << THRES_LOW_POS) & THRES_LOW_MASK;
> +	val |= (4 << THRES_HIGH_POS) & THRES_HIGH_MASK;
> +	mask = THRES_LOW_MASK | THRES_HIGH_MASK | NUM_ROWS_ACTIVE;
> +
> +	ch->rtram_ctrl &= ~mask;
> +	ch->rtram_ctrl |= (val & mask);
> +}
> +
> +static void dcss_dpr_setup_components(struct dcss_dpr_ch *ch,
> +				      const struct drm_format_info *format)
> +{
> +	int a_sel, r_sel, g_sel, b_sel;
> +	bool uv_swap, y_uv_swap;
> +
> +	switch (format->format) {
> +	case DRM_FORMAT_YVYU:
> +		uv_swap = true;
> +		y_uv_swap = true;
> +		break;
> +
> +	case DRM_FORMAT_VYUY:
> +	case DRM_FORMAT_NV21:
> +		uv_swap = true;
> +		y_uv_swap = false;
> +		break;
> +
> +	case DRM_FORMAT_YUYV:
> +		uv_swap = false;
> +		y_uv_swap = true;
> +		break;
> +
> +	default:
> +		uv_swap = false;
> +		y_uv_swap = false;
> +		break;
> +	}
> +
> +	dcss_dpr_uv_swap(ch, uv_swap);
> +
> +	dcss_dpr_y_uv_swap(ch, y_uv_swap);
> +
> +	if (!format->is_yuv) {
> +		if (!to_comp_sel(format->format, &a_sel, &r_sel,
> +				 &g_sel, &b_sel)) {
> +			dcss_dpr_argb_comp_sel(ch, a_sel, r_sel, g_sel, b_sel);
> +		} else {
> +			dcss_dpr_argb_comp_sel(ch, 3, 2, 1, 0);
> +		}
> +	} else {
> +		dcss_dpr_argb_comp_sel(ch, 0, 0, 0, 0);
> +	}
> +}
> +
> +static void dcss_dpr_tile_set(struct dcss_dpr_ch *ch, uint64_t modifier)
> +{
> +	switch (ch->ch_num) {
> +	case 0:
> +		switch (modifier) {
> +		case DRM_FORMAT_MOD_LINEAR:
> +			ch->tile = TILE_LINEAR;
> +			break;
> +		case DRM_FORMAT_MOD_VIVANTE_TILED:
> +			ch->tile = TILE_GPU_STANDARD;
> +			break;
> +		case DRM_FORMAT_MOD_VIVANTE_SUPER_TILED:
> +			ch->tile = TILE_GPU_SUPER;
> +			break;
> +		default:
> +			WARN_ON(1);
> +			break;
> +		}
> +		break;
> +	case 1:
> +	case 2:
> +		ch->tile = TILE_LINEAR;
> +		break;
> +	default:
> +		WARN_ON(1);
> +		return;
> +	}
> +
> +	ch->mode_ctrl &= ~TILE_TYPE_MASK;
> +	ch->mode_ctrl |= ((ch->tile << TILE_TYPE_POS) & TILE_TYPE_MASK);
> +}
> +
> +void dcss_dpr_format_set(struct dcss_dpr *dpr, int ch_num,
> +			 const struct drm_format_info *format, u64 modifier)
> +{
> +	struct dcss_dpr_ch *ch = &dpr->ch[ch_num];
> +
> +	ch->format = *format;
> +
> +	dcss_dpr_yuv_en(ch, format->is_yuv);
> +
> +	dcss_dpr_pix_size_set(ch, format);
> +
> +	dcss_dpr_setup_components(ch, format);
> +
> +	dcss_dpr_2plane_en(ch, format->num_planes == 2);
> +
> +	dcss_dpr_rtram_set(ch, format->format);
> +
> +	dcss_dpr_tile_set(ch, modifier);
> +}
> +
> +/* This function will be called from interrupt context. */
> +void dcss_dpr_write_sysctrl(struct dcss_dpr *dpr)
> +{
> +	int chnum;
> +
> +	for (chnum = 0; chnum < 3; chnum++) {
> +		struct dcss_dpr_ch *ch = &dpr->ch[chnum];
> +
> +		if (ch->sys_ctrl_chgd) {
> +			dcss_ctxld_write_irqsafe(dpr->ctxld, dpr->ctx_id,
> +						 ch->sys_ctrl,
> +						 ch->base_ofs +
> +						 DCSS_DPR_SYSTEM_CTRL0);
> +			ch->sys_ctrl_chgd = false;
> +		}
> +	}
> +}
> +
> +void dcss_dpr_set_rotation(struct dcss_dpr *dpr, int ch_num, u32 rotation)
> +{
> +	struct dcss_dpr_ch *ch = &dpr->ch[ch_num];
> +
> +	ch->frame_ctrl &= ~(HFLIP_EN | VFLIP_EN | ROT_ENC_MASK);
> +
> +	ch->frame_ctrl |= rotation & DRM_MODE_REFLECT_X ? HFLIP_EN : 0;
> +	ch->frame_ctrl |= rotation & DRM_MODE_REFLECT_Y ? VFLIP_EN : 0;
> +
> +	if (rotation & DRM_MODE_ROTATE_90)
> +		ch->frame_ctrl |= 1 << ROT_ENC_POS;
> +	else if (rotation & DRM_MODE_ROTATE_180)
> +		ch->frame_ctrl |= 2 << ROT_ENC_POS;
> +	else if (rotation & DRM_MODE_ROTATE_270)
> +		ch->frame_ctrl |= 3 << ROT_ENC_POS;
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-drv.c b/drivers/gpu/drm/imx/dcss/dcss-drv.c
> new file mode 100644
> index 00000000..109fb38
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-drv.c
> @@ -0,0 +1,181 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/module.h>
> +#include <linux/kernel.h>
> +#include <linux/platform_device.h>
> +#include <linux/component.h>
> +#include <drm/drm_of.h>
> +
> +#include "dcss-dev.h"
> +#include "dcss-kms.h"
> +
> +struct dcss_drv {
> +	struct dcss_dev *dcss;
> +	struct dcss_kms_dev *kms;
> +
> +	bool is_componentized;
> +};
> +
> +struct dcss_dev *dcss_drv_dev_to_dcss(struct device *dev)
> +{
> +	struct dcss_drv *mdrv = dev_get_drvdata(dev);
> +
> +	return mdrv ? mdrv->dcss : NULL;
> +}
> +
> +struct drm_device *dcss_drv_dev_to_drm(struct device *dev)
> +{
> +	struct dcss_drv *mdrv = dev_get_drvdata(dev);
> +
> +	return mdrv ? &mdrv->kms->base : NULL;
> +}
> +
> +static int dcss_drv_init(struct device *dev, bool componentized)
> +{
> +	struct dcss_drv *mdrv;
> +	int err = 0;
> +
> +	mdrv = devm_kzalloc(dev, sizeof(*mdrv), GFP_KERNEL);
> +	if (!mdrv)
> +		return -ENOMEM;
> +
> +	mdrv->is_componentized = componentized;
> +
> +	mdrv->dcss = dcss_dev_create(dev, componentized);
> +	if (IS_ERR(mdrv->dcss)) {
> +		err = PTR_ERR(mdrv->dcss);
> +		goto err;
> +	}
> +
> +	dev_set_drvdata(dev, mdrv);
> +
> +	mdrv->kms = dcss_kms_attach(mdrv->dcss, componentized);
> +	if (IS_ERR(mdrv->kms)) {
> +		err = PTR_ERR(mdrv->kms);
> +		goto dcss_shutoff;
> +	}
> +
> +	return 0;
> +
> +dcss_shutoff:
> +	dcss_dev_destroy(mdrv->dcss);
> +
> +	dev_set_drvdata(dev, NULL);
> +
> +err:
> +	devm_kfree(dev, mdrv);
> +	return err;
> +}
> +
> +static void dcss_drv_deinit(struct device *dev, bool componentized)
> +{
> +	struct dcss_drv *mdrv = dev_get_drvdata(dev);
> +
> +	if (!mdrv)
> +		return;
> +
> +	dcss_kms_detach(mdrv->kms, componentized);
> +	dcss_dev_destroy(mdrv->dcss);
> +
> +	dev_set_drvdata(dev, NULL);
> +}
> +
> +static int dcss_drv_bind(struct device *dev)
> +{
> +	return dcss_drv_init(dev, true);
> +}
> +
> +static void dcss_drv_unbind(struct device *dev)
> +{
> +	return dcss_drv_deinit(dev, true);
> +}
> +
> +static const struct component_master_ops dcss_master_ops = {
> +	.bind	= dcss_drv_bind,
> +	.unbind	= dcss_drv_unbind,
> +};
> +
> +static int compare_of(struct device *dev, void *data)
> +{
> +	return dev->of_node == data;
> +}
> +
> +static int dcss_drv_platform_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct component_match *match = NULL;
> +	struct device_node *remote;
> +
> +	if (!dev->of_node)
> +		return -ENODEV;
> +
> +	remote = of_graph_get_remote_node(dev->of_node, 0, 0);
> +	if (!remote)
> +		return -ENODEV;
> +
> +	if (of_device_is_compatible(remote, "fsl,imx8mq-nwl-dsi")) {
> +		of_node_put(remote);
> +		return dcss_drv_init(dev, false);
> +	}
> +
> +	drm_of_component_match_add(dev, &match, compare_of, remote);
> +	of_node_put(remote);
> +
> +	return component_master_add_with_match(dev, &dcss_master_ops, match);
> +}
> +
> +static int dcss_drv_platform_remove(struct platform_device *pdev)
> +{
> +	struct dcss_drv *mdrv = dev_get_drvdata(&pdev->dev);
> +
> +	if (mdrv->is_componentized)
> +		component_master_del(&pdev->dev, &dcss_master_ops);
> +	else
> +		dcss_drv_deinit(&pdev->dev, false);
> +
> +	return 0;
> +}
> +
> +static struct dcss_type_data dcss_types[] = {
> +	[DCSS_IMX8MQ] = {
> +		.name = "DCSS_IMX8MQ",
> +		.blkctl_ofs = 0x2F000,
> +		.ctxld_ofs = 0x23000,
> +		.dtg_ofs = 0x20000,
> +		.scaler_ofs = 0x1C000,
> +		.ss_ofs = 0x1B000,
> +		.dpr_ofs = 0x18000,
> +	},
> +};
> +
> +static const struct of_device_id dcss_of_match[] = {
> +	{ .compatible = "nxp,imx8mq-dcss", .data = &dcss_types[DCSS_IMX8MQ], },
> +	{},
> +};
> +
> +MODULE_DEVICE_TABLE(of, dcss_of_match);
> +
> +static const struct dev_pm_ops dcss_dev_pm = {
> +	SET_SYSTEM_SLEEP_PM_OPS(dcss_dev_suspend, dcss_dev_resume)
> +	SET_RUNTIME_PM_OPS(dcss_dev_runtime_suspend,
> +			   dcss_dev_runtime_resume, NULL)
> +};
> +
> +static struct platform_driver dcss_platform_driver = {
> +	.probe	= dcss_drv_platform_probe,
> +	.remove	= dcss_drv_platform_remove,
> +	.driver	= {
> +		.name = "imx-dcss",
> +		.of_match_table	= dcss_of_match,
> +		.pm = &dcss_dev_pm,
> +	},
> +};
> +
> +module_platform_driver(dcss_platform_driver);
> +
> +MODULE_AUTHOR("Laurentiu Palcu <laurentiu.palcu@nxp.com>");
> +MODULE_DESCRIPTION("DCSS driver for i.MX8MQ");
> +MODULE_LICENSE("GPL v2");
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-dtg.c b/drivers/gpu/drm/imx/dcss/dcss-dtg.c
> new file mode 100644
> index 00000000..6ab4c67
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-dtg.c
> @@ -0,0 +1,442 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/delay.h>
> +#include <linux/interrupt.h>
> +#include <linux/of.h>
> +#include <linux/platform_device.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_DTG_TC_CONTROL_STATUS			0x00
> +#define   CH3_EN					BIT(0)
> +#define   CH2_EN					BIT(1)
> +#define   CH1_EN					BIT(2)
> +#define   OVL_DATA_MODE					BIT(3)
> +#define   BLENDER_VIDEO_ALPHA_SEL			BIT(7)
> +#define   DTG_START					BIT(8)
> +#define   DBY_MODE_EN					BIT(9)
> +#define   CH1_ALPHA_SEL					BIT(10)
> +#define   CSS_PIX_COMP_SWAP_POS				12
> +#define   CSS_PIX_COMP_SWAP_MASK			GENMASK(14, 12)
> +#define   DEFAULT_FG_ALPHA_POS				24
> +#define   DEFAULT_FG_ALPHA_MASK				GENMASK(31, 24)
> +#define DCSS_DTG_TC_DTG					0x04
> +#define DCSS_DTG_TC_DISP_TOP				0x08
> +#define DCSS_DTG_TC_DISP_BOT				0x0C
> +#define DCSS_DTG_TC_CH1_TOP				0x10
> +#define DCSS_DTG_TC_CH1_BOT				0x14
> +#define DCSS_DTG_TC_CH2_TOP				0x18
> +#define DCSS_DTG_TC_CH2_BOT				0x1C
> +#define DCSS_DTG_TC_CH3_TOP				0x20
> +#define DCSS_DTG_TC_CH3_BOT				0x24
> +#define   TC_X_POS					0
> +#define   TC_X_MASK					GENMASK(12, 0)
> +#define   TC_Y_POS					16
> +#define   TC_Y_MASK					GENMASK(28, 16)
> +#define DCSS_DTG_TC_CTXLD				0x28
> +#define   TC_CTXLD_DB_Y_POS				0
> +#define   TC_CTXLD_DB_Y_MASK				GENMASK(12, 0)
> +#define   TC_CTXLD_SB_Y_POS				16
> +#define   TC_CTXLD_SB_Y_MASK				GENMASK(28, 16)
> +#define DCSS_DTG_TC_CH1_BKRND				0x2C
> +#define DCSS_DTG_TC_CH2_BKRND				0x30
> +#define   BKRND_R_Y_COMP_POS				20
> +#define   BKRND_R_Y_COMP_MASK				GENMASK(29, 20)
> +#define   BKRND_G_U_COMP_POS				10
> +#define   BKRND_G_U_COMP_MASK				GENMASK(19, 10)
> +#define   BKRND_B_V_COMP_POS				0
> +#define   BKRND_B_V_COMP_MASK				GENMASK(9, 0)
> +#define DCSS_DTG_BLENDER_DBY_RANGEINV			0x38
> +#define DCSS_DTG_BLENDER_DBY_RANGEMIN			0x3C
> +#define DCSS_DTG_BLENDER_DBY_BDP			0x40
> +#define DCSS_DTG_BLENDER_BKRND_I			0x44
> +#define DCSS_DTG_BLENDER_BKRND_P			0x48
> +#define DCSS_DTG_BLENDER_BKRND_T			0x4C
> +#define DCSS_DTG_LINE0_INT				0x50
> +#define DCSS_DTG_LINE1_INT				0x54
> +#define DCSS_DTG_BG_ALPHA_DEFAULT			0x58
> +#define DCSS_DTG_INT_STATUS				0x5C
> +#define DCSS_DTG_INT_CONTROL				0x60
> +#define DCSS_DTG_TC_CH3_BKRND				0x64
> +#define DCSS_DTG_INT_MASK				0x68
> +#define   LINE0_IRQ					BIT(0)
> +#define   LINE1_IRQ					BIT(1)
> +#define   LINE2_IRQ					BIT(2)
> +#define   LINE3_IRQ					BIT(3)
> +#define DCSS_DTG_LINE2_INT				0x6C
> +#define DCSS_DTG_LINE3_INT				0x70
> +#define DCSS_DTG_DBY_OL					0x74
> +#define DCSS_DTG_DBY_BL					0x78
> +#define DCSS_DTG_DBY_EL					0x7C
> +
> +struct dcss_dtg {
> +	struct device *dev;
> +	struct dcss_ctxld *ctxld;
> +	void __iomem *base_reg;
> +	u32 base_ofs;
> +
> +	u32 ctx_id;
> +
> +	bool in_use;
> +
> +	u32 dis_ulc_x;
> +	u32 dis_ulc_y;
> +
> +	u32 control_status;
> +	u32 alpha;
> +
> +	int ctxld_kick_irq;
> +	bool ctxld_kick_irq_en;
> +
> +	struct clk *pix_clk;
> +	struct clk *pll_src_clk;
> +	struct clk *pll_phy_ref_clk;

