[PATCH v3 0/6] Enable DMA for daVinci SPI controller

[PATCH v3 0/6] Enable DMA for daVinci SPI controller

[Frode Isaksen]

Enable DMA for the daVinci SPI using the SPI framework DMA mapping
function. 
Add fixes caused by large number of SG entries.
Do not allow vmalloc'ed buffers for DMA.
Also, add some extra options to the spi-loopback-test
module to catch errors caused by many SG entries
and VIVT cache.
This has been tested using spidev_test, spi-loopback-test and
mounting and read/write UBIFS volume over SPI flash.

v2 changes:
- Use SPI_MASTER_MUST_ flag(s) for dummy buffer allocation (thanks
Mark Brown).
- For rx-only transfers, use rx buffer as dummy tx buffer to avoid
errors with more than 20 SG entries.
- Improved comments and use #define for contants.
- Abandon patches to set loopback and limit transfer size in
spi-loopback-test.

v3 changes:
- Disable DMA for vmalloc'ed buffers.
- Reworked 'can_dma' function.
- Add option to skip a check in spi-loopback-test to catch VIVT
cache errors.

Thanks,
Frode

[PATCH v3 1/6] spi: davinci: Use SPI framework to handle DMA mapping

[Frode Isaksen]

From: Fabien Parent <fparent@baylibre.com>

Uppers layers like MTD can pass vmalloc'd buffers to the SPI driver,
and the current implementation will fail to map these kind of buffers.
The SPI framework is able to detect the best way to handle and map
buffers.
This commit updates the davinci SPI driver in order to use the SPI
framework to handle the DMA mapping of buffers coming from an upper
layer.

Signed-off-by: Fabien Parent <fparent@baylibre.com>
Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-davinci.c | 69 +++++++++++++----------------------------------
 1 file changed, 18 insertions(+), 51 deletions(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 02fb967..164cc71 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -467,6 +467,19 @@ static void davinci_spi_cleanup(struct spi_device *spi)
 		kfree(spicfg);
 }
 
+static bool davinci_spi_can_dma(struct spi_master *master,
+				struct spi_device *spi,
+				struct spi_transfer *xfer)
+{
+	struct davinci_spi_config *spicfg = spi->controller_data;
+	bool can_dma = false;
+
+	if (spicfg)
+		can_dma = spicfg->io_type == SPI_IO_TYPE_DMA;
+
+	return can_dma;
+}
+
 static int davinci_spi_check_error(struct davinci_spi *dspi, int int_status)
 {
 	struct device *sdev = dspi->bitbang.master->dev.parent;
@@ -581,8 +594,6 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 	struct davinci_spi_config *spicfg;
 	struct davinci_spi_platform_data *pdata;
 	unsigned uninitialized_var(rx_buf_count);
-	void *dummy_buf = NULL;
-	struct scatterlist sg_rx, sg_tx;
 
 	dspi = spi_master_get_devdata(spi->master);
 	pdata = &dspi->pdata;
@@ -630,51 +641,18 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 		};
 		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
-			buf = t->rx_buf;
-		t->rx_dma = dma_map_single(&spi->dev, buf,
-				t->len, DMA_FROM_DEVICE);
-		if (dma_mapping_error(&spi->dev, !t->rx_dma)) {
-			ret = -EFAULT;
-			goto err_rx_map;
-		}
-		sg_dma_address(&sg_rx) = t->rx_dma;
-		sg_dma_len(&sg_rx) = t->len;
-
-		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_TO_DEVICE);
-		if (dma_mapping_error(&spi->dev, t->tx_dma)) {
-			ret = -EFAULT;
-			goto err_tx_map;
-		}
-		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,
+				t->rx_sg.sgl, t->rx_sg.nents, 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,
+				t->tx_sg.sgl, t->tx_sg.nents, DMA_MEM_TO_DEV,
 				DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
 		if (!txdesc)
 			goto err_desc;
@@ -710,16 +688,9 @@ 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 (spicfg->io_type == SPI_IO_TYPE_DMA)
 		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);
 	set_io_bits(dspi->base + SPIGCR1, SPIGCR1_POWERDOWN_MASK);
 
