LinuxLists.cc - [PATCH v3 0/3] DaVinci DMA engine conversion

2012-08-23 01:09:39

Subject: [PATCH v3 0/3] DaVinci DMA engine conversion

Changes since v1:
- Add virt-dma support. Better error checks
and simplified descriptor handling.
- Fix support for multiple EDMA controllers
Tested on AM18x EVM with WL12xx on MMC1

Changes since v2:
- Set default Kconfig state to off
- Fix whitespace and indent issues
- Simplify struct device * assignments
- Remove unneeded chcnt in edma_chan_init
- Remove unneeded conditional assignment of prep_slave_sg
- Fix devm_* usage

This series begins the conversion of the DaVinci private EDMA API
implementation to a DMA engine driver and converts two of the three
in-kernel users of the private EDMA API to DMA engine.

The approach taken is similar to the recent OMAP DMA Engine
conversion. The EDMA DMA Engine driver is a wrapper around the existing
private EDMA implementation and registers the platform device within
the driver. This allows the conversion series to stand alone with just
the drivers and no changes to platform code. It also allows peripheral
drivers to continue to use the private EDMA implementation until they
are converted.

The EDMA DMA Engine driver supports slave transfers only at this time. It
is planned to add cyclic transfers in support of audio peripherals.

There are three users of the private EDMA API in the kernel now:
davinci_mmc, spi-davinci, and davinci-mcasp. This series provides DMA
Engine conversions for the davinci_mmc and spi-davinci drivers which
use the supported slave transfers.

This series has been tested on an AM18x EVM and Hawkboard with
driver performance comparable to that of the private EDMA API
implementations. Both MMC0 and MMC1 are tested which handles the
DA850/OMAP-L138/AM18x specific case where MMC1 uses DMA channels on
a second EDMA channel controller. All other platforms have a simpler
design with just a single EDMA channel controller.

For those wanting to easily test this series, I've pushed a branch for
each version to my github tree at https://github.com/ohporter/linux. The
current branch is edma-dmaengine-v3.

After this series, the current plan is to complete the mcasp driver
conversion which includes adding cyclic dma support. This will then
enable the removal and refactoring of the private EDMA API functionality
into the EDMA DMA Engine driver. Since EDMA is also used on the AM33xx
family of parts in mach-omap2/, the plan is to enable this driver on
that platform as well.

Matt Porter (3):
dmaengine: add TI EDMA DMA engine driver
mmc: davinci_mmc: convert to DMA engine API
spi: spi-davinci: convert to DMA engine API

drivers/dma/Kconfig | 10 +
drivers/dma/Makefile | 1 +
drivers/dma/edma.c | 671 ++++++++++++++++++++++++++++++++++++++++
drivers/mmc/host/davinci_mmc.c | 271 +++++-----------
drivers/spi/spi-davinci.c | 292 ++++++++---------
include/linux/edma.h | 29 ++
6 files changed, 923 insertions(+), 351 deletions(-)
create mode 100644 drivers/dma/edma.c
create mode 100644 include/linux/edma.h

--
1.7.9.5

2012-08-23 01:10:45

by Matt Porter

[permalink] [raw]

Subject: [PATCH v3 3/3] spi: spi-davinci: convert to DMA engine API

Removes use of the DaVinci EDMA private DMA API and replaces
it with use of the DMA engine API.

Signed-off-by: Matt Porter <[email protected]>
---
drivers/spi/spi-davinci.c | 292 ++++++++++++++++++++-------------------------
1 file changed, 130 insertions(+), 162 deletions(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 9b2901f..c1ec52d 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -25,13 +25,14 @@
#include <linux/platform_device.h>
#include <linux/err.h>
#include <linux/clk.h>
+#include <linux/dmaengine.h>
#include <linux/dma-mapping.h>
+#include <linux/edma.h>
#include <linux/spi/spi.h>
#include <linux/spi/spi_bitbang.h>
#include <linux/slab.h>

#include <mach/spi.h>
-#include <mach/edma.h>

#define SPI_NO_RESOURCE ((resource_size_t)-1)

@@ -113,14 +114,6 @@
#define SPIDEF 0x4c
#define SPIFMT0 0x50

-/* We have 2 DMA channels per CS, one for RX and one for TX */
-struct davinci_spi_dma {
- int tx_channel;
- int rx_channel;
- int dummy_param_slot;
- enum dma_event_q eventq;
-};
-
/* SPI Controller driver's private data. */
struct davinci_spi {
struct spi_bitbang bitbang;
@@ -134,11 +127,14 @@ struct davinci_spi {

const void *tx;
void *rx;
-#define SPI_TMP_BUFSZ (SMP_CACHE_BYTES + 1)
- u8 rx_tmp_buf[SPI_TMP_BUFSZ];
int rcount;
int wcount;
- struct davinci_spi_dma dma;
+
+ struct dma_chan *dma_rx;
+ struct dma_chan *dma_tx;
+ int dma_rx_chnum;
+ int dma_tx_chnum;
+
struct davinci_spi_platform_data *pdata;

void (*get_rx)(u32 rx_data, struct davinci_spi *);
@@ -496,21 +492,23 @@ out:
return errors;
}

-static void davinci_spi_dma_callback(unsigned lch, u16 status, void *data)
+static void davinci_spi_dma_rx_callback(void *data)
{
- struct davinci_spi *dspi = data;
- struct davinci_spi_dma *dma = &dspi->dma;
+ struct davinci_spi *dspi = (struct davinci_spi *)data;

- edma_stop(lch);
+ dspi->rcount = 0;

- if (status == DMA_COMPLETE) {
- if (lch == dma->rx_channel)
- dspi->rcount = 0;
- if (lch == dma->tx_channel)
- dspi->wcount = 0;
- }
+ if (!dspi->wcount && !dspi->rcount)
+ complete(&dspi->done);
+}

- if ((!dspi->wcount && !dspi->rcount) || (status != DMA_COMPLETE))
+static void davinci_spi_dma_tx_callback(void *data)
+{
+ struct davinci_spi *dspi = (struct davinci_spi *)data;
+
+ dspi->wcount = 0;
+
+ if (!dspi->wcount && !dspi->rcount)
complete(&dspi->done);
}

@@ -526,20 +524,20 @@ static void davinci_spi_dma_callback(unsigned lch, u16 status, void *data)
static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
{
struct davinci_spi *dspi;
- int data_type, ret;
+ int data_type, ret = -ENOMEM;
u32 tx_data, spidat1;
u32 errors = 0;
struct davinci_spi_config *spicfg;
struct davinci_spi_platform_data *pdata;
unsigned uninitialized_var(rx_buf_count);
- struct device *sdev;
+ void *dummy_buf = NULL;
+ struct scatterlist sg_rx, sg_tx;

dspi = spi_master_get_devdata(spi->master);
pdata = dspi->pdata;
spicfg = (struct davinci_spi_config *)spi->controller_data;
if (!spicfg)
spicfg = &davinci_spi_default_cfg;
- sdev = dspi->bitbang.master->dev.parent;