This is already present in the struct dcss_dev, better just keep a
parent pointer around, instead of duplicating this state here?

> +
> +	/*
> +	 * This will be passed on by DRM CRTC so that we can signal when DTG has
> +	 * been successfully stopped. Otherwise, any modesetting while DTG is
> +	 * still ON may result in unpredictable behavior.
> +	 */
> +	struct completion *dis_completion;
> +};
> +
> +static void dcss_dtg_write(struct dcss_dtg *dtg, u32 val, u32 ofs)
> +{
> +	if (!dtg->in_use)
> +		dcss_writel(val, dtg->base_reg + ofs);
> +
> +	dcss_ctxld_write(dtg->ctxld, dtg->ctx_id, val, dtg->base_ofs + ofs);
> +}
> +
> +static irqreturn_t dcss_dtg_irq_handler(int irq, void *data)
> +{
> +	struct dcss_dtg *dtg = data;
> +	u32 status;
> +
> +	status = dcss_readl(dtg->base_reg + DCSS_DTG_INT_STATUS);
> +
> +	if (!(status & LINE0_IRQ))
> +		return IRQ_HANDLED;

IRQ_NONE.

> +
> +	dcss_ctxld_kick(dtg->ctxld);
> +
> +	dcss_writel(status & LINE0_IRQ, dtg->base_reg + DCSS_DTG_INT_CONTROL);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static int dcss_dtg_irq_config(struct dcss_dtg *dtg,
> +			       struct platform_device *pdev)
> +{
> +	int ret;
> +
> +	dtg->ctxld_kick_irq = platform_get_irq_byname(pdev, "ctxld_kick");
> +	if (dtg->ctxld_kick_irq < 0) {
> +		dev_err(dtg->dev, "dtg: can't get line2 irq number\n");
> +		return dtg->ctxld_kick_irq;
> +	}
> +
> +	dcss_update(0, LINE0_IRQ | LINE1_IRQ,
> +		    dtg->base_reg + DCSS_DTG_INT_MASK);
> +
> +	ret = devm_request_irq(dtg->dev, dtg->ctxld_kick_irq,
> +			       dcss_dtg_irq_handler,
> +			       IRQF_TRIGGER_HIGH,

Remove trigger flag.

> +			       "dcss_ctxld_kick", dtg);
> +	if (ret) {
> +		dev_err(dtg->dev, "dtg: irq request failed.\n");
> +		return ret;
> +	}
> +
> +	disable_irq(dtg->ctxld_kick_irq);
> +
> +	dtg->ctxld_kick_irq_en = false;
> +
> +	return 0;
> +}
> +
> +int dcss_dtg_init(struct dcss_dev *dcss, unsigned long dtg_base)
> +{
> +	int ret = 0;
> +	struct dcss_dtg *dtg;
> +
> +	dtg = devm_kzalloc(dcss->dev, sizeof(*dtg), GFP_KERNEL);
> +	if (!dtg)
> +		return -ENOMEM;
> +
> +	dcss->dtg = dtg;
> +	dtg->dev = dcss->dev;
> +	dtg->ctxld = dcss->ctxld;
> +
> +	dtg->base_reg = devm_ioremap(dcss->dev, dtg_base, SZ_4K);
> +	if (!dtg->base_reg) {
> +		dev_err(dcss->dev, "dtg: unable to remap dtg base\n");
> +		ret = -ENOMEM;
> +		goto err_ioremap;
> +	}
> +
> +	dtg->base_ofs = dtg_base;
> +	dtg->ctx_id = CTX_DB;
> +
> +	dtg->pix_clk = dcss->pix_clk;
> +	dtg->pll_src_clk = dcss->pll_src_clk;
> +	dtg->pll_phy_ref_clk = dcss->pll_phy_ref_clk;
> +
> +	dtg->alpha = 255;
> +
> +	dtg->control_status |= OVL_DATA_MODE | BLENDER_VIDEO_ALPHA_SEL |
> +		((dtg->alpha << DEFAULT_FG_ALPHA_POS) & DEFAULT_FG_ALPHA_MASK);
> +
> +	ret = dcss_dtg_irq_config(dtg, to_platform_device(dcss->dev));
> +	if (ret)
> +		goto err_irq;
> +
> +	return 0;
> +
> +err_irq:
> +	devm_iounmap(dtg->dev, dtg->base_reg);
> +
> +err_ioremap:
> +	devm_kfree(dtg->dev, dtg);
> +
> +	return ret;
> +}
> +
> +void dcss_dtg_exit(struct dcss_dtg *dtg)
> +{
> +	/* stop DTG */
> +	dcss_writel(DTG_START, dtg->base_reg + DCSS_DTG_TC_CONTROL_STATUS);
> +
> +	devm_free_irq(dtg->dev, dtg->ctxld_kick_irq, dtg);
> +
> +	if (dtg->base_reg)
> +		devm_iounmap(dtg->dev, dtg->base_reg);
> +
> +	devm_kfree(dtg->dev, dtg);
> +}
> +
> +void dcss_dtg_sync_set(struct dcss_dtg *dtg, struct videomode *vm)
> +{
> +	u16 dtg_lrc_x, dtg_lrc_y;
> +	u16 dis_ulc_x, dis_ulc_y;
> +	u16 dis_lrc_x, dis_lrc_y;
> +	u32 sb_ctxld_trig, db_ctxld_trig;
> +	u32 pixclock = vm->pixelclock;
> +	u32 actual_clk;
> +
> +	dtg_lrc_x = vm->hfront_porch + vm->hback_porch + vm->hsync_len +
> +		    vm->hactive - 1;
> +	dtg_lrc_y = vm->vfront_porch + vm->vback_porch + vm->vsync_len +
> +		    vm->vactive - 1;
> +	dis_ulc_x = vm->hsync_len + vm->hback_porch - 1;
> +	dis_ulc_y = vm->vsync_len + vm->vfront_porch + vm->vback_porch - 1;
> +	dis_lrc_x = vm->hsync_len + vm->hback_porch + vm->hactive - 1;
> +	dis_lrc_y = vm->vsync_len + vm->vfront_porch + vm->vback_porch +
> +		    vm->vactive - 1;
> +
> +	clk_disable_unprepare(dtg->pix_clk);
> +	clk_set_rate(dtg->pix_clk, vm->pixelclock);
> +	clk_prepare_enable(dtg->pix_clk);
> +
> +	actual_clk = clk_get_rate(dtg->pix_clk);
> +	if (pixclock != actual_clk) {
> +		dev_info(dtg->dev,
> +			 "Pixel clock set to %u kHz instead of %u kHz.\n",
> +			 (actual_clk / 1000), (pixclock / 1000));

IMHO this is pretty noisy. Maybe allow some deviation and only print
the message if the actual clock differs significantly from the
requested clock?

> +	}
> +
> +	dcss_dtg_write(dtg, ((dtg_lrc_y << TC_Y_POS) | dtg_lrc_x),
> +		       DCSS_DTG_TC_DTG);
> +	dcss_dtg_write(dtg, ((dis_ulc_y << TC_Y_POS) | dis_ulc_x),
> +		       DCSS_DTG_TC_DISP_TOP);
> +	dcss_dtg_write(dtg, ((dis_lrc_y << TC_Y_POS) | dis_lrc_x),
> +		       DCSS_DTG_TC_DISP_BOT);
> +
> +	dtg->dis_ulc_x = dis_ulc_x;
> +	dtg->dis_ulc_y = dis_ulc_y;
> +
> +	sb_ctxld_trig = ((0 * dis_lrc_y / 100) << TC_CTXLD_SB_Y_POS) &
> +							TC_CTXLD_SB_Y_MASK;
> +	db_ctxld_trig = ((99 * dis_lrc_y / 100) << TC_CTXLD_DB_Y_POS) &
> +							TC_CTXLD_DB_Y_MASK;
> +
> +	dcss_dtg_write(dtg, sb_ctxld_trig | db_ctxld_trig, DCSS_DTG_TC_CTXLD);
> +
> +	/* vblank trigger */
> +	dcss_dtg_write(dtg, 0, DCSS_DTG_LINE1_INT);
> +
> +	/* CTXLD trigger */
> +	dcss_dtg_write(dtg, ((90 * dis_lrc_y) / 100) << 16, DCSS_DTG_LINE0_INT);
> +}
> +
> +void dcss_dtg_plane_pos_set(struct dcss_dtg *dtg, int ch_num,
> +			    int px, int py, int pw, int ph)
> +{
> +	u16 p_ulc_x, p_ulc_y;
> +	u16 p_lrc_x, p_lrc_y;
> +
> +	p_ulc_x = dtg->dis_ulc_x + px;
> +	p_ulc_y = dtg->dis_ulc_y + py;
> +	p_lrc_x = p_ulc_x + pw;
> +	p_lrc_y = p_ulc_y + ph;
> +
> +	if (!px && !py && !pw && !ph) {
> +		dcss_dtg_write(dtg, 0, DCSS_DTG_TC_CH1_TOP + 0x8 * ch_num);
> +		dcss_dtg_write(dtg, 0, DCSS_DTG_TC_CH1_BOT + 0x8 * ch_num);
> +	} else {
> +		dcss_dtg_write(dtg, ((p_ulc_y << TC_Y_POS) | p_ulc_x),
> +			       DCSS_DTG_TC_CH1_TOP + 0x8 * ch_num);
> +		dcss_dtg_write(dtg, ((p_lrc_y << TC_Y_POS) | p_lrc_x),
> +			       DCSS_DTG_TC_CH1_BOT + 0x8 * ch_num);
> +	}
> +}
> +
> +bool dcss_dtg_global_alpha_changed(struct dcss_dtg *dtg, int ch_num, int alpha)
> +{
> +	if (ch_num)
> +		return false;
> +
> +	return alpha != dtg->alpha;
> +}
> +
> +void dcss_dtg_plane_alpha_set(struct dcss_dtg *dtg, int ch_num,
> +			      const struct drm_format_info *format, int alpha)
> +{
> +	u32 alpha_val;
> +
> +	/* we care about alpha only when channel 0 is concerned */
> +	if (ch_num)
> +		return;
> +
> +	alpha_val = (alpha << DEFAULT_FG_ALPHA_POS) & DEFAULT_FG_ALPHA_MASK;
> +
> +	/*
> +	 * Use global alpha if pixel format does not have alpha channel or the
> +	 * user explicitly chose to use global alpha (i.e. alpha is not OPAQUE).
> +	 */
> +	if (!format->has_alpha || alpha != 255) {
> +		dtg->control_status &= ~(CH1_ALPHA_SEL | DEFAULT_FG_ALPHA_MASK);
> +		dtg->control_status |= alpha_val;
> +	} else { /* use per-pixel alpha otherwise */
> +		dtg->control_status |= CH1_ALPHA_SEL;
> +	}
> +
> +	dtg->alpha = alpha;
> +}
> +
> +void dcss_dtg_css_set(struct dcss_dtg *dtg)
> +{
> +	dtg->control_status |=
> +			(0x5 << CSS_PIX_COMP_SWAP_POS) & CSS_PIX_COMP_SWAP_MASK;
> +}
> +
> +static void dcss_dtg_disable_callback(void *data)
> +{
> +	struct dcss_dtg *dtg = data;
> +
> +	dtg->control_status &= ~DTG_START;
> +
> +	dcss_writel(dtg->control_status,
> +		    dtg->base_reg + DCSS_DTG_TC_CONTROL_STATUS);
> +
> +	dtg->in_use = false;
> +
> +	complete(dtg->dis_completion);
> +}
> +
> +void dcss_dtg_enable(struct dcss_dtg *dtg, bool en,
> +		     struct completion *dis_completion)
> +{
> +	if (!en) {
> +		dtg->dis_completion = dis_completion;
> +		dcss_ctxld_register_dtg_disable_cb(dtg->ctxld,
> +						   dcss_dtg_disable_callback,
> +						   dtg);
> +		return;
> +	}
> +
> +	dtg->dis_completion = NULL;
> +
> +	dtg->control_status |= DTG_START;
> +
> +	dcss_dtg_write(dtg, dtg->control_status, DCSS_DTG_TC_CONTROL_STATUS);
> +
> +	dtg->in_use = true;
> +}
> +
> +bool dcss_dtg_is_enabled(struct dcss_dtg *dtg)
> +{
> +	return dtg->in_use;
> +}
> +
> +void dcss_dtg_ch_enable(struct dcss_dtg *dtg, int ch_num, bool en)
> +{
> +	u32 ch_en_map[] = {CH1_EN, CH2_EN, CH3_EN};
> +	u32 control_status;
> +
> +	control_status = dtg->control_status & ~ch_en_map[ch_num];
> +	control_status |= en ? ch_en_map[ch_num] : 0;
> +
> +	if (dtg->control_status != control_status)
> +		dcss_dtg_write(dtg, control_status, DCSS_DTG_TC_CONTROL_STATUS);
> +
> +	dtg->control_status = control_status;
> +}
> +
> +void dcss_dtg_vblank_irq_enable(struct dcss_dtg *dtg, bool en)
> +{
> +	u32 status;
> +	u32 mask = en ? LINE1_IRQ : 0;
> +
> +	if (en) {
> +		status = dcss_readl(dtg->base_reg + DCSS_DTG_INT_STATUS);
> +		dcss_writel(status & LINE1_IRQ,
> +			    dtg->base_reg + DCSS_DTG_INT_CONTROL);