@@ -742,12 +713,6 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 	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;
 }
 
@@ -988,8 +953,10 @@ static int davinci_spi_probe(struct platform_device *pdev)
 	master->bus_num = pdev->id;
 	master->num_chipselect = pdata->num_chipselect;
 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(2, 16);
+	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
 	master->setup = davinci_spi_setup;
 	master->cleanup = davinci_spi_cleanup;
+	master->can_dma = davinci_spi_can_dma;
 
 	dspi->bitbang.chipselect = davinci_spi_chipselect;
 	dspi->bitbang.setup_transfer = davinci_spi_setup_transfer;
-- 
2.7.4

Applied "spi: davinci: Use SPI framework to handle DMA mapping" to the spi tree

[Mark Brown]

The patch

   spi: davinci: Use SPI framework to handle DMA mapping

has been applied to the spi tree at

   git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git 

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From 8aedbf580d21121d2a032e4c8ea12d8d2d85e275 Mon Sep 17 00:00:00 2001
From: Fabien Parent <fparent@baylibre.com>
Date: Thu, 23 Feb 2017 19:01:56 +0100
Subject: [PATCH] spi: davinci: Use SPI framework to handle DMA mapping

Uppers layers like MTD can pass vmalloc'd buffers to the SPI driver,
and the current implementation will fail to map these kind of buffers.
The SPI framework is able to detect the best way to handle and map
buffers.
This commit updates the davinci SPI driver in order to use the SPI
framework to handle the DMA mapping of buffers coming from an upper
layer.

Signed-off-by: Fabien Parent <fparent@baylibre.com>
Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-davinci.c | 69 +++++++++++++----------------------------------
 1 file changed, 18 insertions(+), 51 deletions(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 02fb96797ac8..164cc719be54 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -467,6 +467,19 @@ static void davinci_spi_cleanup(struct spi_device *spi)
 		kfree(spicfg);
 }
 
+static bool davinci_spi_can_dma(struct spi_master *master,
+				struct spi_device *spi,
+				struct spi_transfer *xfer)
+{
+	struct davinci_spi_config *spicfg = spi->controller_data;
+	bool can_dma = false;
+
+	if (spicfg)
+		can_dma = spicfg->io_type == SPI_IO_TYPE_DMA;
+
+	return can_dma;
+}
+
 static int davinci_spi_check_error(struct davinci_spi *dspi, int int_status)
 {
 	struct device *sdev = dspi->bitbang.master->dev.parent;
@@ -581,8 +594,6 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 	struct davinci_spi_config *spicfg;
 	struct davinci_spi_platform_data *pdata;
 	unsigned uninitialized_var(rx_buf_count);
-	void *dummy_buf = NULL;
-	struct scatterlist sg_rx, sg_tx;
 
 	dspi = spi_master_get_devdata(spi->master);
 	pdata = &dspi->pdata;
@@ -630,51 +641,18 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 		};
 		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
-			buf = t->rx_buf;
-		t->rx_dma = dma_map_single(&spi->dev, buf,
-				t->len, DMA_FROM_DEVICE);
-		if (dma_mapping_error(&spi->dev, !t->rx_dma)) {
-			ret = -EFAULT;
-			goto err_rx_map;
-		}
-		sg_dma_address(&sg_rx) = t->rx_dma;
-		sg_dma_len(&sg_rx) = t->len;
-
-		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_TO_DEVICE);
-		if (dma_mapping_error(&spi->dev, t->tx_dma)) {
-			ret = -EFAULT;
-			goto err_tx_map;
-		}
-		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,
+				t->rx_sg.sgl, t->rx_sg.nents, 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,
+				t->tx_sg.sgl, t->tx_sg.nents, DMA_MEM_TO_DEV,
 				DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
 		if (!txdesc)
 			goto err_desc;
@@ -710,16 +688,9 @@ 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 (spicfg->io_type == SPI_IO_TYPE_DMA)
 		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);
 	set_io_bits(dspi->base + SPIGCR1, SPIGCR1_POWERDOWN_MASK);
 