/* convert len to words based on bits_per_word */
data_type = dspi->bytes_per_word[spi->chip_select];
@@ -567,112 +565,83 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
spidat1 |= tx_data & 0xFFFF;
iowrite32(spidat1, dspi->base + SPIDAT1);
} else {
- struct davinci_spi_dma *dma;
- unsigned long tx_reg, rx_reg;
- struct edmacc_param param;
- void *rx_buf;
- int b, c;
-
- dma = &dspi->dma;
-
- tx_reg = (unsigned long)dspi->pbase + SPIDAT1;
- rx_reg = (unsigned long)dspi->pbase + SPIBUF;
-
- /*
- * Transmit DMA setup
- *
- * If there is transmit data, map the transmit buffer, set it
- * as the source of data and set the source B index to data
- * size. If there is no transmit data, set the transmit register
- * as the source of data, and set the source B index to zero.
- *
- * The destination is always the transmit register itself. And
- * the destination never increments.
- */
-
- if (t->tx_buf) {
- t->tx_dma = dma_map_single(&spi->dev, (void *)t->tx_buf,
- t->len, DMA_TO_DEVICE);
- if (dma_mapping_error(&spi->dev, t->tx_dma)) {
- dev_dbg(sdev, "Unable to DMA map %d bytes"
- "TX buffer\n", t->len);
- return -ENOMEM;
- }
- }
-
- /*
- * If number of words is greater than 65535, then we need
- * to configure a 3 dimension transfer. Use the BCNTRLD
- * feature to allow for transfers that aren't even multiples
- * of 65535 (or any other possible b size) by first transferring
- * the remainder amount then grabbing the next N blocks of
- * 65535 words.
- */
-
- c = dspi->wcount / (SZ_64K - 1); /* N 65535 Blocks */
- b = dspi->wcount - c * (SZ_64K - 1); /* Remainder */
- if (b)
- c++;
+ struct dma_slave_config dma_rx_conf = {
+ .direction = DMA_DEV_TO_MEM,
+ .src_addr = (unsigned long)dspi->pbase + SPIBUF,
+ .src_addr_width = data_type,
+ .src_maxburst = 1,
+ };
+ struct dma_slave_config dma_tx_conf = {
+ .direction = DMA_MEM_TO_DEV,
+ .dst_addr = (unsigned long)dspi->pbase + SPIDAT1,
+ .dst_addr_width = data_type,
+ .dst_maxburst = 1,
+ };
+ struct dma_async_tx_descriptor *rxdesc;
+ struct dma_async_tx_descriptor *txdesc;
+ void *buf;
+
+ dummy_buf = kzalloc(t->len, GFP_KERNEL);
+ if (!dummy_buf)
+ goto err_alloc_dummy_buf;
+
+ dmaengine_slave_config(dspi->dma_rx, &dma_rx_conf);
+ dmaengine_slave_config(dspi->dma_tx, &dma_tx_conf);
+
+ sg_init_table(&sg_rx, 1);
+ if (!t->rx_buf)
+ buf = dummy_buf;
else
- b = SZ_64K - 1;
-
- param.opt = TCINTEN | EDMA_TCC(dma->tx_channel);
- param.src = t->tx_buf ? t->tx_dma : tx_reg;
- param.a_b_cnt = b << 16 | data_type;
- param.dst = tx_reg;
- param.src_dst_bidx = t->tx_buf ? data_type : 0;
- param.link_bcntrld = 0xffffffff;
- param.src_dst_cidx = t->tx_buf ? data_type : 0;
- param.ccnt = c;
- edma_write_slot(dma->tx_channel, &param);
- edma_link(dma->tx_channel, dma->dummy_param_slot);
-
- /*
- * Receive DMA setup
- *
- * If there is receive buffer, use it to receive data. If there
- * is none provided, use a temporary receive buffer. Set the
- * destination B index to 0 so effectively only one byte is used
- * in the temporary buffer (address does not increment).
- *
- * The source of receive data is the receive data register. The
- * source address never increments.
- */
-
- if (t->rx_buf) {
- rx_buf = t->rx_buf;
- rx_buf_count = t->len;
- } else {
- rx_buf = dspi->rx_tmp_buf;
- rx_buf_count = sizeof(dspi->rx_tmp_buf);
+ buf = t->rx_buf;
+ t->rx_dma = dma_map_single(&spi->dev, buf,
+ t->len, DMA_FROM_DEVICE);
+ if (!t->rx_dma) {
+ ret = -EFAULT;
+ goto err_rx_map;
}
+ sg_dma_address(&sg_rx) = t->rx_dma;
+ sg_dma_len(&sg_rx) = t->len;

- t->rx_dma = dma_map_single(&spi->dev, rx_buf, rx_buf_count,
- DMA_FROM_DEVICE);
- if (dma_mapping_error(&spi->dev, t->rx_dma)) {
- dev_dbg(sdev, "Couldn't DMA map a %d bytes RX buffer\n",
- rx_buf_count);
- if (t->tx_buf)
- dma_unmap_single(&spi->dev, t->tx_dma, t->len,
- DMA_TO_DEVICE);
- return -ENOMEM;
+ sg_init_table(&sg_tx, 1);
+ if (!t->tx_buf)
+ buf = dummy_buf;
+ else
+ buf = (void *)t->tx_buf;
+ t->tx_dma = dma_map_single(&spi->dev, buf,
+ t->len, DMA_FROM_DEVICE);
+ if (!t->tx_dma) {
+ ret = -EFAULT;
+ goto err_tx_map;
}
-
- param.opt = TCINTEN | EDMA_TCC(dma->rx_channel);
- param.src = rx_reg;
- param.a_b_cnt = b << 16 | data_type;
- param.dst = t->rx_dma;
- param.src_dst_bidx = (t->rx_buf ? data_type : 0) << 16;
- param.link_bcntrld = 0xffffffff;
- param.src_dst_cidx = (t->rx_buf ? data_type : 0) << 16;
- param.ccnt = c;
- edma_write_slot(dma->rx_channel, &param);
+ sg_dma_address(&sg_tx) = t->tx_dma;
+ sg_dma_len(&sg_tx) = t->len;
+
+ rxdesc = dmaengine_prep_slave_sg(dspi->dma_rx,
+ &sg_rx, 1, DMA_DEV_TO_MEM,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!rxdesc)
+ goto err_desc;
+
+ txdesc = dmaengine_prep_slave_sg(dspi->dma_tx,
+ &sg_tx, 1, DMA_MEM_TO_DEV,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!txdesc)
+ goto err_desc;
+
+ rxdesc->callback = davinci_spi_dma_rx_callback;
+ rxdesc->callback_param = (void *)dspi;
+ txdesc->callback = davinci_spi_dma_tx_callback;
+ txdesc->callback_param = (void *)dspi;

if (pdata->cshold_bug)
iowrite16(spidat1 >> 16, dspi->base + SPIDAT1 + 2);

- edma_start(dma->rx_channel);
- edma_start(dma->tx_channel);
+ dmaengine_submit(rxdesc);
+ dmaengine_submit(txdesc);
+
+ dma_async_issue_pending(dspi->dma_rx);
+ dma_async_issue_pending(dspi->dma_tx);
+
set_io_bits(dspi->base + SPIINT, SPIINT_DMA_REQ_EN);
}