This clobbers the DCSS_DTG_INT_CONTROL register in the !en case, as it
forces all bits to 0, instead if just the LINE1_IRQ bit.

> +	}
> +
> +	dcss_update(mask, LINE1_IRQ, dtg->base_reg + DCSS_DTG_INT_MASK);
> +}
> +
> +void dcss_dtg_ctxld_kick_irq_enable(struct dcss_dtg *dtg, bool en)
> +{
> +	u32 status;
> +	u32 mask = en ? LINE0_IRQ : 0;
> +
> +	if (en) {
> +		status = dcss_readl(dtg->base_reg + DCSS_DTG_INT_STATUS);
> +
> +		if (!dtg->ctxld_kick_irq_en) {
> +			dcss_writel(status & LINE0_IRQ,
> +				    dtg->base_reg + DCSS_DTG_INT_CONTROL);
> +			enable_irq(dtg->ctxld_kick_irq);
> +			dtg->ctxld_kick_irq_en = true;
> +			goto mask_unmask;

This goto only saves one line of code and doesn't make this function
more readable. Please just duplicate the dcss_update line here and drop
the goto.

> +		}
> +
> +		return;
> +	}
> +
> +	if (!dtg->ctxld_kick_irq_en)
> +		return;
> +
> +	disable_irq_nosync(dtg->ctxld_kick_irq);
> +	dtg->ctxld_kick_irq_en = false;
> +
> +mask_unmask:
> +	dcss_update(mask, LINE0_IRQ, dtg->base_reg + DCSS_DTG_INT_MASK);
> +}
> +
> +void dcss_dtg_vblank_irq_clear(struct dcss_dtg *dtg)
> +{
> +	dcss_update(LINE1_IRQ, LINE1_IRQ, dtg->base_reg + DCSS_DTG_INT_CONTROL);
> +}
> +
> +bool dcss_dtg_vblank_irq_valid(struct dcss_dtg *dtg)
> +{
> +	return !!(dcss_readl(dtg->base_reg + DCSS_DTG_INT_STATUS) & LINE1_IRQ);
> +}
> +
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-kms.c b/drivers/gpu/drm/imx/dcss/dcss-kms.c
> new file mode 100644
> index 00000000..a33b3e4
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-kms.c
> @@ -0,0 +1,322 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <drm/drm_atomic.h>
> +#include <drm/drm_atomic_helper.h>
> +#include <drm/drm_drv.h>
> +#include <drm/drm_fb_helper.h>
> +#include <drm/drm_gem_cma_helper.h>
> +#include <drm/drm_gem_framebuffer_helper.h>
> +#include <drm/drm_of.h>
> +#include <drm/drm_probe_helper.h>
> +#include <drm/drm_vblank.h>
> +#include <linux/component.h>
> +
> +#include "dcss-dev.h"
> +#include "dcss-kms.h"
> +
> +DEFINE_DRM_GEM_CMA_FOPS(dcss_cma_fops);
> +
> +struct dcss_drm_commit {
> +	struct work_struct work;
> +	struct drm_device *drm;
> +	struct drm_atomic_state *state;
> +};
> +
> +static void dcss_drm_atomic_commit_tail(struct dcss_drm_commit *commit)
> +{
> +	struct drm_atomic_state *state = commit->state;
> +	struct drm_device *drm = commit->drm;
> +	struct dcss_kms_dev *kms = container_of(drm, struct dcss_kms_dev, base);
> +
> +	drm_atomic_helper_wait_for_fences(drm, state, false);
> +
> +	drm_atomic_helper_wait_for_dependencies(state);
> +
> +	drm_atomic_helper_commit_modeset_disables(drm, state);
> +
> +	drm_atomic_helper_commit_modeset_enables(drm, state);
> +
> +	drm_atomic_helper_commit_planes(drm, state,
> +					DRM_PLANE_COMMIT_ACTIVE_ONLY);
> +
> +	drm_atomic_helper_commit_hw_done(state);
> +
> +	drm_atomic_helper_wait_for_vblanks(drm, state);
> +
> +	drm_atomic_helper_cleanup_planes(drm, state);
> +
> +	drm_atomic_helper_commit_cleanup_done(state);
> +
> +	drm_atomic_state_put(state);
> +
> +	spin_lock(&kms->commit.wait.lock);
> +	kms->commit.pending = false;
> +	wake_up_all_locked(&kms->commit.wait);
> +	spin_unlock(&kms->commit.wait.lock);

What is the reason for this custom waitqueue in the driver?

drm_atomic_helper_wait_for_dependencies should already synchronize new
atomic commits against already running ones.