@@ -742,12 +713,6 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 	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;
 }
 
@@ -988,8 +953,10 @@ static int davinci_spi_probe(struct platform_device *pdev)
 	master->bus_num = pdev->id;
 	master->num_chipselect = pdata->num_chipselect;
 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(2, 16);
+	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
 	master->setup = davinci_spi_setup;
 	master->cleanup = davinci_spi_cleanup;
+	master->can_dma = davinci_spi_can_dma;
 
 	dspi->bitbang.chipselect = davinci_spi_chipselect;
 	dspi->bitbang.setup_transfer = davinci_spi_setup_transfer;
-- 
2.11.0

[PATCH v3 2/6] spi: davinci: enable DMA when channels are defined in DT

[Frode Isaksen]

From: Fabien Parent <fparent@baylibre.com>

When booting with DT the SPI driver is always using
the SPI_IO_TYPE_INTR mode to transfer data even if DMA channels are
defined in the DT.

This commit changes the behaviour to select the SPI_IO_TYPE_DMA mode
if DMA channels are defined in the DT and will keep SPI_IO_TYPE_INTR
if the channels are not defined in it.

Signed-off-by: Fabien Parent <fparent@baylibre.com>
Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-davinci.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 164cc71..1e24395 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -389,6 +389,7 @@ static int davinci_spi_of_setup(struct spi_device *spi)
 {
 	struct davinci_spi_config *spicfg = spi->controller_data;
 	struct device_node *np = spi->dev.of_node;
+	struct davinci_spi *dspi = spi_master_get_devdata(spi->master);
 	u32 prop;
 
 	if (spicfg == NULL && np) {
@@ -400,6 +401,9 @@ static int davinci_spi_of_setup(struct spi_device *spi)
 		if (!of_property_read_u32(np, "ti,spi-wdelay", &prop))
 			spicfg->wdelay = (u8)prop;
 		spi->controller_data = spicfg;
+
+		if (dspi->dma_rx && dspi->dma_tx)
+			spicfg->io_type = SPI_IO_TYPE_DMA;
 	}
 
 	return 0;
-- 
2.7.4

[PATCH v3 3/6] spi: davinci: use rx buffer as dummy tx buffer

[Frode Isaksen]

When doing rx-only transfer, the transfer will fail
if the number of SG entries exceeds 20.
This happens because the eDMA DMA engine is limited
to 20 SG entries in one transaction, and when the
DMA transcation is resumed (which takes > 150us),
rx errors occurs because the slave is still transmitting.
Fix this by using the rx buffer as the dummy tx buffer,
so that resuming the rx transcation happens at the same
time as resuming the tx transcation.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-davinci.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 1e24395..ca12216 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -655,6 +655,12 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 		if (!rxdesc)
 			goto err_desc;
 
+		if (!t->tx_buf) {
+			/* use rx buffer as dummy tx buffer */
+			t->tx_sg.sgl = t->rx_sg.sgl;
+			t->tx_sg.nents = t->rx_sg.nents;
+		}
+
 		txdesc = dmaengine_prep_slave_sg(dspi->dma_tx,
 				t->tx_sg.sgl, t->tx_sg.nents, DMA_MEM_TO_DEV,
 				DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
@@ -957,7 +963,7 @@ static int davinci_spi_probe(struct platform_device *pdev)
 	master->bus_num = pdev->id;
 	master->num_chipselect = pdata->num_chipselect;
 	master->bits_per_word_mask = SPI_BPW_RANGE_MASK(2, 16);
-	master->flags = (SPI_MASTER_MUST_RX | SPI_MASTER_MUST_TX);
+	master->flags = SPI_MASTER_MUST_RX;
 	master->setup = davinci_spi_setup;
 	master->cleanup = davinci_spi_cleanup;
 	master->can_dma = davinci_spi_can_dma;
-- 
2.7.4