@@ -690,15 +659,13 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)

clear_io_bits(dspi->base + SPIINT, SPIINT_MASKALL);
if (spicfg->io_type == SPI_IO_TYPE_DMA) {
-
- if (t->tx_buf)
- dma_unmap_single(&spi->dev, t->tx_dma, t->len,
- DMA_TO_DEVICE);
-
- dma_unmap_single(&spi->dev, t->rx_dma, rx_buf_count,
- DMA_FROM_DEVICE);
-
clear_io_bits(dspi->base + SPIINT, SPIINT_DMA_REQ_EN);
+
+ dma_unmap_single(&spi->dev, t->rx_dma,
+ t->len, DMA_FROM_DEVICE);
+ dma_unmap_single(&spi->dev, t->tx_dma,
+ t->len, DMA_TO_DEVICE);
+ kfree(dummy_buf);
}

clear_io_bits(dspi->base + SPIGCR1, SPIGCR1_SPIENA_MASK);
@@ -716,11 +683,20 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
}

if (dspi->rcount != 0 || dspi->wcount != 0) {
- dev_err(sdev, "SPI data transfer error\n");
+ dev_err(&spi->dev, "SPI data transfer error\n");
return -EIO;
}

return t->len;
+
+err_desc:
+ dma_unmap_single(&spi->dev, t->tx_dma, t->len, DMA_TO_DEVICE);
+err_tx_map:
+ dma_unmap_single(&spi->dev, t->rx_dma, t->len, DMA_FROM_DEVICE);
+err_rx_map:
+ kfree(dummy_buf);
+err_alloc_dummy_buf:
+ return ret;
}

/**
@@ -751,39 +727,33 @@ static irqreturn_t davinci_spi_irq(s32 irq, void *data)

static int davinci_spi_request_dma(struct davinci_spi *dspi)
{
+ dma_cap_mask_t mask;
+ struct device *sdev = dspi->bitbang.master->dev.parent;
int r;
- struct davinci_spi_dma *dma = &dspi->dma;

- r = edma_alloc_channel(dma->rx_channel, davinci_spi_dma_callback, dspi,
- dma->eventq);
- if (r < 0) {
- pr_err("Unable to request DMA channel for SPI RX\n");
- r = -EAGAIN;
+ dma_cap_zero(mask);
+ dma_cap_set(DMA_SLAVE, mask);
+
+ dspi->dma_rx = dma_request_channel(mask, edma_filter_fn,
+ &dspi->dma_rx_chnum);
+ if (!dspi->dma_rx) {
+ dev_err(sdev, "request RX DMA channel failed\n");
+ r = -ENODEV;
goto rx_dma_failed;
}

- r = edma_alloc_channel(dma->tx_channel, davinci_spi_dma_callback, dspi,
- dma->eventq);
- if (r < 0) {
- pr_err("Unable to request DMA channel for SPI TX\n");
- r = -EAGAIN;
+ dspi->dma_tx = dma_request_channel(mask, edma_filter_fn,
+ &dspi->dma_tx_chnum);
+ if (!dspi->dma_tx) {
+ dev_err(sdev, "request TX DMA channel failed\n");
+ r = -ENODEV;
goto tx_dma_failed;
}

- r = edma_alloc_slot(EDMA_CTLR(dma->tx_channel), EDMA_SLOT_ANY);
- if (r < 0) {
- pr_err("Unable to request SPI TX DMA param slot\n");
- r = -EAGAIN;
- goto param_failed;
- }
- dma->dummy_param_slot = r;
- edma_link(dma->dummy_param_slot, dma->dummy_param_slot);
-
return 0;
-param_failed:
- edma_free_channel(dma->tx_channel);
+
tx_dma_failed:
- edma_free_channel(dma->rx_channel);
+ dma_release_channel(dspi->dma_rx);
rx_dma_failed:
return r;
}
@@ -898,9 +868,8 @@ static int __devinit davinci_spi_probe(struct platform_device *pdev)
dspi->bitbang.txrx_bufs = davinci_spi_bufs;
if (dma_rx_chan != SPI_NO_RESOURCE &&
dma_tx_chan != SPI_NO_RESOURCE) {
- dspi->dma.rx_channel = dma_rx_chan;
- dspi->dma.tx_channel = dma_tx_chan;
- dspi->dma.eventq = pdata->dma_event_q;
+ dspi->dma_rx_chnum = dma_rx_chan;
+ dspi->dma_tx_chnum = dma_tx_chan;

ret = davinci_spi_request_dma(dspi);
if (ret)
@@ -955,9 +924,8 @@ static int __devinit davinci_spi_probe(struct platform_device *pdev)
return ret;

free_dma:
- edma_free_channel(dspi->dma.tx_channel);
- edma_free_channel(dspi->dma.rx_channel);
- edma_free_slot(dspi->dma.dummy_param_slot);
+ dma_release_channel(dspi->dma_rx);
+ dma_release_channel(dspi->dma_tx);
free_clk:
clk_disable(dspi->clk);
clk_put(dspi->clk);
--
1.7.9.5

2012-08-23 01:10:50

by Matt Porter

[permalink] [raw]

Subject: [PATCH v3 1/3] dmaengine: add TI EDMA DMA engine driver

Add a DMA engine driver for the TI EDMA controller. This driver
is implemented as a wrapper around the existing DaVinci private
DMA implementation. This approach allows for incremental conversion
of each peripheral driver to the DMA engine API. The EDMA driver
supports slave transfers but does not yet support cyclic transfers.

Signed-off-by: Matt Porter <[email protected]>
---
drivers/dma/Kconfig | 10 +
drivers/dma/Makefile | 1 +
drivers/dma/edma.c | 671 ++++++++++++++++++++++++++++++++++++++++++++++++++
include/linux/edma.h | 29 +++
4 files changed, 711 insertions(+)
create mode 100644 drivers/dma/edma.c
create mode 100644 include/linux/edma.h

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index d06ea29..0351719 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -208,6 +208,16 @@ config SIRF_DMA
help
Enable support for the CSR SiRFprimaII DMA engine.

+config TI_EDMA
+ tristate "TI EDMA support"
+ depends on ARCH_DAVINCI
+ select DMA_ENGINE
+ select DMA_VIRTUAL_CHANNELS
+ default n
+ help
+ Enable support for the TI EDMA controller. This DMA
+ engine is found on TI DaVinci and AM33xx parts.
+
config ARCH_HAS_ASYNC_TX_FIND_CHANNEL
bool

diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index 4cf6b12..f5cf310 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -23,6 +23,7 @@ obj-$(CONFIG_IMX_DMA) += imx-dma.o
obj-$(CONFIG_MXS_DMA) += mxs-dma.o
obj-$(CONFIG_TIMB_DMA) += timb_dma.o
obj-$(CONFIG_SIRF_DMA) += sirf-dma.o
+obj-$(CONFIG_TI_EDMA) += edma.o
obj-$(CONFIG_STE_DMA40) += ste_dma40.o ste_dma40_ll.o
obj-$(CONFIG_TEGRA20_APB_DMA) += tegra20-apb-dma.o
obj-$(CONFIG_PL330_DMA) += pl330.o
diff --git a/drivers/dma/edma.c b/drivers/dma/edma.c
new file mode 100644
index 0000000..05aea3c
--- /dev/null
+++ b/drivers/dma/edma.c
@@ -0,0 +1,671 @@
+/*
+ * TI EDMA DMA engine driver
+ *
+ * Copyright 2012 Texas Instruments
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#include <mach/edma.h>
+
+#include "dmaengine.h"
+#include "virt-dma.h"
+
+/*
+ * This will go away when the private EDMA API is folded
+ * into this driver and the platform device(s) are
+ * instantiated in the arch code. We can only get away
+ * with this simplification because DA8XX may not be built
+ * in the same kernel image with other DaVinci parts. This
+ * avoids having to sprinkle dmaengine driver platform devices
+ * and data throughout all the existing board files.
+ */
+#ifdef CONFIG_ARCH_DAVINCI_DA8XX
+#define EDMA_CTLRS 2
+#define EDMA_CHANS 32
+#else
+#define EDMA_CTLRS 1
+#define EDMA_CHANS 64
+#endif /* CONFIG_ARCH_DAVINCI_DA8XX */
+
+/* Max of 16 segments per channel to conserve PaRAM slots */
+#define MAX_NR_SG 16
+#define EDMA_MAX_SLOTS MAX_NR_SG
+#define EDMA_DESCRIPTORS 16
+
+struct edma_desc {
+ struct virt_dma_desc vdesc;
+ struct list_head node;
+ int absync;
+ int pset_nr;
+ struct edmacc_param pset[0];
+};
+
+struct edma_cc;
+
+struct edma_chan {
+ struct virt_dma_chan vchan;
+ struct list_head node;
+ struct edma_desc *edesc;
+ struct edma_cc *ecc;
+ int ch_num;
+ bool alloced;
+ int slot[EDMA_MAX_SLOTS];
+ dma_addr_t addr;
+ int addr_width;
+ int maxburst;
+};
+
+struct edma_cc {
+ int ctlr;
+ struct dma_device dma_slave;
+ struct edma_chan slave_chans[EDMA_CHANS];
+ int num_slave_chans;
+ int dummy_slot;
+};
+
+static inline struct edma_cc *to_edma_cc(struct dma_device *d)
+{
+ return container_of(d, struct edma_cc, dma_slave);
+}
+
+static inline struct edma_chan *to_edma_chan(struct dma_chan *c)
+{
+ return container_of(c, struct edma_chan, vchan.chan);
+}
+
+static inline struct edma_desc
+*to_edma_desc(struct dma_async_tx_descriptor *tx)
+{
+ return container_of(tx, struct edma_desc, vdesc.tx);
+}
+
+static void edma_desc_free(struct virt_dma_desc *vdesc)
+{
+ kfree(container_of(vdesc, struct edma_desc, vdesc));
+}
+
+/* Dispatch a queued descriptor to the controller (caller holds lock) */
+static void edma_execute(struct edma_chan *echan)
+{
+ struct virt_dma_desc *vdesc = vchan_next_desc(&echan->vchan);
+ struct edma_desc *edesc;
+ int i;
+
+ if (!vdesc) {
+ echan->edesc = NULL;
+ return;
+ }
+
+ list_del(&vdesc->node);
+
+ echan->edesc = edesc = to_edma_desc(&vdesc->tx);
+
+ /* Write descriptor PaRAM set(s) */
+ for (i = 0; i < edesc->pset_nr; i++) {
+ edma_write_slot(echan->slot[i], &edesc->pset[i]);
+ dev_dbg(echan->vchan.chan.device->dev,
+ "\n pset[%d]:\n"
+ " chnum\t%d\n"
+ " slot\t%d\n"
+ " opt\t%08x\n"
+ " src\t%08x\n"
+ " dst\t%08x\n"
+ " abcnt\t%08x\n"
+ " ccnt\t%08x\n"
+ " bidx\t%08x\n"
+ " cidx\t%08x\n"
+ " lkrld\t%08x\n",
+ i, echan->ch_num, echan->slot[i],
+ edesc->pset[i].opt,
+ edesc->pset[i].src,
+ edesc->pset[i].dst,
+ edesc->pset[i].a_b_cnt,
+ edesc->pset[i].ccnt,
+ edesc->pset[i].src_dst_bidx,
+ edesc->pset[i].src_dst_cidx,
+ edesc->pset[i].link_bcntrld);
+ /* Link to the previous slot if not the last set */
+ if (i != (edesc->pset_nr - 1))
+ edma_link(echan->slot[i], echan->slot[i+1]);
+ /* Final pset links to the dummy pset */
+ else
+ edma_link(echan->slot[i], echan->ecc->dummy_slot);
+ }
+
+ edma_start(echan->ch_num);
+}
+
+static int edma_terminate_all(struct edma_chan *echan)
+{
+ unsigned long flags;
+ LIST_HEAD(head);
+
+ spin_lock_irqsave(&echan->vchan.lock, flags);
+
+ /*
+ * Stop DMA activity: we assume the callback will not be called
+ * after edma_dma() returns (even if it does, it will see
+ * echan->edesc is NULL and exit.)
+ */
+ if (echan->edesc) {
+ echan->edesc = NULL;
+ edma_stop(echan->ch_num);
+ }
+
+ vchan_get_all_descriptors(&echan->vchan, &head);
+ spin_unlock_irqrestore(&echan->vchan.lock, flags);
+ vchan_dma_desc_free_list(&echan->vchan, &head);
+
+ return 0;
+}
+
+
+static int edma_slave_config(struct edma_chan *echan,
+ struct dma_slave_config *config)
+{
+ if ((config->src_addr_width > DMA_SLAVE_BUSWIDTH_4_BYTES) ||
+ (config->dst_addr_width > DMA_SLAVE_BUSWIDTH_4_BYTES))
+ return -EINVAL;
+
+ if (config->direction == DMA_MEM_TO_DEV) {
+ if (config->dst_addr)
+ echan->addr = config->dst_addr;
+ if (config->dst_addr_width)
+ echan->addr_width = config->dst_addr_width;
+ if (config->dst_maxburst)
+ echan->maxburst = config->dst_maxburst;
+ } else if (config->direction == DMA_DEV_TO_MEM) {
+ if (config->src_addr)
+ echan->addr = config->src_addr;
+ if (config->src_addr_width)
+ echan->addr_width = config->src_addr_width;
+ if (config->src_maxburst)
+ echan->maxburst = config->src_maxburst;
+ }
+
+ return 0;
+}
+
+static int edma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
+ unsigned long arg)
+{
+ int ret = 0;
+ struct dma_slave_config *config;
+ struct edma_chan *echan = to_edma_chan(chan);
+
+ switch (cmd) {
+ case DMA_TERMINATE_ALL:
+ edma_terminate_all(echan);
+ break;
+ case DMA_SLAVE_CONFIG:
+ config = (struct dma_slave_config *)arg;
+ ret = edma_slave_config(echan, config);
+ break;
+ default:
+ ret = -ENOSYS;
+ }
+
+ return ret;
+}
+
+static struct dma_async_tx_descriptor *edma_prep_slave_sg(
+ struct dma_chan *chan, struct scatterlist *sgl,