> +
> +	kfree(commit);
> +}
> +
> +static void dcss_commit_work(struct work_struct *work)
> +{
> +	struct dcss_drm_commit *commit = container_of(work,
> +						      struct dcss_drm_commit,
> +						      work);
> +
> +	dcss_drm_atomic_commit_tail(commit);
> +}
> +
> +static int dcss_drm_atomic_commit(struct drm_device *drm,
> +				  struct drm_atomic_state *state,
> +				  bool nonblock)
> +{
> +	int ret;
> +	struct dcss_kms_dev *kms = container_of(drm, struct dcss_kms_dev, base);
> +	struct dcss_drm_commit *commit;
> +
> +	if (state->async_update) {
> +		ret = drm_atomic_helper_prepare_planes(drm, state);
> +		if (ret)
> +			return ret;
> +
> +		drm_atomic_helper_async_commit(drm, state);
> +		drm_atomic_helper_cleanup_planes(drm, state);
> +
> +		return 0;
> +	}
> +
> +	commit = kzalloc(sizeof(*commit), GFP_KERNEL);
> +	if (!commit)
> +		return -ENOMEM;
> +
> +	commit->drm = drm;
> +	commit->state = state;
> +
> +	ret = drm_atomic_helper_setup_commit(state, nonblock);
> +	if (ret)
> +		goto err_free;
> +
> +	INIT_WORK(&commit->work, dcss_commit_work);
> +
> +	ret = drm_atomic_helper_prepare_planes(drm, state);
> +	if (ret)
> +		goto err_free;
> +
> +	if (!nonblock) {
> +		ret = drm_atomic_helper_wait_for_fences(drm, state, true);
> +		if (ret)
> +			goto err;
> +	}
> +
> +	spin_lock(&kms->commit.wait.lock);
> +	ret = wait_event_interruptible_locked(kms->commit.wait,
> +					      !kms->commit.pending);
> +	if (ret == 0)
> +		kms->commit.pending = true;
> +	spin_unlock(&kms->commit.wait.lock);
> +
> +	if (ret)
> +		goto err;
> +
> +	ret = drm_atomic_helper_swap_state(state, true);
> +	if (ret)
> +		goto err;
> +
> +	drm_atomic_state_get(state);
> +	if (nonblock)
> +		queue_work(kms->commit_wq, &commit->work);
> +	else
> +		dcss_drm_atomic_commit_tail(commit);
> +
> +	return 0;
> +
> +err:
> +	drm_atomic_helper_cleanup_planes(drm, state);
> +
> +err_free:
> +	kfree(commit);
> +	return ret;
> +}
> +
> +const struct drm_mode_config_funcs dcss_drm_mode_config_funcs = {
> +	.fb_create = drm_gem_fb_create,
> +	.output_poll_changed = drm_fb_helper_output_poll_changed,
> +	.atomic_check = drm_atomic_helper_check,
> +	.atomic_commit = dcss_drm_atomic_commit,
> +};
> +
> +static struct drm_driver dcss_kms_driver = {
> +	.driver_features	= DRIVER_MODESET | DRIVER_GEM | DRIVER_ATOMIC,
> +	.gem_free_object_unlocked = drm_gem_cma_free_object,
> +	.gem_vm_ops		= &drm_gem_cma_vm_ops,
> +	.dumb_create		= drm_gem_cma_dumb_create,
> +
> +	.prime_handle_to_fd	= drm_gem_prime_handle_to_fd,
> +	.prime_fd_to_handle	= drm_gem_prime_fd_to_handle,
> +	.gem_prime_import	= drm_gem_prime_import,
> +	.gem_prime_export	= drm_gem_prime_export,
> +	.gem_prime_get_sg_table	= drm_gem_cma_prime_get_sg_table,
> +	.gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table,
> +	.gem_prime_vmap		= drm_gem_cma_prime_vmap,
> +	.gem_prime_vunmap	= drm_gem_cma_prime_vunmap,
> +	.gem_prime_mmap		= drm_gem_cma_prime_mmap,
> +	.fops			= &dcss_cma_fops,
> +	.name			= "imx-dcss",
> +	.desc			= "i.MX8MQ Display Subsystem",
> +	.date			= "20190917",
> +	.major			= 1,
> +	.minor			= 0,
> +	.patchlevel		= 0,
> +};
> +
> +static const struct drm_mode_config_helper_funcs dcss_mode_config_helpers = {
> +	.atomic_commit_tail = drm_atomic_helper_commit_tail_rpm,
> +};
> +
> +static void dcss_kms_mode_config_init(struct dcss_kms_dev *kms)
> +{
> +	struct drm_mode_config *config = &kms->base.mode_config;
> +
> +	drm_mode_config_init(&kms->base);
> +
> +	config->min_width = 1;
> +	config->min_height = 1;
> +	config->max_width = 4096;
> +	config->max_height = 4096;
> +	config->allow_fb_modifiers = true;
> +	config->normalize_zpos = true;
> +
> +	config->funcs = &dcss_drm_mode_config_funcs;
> +	config->helper_private = &dcss_mode_config_helpers;
> +}
> +
> +static const struct drm_encoder_funcs dcss_kms_simple_encoder_funcs = {
> +	.destroy = drm_encoder_cleanup,
> +};
> +
> +static int dcss_kms_setup_encoder(struct dcss_kms_dev *kms)
> +{
> +	struct drm_device *ddev = &kms->base;
> +	struct drm_encoder *encoder = &kms->encoder;
> +	struct drm_crtc *crtc = (struct drm_crtc *)&kms->crtc;
> +	struct drm_panel *panel;
> +	struct drm_bridge *bridge;
> +	int ret;
> +
> +	ret = drm_of_find_panel_or_bridge(ddev->dev->of_node, 0, 0,
> +					  &panel, &bridge);
> +	if (ret)
> +		return ret;
> +
> +	if (!bridge) {
> +		dev_err(ddev->dev, "No bridge found %d.\n", ret);
> +		return -ENODEV;
> +	}
> +
> +	encoder->possible_crtcs = drm_crtc_mask(crtc);
> +
> +	ret = drm_encoder_init(&kms->base, encoder,
> +			       &dcss_kms_simple_encoder_funcs,
> +			       DRM_MODE_ENCODER_NONE, NULL);
> +	if (ret) {
> +		dev_err(ddev->dev, "Failed initializing encoder %d.\n", ret);
> +		return ret;
> +	}
> +
> +	return drm_bridge_attach(encoder, bridge, NULL);
> +}
> +
> +struct dcss_kms_dev *dcss_kms_attach(struct dcss_dev *dcss, bool componentized)
> +{
> +	struct dcss_kms_dev *kms = kzalloc(sizeof(*kms), GFP_KERNEL);
> +	struct drm_device *drm;
> +	struct dcss_crtc *crtc;
> +	int ret;
> +
> +	if (!kms)
> +		return ERR_PTR(-ENOMEM);
> +
> +	drm = &kms->base;
> +	crtc = &kms->crtc;
> +	ret = drm_dev_init(drm, &dcss_kms_driver, dcss->dev);
> +	if (ret)
> +		goto free_kms;
> +
> +	drm->dev_private = dcss;
> +
> +	dcss_kms_mode_config_init(kms);
> +
> +	ret = drm_vblank_init(drm, 1);
> +	if (ret)
> +		goto cleanup_mode_config;
> +
> +	drm->irq_enabled = true;
> +
> +	ret = dcss_crtc_init(crtc, drm);
> +	if (ret)
> +		goto cleanup_mode_config;
> +
> +	kms->commit_wq = alloc_ordered_workqueue("dcss_nonblock_commit_wq", 0);
> +	if (!kms->commit_wq) {
> +		ret = -ENOMEM;
> +		goto cleanup_crtc;
> +	}
> +
> +	init_waitqueue_head(&kms->commit.wait);
> +
> +	if (componentized)
> +		ret = component_bind_all(dcss->dev, kms);
> +	else
> +		ret = dcss_kms_setup_encoder(kms);
> +
> +	if (ret)
> +		goto cleanup_wq;
> +
> +	drm_mode_config_reset(drm);
> +
> +	drm_kms_helper_poll_init(drm);
> +
> +	ret = drm_dev_register(drm, 0);
> +	if (ret)
> +		goto cleanup_wq;
> +
> +	drm_fbdev_generic_setup(drm, 32);
> +
> +	return kms;
> +
> +cleanup_wq:
> +	drm_kms_helper_poll_fini(drm);
> +	destroy_workqueue(kms->commit_wq);
> +
> +cleanup_crtc:
> +	dcss_crtc_deinit(crtc, drm);
> +
> +cleanup_mode_config:
> +	drm_mode_config_cleanup(drm);
> +
> +free_kms:
> +	kfree(kms);
> +	return ERR_PTR(ret);
> +}
> +
> +void dcss_kms_detach(struct dcss_kms_dev *kms, bool componentized)
> +{
> +	struct drm_device *drm = &kms->base;
> +	struct dcss_dev *dcss = drm->dev_private;
> +
> +	drm_dev_unregister(drm);
> +	drm_kms_helper_poll_fini(drm);
> +	drm_atomic_helper_shutdown(drm);
> +	drm_crtc_vblank_off(&kms->crtc.base);
> +	drm->irq_enabled = false;
> +	drm_mode_config_cleanup(drm);
> +	destroy_workqueue(kms->commit_wq);
> +	dcss_crtc_deinit(&kms->crtc, drm);
> +	if (componentized)
> +		component_unbind_all(dcss->dev, drm);
> +	drm->dev_private = NULL;
> +	drm_dev_put(drm);
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-kms.h b/drivers/gpu/drm/imx/dcss/dcss-kms.h
> new file mode 100644
> index 00000000..6297005
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-kms.h
> @@ -0,0 +1,52 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#ifndef _DCSS_KMS_H_
> +#define _DCSS_KMS_H_
> +
> +#include <drm/drm_encoder.h>
> +
> +struct dcss_plane {
> +	struct drm_plane base;
> +
> +	int ch_num;
> +};
> +
> +struct dcss_crtc {
> +	struct drm_crtc		base;
> +	struct drm_crtc_state	*state;
> +
> +	struct dcss_plane	*plane[3];
> +
> +	int			irq;
> +	bool			irq_enabled;
> +
> +	struct completion en_completion;
> +	struct completion dis_completion;
> +};
> +
> +struct commit {
> +	wait_queue_head_t wait;
> +	bool pending;
> +};
> +
> +struct dcss_kms_dev {
> +	struct drm_device base;
> +	struct dcss_crtc crtc;
> +	struct drm_encoder encoder;
> +	struct workqueue_struct *commit_wq;
> +	struct commit commit;
> +};
> +
> +struct dcss_kms_dev *dcss_kms_attach(struct dcss_dev *dcss, bool componentized);
> +void dcss_kms_detach(struct dcss_kms_dev *kms, bool componentized);
> +int dcss_crtc_init(struct dcss_crtc *crtc, struct drm_device *drm);
> +void dcss_crtc_deinit(struct dcss_crtc *crtc, struct drm_device *drm);
> +struct dcss_plane *dcss_plane_init(struct drm_device *drm,
> +				   unsigned int possible_crtcs,
> +				   enum drm_plane_type type,
> +				   unsigned int zpos);
> +
> +#endif
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-plane.c b/drivers/gpu/drm/imx/dcss/dcss-plane.c
> new file mode 100644
> index 00000000..cb18c3e
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-plane.c
> @@ -0,0 +1,418 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <drm/drm_atomic.h>
> +#include <drm/drm_atomic_helper.h>
> +#include <drm/drm_fb_cma_helper.h>
> +#include <drm/drm_gem_framebuffer_helper.h>
> +#include <drm/drm_gem_cma_helper.h>
> +
> +#include "dcss-dev.h"
> +#include "dcss-kms.h"
> +
> +static const u32 dcss_common_formats[] = {
> +	/* RGB */
> +	DRM_FORMAT_ARGB8888,
> +	DRM_FORMAT_XRGB8888,
> +	DRM_FORMAT_ABGR8888,
> +	DRM_FORMAT_XBGR8888,
> +	DRM_FORMAT_RGBA8888,
> +	DRM_FORMAT_RGBX8888,
> +	DRM_FORMAT_BGRA8888,
> +	DRM_FORMAT_BGRX8888,
> +	DRM_FORMAT_XRGB2101010,
> +	DRM_FORMAT_XBGR2101010,
> +	DRM_FORMAT_RGBX1010102,
> +	DRM_FORMAT_BGRX1010102,
> +	DRM_FORMAT_ARGB2101010,
> +	DRM_FORMAT_ABGR2101010,
> +	DRM_FORMAT_RGBA1010102,
> +	DRM_FORMAT_BGRA1010102,
> +
> +	/* YUV444 */
> +	DRM_FORMAT_AYUV,
> +
> +	/* YUV422 */
> +	DRM_FORMAT_UYVY,
> +	DRM_FORMAT_VYUY,
> +	DRM_FORMAT_YUYV,
> +	DRM_FORMAT_YVYU,
> +
> +	/* YUV420 */
> +	DRM_FORMAT_NV12,
> +	DRM_FORMAT_NV21,
> +};
> +
> +static const u64 dcss_video_format_modifiers[] = {
> +	DRM_FORMAT_MOD_LINEAR,
> +	DRM_FORMAT_MOD_INVALID,
> +};
> +
> +static const u64 dcss_graphics_format_modifiers[] = {
> +	DRM_FORMAT_MOD_VIVANTE_TILED,
> +	DRM_FORMAT_MOD_VIVANTE_SUPER_TILED,
> +	DRM_FORMAT_MOD_LINEAR,
> +	DRM_FORMAT_MOD_INVALID,
> +};
> +
> +static inline struct dcss_plane *to_dcss_plane(struct drm_plane *p)
> +{
> +	return container_of(p, struct dcss_plane, base);
> +}
> +
> +static inline bool dcss_plane_fb_is_linear(const struct drm_framebuffer *fb)
> +{
> +	return ((fb->flags & DRM_MODE_FB_MODIFIERS) == 0) ||
> +	       ((fb->flags & DRM_MODE_FB_MODIFIERS) != 0 &&
> +		fb->modifier == DRM_FORMAT_MOD_LINEAR);
> +}
> +
> +static void dcss_plane_destroy(struct drm_plane *plane)
> +{
> +	struct dcss_plane *dcss_plane = container_of(plane, struct dcss_plane,
> +						     base);
> +
> +	drm_plane_cleanup(plane);
> +	kfree(dcss_plane);
> +}
> +
> +static bool dcss_plane_format_mod_supported(struct drm_plane *plane,
> +					    u32 format,
> +					    u64 modifier)
> +{
> +	switch (plane->type) {
> +	case DRM_PLANE_TYPE_PRIMARY:
> +		switch (format) {
> +		case DRM_FORMAT_ARGB8888:
> +		case DRM_FORMAT_XRGB8888:
> +		case DRM_FORMAT_ARGB2101010:
> +			return modifier == DRM_FORMAT_MOD_LINEAR ||
> +			       modifier == DRM_FORMAT_MOD_VIVANTE_TILED ||
> +			       modifier == DRM_FORMAT_MOD_VIVANTE_SUPER_TILED;
> +		default:
> +			return modifier == DRM_FORMAT_MOD_LINEAR;
> +		}
> +		break;
> +	case DRM_PLANE_TYPE_OVERLAY:
> +		return modifier == DRM_FORMAT_MOD_LINEAR;
> +	default:
> +		return false;
> +	}
> +}
> +
> +static const struct drm_plane_funcs dcss_plane_funcs = {
> +	.update_plane		= drm_atomic_helper_update_plane,
> +	.disable_plane		= drm_atomic_helper_disable_plane,
> +	.destroy		= dcss_plane_destroy,
> +	.reset			= drm_atomic_helper_plane_reset,
> +	.atomic_duplicate_state	= drm_atomic_helper_plane_duplicate_state,
> +	.atomic_destroy_state	= drm_atomic_helper_plane_destroy_state,
> +	.format_mod_supported	= dcss_plane_format_mod_supported,
> +};
> +
> +static bool dcss_plane_can_rotate(const struct drm_format_info *format,
> +				  bool mod_present, u64 modifier,
> +				  unsigned int rotation)
> +{
> +	bool linear_format = !mod_present ||