[PATCH v3 3/6] spi: davinci: use rx buffer as dummy tx buffer

[Mark Brown]

On Thu, Feb 23, 2017 at 07:01:58PM +0100, Frode Isaksen wrote:
> When doing rx-only transfer, the transfer will fail
> if the number of SG entries exceeds 20.
> This happens because the eDMA DMA engine is limited
> to 20 SG entries in one transaction, and when the
> DMA transcation is resumed (which takes > 150us),
> rx errors occurs because the slave is still transmitting.
> Fix this by using the rx buffer as the dummy tx buffer,
> so that resuming the rx transcation happens at the same
> time as resuming the tx transcation.

Please submit a followup patch adding a comment explaing what's going
in the actual code as someone's going to clean this up at some point.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20170315/2cfc400d/attachment.sig>

[PATCH v3 4/6] spi: davinci: do not use DMA if transfer length is less than 16

[Frode Isaksen]

Higher bitrate and lower CPU load if using PIO in this case.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-davinci.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index ca12216..75c658e 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -109,6 +109,8 @@
 #define SPIDEF		0x4c
 #define SPIFMT0		0x50
 
+#define DMA_MIN_BYTES	16
+
 /* SPI Controller driver's private data. */
 struct davinci_spi {
 	struct spi_bitbang	bitbang;
@@ -479,7 +481,8 @@ static bool davinci_spi_can_dma(struct spi_master *master,
 	bool can_dma = false;
 
 	if (spicfg)
-		can_dma = spicfg->io_type == SPI_IO_TYPE_DMA;
+		can_dma = (spicfg->io_type == SPI_IO_TYPE_DMA) &&
+			(xfer->len >= DMA_MIN_BYTES);
 
 	return can_dma;
 }
@@ -620,10 +623,9 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 
 	reinit_completion(&dspi->done);
 
-	if (spicfg->io_type == SPI_IO_TYPE_INTR)
-		set_io_bits(dspi->base + SPIINT, SPIINT_MASKINT);
-
-	if (spicfg->io_type != SPI_IO_TYPE_DMA) {
+	if (!davinci_spi_can_dma(spi->master, spi, t)) {
+		if (spicfg->io_type != SPI_IO_TYPE_POLL)
+			set_io_bits(dspi->base + SPIINT, SPIINT_MASKINT);
 		/* start the transfer */
 		dspi->wcount--;
 		tx_data = dspi->get_tx(dspi);
@@ -698,7 +700,7 @@ 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 (davinci_spi_can_dma(spi->master, spi, t))
 		clear_io_bits(dspi->base + SPIINT, SPIINT_DMA_REQ_EN);
 
 	clear_io_bits(dspi->base + SPIGCR1, SPIGCR1_SPIENA_MASK);
-- 
2.7.4

Applied "spi: davinci: do not use DMA if transfer length is less than 16" to the spi tree

[Mark Brown]

The patch

   spi: davinci: do not use DMA if transfer length is less than 16

has been applied to the spi tree at

   git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git 

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From 0718b764880434ac7a5b7c0f5cb2c805c589a807 Mon Sep 17 00:00:00 2001
From: Frode Isaksen <fisaksen@baylibre.com>
Date: Thu, 23 Feb 2017 19:01:59 +0100
Subject: [PATCH] spi: davinci: do not use DMA if transfer length is less than
 16

Higher bitrate and lower CPU load if using PIO in this case.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-davinci.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index ca122165a3c6..75c658e4e487 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -109,6 +109,8 @@
 #define SPIDEF		0x4c
 #define SPIFMT0		0x50
 