+ unsigned int sg_len, enum dma_transfer_direction direction,
+ unsigned long tx_flags, void *context)
+{
+ struct edma_chan *echan = to_edma_chan(chan);
+ struct device *dev = chan->device->dev;
+ struct edma_desc *edesc;
+ struct scatterlist *sg;
+ int i;
+ int acnt, bcnt, ccnt, src, dst, cidx;
+ int src_bidx, dst_bidx, src_cidx, dst_cidx;
+
+ if (unlikely(!echan || !sgl || !sg_len))
+ return NULL;
+
+ if (echan->addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) {
+ dev_err(dev, "Undefined slave buswidth\n");
+ return NULL;
+ }
+
+ if (sg_len > MAX_NR_SG) {
+ dev_err(dev, "Exceeded max SG segments %d > %d\n",
+ sg_len, MAX_NR_SG);
+ return NULL;
+ }
+
+ edesc = kzalloc(sizeof(*edesc) + sg_len *
+ sizeof(edesc->pset[0]), GFP_ATOMIC);
+ if (!edesc) {
+ dev_dbg(dev, "Failed to allocate a descriptor\n");
+ return NULL;
+ }
+
+ edesc->pset_nr = sg_len;
+
+ for_each_sg(sgl, sg, sg_len, i) {
+ /* Allocate a PaRAM slot, if needed */
+ if (echan->slot[i] < 0) {
+ echan->slot[i] =
+ edma_alloc_slot(EDMA_CTLR(echan->ch_num),
+ EDMA_SLOT_ANY);
+ if (echan->slot[i] < 0) {
+ dev_err(dev, "Failed to allocate slot\n");
+ return NULL;
+ }
+ }
+
+ acnt = echan->addr_width;
+
+ /*
+ * If the maxburst is equal to the fifo width, use
+ * A-synced transfers. This allows for large contiguous
+ * buffer transfers using only one PaRAM set.
+ */
+ if (echan->maxburst == 1) {
+ edesc->absync = false;
+ ccnt = sg_dma_len(sg) / acnt / (SZ_64K - 1);
+ bcnt = sg_dma_len(sg) / acnt - ccnt * (SZ_64K - 1);
+ if (bcnt)
+ ccnt++;
+ else
+ bcnt = SZ_64K - 1;
+ cidx = acnt;
+ /*
+ * If maxburst is greater than the fifo address_width,
+ * use AB-synced transfers where A count is the fifo
+ * address_width and B count is the maxburst. In this
+ * case, we are limited to transfers of C count frames
+ * of (address_width * maxburst) where C count is limited
+ * to SZ_64K-1. This places an upper bound on the length
+ * of an SG segment that can be handled.
+ */
+ } else {
+ edesc->absync = true;
+ bcnt = echan->maxburst;
+ ccnt = sg_dma_len(sg) / (acnt * bcnt);
+ if (ccnt > (SZ_64K - 1)) {
+ dev_err(dev, "Exceeded max SG segment size\n");
+ return NULL;
+ }
+ cidx = acnt * bcnt;
+ }
+
+ if (direction == DMA_MEM_TO_DEV) {
+ src = sg_dma_address(sg);
+ dst = echan->addr;
+ src_bidx = acnt;
+ src_cidx = cidx;
+ dst_bidx = 0;
+ dst_cidx = 0;
+ } else {
+ src = echan->addr;
+ dst = sg_dma_address(sg);
+ src_bidx = 0;
+ src_cidx = 0;
+ dst_bidx = acnt;
+ dst_cidx = cidx;
+ }
+
+ edesc->pset[i].opt = EDMA_TCC(EDMA_CHAN_SLOT(echan->ch_num));
+ /* Configure A or AB synchronized transfers */
+ if (edesc->absync)
+ edesc->pset[i].opt |= SYNCDIM;
+ /* If this is the last set, enable completion interrupt flag */
+ if (i == sg_len - 1)
+ edesc->pset[i].opt |= TCINTEN;
+
+ edesc->pset[i].src = src;
+ edesc->pset[i].dst = dst;
+
+ edesc->pset[i].src_dst_bidx = (dst_bidx << 16) | src_bidx;
+ edesc->pset[i].src_dst_cidx = (dst_cidx << 16) | src_cidx;
+
+ edesc->pset[i].a_b_cnt = bcnt << 16 | acnt;
+ edesc->pset[i].ccnt = ccnt;
+ edesc->pset[i].link_bcntrld = 0xffffffff;
+
+ }
+
+ return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
+}
+
+static void edma_callback(unsigned ch_num, u16 ch_status, void *data)
+{
+ struct edma_chan *echan = data;
+ struct device *dev = echan->vchan.chan.device->dev;
+ struct edma_desc *edesc;
+ unsigned long flags;
+
+ /* Stop the channel */
+ edma_stop(echan->ch_num);
+
+ switch (ch_status) {
+ case DMA_COMPLETE:
+ dev_dbg(dev, "transfer complete on channel %d\n", ch_num);
+
+ spin_lock_irqsave(&echan->vchan.lock, flags);
+
+ edesc = echan->edesc;
+ if (edesc) {
+ edma_execute(echan);
+ vchan_cookie_complete(&edesc->vdesc);
+ }
+
+ spin_unlock_irqrestore(&echan->vchan.lock, flags);
+
+ break;
+ case DMA_CC_ERROR:
+ dev_dbg(dev, "transfer error on channel %d\n", ch_num);
+ break;
+ default:
+ break;
+ }
+}
+
+/* Alloc channel resources */
+static int edma_alloc_chan_resources(struct dma_chan *chan)
+{
+ struct edma_chan *echan = to_edma_chan(chan);
+ struct device *dev = chan->device->dev;
+ int ret;
+ int a_ch_num;
+ LIST_HEAD(descs);
+
+ a_ch_num = edma_alloc_channel(echan->ch_num, edma_callback,
+ chan, EVENTQ_DEFAULT);
+
+ if (a_ch_num < 0) {
+ ret = -ENODEV;
+ goto err_no_chan;
+ }
+
+ if (a_ch_num != echan->ch_num) {
+ dev_err(dev, "failed to allocate requested channel %u:%u\n",
+ EDMA_CTLR(echan->ch_num),
+ EDMA_CHAN_SLOT(echan->ch_num));
+ ret = -ENODEV;
+ goto err_wrong_chan;
+ }
+
+ echan->alloced = true;
+ echan->slot[0] = echan->ch_num;
+
+ dev_info(dev, "allocated channel for %u:%u\n",
+ EDMA_CTLR(echan->ch_num), EDMA_CHAN_SLOT(echan->ch_num));
+
+ return 0;
+
+err_wrong_chan:
+ edma_free_channel(a_ch_num);
+err_no_chan:
+ return ret;
+}
+
+/* Free channel resources */
+static void edma_free_chan_resources(struct dma_chan *chan)
+{
+ struct edma_chan *echan = to_edma_chan(chan);
+ struct device *dev = chan->device->dev;
+ int i;
+
+ /* Terminate transfers */
+ edma_stop(echan->ch_num);
+
+ vchan_free_chan_resources(&echan->vchan);
+
+ /* Free EDMA PaRAM slots */
+ for (i = 1; i < EDMA_MAX_SLOTS; i++) {
+ if (echan->slot[i] >= 0) {
+ edma_free_slot(echan->slot[i]);
+ echan->slot[i] = -1;
+ }
+ }
+
+ /* Free EDMA channel */
+ if (echan->alloced) {
+ edma_free_channel(echan->ch_num);
+ echan->alloced = false;
+ }
+
+ dev_info(dev, "freeing channel for %u\n", echan->ch_num);
+}
+
+/* Send pending descriptor to hardware */
+static void edma_issue_pending(struct dma_chan *chan)
+{
+ struct edma_chan *echan = to_edma_chan(chan);
+ unsigned long flags;
+
+ spin_lock_irqsave(&echan->vchan.lock, flags);
+ if (vchan_issue_pending(&echan->vchan) && !echan->edesc)
+ edma_execute(echan);
+ spin_unlock_irqrestore(&echan->vchan.lock, flags);
+}
+
+static size_t edma_desc_size(struct edma_desc *edesc)
+{
+ int i;