> +			     (mod_present && modifier == DRM_FORMAT_MOD_LINEAR);
> +	u32 supported_rotation = DRM_MODE_ROTATE_0;
> +
> +	if (!format->is_yuv && linear_format)
> +		supported_rotation = DRM_MODE_ROTATE_0 | DRM_MODE_ROTATE_180 |
> +				     DRM_MODE_REFLECT_MASK;
> +	else if (!format->is_yuv &&
> +		 modifier == DRM_FORMAT_MOD_VIVANTE_TILED)
> +		supported_rotation = DRM_MODE_ROTATE_MASK |
> +				     DRM_MODE_REFLECT_MASK;
> +	else if (format->is_yuv && linear_format &&
> +		 (format->format == DRM_FORMAT_NV12 ||
> +		  format->format == DRM_FORMAT_NV21))
> +		supported_rotation = DRM_MODE_ROTATE_0 | DRM_MODE_ROTATE_180 |
> +				     DRM_MODE_REFLECT_MASK;
> +
> +	return !!(rotation & supported_rotation);
> +}
> +
> +static bool dcss_plane_is_source_size_allowed(u16 src_w, u16 src_h, u32 pix_fmt)
> +{
> +	if (src_w < 64 &&
> +	    (pix_fmt == DRM_FORMAT_NV12 || pix_fmt == DRM_FORMAT_NV21))
> +		return false;
> +	else if (src_w < 32 &&
> +		 (pix_fmt == DRM_FORMAT_UYVY || pix_fmt == DRM_FORMAT_VYUY ||
> +		  pix_fmt == DRM_FORMAT_YUYV || pix_fmt == DRM_FORMAT_YVYU))
> +		return false;
> +
> +	return src_w >= 16 && src_h >= 8;
> +}
> +
> +static int dcss_plane_atomic_check(struct drm_plane *plane,
> +				   struct drm_plane_state *state)
> +{
> +	struct dcss_plane *dcss_plane = to_dcss_plane(plane);
> +	struct dcss_dev *dcss = plane->dev->dev_private;
> +	struct drm_framebuffer *fb = state->fb;
> +	bool is_primary_plane = plane->type == DRM_PLANE_TYPE_PRIMARY;
> +	struct drm_gem_cma_object *cma_obj;
> +	struct drm_crtc_state *crtc_state;
> +	int hdisplay, vdisplay;
> +	int min, max;
> +	int ret;
> +
> +	if (!fb || !state->crtc)
> +		return 0;
> +
> +	cma_obj = drm_fb_cma_get_gem_obj(fb, 0);
> +	WARN_ON(!cma_obj);
> +
> +	crtc_state = drm_atomic_get_existing_crtc_state(state->state,
> +							state->crtc);
> +
> +	hdisplay = crtc_state->adjusted_mode.hdisplay;
> +	vdisplay = crtc_state->adjusted_mode.vdisplay;
> +
> +	if (!dcss_plane_is_source_size_allowed(state->src_w >> 16,
> +					       state->src_h >> 16,
> +					       fb->format->format)) {
> +		DRM_DEBUG_KMS("Source plane size is not allowed!\n");
> +		return -EINVAL;
> +	}
> +
> +	dcss_scaler_get_min_max_ratios(dcss->scaler, dcss_plane->ch_num,
> +				       &min, &max);
> +
> +	ret = drm_atomic_helper_check_plane_state(state, crtc_state,
> +						  min, max, !is_primary_plane,
> +						  false);
> +	if (ret)
> +		return ret;
> +
> +	if (!state->visible)
> +		return 0;
> +
> +	if (!dcss_plane_can_rotate(fb->format,
> +				   !!(fb->flags & DRM_MODE_FB_MODIFIERS),
> +				   fb->modifier,
> +				   state->rotation)) {
> +		DRM_DEBUG_KMS("requested rotation is not allowed!\n");
> +		return -EINVAL;
> +	}
> +
> +	if ((state->crtc_x < 0 || state->crtc_y < 0 ||
> +	     state->crtc_x + state->crtc_w > hdisplay ||
> +	     state->crtc_y + state->crtc_h > vdisplay) &&
> +	    !dcss_plane_fb_is_linear(fb)) {
> +		DRM_DEBUG_KMS("requested cropping operation is not allowed!\n");
> +		return -EINVAL;
> +	}
> +
> +	if ((fb->flags & DRM_MODE_FB_MODIFIERS) &&
> +	    !plane->funcs->format_mod_supported(plane,
> +				fb->format->format,
> +				fb->modifier)) {
> +		DRM_DEBUG_KMS("Invalid modifier: %llx", fb->modifier);
> +		return -EINVAL;
> +	}
> +
> +	return 0;
> +}
> +
> +static void dcss_plane_atomic_set_base(struct dcss_plane *dcss_plane)
> +{
> +	struct drm_plane *plane = &dcss_plane->base;
> +	struct drm_plane_state *state = plane->state;
> +	struct dcss_dev *dcss = plane->dev->dev_private;
> +	struct drm_framebuffer *fb = state->fb;
> +	const struct drm_format_info *format = fb->format;
> +	struct drm_gem_cma_object *cma_obj = drm_fb_cma_get_gem_obj(fb, 0);
> +	unsigned long p1_ba = 0, p2_ba = 0;
> +
> +	if (!format->is_yuv ||
> +	    format->format == DRM_FORMAT_NV12 ||
> +	    format->format == DRM_FORMAT_NV21)
> +		p1_ba = cma_obj->paddr + fb->offsets[0] +
> +			fb->pitches[0] * (state->src.y1 >> 16) +
> +			format->char_per_block[0] * (state->src.x1 >> 16);
> +	else if (format->format == DRM_FORMAT_UYVY ||
> +		 format->format == DRM_FORMAT_VYUY ||
> +		 format->format == DRM_FORMAT_YUYV ||
> +		 format->format == DRM_FORMAT_YVYU)
> +		p1_ba = cma_obj->paddr + fb->offsets[0] +
> +			fb->pitches[0] * (state->src.y1 >> 16) +
> +			2 * format->char_per_block[0] * (state->src.x1 >> 17);
> +
> +	if (format->format == DRM_FORMAT_NV12 ||
> +	    format->format == DRM_FORMAT_NV21)
> +		p2_ba = cma_obj->paddr + fb->offsets[1] +
> +			(((fb->pitches[1] >> 1) * (state->src.y1 >> 17) +
> +			(state->src.x1 >> 17)) << 1);
> +
> +	dcss_dpr_addr_set(dcss->dpr, dcss_plane->ch_num, p1_ba, p2_ba,
> +			  fb->pitches[0]);
> +}
> +
> +static bool dcss_plane_needs_setup(struct drm_plane_state *state,
> +				   struct drm_plane_state *old_state)
> +{
> +	struct drm_framebuffer *fb = state->fb;
> +	struct drm_framebuffer *old_fb = old_state->fb;
> +
> +	return state->crtc_x != old_state->crtc_x ||
> +	       state->crtc_y != old_state->crtc_y ||
> +	       state->crtc_w != old_state->crtc_w ||
> +	       state->crtc_h != old_state->crtc_h ||
> +	       state->src_x  != old_state->src_x  ||
> +	       state->src_y  != old_state->src_y  ||
> +	       state->src_w  != old_state->src_w  ||
> +	       state->src_h  != old_state->src_h  ||
> +	       fb->format->format != old_fb->format->format ||
> +	       fb->modifier  != old_fb->modifier ||
> +	       state->rotation != old_state->rotation;
> +}
> +
> +static void dcss_plane_atomic_update(struct drm_plane *plane,
> +				     struct drm_plane_state *old_state)
> +{
> +	struct drm_plane_state *state = plane->state;
> +	struct dcss_plane *dcss_plane = to_dcss_plane(plane);
> +	struct dcss_dev *dcss = plane->dev->dev_private;
> +	struct drm_framebuffer *fb = state->fb;
> +	u32 pixel_format;
> +	struct drm_crtc_state *crtc_state;
> +	bool modifiers_present;
> +	u32 src_w, src_h, dst_w, dst_h;
> +	struct drm_rect src, dst;
> +	bool enable = true;
> +
> +	if (!fb || !state->crtc || !state->visible)
> +		return;
> +
> +	pixel_format = state->fb->format->format;
> +	crtc_state = state->crtc->state;
> +	modifiers_present = !!(fb->flags & DRM_MODE_FB_MODIFIERS);
> +
> +	if (old_state->fb && !drm_atomic_crtc_needs_modeset(crtc_state) &&
> +	    !dcss_plane_needs_setup(state, old_state)) {
> +		dcss_plane_atomic_set_base(dcss_plane);
> +		return;
> +	}
> +
> +	src = plane->state->src;
> +	dst = plane->state->dst;
> +
> +	/*
> +	 * The width and height after clipping.
> +	 */
> +	src_w = drm_rect_width(&src) >> 16;
> +	src_h = drm_rect_height(&src) >> 16;
> +	dst_w = drm_rect_width(&dst);
> +	dst_h = drm_rect_height(&dst);
> +
> +	if (plane->type == DRM_PLANE_TYPE_OVERLAY &&
> +	    modifiers_present && fb->modifier == DRM_FORMAT_MOD_LINEAR)
> +		modifiers_present = false;
> +
> +	dcss_dpr_format_set(dcss->dpr, dcss_plane->ch_num, state->fb->format,
> +			    modifiers_present ? fb->modifier :
> +						DRM_FORMAT_MOD_LINEAR);
> +	dcss_dpr_set_res(dcss->dpr, dcss_plane->ch_num, src_w, src_h);
> +	dcss_dpr_set_rotation(dcss->dpr, dcss_plane->ch_num,
> +			      state->rotation);
> +
> +	dcss_plane_atomic_set_base(dcss_plane);
> +
> +	dcss_scaler_setup(dcss->scaler, dcss_plane->ch_num,
> +			  state->fb->format, src_w, src_h,
> +			  dst_w, dst_h,
> +			  drm_mode_vrefresh(&crtc_state->mode));
> +
> +	dcss_dtg_plane_pos_set(dcss->dtg, dcss_plane->ch_num,
> +			       dst.x1, dst.y1, dst_w, dst_h);
> +	dcss_dtg_plane_alpha_set(dcss->dtg, dcss_plane->ch_num,
> +				 fb->format, state->alpha >> 8);
> +
> +	if (!dcss_plane->ch_num && (state->alpha >> 8) == 0)
> +		enable = false;
> +
> +	dcss_dpr_enable(dcss->dpr, dcss_plane->ch_num, enable);
> +	dcss_scaler_ch_enable(dcss->scaler, dcss_plane->ch_num, enable);
> +
> +	if (!enable)
> +		dcss_dtg_plane_pos_set(dcss->dtg, dcss_plane->ch_num,
> +				       0, 0, 0, 0);
> +
> +	dcss_dtg_ch_enable(dcss->dtg, dcss_plane->ch_num, enable);
> +}
> +
> +static void dcss_plane_atomic_disable(struct drm_plane *plane,
> +				      struct drm_plane_state *old_state)
> +{
> +	struct dcss_plane *dcss_plane = to_dcss_plane(plane);
> +	struct dcss_dev *dcss = plane->dev->dev_private;
> +
> +	dcss_dpr_enable(dcss->dpr, dcss_plane->ch_num, false);
> +	dcss_scaler_ch_enable(dcss->scaler, dcss_plane->ch_num, false);
> +	dcss_dtg_plane_pos_set(dcss->dtg, dcss_plane->ch_num, 0, 0, 0, 0);
> +	dcss_dtg_ch_enable(dcss->dtg, dcss_plane->ch_num, false);
> +}
> +
> +static const struct drm_plane_helper_funcs dcss_plane_helper_funcs = {
> +	.prepare_fb = drm_gem_fb_prepare_fb,
> +	.atomic_check = dcss_plane_atomic_check,
> +	.atomic_update = dcss_plane_atomic_update,
> +	.atomic_disable = dcss_plane_atomic_disable,
> +};
> +
> +struct dcss_plane *dcss_plane_init(struct drm_device *drm,
> +				   unsigned int possible_crtcs,
> +				   enum drm_plane_type type,
> +				   unsigned int zpos)
> +{
> +	struct dcss_plane *dcss_plane;
> +	const u64 *format_modifiers = dcss_video_format_modifiers;
> +	int ret;
> +
> +	if (zpos > 2)
> +		return ERR_PTR(-EINVAL);
> +
> +	dcss_plane = kzalloc(sizeof(*dcss_plane), GFP_KERNEL);
> +	if (!dcss_plane) {
> +		DRM_ERROR("failed to allocate plane\n");
> +		return ERR_PTR(-ENOMEM);
> +	}
> +
> +	if (type == DRM_PLANE_TYPE_PRIMARY)
> +		format_modifiers = dcss_graphics_format_modifiers;
> +
> +	ret = drm_universal_plane_init(drm, &dcss_plane->base, possible_crtcs,
> +				       &dcss_plane_funcs, dcss_common_formats,
> +				       ARRAY_SIZE(dcss_common_formats),
> +				       format_modifiers, type, NULL);
> +	if (ret) {
> +		DRM_ERROR("failed to initialize plane\n");
> +		kfree(dcss_plane);
> +		return ERR_PTR(ret);
> +	}
> +
> +	drm_plane_helper_add(&dcss_plane->base, &dcss_plane_helper_funcs);
> +
> +	ret = drm_plane_create_zpos_immutable_property(&dcss_plane->base, zpos);
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	drm_plane_create_rotation_property(&dcss_plane->base,
> +					   DRM_MODE_ROTATE_0,
> +					   DRM_MODE_ROTATE_0   |
> +					   DRM_MODE_ROTATE_90  |
> +					   DRM_MODE_ROTATE_180 |
> +					   DRM_MODE_ROTATE_270 |
> +					   DRM_MODE_REFLECT_X  |
> +					   DRM_MODE_REFLECT_Y);
> +
> +	dcss_plane->ch_num = zpos;
> +
> +	return dcss_plane;
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-scaler.c b/drivers/gpu/drm/imx/dcss/dcss-scaler.c
> new file mode 100644
> index 00000000..56fc872
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-scaler.c
> @@ -0,0 +1,826 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + *
> + * Scaling algorithms were contributed by Dzung Hoang <dzung.hoang@nxp.com>
> + */
> +
> +#include <linux/device.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_SCALER_CTRL			0x00
> +#define   SCALER_EN				BIT(0)
> +#define   REPEAT_EN				BIT(4)
> +#define   SCALE2MEM_EN				BIT(8)
> +#define   MEM2OFIFO_EN				BIT(12)
> +#define DCSS_SCALER_OFIFO_CTRL			0x04
> +#define   OFIFO_LOW_THRES_POS			0
> +#define   OFIFO_LOW_THRES_MASK			GENMASK(9, 0)
> +#define   OFIFO_HIGH_THRES_POS			16
> +#define   OFIFO_HIGH_THRES_MASK			GENMASK(25, 16)
> +#define   UNDERRUN_DETECT_CLR			BIT(26)
> +#define   LOW_THRES_DETECT_CLR			BIT(27)
> +#define   HIGH_THRES_DETECT_CLR			BIT(28)
> +#define   UNDERRUN_DETECT_EN			BIT(29)
> +#define   LOW_THRES_DETECT_EN			BIT(30)
> +#define   HIGH_THRES_DETECT_EN			BIT(31)
> +#define DCSS_SCALER_SDATA_CTRL			0x08
> +#define   YUV_EN				BIT(0)
> +#define   RTRAM_8LINES				BIT(1)
> +#define   Y_UV_BYTE_SWAP			BIT(4)
> +#define   A2R10G10B10_FORMAT_POS		8
> +#define   A2R10G10B10_FORMAT_MASK		GENMASK(11, 8)
> +#define DCSS_SCALER_BIT_DEPTH			0x0C
> +#define   LUM_BIT_DEPTH_POS			0
> +#define   LUM_BIT_DEPTH_MASK			GENMASK(1, 0)
> +#define   CHR_BIT_DEPTH_POS			4
> +#define   CHR_BIT_DEPTH_MASK			GENMASK(5, 4)
> +#define DCSS_SCALER_SRC_FORMAT			0x10
> +#define DCSS_SCALER_DST_FORMAT			0x14
> +#define   FORMAT_MASK				GENMASK(1, 0)