+#define DMA_MIN_BYTES	16
+
 /* SPI Controller driver's private data. */
 struct davinci_spi {
 	struct spi_bitbang	bitbang;
@@ -479,7 +481,8 @@ static bool davinci_spi_can_dma(struct spi_master *master,
 	bool can_dma = false;
 
 	if (spicfg)
-		can_dma = spicfg->io_type == SPI_IO_TYPE_DMA;
+		can_dma = (spicfg->io_type == SPI_IO_TYPE_DMA) &&
+			(xfer->len >= DMA_MIN_BYTES);
 
 	return can_dma;
 }
@@ -620,10 +623,9 @@ static int davinci_spi_bufs(struct spi_device *spi, struct spi_transfer *t)
 
 	reinit_completion(&dspi->done);
 
-	if (spicfg->io_type == SPI_IO_TYPE_INTR)
-		set_io_bits(dspi->base + SPIINT, SPIINT_MASKINT);
-
-	if (spicfg->io_type != SPI_IO_TYPE_DMA) {
+	if (!davinci_spi_can_dma(spi->master, spi, t)) {
+		if (spicfg->io_type != SPI_IO_TYPE_POLL)
+			set_io_bits(dspi->base + SPIINT, SPIINT_MASKINT);
 		/* start the transfer */
 		dspi->wcount--;
 		tx_data = dspi->get_tx(dspi);
@@ -698,7 +700,7 @@ 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 (davinci_spi_can_dma(spi->master, spi, t))
 		clear_io_bits(dspi->base + SPIINT, SPIINT_DMA_REQ_EN);
 
 	clear_io_bits(dspi->base + SPIGCR1, SPIGCR1_SPIENA_MASK);
-- 
2.11.0

[PATCH v3 5/6] spi: davinci: do not use DMA for vmalloc'ed buffers

[Frode Isaksen]

Using vmalloc'ed buffers will fail since daVinci has
VIVT cache and only the kernel lowmem virtual address
is invalidated/flushed when performing DMA.
The virtual address returned from vmalloc() is not
invalidated/flushed and may contain stale data when
returning from spi_sync().
Fixes errors when running UBIFS over SPI NOR.
Revert this when all upper layer users of vmalloc'ed
buffers sent to SPI handles cache flushing/invalidating.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-davinci.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 75c658e..f37bbdd 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -482,7 +482,9 @@ static bool davinci_spi_can_dma(struct spi_master *master,
 
 	if (spicfg)
 		can_dma = (spicfg->io_type == SPI_IO_TYPE_DMA) &&
-			(xfer->len >= DMA_MIN_BYTES);
+			(xfer->len >= DMA_MIN_BYTES) &&
+			!is_vmalloc_addr(xfer->rx_buf) &&
+			!is_vmalloc_addr(xfer->tx_buf);
 
 	return can_dma;
 }
-- 
2.7.4

Applied "spi: davinci: do not use DMA for vmalloc'ed buffers" to the spi tree

[Mark Brown]

The patch

   spi: davinci: do not use DMA for vmalloc'ed buffers

has been applied to the spi tree at

   git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git 

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From 4dd9becbce4f10009322c3e2297f9db3ace94a10 Mon Sep 17 00:00:00 2001
From: Frode Isaksen <fisaksen@baylibre.com>
Date: Thu, 23 Feb 2017 19:02:00 +0100
Subject: [PATCH] spi: davinci: do not use DMA for vmalloc'ed buffers

Using vmalloc'ed buffers will fail since daVinci has
VIVT cache and only the kernel lowmem virtual address
is invalidated/flushed when performing DMA.
The virtual address returned from vmalloc() is not
invalidated/flushed and may contain stale data when
returning from spi_sync().
Fixes errors when running UBIFS over SPI NOR.
Revert this when all upper layer users of vmalloc'ed
buffers sent to SPI handles cache flushing/invalidating.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-davinci.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/spi/spi-davinci.c b/drivers/spi/spi-davinci.c
index 75c658e4e487..f37bbdd18d61 100644
--- a/drivers/spi/spi-davinci.c
+++ b/drivers/spi/spi-davinci.c
@@ -482,7 +482,9 @@ static bool davinci_spi_can_dma(struct spi_master *master,
 
 	if (spicfg)
 		can_dma = (spicfg->io_type == SPI_IO_TYPE_DMA) &&
-			(xfer->len >= DMA_MIN_BYTES);
+			(xfer->len >= DMA_MIN_BYTES) &&
+			!is_vmalloc_addr(xfer->rx_buf) &&
+			!is_vmalloc_addr(xfer->tx_buf);
 
 	return can_dma;
 }
-- 
2.11.0

[PATCH v3 6/6] spi: loopback-test: add option to use vmalloc'ed buffers

[Frode Isaksen]