+ size_t size;
+
+ if (edesc->absync)
+ for (size = i = 0; i < edesc->pset_nr; i++)
+ size += (edesc->pset[i].a_b_cnt & 0xffff) *
+ (edesc->pset[i].a_b_cnt >> 16) *
+ edesc->pset[i].ccnt;
+ else
+ size = (edesc->pset[0].a_b_cnt & 0xffff) *
+ (edesc->pset[0].a_b_cnt >> 16) +
+ (edesc->pset[0].a_b_cnt & 0xffff) *
+ (SZ_64K - 1) * edesc->pset[0].ccnt;
+
+ return size;
+}
+
+/* Check request completion status */
+static enum dma_status edma_tx_status(struct dma_chan *chan,
+ dma_cookie_t cookie,
+ struct dma_tx_state *txstate)
+{
+ struct edma_chan *echan = to_edma_chan(chan);
+ struct virt_dma_desc *vdesc;
+ enum dma_status ret;
+ unsigned long flags;
+
+ ret = dma_cookie_status(chan, cookie, txstate);
+ if (ret == DMA_SUCCESS || !txstate)
+ return ret;
+
+ spin_lock_irqsave(&echan->vchan.lock, flags);
+ vdesc = vchan_find_desc(&echan->vchan, cookie);
+ if (vdesc) {
+ txstate->residue = edma_desc_size(to_edma_desc(&vdesc->tx));
+ } else if (echan->edesc && echan->edesc->vdesc.tx.cookie == cookie) {
+ struct edma_desc *edesc = echan->edesc;
+ txstate->residue = edma_desc_size(edesc);
+ } else {
+ txstate->residue = 0;
+ }
+ spin_unlock_irqrestore(&echan->vchan.lock, flags);
+
+ return ret;
+}
+
+static void __init edma_chan_init(struct edma_cc *ecc,
+ struct dma_device *dma,
+ struct edma_chan *echans)
+{
+ int i, j;
+
+ for (i = 0; i < EDMA_CHANS; i++) {
+ struct edma_chan *echan = &echans[i];
+ echan->ch_num = EDMA_CTLR_CHAN(ecc->ctlr, i);
+ echan->ecc = ecc;
+ echan->vchan.desc_free = edma_desc_free;
+
+ vchan_init(&echan->vchan, dma);
+
+ INIT_LIST_HEAD(&echan->node);
+ for (j = 0; j < EDMA_MAX_SLOTS; j++)
+ echan->slot[j] = -1;
+ }
+}
+
+static void edma_dma_init(struct edma_cc *ecc, struct dma_device *dma,
+ struct device *dev)
+{
+ dma->device_prep_slave_sg = edma_prep_slave_sg;
+ dma->device_alloc_chan_resources = edma_alloc_chan_resources;
+ dma->device_free_chan_resources = edma_free_chan_resources;
+ dma->device_issue_pending = edma_issue_pending;
+ dma->device_tx_status = edma_tx_status;
+ dma->device_control = edma_control;
+ dma->dev = dev;
+
+ INIT_LIST_HEAD(&dma->channels);
+}
+
+static int __devinit edma_probe(struct platform_device *pdev)
+{
+ struct edma_cc *ecc;
+ int ret;
+
+ ecc = devm_kzalloc(&pdev->dev, sizeof(*ecc), GFP_KERNEL);
+ if (!ecc) {
+ dev_err(&pdev->dev, "Can't allocate controller\n");
+ return -ENOMEM;
+ }
+
+ ecc->ctlr = pdev->id;
+ ecc->dummy_slot = edma_alloc_slot(ecc->ctlr, EDMA_SLOT_ANY);
+ if (ecc->dummy_slot < 0) {
+ dev_err(&pdev->dev, "Can't allocate PaRAM dummy slot\n");
+ return -EIO;
+ }
+
+ dma_cap_zero(ecc->dma_slave.cap_mask);
+ dma_cap_set(DMA_SLAVE, ecc->dma_slave.cap_mask);
+
+ edma_dma_init(ecc, &ecc->dma_slave, &pdev->dev);
+
+ edma_chan_init(ecc, &ecc->dma_slave, ecc->slave_chans);
+
+ ret = dma_async_device_register(&ecc->dma_slave);
+ if (ret)
+ goto err_reg1;
+
+ platform_set_drvdata(pdev, ecc);
+
+ dev_info(&pdev->dev, "TI EDMA DMA engine driver\n");
+
+ return 0;
+
+err_reg1:
+ edma_free_slot(ecc->dummy_slot);
+ return ret;
+}
+
+static int __devexit edma_remove(struct platform_device *pdev)
+{
+ struct device *dev = &pdev->dev;
+ struct edma_cc *ecc = dev_get_drvdata(dev);
+
+ dma_async_device_unregister(&ecc->dma_slave);
+ edma_free_slot(ecc->dummy_slot);
+
+ return 0;
+}
+
+static struct platform_driver edma_driver = {
+ .probe = edma_probe,
+ .remove = __devexit_p(edma_remove),
+ .driver = {
+ .name = "edma-dma-engine",
+ .owner = THIS_MODULE,
+ },
+};
+
+bool edma_filter_fn(struct dma_chan *chan, void *param)
+{
+ if (chan->device->dev->driver == &edma_driver.driver) {
+ struct edma_chan *echan = to_edma_chan(chan);
+ unsigned ch_req = *(unsigned *)param;
+ return ch_req == echan->ch_num;
+ }
+ return false;
+}
+EXPORT_SYMBOL(edma_filter_fn);
+
+static struct platform_device *pdev0, *pdev1;
+
+static const struct platform_device_info edma_dev_info0 = {
+ .name = "edma-dma-engine",
+ .id = 0,
+ .dma_mask = DMA_BIT_MASK(32),
+};
+
+static const struct platform_device_info edma_dev_info1 = {
+ .name = "edma-dma-engine",
+ .id = 1,
+ .dma_mask = DMA_BIT_MASK(32),
+};
+
+static int edma_init(void)
+{
+ int ret = platform_driver_register(&edma_driver);
+
+ if (ret == 0) {
+ pdev0 = platform_device_register_full(&edma_dev_info0);
+ if (IS_ERR(pdev0)) {
+ platform_driver_unregister(&edma_driver);
+ ret = PTR_ERR(pdev0);
+ goto out;
+ }
+ }
+
+ if (EDMA_CTLRS == 2) {
+ pdev1 = platform_device_register_full(&edma_dev_info1);
+ if (IS_ERR(pdev1)) {
+ platform_driver_unregister(&edma_driver);
+ platform_device_unregister(pdev0);
+ ret = PTR_ERR(pdev1);
+ }
+ }
+
+out:
+ return ret;
+}
+subsys_initcall(edma_init);
+
+static void __exit edma_exit(void)
+{
+ platform_device_unregister(pdev0);
+ if (pdev1)
+ platform_device_unregister(pdev1);
+ platform_driver_unregister(&edma_driver);
+}
+module_exit(edma_exit);
+
+MODULE_AUTHOR("Matt Porter <[email protected]>");
+MODULE_DESCRIPTION("TI EDMA DMA engine driver");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/edma.h b/include/linux/edma.h
new file mode 100644
index 0000000..a1307e7
--- /dev/null
+++ b/include/linux/edma.h
@@ -0,0 +1,29 @@
+/*
+ * TI EDMA DMA engine driver
+ *
+ * Copyright 2012 Texas Instruments
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation version 2.
+ *
+ * This program is distributed "as is" WITHOUT ANY WARRANTY of any
+ * kind, whether express or implied; without even the implied warranty
+ * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ */
+#ifndef __LINUX_EDMA_H
+#define __LINUX_EDMA_H
+
+struct dma_chan;
+
+#if defined(CONFIG_TI_EDMA) || defined(CONFIG_TI_EDMA_MODULE)
+bool edma_filter_fn(struct dma_chan *, void *);
+#else
+static inline bool edma_filter_fn(struct dma_chan *chan, void *param)
+{
+ return false;
+}
+#endif
+
+#endif
--
1.7.9.5