> +#define DCSS_SCALER_SRC_LUM_RES			0x18
> +#define DCSS_SCALER_SRC_CHR_RES			0x1C
> +#define DCSS_SCALER_DST_LUM_RES			0x20
> +#define DCSS_SCALER_DST_CHR_RES			0x24
> +#define   WIDTH_POS				0
> +#define   WIDTH_MASK				GENMASK(11, 0)
> +#define   HEIGHT_POS				16
> +#define   HEIGHT_MASK				GENMASK(27, 16)
> +#define DCSS_SCALER_V_LUM_START			0x48
> +#define   V_START_MASK				GENMASK(15, 0)
> +#define DCSS_SCALER_V_LUM_INC			0x4C
> +#define   V_INC_MASK				GENMASK(15, 0)
> +#define DCSS_SCALER_H_LUM_START			0x50
> +#define   H_START_MASK				GENMASK(18, 0)
> +#define DCSS_SCALER_H_LUM_INC			0x54
> +#define   H_INC_MASK				GENMASK(15, 0)
> +#define DCSS_SCALER_V_CHR_START			0x58
> +#define DCSS_SCALER_V_CHR_INC			0x5C
> +#define DCSS_SCALER_H_CHR_START			0x60
> +#define DCSS_SCALER_H_CHR_INC			0x64
> +#define DCSS_SCALER_COEF_VLUM			0x80
> +#define DCSS_SCALER_COEF_HLUM			0x140
> +#define DCSS_SCALER_COEF_VCHR			0x200
> +#define DCSS_SCALER_COEF_HCHR			0x300
> +
> +struct dcss_scaler_ch {
> +	void __iomem *base_reg;
> +	u32 base_ofs;
> +	struct dcss_scaler *scl;
> +
> +	u32 sdata_ctrl;
> +	u32 scaler_ctrl;
> +
> +	bool scaler_ctrl_chgd;
> +
> +	u32 c_vstart;
> +	u32 c_hstart;
> +};
> +
> +struct dcss_scaler {
> +	struct device *dev;
> +
> +	struct dcss_ctxld *ctxld;
> +	u32 ctx_id;
> +
> +	struct dcss_scaler_ch ch[3];
> +};
> +
> +/* scaler coefficients generator */
> +#define PSC_FRAC_BITS 30
> +#define PSC_FRAC_SCALE BIT(PSC_FRAC_BITS)
> +#define PSC_BITS_FOR_PHASE 4
> +#define PSC_NUM_PHASES 16
> +#define PSC_STORED_PHASES (PSC_NUM_PHASES / 2 + 1)
> +#define PSC_NUM_TAPS 7
> +#define PSC_NUM_TAPS_RGBA 5
> +#define PSC_COEFF_PRECISION 10
> +#define PSC_PHASE_FRACTION_BITS 13
> +#define PSC_PHASE_MASK (PSC_NUM_PHASES - 1)
> +#define PSC_Q_FRACTION 19
> +#define PSC_Q_ROUND_OFFSET (1 << (PSC_Q_FRACTION - 1))
> +
> +/**
> + * mult_q() - Performs fixed-point multiplication.
> + * @A: multiplier
> + * @B: multiplicand
> + */
> +static int mult_q(int A, int B)
> +{
> +	int result;
> +	s64 temp;
> +
> +	temp = (int64_t)A * (int64_t)B;
> +	temp += PSC_Q_ROUND_OFFSET;
> +	result = (int)(temp >> PSC_Q_FRACTION);
> +	return result;
> +}
> +
> +/**
> + * div_q() - Performs fixed-point division.
> + * @A: dividend
> + * @B: divisor
> + */
> +static int div_q(int A, int B)
> +{
> +	int result;
> +	s64 temp;
> +
> +	temp = (int64_t)A << PSC_Q_FRACTION;
> +	if ((temp >= 0 && B >= 0) || (temp < 0 && B < 0))
> +		temp += B / 2;
> +	else
> +		temp -= B / 2;
> +
> +	result = (int)(temp / B);
> +	return result;
> +}
> +
> +/**
> + * exp_approx_q() - Compute approximation to exp(x) function using Taylor
> + *		    series.
> + * @x: fixed-point argument of exp function
> + */
> +static int exp_approx_q(int x)
> +{
> +	int sum = 1 << PSC_Q_FRACTION;
> +	int term = 1 << PSC_Q_FRACTION;
> +
> +	term = mult_q(term, div_q(x, 1 << PSC_Q_FRACTION));
> +	sum += term;
> +	term = mult_q(term, div_q(x, 2 << PSC_Q_FRACTION));
> +	sum += term;
> +	term = mult_q(term, div_q(x, 3 << PSC_Q_FRACTION));
> +	sum += term;
> +	term = mult_q(term, div_q(x, 4 << PSC_Q_FRACTION));
> +	sum += term;
> +
> +	return sum;
> +}
> +
> +/**
> + * dcss_scaler_gaussian_filter() - Generate gaussian prototype filter.
> + * @fc_q: fixed-point cutoff frequency normalized to range [0, 1]
> + * @use_5_taps: indicates whether to use 5 taps or 7 taps
> + * @coef: output filter coefficients
> + */
> +static void dcss_scaler_gaussian_filter(int fc_q, bool use_5_taps,
> +					bool phase0_identity,
> +					int coef[][PSC_NUM_TAPS])
> +{
> +	int sigma_q, g0_q, g1_q, g2_q;
> +	int tap_cnt1, tap_cnt2, tap_idx, phase_cnt;
> +	int mid;
> +	int phase;
> +	int i;
> +	int taps;
> +
> +	if (use_5_taps)
> +		for (phase = 0; phase < PSC_STORED_PHASES; phase++) {
> +			coef[phase][0] = 0;
> +			coef[phase][PSC_NUM_TAPS - 1] = 0;
> +		}
> +
> +	/* seed coefficient scanner */
> +	taps = use_5_taps ? PSC_NUM_TAPS_RGBA : PSC_NUM_TAPS;
> +	mid = (PSC_NUM_PHASES * taps) / 2 - 1;
> +	phase_cnt = (PSC_NUM_PHASES * (PSC_NUM_TAPS + 1)) / 2;
> +	tap_cnt1 = (PSC_NUM_PHASES * PSC_NUM_TAPS) / 2;
> +	tap_cnt2 = (PSC_NUM_PHASES * PSC_NUM_TAPS) / 2;
> +
> +	/* seed gaussian filter generator */
> +	sigma_q = div_q(PSC_Q_ROUND_OFFSET, fc_q);
> +	g0_q = 1 << PSC_Q_FRACTION;
> +	g1_q = exp_approx_q(div_q(-PSC_Q_ROUND_OFFSET,
> +				  mult_q(sigma_q, sigma_q)));
> +	g2_q = mult_q(g1_q, g1_q);
> +	coef[phase_cnt & PSC_PHASE_MASK][tap_cnt1 >> PSC_BITS_FOR_PHASE] = g0_q;
> +
> +	for (i = 0; i < mid; i++) {
> +		phase_cnt++;
> +		tap_cnt1--;
> +		tap_cnt2++;
> +
> +		g0_q = mult_q(g0_q, g1_q);
> +		g1_q = mult_q(g1_q, g2_q);
> +
> +		if ((phase_cnt & PSC_PHASE_MASK) <= 8) {
> +			tap_idx = tap_cnt1 >> PSC_BITS_FOR_PHASE;
> +			coef[phase_cnt & PSC_PHASE_MASK][tap_idx] = g0_q;
> +		}
> +		if (((-phase_cnt) & PSC_PHASE_MASK) <= 8) {
> +			tap_idx = tap_cnt2 >> PSC_BITS_FOR_PHASE;
> +			coef[(-phase_cnt) & PSC_PHASE_MASK][tap_idx] = g0_q;
> +		}
> +	}
> +
> +	phase_cnt++;
> +	tap_cnt1--;
> +	coef[phase_cnt & PSC_PHASE_MASK][tap_cnt1 >> PSC_BITS_FOR_PHASE] = 0;
> +
> +	/* override phase 0 with identity filter if specified */
> +	if (phase0_identity)
> +		for (i = 0; i < PSC_NUM_TAPS; i++)
> +			coef[0][i] = i == (PSC_NUM_TAPS >> 1) ?
> +						(1 << PSC_COEFF_PRECISION) : 0;
> +
> +	/* normalize coef */
> +	for (phase = 0; phase < PSC_STORED_PHASES; phase++) {
> +		int sum = 0;
> +		s64 ll_temp;
> +
> +		for (i = 0; i < PSC_NUM_TAPS; i++)
> +			sum += coef[phase][i];
> +		for (i = 0; i < PSC_NUM_TAPS; i++) {
> +			ll_temp = coef[phase][i];
> +			ll_temp <<= PSC_COEFF_PRECISION;
> +			ll_temp += sum >> 1;
> +			ll_temp /= sum;
> +			coef[phase][i] = (int)ll_temp;
> +		}
> +	}
> +}
> +
> +/**
> + * dcss_scaler_filter_design() - Compute filter coefficients using
> + *				 Gaussian filter.
> + * @src_length: length of input
> + * @dst_length: length of output
> + * @use_5_taps: 0 for 7 taps per phase, 1 for 5 taps
> + * @coef: output coefficients
> + */
> +static void dcss_scaler_filter_design(int src_length, int dst_length,
> +				      bool use_5_taps, bool phase0_identity,
> +				      int coef[][PSC_NUM_TAPS])
> +{
> +	int fc_q;
> +
> +	/* compute cutoff frequency */
> +	if (dst_length >= src_length)
> +		fc_q = div_q(1, PSC_NUM_PHASES);
> +	else
> +		fc_q = div_q(dst_length, src_length * PSC_NUM_PHASES);
> +
> +	/* compute gaussian filter coefficients */
> +	dcss_scaler_gaussian_filter(fc_q, use_5_taps, phase0_identity, coef);
> +}
> +
> +static void dcss_scaler_write(struct dcss_scaler_ch *ch, u32 val, u32 ofs)
> +{
> +	struct dcss_scaler *scl = ch->scl;
> +
> +	dcss_ctxld_write(scl->ctxld, scl->ctx_id, val, ch->base_ofs + ofs);
> +}
> +
> +static int dcss_scaler_ch_init_all(struct dcss_scaler *scl,
> +				   unsigned long scaler_base)
> +{
> +	struct dcss_scaler_ch *ch;
> +	int i;
> +
> +	for (i = 0; i < 3; i++) {
> +		ch = &scl->ch[i];
> +
> +		ch->base_ofs = scaler_base + i * 0x400;
> +
> +		ch->base_reg = devm_ioremap(scl->dev, ch->base_ofs, SZ_4K);
> +		if (!ch->base_reg) {
> +			dev_err(scl->dev, "scaler: unable to remap ch base\n");
> +			return -ENOMEM;
> +		}
> +
> +		ch->scl = scl;
> +	}
> +
> +	return 0;
> +}
> +
> +int dcss_scaler_init(struct dcss_dev *dcss, unsigned long scaler_base)
> +{
> +	struct dcss_scaler *scaler;
> +
> +	scaler = devm_kzalloc(dcss->dev, sizeof(*scaler), GFP_KERNEL);
> +	if (!scaler)
> +		return -ENOMEM;
> +
> +	dcss->scaler = scaler;
> +	scaler->dev = dcss->dev;
> +	scaler->ctxld = dcss->ctxld;
> +	scaler->ctx_id = CTX_SB_HP;
> +
> +	if (dcss_scaler_ch_init_all(scaler, scaler_base)) {
> +		int i;
> +
> +		for (i = 0; i < 3; i++) {
> +			if (scaler->ch[i].base_reg)
> +				devm_iounmap(scaler->dev,
> +					     scaler->ch[i].base_reg);
> +		}
> +
> +		devm_kfree(scaler->dev, scaler);
> +
> +		return -ENOMEM;
> +	}
> +
> +	return 0;
> +}
> +
> +void dcss_scaler_exit(struct dcss_scaler *scl)
> +{
> +	int ch_no;
> +
> +	for (ch_no = 0; ch_no < 3; ch_no++) {
> +		struct dcss_scaler_ch *ch = &scl->ch[ch_no];
> +
> +		dcss_writel(0, ch->base_reg + DCSS_SCALER_CTRL);
> +
> +		if (ch->base_reg)
> +			devm_iounmap(scl->dev, ch->base_reg);
> +	}
> +
> +	devm_kfree(scl->dev, scl);
> +}
> +
> +void dcss_scaler_ch_enable(struct dcss_scaler *scl, int ch_num, bool en)
> +{
> +	struct dcss_scaler_ch *ch = &scl->ch[ch_num];
> +	u32 scaler_ctrl;
> +
> +	scaler_ctrl = en ? SCALER_EN | REPEAT_EN : 0;
> +
> +	if (en)
> +		dcss_scaler_write(ch, ch->sdata_ctrl, DCSS_SCALER_SDATA_CTRL);
> +
> +	if (ch->scaler_ctrl != scaler_ctrl)
> +		ch->scaler_ctrl_chgd = true;
> +
> +	ch->scaler_ctrl = scaler_ctrl;
> +}
> +
> +static void dcss_scaler_yuv_enable(struct dcss_scaler_ch *ch, bool en)
> +{
> +	ch->sdata_ctrl &= ~YUV_EN;
> +	ch->sdata_ctrl |= en ? YUV_EN : 0;
> +}
> +
> +static void dcss_scaler_rtr_8lines_enable(struct dcss_scaler_ch *ch, bool en)
> +{
> +	ch->sdata_ctrl &= ~RTRAM_8LINES;
> +	ch->sdata_ctrl |= en ? RTRAM_8LINES : 0;
> +}
> +
> +static void dcss_scaler_bit_depth_set(struct dcss_scaler_ch *ch, int depth)
> +{
> +	u32 val;
> +
> +	val = depth == 30 ? 2 : 0;
> +
> +	dcss_scaler_write(ch,
> +			  ((val << CHR_BIT_DEPTH_POS) & CHR_BIT_DEPTH_MASK) |
> +			  ((val << LUM_BIT_DEPTH_POS) & LUM_BIT_DEPTH_MASK),
> +			  DCSS_SCALER_BIT_DEPTH);
> +}
> +
> +enum buffer_format {
> +	BUF_FMT_YUV420,
> +	BUF_FMT_YUV422,
> +	BUF_FMT_ARGB8888_YUV444,
> +};
> +
> +enum chroma_location {
> +	PSC_LOC_HORZ_0_VERT_1_OVER_4 = 0,
> +	PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_4 = 1,
> +	PSC_LOC_HORZ_0_VERT_0 = 2,
> +	PSC_LOC_HORZ_1_OVER_4_VERT_0 = 3,
> +	PSC_LOC_HORZ_0_VERT_1_OVER_2 = 4,
> +	PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_2 = 5
> +};
> +
> +static void dcss_scaler_format_set(struct dcss_scaler_ch *ch,
> +				   enum buffer_format src_fmt,
> +				   enum buffer_format dst_fmt)
> +{
> +	dcss_scaler_write(ch, src_fmt, DCSS_SCALER_SRC_FORMAT);
> +	dcss_scaler_write(ch, dst_fmt, DCSS_SCALER_DST_FORMAT);
> +}
> +
> +static void dcss_scaler_res_set(struct dcss_scaler_ch *ch,
> +				int src_xres, int src_yres,
> +				int dst_xres, int dst_yres,
> +				u32 pix_format, enum buffer_format dst_format)
> +{
> +	u32 lsrc_xres, lsrc_yres, csrc_xres, csrc_yres;
> +	u32 ldst_xres, ldst_yres, cdst_xres, cdst_yres;
> +	bool src_is_444 = true;
> +
> +	lsrc_xres = src_xres;
> +	csrc_xres = src_xres;
> +	lsrc_yres = src_yres;
> +	csrc_yres = src_yres;
> +	ldst_xres = dst_xres;
> +	cdst_xres = dst_xres;
> +	ldst_yres = dst_yres;
> +	cdst_yres = dst_yres;
> +
> +	if (pix_format == DRM_FORMAT_UYVY || pix_format == DRM_FORMAT_VYUY ||
> +	    pix_format == DRM_FORMAT_YUYV || pix_format == DRM_FORMAT_YVYU) {
> +		csrc_xres >>= 1;
> +		src_is_444 = false;
> +	} else if (pix_format == DRM_FORMAT_NV12 ||
> +		   pix_format == DRM_FORMAT_NV21) {
> +		csrc_xres >>= 1;
> +		csrc_yres >>= 1;
> +		src_is_444 = false;
> +	}
> +
> +	if (dst_format == BUF_FMT_YUV422)
> +		cdst_xres >>= 1;
> +
> +	/* for 4:4:4 to 4:2:2 conversion, source height should be 1 less */
> +	if (src_is_444 && dst_format == BUF_FMT_YUV422) {
> +		lsrc_yres--;
> +		csrc_yres--;
> +	}
> +
> +	dcss_scaler_write(ch, (((lsrc_yres - 1) << HEIGHT_POS) & HEIGHT_MASK) |
> +			       (((lsrc_xres - 1) << WIDTH_POS) & WIDTH_MASK),
> +			  DCSS_SCALER_SRC_LUM_RES);
> +	dcss_scaler_write(ch, (((csrc_yres - 1) << HEIGHT_POS) & HEIGHT_MASK) |
> +			       (((csrc_xres - 1) << WIDTH_POS) & WIDTH_MASK),
> +			  DCSS_SCALER_SRC_CHR_RES);
> +	dcss_scaler_write(ch, (((ldst_yres - 1) << HEIGHT_POS) & HEIGHT_MASK) |
> +			       (((ldst_xres - 1) << WIDTH_POS) & WIDTH_MASK),
> +			  DCSS_SCALER_DST_LUM_RES);
> +	dcss_scaler_write(ch, (((cdst_yres - 1) << HEIGHT_POS) & HEIGHT_MASK) |
> +			       (((cdst_xres - 1) << WIDTH_POS) & WIDTH_MASK),
> +			  DCSS_SCALER_DST_CHR_RES);
> +}
> +
> +#define max_downscale(ratio)		((ratio) << 16)