Using vmalloc'ed buffers will use one SG entry for each page,
that may provoke DMA errors for large transfers.
Also vmalloc'ed buffers may cause errors on CPU's with VIVT cache.
Add this option to catch these errors when testing.
Note that to catch VIVT cache errors, checking the rx range
has to be disabled, so this option has been added as well.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
---
 drivers/spi/spi-loopback-test.c | 34 +++++++++++++++++++++++++++-------
 1 file changed, 27 insertions(+), 7 deletions(-)

diff --git a/drivers/spi/spi-loopback-test.c b/drivers/spi/spi-loopback-test.c
index 50e620f..3596051 100644
--- a/drivers/spi/spi-loopback-test.c
+++ b/drivers/spi/spi-loopback-test.c
@@ -55,6 +55,18 @@ module_param(run_only_test, int, 0);
 MODULE_PARM_DESC(run_only_test,
 		 "only run the test with this number (0-based !)");
 
+/* use vmalloc'ed buffers */
+int use_vmalloc;
+module_param(use_vmalloc, int, 0644);
+MODULE_PARM_DESC(use_vmalloc,
+		 "use vmalloc'ed buffers instead of kmalloc'ed");
+
+/* check rx ranges */
+int check_ranges = 1;
+module_param(check_ranges, int, 0644);
+MODULE_PARM_DESC(check_ranges,
+		 "checks rx_buffer pattern are valid");
+
 /* the actual tests to execute */
 static struct spi_test spi_tests[] = {
 	{
@@ -492,9 +504,11 @@ static int spi_test_check_loopback_result(struct spi_device *spi,
 	int ret;
 
 	/* checks rx_buffer pattern are valid with loopback or without */
-	ret = spi_check_rx_ranges(spi, msg, rx);
-	if (ret)
-		return ret;
+	if (check_ranges) {
+		ret = spi_check_rx_ranges(spi, msg, rx);
+		if (ret)
+			return ret;
+	}
 
 	/* if we run without loopback, then return now */
 	if (!loopback)
@@ -965,13 +979,19 @@ int spi_test_run_tests(struct spi_device *spi,
 	/* allocate rx/tx buffers of 128kB size without devm
 	 * in the hope that is on a page boundary
 	 */
-	rx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
+	if (use_vmalloc)
+		rx = vmalloc(SPI_TEST_MAX_SIZE_PLUS);
+	else
+		rx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
 	if (!rx) {
 		ret = -ENOMEM;
 		goto out;
 	}
 
-	tx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
+	if (use_vmalloc)
+		tx = vmalloc(SPI_TEST_MAX_SIZE_PLUS);
+	else
+		tx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
 	if (!tx) {
 		ret = -ENOMEM;
 		goto out;
@@ -999,8 +1019,8 @@ int spi_test_run_tests(struct spi_device *spi,
 	}
 
 out:
-	kfree(rx);
-	kfree(tx);
+	kvfree(rx);
+	kvfree(tx);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(spi_test_run_tests);
-- 
2.7.4

Applied "spi: loopback-test: add option to use vmalloc'ed buffers" to the spi tree

[Mark Brown]

The patch

   spi: loopback-test: add option to use vmalloc'ed buffers

has been applied to the spi tree at

   git://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi.git 

All being well this means that it will be integrated into the linux-next
tree (usually sometime in the next 24 hours) and sent to Linus during
the next merge window (or sooner if it is a bug fix), however if
problems are discovered then the patch may be dropped or reverted.  