2012-08-23 01:10:42

by Matt Porter

[permalink] [raw]

Subject: [PATCH v3 2/3] mmc: davinci_mmc: convert to DMA engine API

Removes use of the DaVinci EDMA private DMA API and replaces
it with use of the DMA engine API.

Signed-off-by: Matt Porter <[email protected]>
---
drivers/mmc/host/davinci_mmc.c | 271 ++++++++++++----------------------------
1 file changed, 82 insertions(+), 189 deletions(-)

diff --git a/drivers/mmc/host/davinci_mmc.c b/drivers/mmc/host/davinci_mmc.c
index 7cf6c62..c5e1eeb 100644
--- a/drivers/mmc/host/davinci_mmc.c
+++ b/drivers/mmc/host/davinci_mmc.c
@@ -30,11 +30,12 @@
#include <linux/io.h>
#include <linux/irq.h>
#include <linux/delay.h>
+#include <linux/dmaengine.h>
#include <linux/dma-mapping.h>
+#include <linux/edma.h>
#include <linux/mmc/mmc.h>

#include <mach/mmc.h>
-#include <mach/edma.h>

/*
* Register Definitions
@@ -200,21 +201,13 @@ struct mmc_davinci_host {
u32 bytes_left;

u32 rxdma, txdma;
+ struct dma_chan *dma_tx;
+ struct dma_chan *dma_rx;
bool use_dma;
bool do_dma;
bool sdio_int;
bool active_request;

- /* Scatterlist DMA uses one or more parameter RAM entries:
- * the main one (associated with rxdma or txdma) plus zero or
- * more links. The entries for a given transfer differ only
- * by memory buffer (address, length) and link field.
- */
- struct edmacc_param tx_template;
- struct edmacc_param rx_template;
- unsigned n_link;
- u32 links[MAX_NR_SG - 1];
-
/* For PIO we walk scatterlists one segment at a time. */
unsigned int sg_len;
struct scatterlist *sg;
@@ -410,153 +403,74 @@ static void mmc_davinci_start_command(struct mmc_davinci_host *host,

static void davinci_abort_dma(struct mmc_davinci_host *host)
{
- int sync_dev;
+ struct dma_chan *sync_dev;

if (host->data_dir == DAVINCI_MMC_DATADIR_READ)
- sync_dev = host->rxdma;
+ sync_dev = host->dma_rx;
else
- sync_dev = host->txdma;
-
- edma_stop(sync_dev);
- edma_clean_channel(sync_dev);
-}
-
-static void
-mmc_davinci_xfer_done(struct mmc_davinci_host *host, struct mmc_data *data);
-
-static void mmc_davinci_dma_cb(unsigned channel, u16 ch_status, void *data)
-{
- if (DMA_COMPLETE != ch_status) {
- struct mmc_davinci_host *host = data;
-
- /* Currently means: DMA Event Missed, or "null" transfer
- * request was seen. In the future, TC errors (like bad
- * addresses) might be presented too.
- */
- dev_warn(mmc_dev(host->mmc), "DMA %s error\n",
- (host->data->flags & MMC_DATA_WRITE)
- ? "write" : "read");
- host->data->error = -EIO;
- mmc_davinci_xfer_done(host, host->data);
- }
-}
-
-/* Set up tx or rx template, to be modified and updated later */
-static void __init mmc_davinci_dma_setup(struct mmc_davinci_host *host,
- bool tx, struct edmacc_param *template)
-{
- unsigned sync_dev;
- const u16 acnt = 4;
- const u16 bcnt = rw_threshold >> 2;
- const u16 ccnt = 0;
- u32 src_port = 0;
- u32 dst_port = 0;
- s16 src_bidx, dst_bidx;
- s16 src_cidx, dst_cidx;
-
- /*
- * A-B Sync transfer: each DMA request is for one "frame" of
- * rw_threshold bytes, broken into "acnt"-size chunks repeated
- * "bcnt" times. Each segment needs "ccnt" such frames; since
- * we tell the block layer our mmc->max_seg_size limit, we can
- * trust (later) that it's within bounds.
- *
- * The FIFOs are read/written in 4-byte chunks (acnt == 4) and
- * EDMA will optimize memory operations to use larger bursts.
- */
- if (tx) {
- sync_dev = host->txdma;
-
- /* src_prt, ccnt, and link to be set up later */
- src_bidx = acnt;
- src_cidx = acnt * bcnt;
-
- dst_port = host->mem_res->start + DAVINCI_MMCDXR;
- dst_bidx = 0;
- dst_cidx = 0;
- } else {
- sync_dev = host->rxdma;
-
- src_port = host->mem_res->start + DAVINCI_MMCDRR;
- src_bidx = 0;
- src_cidx = 0;
-
- /* dst_prt, ccnt, and link to be set up later */
- dst_bidx = acnt;
- dst_cidx = acnt * bcnt;
- }
-
- /*
- * We can't use FIFO mode for the FIFOs because MMC FIFO addresses
- * are not 256-bit (32-byte) aligned. So we use INCR, and the W8BIT
- * parameter is ignored.
- */
- edma_set_src(sync_dev, src_port, INCR, W8BIT);
- edma_set_dest(sync_dev, dst_port, INCR, W8BIT);
+ sync_dev = host->dma_tx;

- edma_set_src_index(sync_dev, src_bidx, src_cidx);
- edma_set_dest_index(sync_dev, dst_bidx, dst_cidx);
-
- edma_set_transfer_params(sync_dev, acnt, bcnt, ccnt, 8, ABSYNC);
-
- edma_read_slot(sync_dev, template);
-
- /* don't bother with irqs or chaining */
- template->opt |= EDMA_CHAN_SLOT(sync_dev) << 12;
+ dmaengine_terminate_all(sync_dev);
}

-static void mmc_davinci_send_dma_request(struct mmc_davinci_host *host,
+static int mmc_davinci_send_dma_request(struct mmc_davinci_host *host,
struct mmc_data *data)
{
- struct edmacc_param *template;
- int channel, slot;
- unsigned link;
- struct scatterlist *sg;
- unsigned sg_len;
- unsigned bytes_left = host->bytes_left;
- const unsigned shift = ffs(rw_threshold) - 1;
+ struct dma_chan *chan;
+ struct dma_async_tx_descriptor *desc;
+ int ret = 0;