> +#define max_upscale(ratio)		((1 << 16) / (ratio))
> +
> +struct dcss_scaler_ratios {
> +	int downscale;
> +	int upscale;
> +};
> +
> +static const struct dcss_scaler_ratios dcss_scaler_ratios[] = {
> +	{max_downscale(3), max_upscale(8)},
> +	{max_downscale(5), max_upscale(8)},
> +	{max_downscale(5), max_upscale(8)},
> +};
> +
> +static void dcss_scaler_fractions_set(struct dcss_scaler_ch *ch,
> +				      int src_xres, int src_yres,
> +				      int dst_xres, int dst_yres,
> +				      u32 src_format, u32 dst_format,
> +				      enum chroma_location src_chroma_loc)
> +{
> +	int src_c_xres, src_c_yres, dst_c_xres, dst_c_yres;
> +	u32 l_vinc, l_hinc, c_vinc, c_hinc;
> +	u32 c_vstart, c_hstart;
> +
> +	src_c_xres = src_xres;
> +	src_c_yres = src_yres;
> +	dst_c_xres = dst_xres;
> +	dst_c_yres = dst_yres;
> +
> +	c_vstart = 0;
> +	c_hstart = 0;
> +
> +	/* adjustments for source chroma location */
> +	if (src_format == BUF_FMT_YUV420) {
> +		/* vertical input chroma position adjustment */
> +		switch (src_chroma_loc) {
> +		case PSC_LOC_HORZ_0_VERT_1_OVER_4:
> +		case PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_4:
> +			/*
> +			 * move chroma up to first luma line
> +			 * (1/4 chroma input line spacing)
> +			 */
> +			c_vstart -= (1 << (PSC_PHASE_FRACTION_BITS - 2));
> +			break;
> +		case PSC_LOC_HORZ_0_VERT_1_OVER_2:
> +		case PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_2:
> +			/*
> +			 * move chroma up to first luma line
> +			 * (1/2 chroma input line spacing)
> +			 */
> +			c_vstart -= (1 << (PSC_PHASE_FRACTION_BITS - 1));
> +			break;
> +		default:
> +			break;
> +		}
> +		/* horizontal input chroma position adjustment */
> +		switch (src_chroma_loc) {
> +		case PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_4:
> +		case PSC_LOC_HORZ_1_OVER_4_VERT_0:
> +		case PSC_LOC_HORZ_1_OVER_4_VERT_1_OVER_2:
> +			/* move chroma left 1/4 chroma input sample spacing */
> +			c_hstart -= (1 << (PSC_PHASE_FRACTION_BITS - 2));
> +			break;
> +		default:
> +			break;
> +		}
> +	}
> +
> +	/* adjustments to chroma resolution */
> +	if (src_format == BUF_FMT_YUV420) {
> +		src_c_xres >>= 1;
> +		src_c_yres >>= 1;
> +	} else if (src_format == BUF_FMT_YUV422) {
> +		src_c_xres >>= 1;
> +	}
> +
> +	if (dst_format == BUF_FMT_YUV422)
> +		dst_c_xres >>= 1;
> +
> +	l_vinc = ((src_yres << 13) + (dst_yres >> 1)) / dst_yres;
> +	c_vinc = ((src_c_yres << 13) + (dst_c_yres >> 1)) / dst_c_yres;
> +	l_hinc = ((src_xres << 13) + (dst_xres >> 1)) / dst_xres;
> +	c_hinc = ((src_c_xres << 13) + (dst_c_xres >> 1)) / dst_c_xres;
> +
> +	/* save chroma start phase */
> +	ch->c_vstart = c_vstart;
> +	ch->c_hstart = c_hstart;
> +
> +	dcss_scaler_write(ch, 0, DCSS_SCALER_V_LUM_START);
> +	dcss_scaler_write(ch, l_vinc, DCSS_SCALER_V_LUM_INC);
> +
> +	dcss_scaler_write(ch, 0, DCSS_SCALER_H_LUM_START);
> +	dcss_scaler_write(ch, l_hinc, DCSS_SCALER_H_LUM_INC);
> +
> +	dcss_scaler_write(ch, c_vstart, DCSS_SCALER_V_CHR_START);
> +	dcss_scaler_write(ch, c_vinc, DCSS_SCALER_V_CHR_INC);
> +
> +	dcss_scaler_write(ch, c_hstart, DCSS_SCALER_H_CHR_START);
> +	dcss_scaler_write(ch, c_hinc, DCSS_SCALER_H_CHR_INC);
> +}
> +
> +int dcss_scaler_get_min_max_ratios(struct dcss_scaler *scl, int ch_num,
> +				   int *min, int *max)
> +{
> +	*min = dcss_scaler_ratios[ch_num].upscale;
> +	*max = dcss_scaler_ratios[ch_num].downscale;
> +
> +	return 0;
> +}
> +
> +static void dcss_scaler_program_5_coef_set(struct dcss_scaler_ch *ch,
> +					   int base_addr,
> +					   int coef[][PSC_NUM_TAPS])
> +{
> +	int i, phase;
> +
> +	for (i = 0; i < PSC_STORED_PHASES; i++) {
> +		dcss_scaler_write(ch, ((coef[i][1] & 0xfff) << 16 |
> +				       (coef[i][2] & 0xfff) << 4  |
> +				       (coef[i][3] & 0xf00) >> 8),
> +				  base_addr + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[i][3] & 0x0ff) << 20 |
> +				       (coef[i][4] & 0xfff) << 8  |
> +				       (coef[i][5] & 0xff0) >> 4),
> +				  base_addr + 0x40 + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[i][5] & 0x00f) << 24),
> +				  base_addr + 0x80 + i * sizeof(u32));
> +	}
> +
> +	/* reverse both phase and tap orderings */
> +	for (phase = (PSC_NUM_PHASES >> 1) - 1;
> +			i < PSC_NUM_PHASES; i++, phase--) {
> +		dcss_scaler_write(ch, ((coef[phase][5] & 0xfff) << 16 |
> +				       (coef[phase][4] & 0xfff) << 4  |
> +				       (coef[phase][3] & 0xf00) >> 8),
> +				  base_addr + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[phase][3] & 0x0ff) << 20 |
> +				       (coef[phase][2] & 0xfff) << 8  |
> +				       (coef[phase][1] & 0xff0) >> 4),
> +				  base_addr + 0x40 + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[phase][1] & 0x00f) << 24),
> +				  base_addr + 0x80 + i * sizeof(u32));
> +	}
> +}
> +
> +static void dcss_scaler_program_7_coef_set(struct dcss_scaler_ch *ch,
> +					   int base_addr,
> +					   int coef[][PSC_NUM_TAPS])
> +{
> +	int i, phase;
> +
> +	for (i = 0; i < PSC_STORED_PHASES; i++) {
> +		dcss_scaler_write(ch, ((coef[i][0] & 0xfff) << 16 |
> +				       (coef[i][1] & 0xfff) << 4  |
> +				       (coef[i][2] & 0xf00) >> 8),
> +				  base_addr + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[i][2] & 0x0ff) << 20 |
> +				       (coef[i][3] & 0xfff) << 8  |
> +				       (coef[i][4] & 0xff0) >> 4),
> +				  base_addr + 0x40 + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[i][4] & 0x00f) << 24 |
> +				       (coef[i][5] & 0xfff) << 12 |
> +				       (coef[i][6] & 0xfff)),
> +				  base_addr + 0x80 + i * sizeof(u32));
> +	}
> +
> +	/* reverse both phase and tap orderings */
> +	for (phase = (PSC_NUM_PHASES >> 1) - 1;
> +			i < PSC_NUM_PHASES; i++, phase--) {
> +		dcss_scaler_write(ch, ((coef[phase][6] & 0xfff) << 16 |
> +				       (coef[phase][5] & 0xfff) << 4  |
> +				       (coef[phase][4] & 0xf00) >> 8),
> +				  base_addr + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[phase][4] & 0x0ff) << 20 |
> +				       (coef[phase][3] & 0xfff) << 8  |
> +				       (coef[phase][2] & 0xff0) >> 4),
> +				  base_addr + 0x40 + i * sizeof(u32));
> +		dcss_scaler_write(ch, ((coef[phase][2] & 0x00f) << 24 |
> +				       (coef[phase][1] & 0xfff) << 12 |
> +				       (coef[phase][0] & 0xfff)),
> +				  base_addr + 0x80 + i * sizeof(u32));
> +	}
> +}
> +
> +static void dcss_scaler_yuv_coef_set(struct dcss_scaler_ch *ch,
> +				     enum buffer_format src_format,
> +				     enum buffer_format dst_format,
> +				     bool use_5_taps,
> +				     int src_xres, int src_yres, int dst_xres,
> +				     int dst_yres)
> +{
> +	int coef[PSC_STORED_PHASES][PSC_NUM_TAPS];
> +	bool program_5_taps = use_5_taps ||
> +			      (dst_format == BUF_FMT_YUV422 &&
> +			       src_format == BUF_FMT_ARGB8888_YUV444);
> +
> +	/* horizontal luma */
> +	dcss_scaler_filter_design(src_xres, dst_xres, false,
> +				  src_xres == dst_xres, coef);
> +	dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_HLUM, coef);
> +
> +	/* vertical luma */
> +	dcss_scaler_filter_design(src_yres, dst_yres, program_5_taps,
> +				  src_yres == dst_yres, coef);
> +
> +	if (program_5_taps)
> +		dcss_scaler_program_5_coef_set(ch, DCSS_SCALER_COEF_VLUM, coef);
> +	else
> +		dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_VLUM, coef);
> +
> +	/* adjust chroma resolution */
> +	if (src_format != BUF_FMT_ARGB8888_YUV444)
> +		src_xres >>= 1;
> +	if (src_format == BUF_FMT_YUV420)
> +		src_yres >>= 1;
> +	if (dst_format != BUF_FMT_ARGB8888_YUV444)
> +		dst_xres >>= 1;
> +	if (dst_format == BUF_FMT_YUV420) /* should not happen */
> +		dst_yres >>= 1;
> +
> +	/* horizontal chroma */
> +	dcss_scaler_filter_design(src_xres, dst_xres, false,
> +				  (src_xres == dst_xres) && (ch->c_hstart == 0),
> +				  coef);
> +
> +	dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_HCHR, coef);
> +
> +	/* vertical chroma */
> +	dcss_scaler_filter_design(src_yres, dst_yres, program_5_taps,
> +				  (src_yres == dst_yres) && (ch->c_vstart == 0),
> +				  coef);
> +	if (program_5_taps)
> +		dcss_scaler_program_5_coef_set(ch, DCSS_SCALER_COEF_VCHR, coef);
> +	else
> +		dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_VCHR, coef);
> +}
> +
> +static void dcss_scaler_rgb_coef_set(struct dcss_scaler_ch *ch,
> +				     int src_xres, int src_yres, int dst_xres,
> +				     int dst_yres)
> +{
> +	int coef[PSC_STORED_PHASES][PSC_NUM_TAPS];
> +
> +	/* horizontal RGB */
> +	dcss_scaler_filter_design(src_xres, dst_xres, false,
> +				  src_xres == dst_xres, coef);
> +	dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_HLUM, coef);
> +
> +	/* vertical RGB */
> +	dcss_scaler_filter_design(src_yres, dst_yres, false,
> +				  src_yres == dst_yres, coef);
> +	dcss_scaler_program_7_coef_set(ch, DCSS_SCALER_COEF_VLUM, coef);
> +}
> +
> +static void dcss_scaler_set_rgb10_order(struct dcss_scaler_ch *ch,
> +					const struct drm_format_info *format)
> +{
> +	u32 a2r10g10b10_format;
> +
> +	if (format->is_yuv)
> +		return;
> +
> +	ch->sdata_ctrl &= ~A2R10G10B10_FORMAT_MASK;
> +
> +	if (format->depth != 30)
> +		return;
> +
> +	switch (format->format) {
> +	case DRM_FORMAT_ARGB2101010:
> +	case DRM_FORMAT_XRGB2101010:
> +		a2r10g10b10_format = 0;
> +		break;
> +
> +	case DRM_FORMAT_ABGR2101010:
> +	case DRM_FORMAT_XBGR2101010:
> +		a2r10g10b10_format = 5;
> +		break;
> +
> +	case DRM_FORMAT_RGBA1010102:
> +	case DRM_FORMAT_RGBX1010102:
> +		a2r10g10b10_format = 6;
> +		break;
> +
> +	case DRM_FORMAT_BGRA1010102:
> +	case DRM_FORMAT_BGRX1010102:
> +		a2r10g10b10_format = 11;
> +		break;
> +
> +	default:
> +		a2r10g10b10_format = 0;
> +		break;
> +	}
> +
> +	ch->sdata_ctrl |= a2r10g10b10_format << A2R10G10B10_FORMAT_POS;
> +}
> +
> +void dcss_scaler_setup(struct dcss_scaler *scl, int ch_num,
> +		       const struct drm_format_info *format,
> +		       int src_xres, int src_yres, int dst_xres, int dst_yres,
> +		       u32 vrefresh_hz)
> +{
> +	struct dcss_scaler_ch *ch = &scl->ch[ch_num];
> +	unsigned int pixel_depth = 0;
> +	bool rtr_8line_en = false;
> +	bool use_5_taps = false;
> +	enum buffer_format src_format = BUF_FMT_ARGB8888_YUV444;
> +	enum buffer_format dst_format = BUF_FMT_ARGB8888_YUV444;
> +	u32 pix_format = format->format;
> +
> +	if (format->is_yuv) {
> +		dcss_scaler_yuv_enable(ch, true);
> +
> +		if (pix_format == DRM_FORMAT_NV12 ||
> +		    pix_format == DRM_FORMAT_NV21) {
> +			rtr_8line_en = true;
> +			src_format = BUF_FMT_YUV420;
> +		} else if (pix_format == DRM_FORMAT_UYVY ||
> +			   pix_format == DRM_FORMAT_VYUY ||
> +			   pix_format == DRM_FORMAT_YUYV ||
> +			   pix_format == DRM_FORMAT_YVYU) {
> +			src_format = BUF_FMT_YUV422;
> +		}
> +
> +		use_5_taps = !rtr_8line_en;
> +	} else {
> +		dcss_scaler_yuv_enable(ch, false);
> +
> +		pixel_depth = format->depth;
> +	}
> +
> +	dcss_scaler_fractions_set(ch, src_xres, src_yres, dst_xres,
> +				  dst_yres, src_format, dst_format,
> +				  PSC_LOC_HORZ_0_VERT_1_OVER_4);
> +
> +	if (format->is_yuv)
> +		dcss_scaler_yuv_coef_set(ch, src_format, dst_format,
> +					 use_5_taps, src_xres, src_yres,
> +					 dst_xres, dst_yres);
> +	else
> +		dcss_scaler_rgb_coef_set(ch, src_xres, src_yres,
> +					 dst_xres, dst_yres);
> +
> +	dcss_scaler_rtr_8lines_enable(ch, rtr_8line_en);
> +	dcss_scaler_bit_depth_set(ch, pixel_depth);
> +	dcss_scaler_set_rgb10_order(ch, format);
> +	dcss_scaler_format_set(ch, src_format, dst_format);
> +	dcss_scaler_res_set(ch, src_xres, src_yres, dst_xres, dst_yres,
> +			    pix_format, dst_format);
> +}
> +
> +/* This function will be called from interrupt context. */
> +void dcss_scaler_write_sclctrl(struct dcss_scaler *scl)
> +{
> +	int chnum;
> +
> +	for (chnum = 0; chnum < 3; chnum++) {
> +		struct dcss_scaler_ch *ch = &scl->ch[chnum];
> +
> +		if (ch->scaler_ctrl_chgd) {
> +			dcss_ctxld_write_irqsafe(scl->ctxld, scl->ctx_id,
> +						 ch->scaler_ctrl,
> +						 ch->base_ofs +
> +						 DCSS_SCALER_CTRL);