You may get further e-mails resulting from automated or manual testing
and review of the tree, please engage with people reporting problems and
send followup patches addressing any issues that are reported if needed.

If any updates are required or you are submitting further changes they
should be sent as incremental updates against current git, existing
patches will not be replaced.

Please add any relevant lists and maintainers to the CCs when replying
to this mail.

Thanks,
Mark

>From 576333a1fb940767e8638fafea908214682e8acd Mon Sep 17 00:00:00 2001
From: Frode Isaksen <fisaksen@baylibre.com>
Date: Thu, 23 Feb 2017 19:02:01 +0100
Subject: [PATCH] spi: loopback-test: add option to use vmalloc'ed buffers

Using vmalloc'ed buffers will use one SG entry for each page,
that may provoke DMA errors for large transfers.
Also vmalloc'ed buffers may cause errors on CPU's with VIVT cache.
Add this option to catch these errors when testing.
Note that to catch VIVT cache errors, checking the rx range
has to be disabled, so this option has been added as well.

Signed-off-by: Frode Isaksen <fisaksen@baylibre.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
---
 drivers/spi/spi-loopback-test.c | 34 +++++++++++++++++++++++++++-------
 1 file changed, 27 insertions(+), 7 deletions(-)

diff --git a/drivers/spi/spi-loopback-test.c b/drivers/spi/spi-loopback-test.c
index 50e620f4e8fe..35960515382e 100644
--- a/drivers/spi/spi-loopback-test.c
+++ b/drivers/spi/spi-loopback-test.c
@@ -55,6 +55,18 @@ module_param(run_only_test, int, 0);
 MODULE_PARM_DESC(run_only_test,
 		 "only run the test with this number (0-based !)");
 
+/* use vmalloc'ed buffers */
+int use_vmalloc;
+module_param(use_vmalloc, int, 0644);
+MODULE_PARM_DESC(use_vmalloc,
+		 "use vmalloc'ed buffers instead of kmalloc'ed");
+
+/* check rx ranges */
+int check_ranges = 1;
+module_param(check_ranges, int, 0644);
+MODULE_PARM_DESC(check_ranges,
+		 "checks rx_buffer pattern are valid");
+
 /* the actual tests to execute */
 static struct spi_test spi_tests[] = {
 	{
@@ -492,9 +504,11 @@ static int spi_test_check_loopback_result(struct spi_device *spi,
 	int ret;
 
 	/* checks rx_buffer pattern are valid with loopback or without */
-	ret = spi_check_rx_ranges(spi, msg, rx);
-	if (ret)
-		return ret;
+	if (check_ranges) {
+		ret = spi_check_rx_ranges(spi, msg, rx);
+		if (ret)
+			return ret;
+	}
 
 	/* if we run without loopback, then return now */
 	if (!loopback)
@@ -965,13 +979,19 @@ int spi_test_run_tests(struct spi_device *spi,
 	/* allocate rx/tx buffers of 128kB size without devm
 	 * in the hope that is on a page boundary
 	 */
-	rx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
+	if (use_vmalloc)
+		rx = vmalloc(SPI_TEST_MAX_SIZE_PLUS);
+	else
+		rx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
 	if (!rx) {
 		ret = -ENOMEM;
 		goto out;
 	}
 
-	tx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
+	if (use_vmalloc)
+		tx = vmalloc(SPI_TEST_MAX_SIZE_PLUS);
+	else
+		tx = kzalloc(SPI_TEST_MAX_SIZE_PLUS, GFP_KERNEL);
 	if (!tx) {
 		ret = -ENOMEM;
 		goto out;
@@ -999,8 +1019,8 @@ int spi_test_run_tests(struct spi_device *spi,
 	}
 
 out:
-	kfree(rx);
-	kfree(tx);
+	kvfree(rx);
+	kvfree(tx);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(spi_test_run_tests);
-- 
2.11.0