if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE) {
- template = &host->tx_template;
- channel = host->txdma;
+ struct dma_slave_config dma_tx_conf = {
+ .direction = DMA_MEM_TO_DEV,
+ .dst_addr = host->mem_res->start + DAVINCI_MMCDXR,
+ .dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
+ .dst_maxburst =
+ rw_threshold / DMA_SLAVE_BUSWIDTH_4_BYTES,
+ };
+ chan = host->dma_tx;
+ dmaengine_slave_config(host->dma_tx, &dma_tx_conf);
+
+ desc = dmaengine_prep_slave_sg(host->dma_tx,
+ data->sg,
+ host->sg_len,
+ DMA_MEM_TO_DEV,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!desc) {
+ dev_dbg(mmc_dev(host->mmc),
+ "failed to allocate DMA TX descriptor");
+ ret = -1;
+ goto out;
+ }
} else {
- template = &host->rx_template;
- channel = host->rxdma;
- }
-
- /* We know sg_len and ccnt will never be out of range because
- * we told the mmc layer which in turn tells the block layer
- * to ensure that it only hands us one scatterlist segment
- * per EDMA PARAM entry. Update the PARAM
- * entries needed for each segment of this scatterlist.
- */
- for (slot = channel, link = 0, sg = data->sg, sg_len = host->sg_len;
- sg_len-- != 0 && bytes_left;
- sg = sg_next(sg), slot = host->links[link++]) {
- u32 buf = sg_dma_address(sg);
- unsigned count = sg_dma_len(sg);
-
- template->link_bcntrld = sg_len
- ? (EDMA_CHAN_SLOT(host->links[link]) << 5)
- : 0xffff;
-
- if (count > bytes_left)
- count = bytes_left;
- bytes_left -= count;
-
- if (host->data_dir == DAVINCI_MMC_DATADIR_WRITE)
- template->src = buf;
- else
- template->dst = buf;
- template->ccnt = count >> shift;
-
- edma_write_slot(slot, template);
+ struct dma_slave_config dma_rx_conf = {
+ .direction = DMA_DEV_TO_MEM,
+ .src_addr = host->mem_res->start + DAVINCI_MMCDRR,
+ .src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES,
+ .src_maxburst =
+ rw_threshold / DMA_SLAVE_BUSWIDTH_4_BYTES,
+ };
+ chan = host->dma_rx;
+ dmaengine_slave_config(host->dma_rx, &dma_rx_conf);
+
+ desc = dmaengine_prep_slave_sg(host->dma_rx,
+ data->sg,
+ host->sg_len,
+ DMA_DEV_TO_MEM,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!desc) {
+ dev_dbg(mmc_dev(host->mmc),
+ "failed to allocate DMA RX descriptor");
+ ret = -1;
+ goto out;
+ }
}

- if (host->version == MMC_CTLR_VERSION_2)
- edma_clear_event(channel);
+ dmaengine_submit(desc);
+ dma_async_issue_pending(chan);

- edma_start(channel);
+out:
+ return ret;
}

static int mmc_davinci_start_dma_transfer(struct mmc_davinci_host *host,
@@ -564,6 +478,7 @@ static int mmc_davinci_start_dma_transfer(struct mmc_davinci_host *host,
{
int i;
int mask = rw_threshold - 1;
+ int ret = 0;

host->sg_len = dma_map_sg(mmc_dev(host->mmc), data->sg, data->sg_len,
((data->flags & MMC_DATA_WRITE)
@@ -583,70 +498,48 @@ static int mmc_davinci_start_dma_transfer(struct mmc_davinci_host *host,
}

host->do_dma = 1;
- mmc_davinci_send_dma_request(host, data);
+ ret = mmc_davinci_send_dma_request(host, data);

- return 0;
+ return ret;
}

static void __init_or_module
davinci_release_dma_channels(struct mmc_davinci_host *host)
{
- unsigned i;
-
if (!host->use_dma)
return;

- for (i = 0; i < host->n_link; i++)
- edma_free_slot(host->links[i]);
-
- edma_free_channel(host->txdma);
- edma_free_channel(host->rxdma);
+ dma_release_channel(host->dma_tx);
+ dma_release_channel(host->dma_rx);
}

static int __init davinci_acquire_dma_channels(struct mmc_davinci_host *host)
{
- u32 link_size;
- int r, i;
-
- /* Acquire master DMA write channel */
- r = edma_alloc_channel(host->txdma, mmc_davinci_dma_cb, host,
- EVENTQ_DEFAULT);
- if (r < 0) {
- dev_warn(mmc_dev(host->mmc), "alloc %s channel err %d\n",
- "tx", r);
- return r;
- }
- mmc_davinci_dma_setup(host, true, &host->tx_template);
-
- /* Acquire master DMA read channel */
- r = edma_alloc_channel(host->rxdma, mmc_davinci_dma_cb, host,
- EVENTQ_DEFAULT);
- if (r < 0) {
- dev_warn(mmc_dev(host->mmc), "alloc %s channel err %d\n",
- "rx", r);
- goto free_master_write;
+ int r;
+ dma_cap_mask_t mask;
+
+ dma_cap_zero(mask);
+ dma_cap_set(DMA_SLAVE, mask);
+
+ host->dma_tx =
+ dma_request_channel(mask, edma_filter_fn, &host->txdma);
+ if (!host->dma_tx) {
+ dev_err(mmc_dev(host->mmc), "Can't get dma_tx channel\n");
+ return -ENODEV;
}
- mmc_davinci_dma_setup(host, false, &host->rx_template);

- /* Allocate parameter RAM slots, which will later be bound to a
- * channel as needed to handle a scatterlist.
- */
- link_size = min_t(unsigned, host->nr_sg, ARRAY_SIZE(host->links));
- for (i = 0; i < link_size; i++) {
- r = edma_alloc_slot(EDMA_CTLR(host->txdma), EDMA_SLOT_ANY);
- if (r < 0) {
- dev_dbg(mmc_dev(host->mmc), "dma PaRAM alloc --> %d\n",
- r);
- break;
- }
- host->links[i] = r;
+ host->dma_rx =
+ dma_request_channel(mask, edma_filter_fn, &host->rxdma);
+ if (!host->dma_rx) {
+ dev_err(mmc_dev(host->mmc), "Can't get dma_rx channel\n");
+ r = -ENODEV;
+ goto free_master_write;
}
- host->n_link = i;

return 0;

free_master_write:
- edma_free_channel(host->txdma);
+ dma_release_channel(host->dma_tx);

return r;
}
@@ -1359,7 +1252,7 @@ static int __init davinci_mmcsd_probe(struct platform_device *pdev)
* Each hw_seg uses one EDMA parameter RAM slot, always one
* channel and then usually some linked slots.
*/
- mmc->max_segs = 1 + host->n_link;
+ mmc->max_segs = MAX_NR_SG;

/* EDMA limit per hw segment (one or two MBytes) */
mmc->max_seg_size = MAX_CCNT * rw_threshold;
--
1.7.9.5

2012-08-23 06:01:32

On Fri, 2012-08-31 at 22:32 +0530, Vinod Koul wrote:
> On Fri, 2012-08-31 at 22:01 +0530, Sekhar Nori wrote:
> > >> Yes, this was the problem. Since the SPI driver now depends on
> > >> CONFIG_TI_EDMA for basic operation may be select CONFIG_TI_EDMA in
> > >> Kconfig if SPI is enabled? That should do until the fallback to PIO
> > is
> > >> implemented.
> > > Then this should be enabled for SPI. Care to send a patch
> >
> > By 'this' you mean the Kconfig select? Then there should be no need of
> > a
> > new patch for this. It can be part of this patch itself, no?
> Either way is okay for me.
And am in good mood :)

Applied all with this as well

>From b5f14330590118e6a0659255476c0f24ab681e05 Mon Sep 17 00:00:00 2001
From: Vinod Koul <[email protected]>
Date: Sat, 1 Sep 2012 06:29:22 +0530
Subject: [PATCH] spi: davici - make davinci select edma

Reported-by: Sekhar Nori <[email protected]>
Signed-off-by: Vinod Koul <[email protected]>
---
drivers/spi/Kconfig | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/spi/Kconfig b/drivers/spi/Kconfig
index 5f84b55..b9fb9a1 100644
--- a/drivers/spi/Kconfig
+++ b/drivers/spi/Kconfig
@@ -134,6 +134,7 @@ config SPI_DAVINCI
tristate "Texas Instruments DaVinci/DA8x/OMAP-L/AM1x SoC SPI
controller"
depends on ARCH_DAVINCI
select SPI_BITBANG
+ select TI_EDMA
help
SPI master controller for DaVinci/DA8x/OMAP-L/AM1x SPI modules.

--
1.7.9.5

--
~Vinod Koul
Intel Corp.