Why is this using the _irqsafe variant without any locking? Won't this
lead to potential internal state corruption? dcss_ctxld_write is using
the _irqsave locking variants, so it fine with being called from IRQ
context.

> +			ch->scaler_ctrl_chgd = false;
> +		}
> +	}
> +}
> diff --git a/drivers/gpu/drm/imx/dcss/dcss-ss.c b/drivers/gpu/drm/imx/dcss/dcss-ss.c
> new file mode 100644
> index 00000000..65a05c2
> --- /dev/null
> +++ b/drivers/gpu/drm/imx/dcss/dcss-ss.c
> @@ -0,0 +1,179 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Copyright 2019 NXP.
> + */
> +
> +#include <linux/device.h>
> +
> +#include "dcss-dev.h"
> +
> +#define DCSS_SS_SYS_CTRL			0x00
> +#define   RUN_EN				BIT(0)
> +#define DCSS_SS_DISPLAY				0x10
> +#define   LRC_X_POS				0
> +#define   LRC_X_MASK				GENMASK(12, 0)
> +#define   LRC_Y_POS				16
> +#define   LRC_Y_MASK				GENMASK(28, 16)
> +#define DCSS_SS_HSYNC				0x20
> +#define DCSS_SS_VSYNC				0x30
> +#define   SYNC_START_POS			0
> +#define   SYNC_START_MASK			GENMASK(12, 0)
> +#define   SYNC_END_POS				16
> +#define   SYNC_END_MASK				GENMASK(28, 16)
> +#define   SYNC_POL				BIT(31)
> +#define DCSS_SS_DE_ULC				0x40
> +#define   ULC_X_POS				0
> +#define   ULC_X_MASK				GENMASK(12, 0)
> +#define   ULC_Y_POS				16
> +#define   ULC_Y_MASK				GENMASK(28, 16)
> +#define   ULC_POL				BIT(31)
> +#define DCSS_SS_DE_LRC				0x50
> +#define DCSS_SS_MODE				0x60
> +#define   PIPE_MODE_POS				0
> +#define   PIPE_MODE_MASK			GENMASK(1, 0)
> +#define DCSS_SS_COEFF				0x70
> +#define   HORIZ_A_POS				0
> +#define   HORIZ_A_MASK				GENMASK(3, 0)
> +#define   HORIZ_B_POS				4
> +#define   HORIZ_B_MASK				GENMASK(7, 4)
> +#define   HORIZ_C_POS				8
> +#define   HORIZ_C_MASK				GENMASK(11, 8)
> +#define   HORIZ_H_NORM_POS			12
> +#define   HORIZ_H_NORM_MASK			GENMASK(14, 12)
> +#define   VERT_A_POS				16
> +#define   VERT_A_MASK				GENMASK(19, 16)
> +#define   VERT_B_POS				20
> +#define   VERT_B_MASK				GENMASK(23, 20)
> +#define   VERT_C_POS				24
> +#define   VERT_C_MASK				GENMASK(27, 24)
> +#define   VERT_H_NORM_POS			28
> +#define   VERT_H_NORM_MASK			GENMASK(30, 28)
> +#define DCSS_SS_CLIP_CB				0x80
> +#define DCSS_SS_CLIP_CR				0x90
> +#define   CLIP_MIN_POS				0
> +#define   CLIP_MIN_MASK				GENMASK(9, 0)
> +#define   CLIP_MAX_POS				0
> +#define   CLIP_MAX_MASK				GENMASK(23, 16)
> +#define DCSS_SS_INTER_MODE			0xA0
> +#define   INT_EN				BIT(0)
> +#define   VSYNC_SHIFT				BIT(1)
> +
> +struct dcss_ss {
> +	struct device *dev;
> +	void __iomem *base_reg;
> +	u32 base_ofs;
> +
> +	struct dcss_ctxld *ctxld;
> +	u32 ctx_id;
> +
> +	bool in_use;
> +};
> +
> +static void dcss_ss_write(struct dcss_ss *ss, u32 val, u32 ofs)
> +{
> +	if (!ss->in_use)
> +		dcss_writel(val, ss->base_reg + ofs);
> +
> +	dcss_ctxld_write(ss->ctxld, ss->ctx_id, val,
> +			 ss->base_ofs + ofs);
> +}
> +
> +int dcss_ss_init(struct dcss_dev *dcss, unsigned long ss_base)
> +{
> +	struct dcss_ss *ss;
> +
> +	ss = devm_kzalloc(dcss->dev, sizeof(*ss), GFP_KERNEL);
> +	if (!ss)
> +		return -ENOMEM;
> +
> +	dcss->ss = ss;
> +	ss->dev = dcss->dev;
> +	ss->ctxld = dcss->ctxld;
> +
> +	ss->base_reg = devm_ioremap(dcss->dev, ss_base, SZ_4K);
> +	if (!ss->base_reg) {
> +		dev_err(dcss->dev, "ss: unable to remap ss base\n");
> +		devm_kfree(ss->dev, ss);
> +		return -ENOMEM;
> +	}
> +
> +	ss->base_ofs = ss_base;
> +	ss->ctx_id = CTX_SB_HP;
> +
> +	return 0;
> +}
> +
> +void dcss_ss_exit(struct dcss_ss *ss)
> +{
> +	/* stop SS */
> +	dcss_writel(0, ss->base_reg + DCSS_SS_SYS_CTRL);
> +
> +	if (ss->base_reg)
> +		devm_iounmap(ss->dev, ss->base_reg);
> +
> +	devm_kfree(ss->dev, ss);
> +}
> +
> +void dcss_ss_subsam_set(struct dcss_ss *ss)
> +{
> +	dcss_ss_write(ss, 0x41614161, DCSS_SS_COEFF);
> +	dcss_ss_write(ss, 0, DCSS_SS_MODE);
> +	dcss_ss_write(ss, 0x03ff0000, DCSS_SS_CLIP_CB);
> +	dcss_ss_write(ss, 0x03ff0000, DCSS_SS_CLIP_CR);
> +}
> +
> +void dcss_ss_sync_set(struct dcss_ss *ss, struct videomode *vm,
> +		      bool phsync, bool pvsync)
> +{
> +	u16 lrc_x, lrc_y;
> +	u16 hsync_start, hsync_end;
> +	u16 vsync_start, vsync_end;
> +	u16 de_ulc_x, de_ulc_y;
> +	u16 de_lrc_x, de_lrc_y;
> +
> +	lrc_x = vm->hfront_porch + vm->hback_porch + vm->hsync_len +
> +		vm->hactive - 1;
> +	lrc_y = vm->vfront_porch + vm->vback_porch + vm->vsync_len +
> +		vm->vactive - 1;
> +
> +	dcss_ss_write(ss, (lrc_y << LRC_Y_POS) | lrc_x, DCSS_SS_DISPLAY);
> +
> +	hsync_start = vm->hfront_porch + vm->hback_porch + vm->hsync_len +
> +		      vm->hactive - 1;
> +	hsync_end = vm->hsync_len - 1;
> +
> +	dcss_ss_write(ss, (phsync ? SYNC_POL : 0) |
> +		      ((u32)hsync_end << SYNC_END_POS) | hsync_start,
> +		      DCSS_SS_HSYNC);
> +
> +	vsync_start = vm->vfront_porch - 1;
> +	vsync_end = vm->vfront_porch + vm->vsync_len - 1;
> +
> +	dcss_ss_write(ss, (pvsync ? SYNC_POL : 0) |
> +		      ((u32)vsync_end << SYNC_END_POS) | vsync_start,
> +		      DCSS_SS_VSYNC);
> +
> +	de_ulc_x = vm->hsync_len + vm->hback_porch - 1;
> +	de_ulc_y = vm->vsync_len + vm->vfront_porch + vm->vback_porch;
> +
> +	dcss_ss_write(ss, SYNC_POL | ((u32)de_ulc_y << ULC_Y_POS) | de_ulc_x,
> +		      DCSS_SS_DE_ULC);
> +
> +	de_lrc_x = vm->hsync_len + vm->hback_porch + vm->hactive - 1;
> +	de_lrc_y = vm->vsync_len + vm->vfront_porch + vm->vback_porch +
> +		   vm->vactive - 1;
> +
> +	dcss_ss_write(ss, (de_lrc_y << LRC_Y_POS) | de_lrc_x, DCSS_SS_DE_LRC);
> +}
> +
> +void dcss_ss_enable(struct dcss_ss *ss)
> +{
> +	dcss_ss_write(ss, RUN_EN, DCSS_SS_SYS_CTRL);
> +	ss->in_use = true;
> +}
> +
> +void dcss_ss_disable(struct dcss_ss *ss)
> +{
> +	dcss_ss_write(ss, 0, DCSS_SS_SYS_CTRL);
> +	ss->in_use = false;
> +}