diff mbox

spi: qup: Add DMA capabilities

Message ID 1403816781-31008-1-git-send-email-agross@codeaurora.org (mailing list archive)
State New, archived
Headers show

Commit Message

Andy Gross June 26, 2014, 9:06 p.m. UTC 
This patch adds DMA capabilities to the spi-qup driver.  If DMA channels are
present, the QUP will use DMA instead of block mode for transfers to/from SPI
peripherals for transactions larger than the length of a block.

Signed-off-by: Andy Gross <agross@codeaurora.org>
---
 .../devicetree/bindings/spi/qcom,spi-qup.txt       |   10 +
 drivers/spi/spi-qup.c                              |  361 ++++++++++++++++++--
 2 files changed, 350 insertions(+), 21 deletions(-)

Comments

Mark Brown June 27, 2014, 10:50 a.m. UTC | #1 
On Thu, Jun 26, 2014 at 04:06:21PM -0500, Andy Gross wrote:

> +	if (xfer->rx_buf) {
> +		rx_dma = dma_map_single(controller->dev, xfer->rx_buf,
> +			xfer->len, DMA_FROM_DEVICE);

It would be better to use the core DMA mapping code rather than open
coding.  This code won't work for vmalloc()ed addresses, or physically
non-contiguous addresses unless there's an IOMMU fixing things up.
Andy Gross June 27, 2014, 3:54 p.m. UTC | #2 
On Fri, Jun 27, 2014 at 11:50:57AM +0100, Mark Brown wrote:
> On Thu, Jun 26, 2014 at 04:06:21PM -0500, Andy Gross wrote:
> 
> > +	if (xfer->rx_buf) {
> > +		rx_dma = dma_map_single(controller->dev, xfer->rx_buf,
> > +			xfer->len, DMA_FROM_DEVICE);
> 
> It would be better to use the core DMA mapping code rather than open
> coding.  This code won't work for vmalloc()ed addresses, or physically
> non-contiguous addresses unless there's an IOMMU fixing things up.

Ah, ok.  So I just need a to setup the scatter gather page list and then do a
dma_map_sg.  I'll resend once I have this in place.
Russell King - ARM Linux June 27, 2014, 4:24 p.m. UTC | #3 
On Fri, Jun 27, 2014 at 10:54:22AM -0500, Andy Gross wrote:
> On Fri, Jun 27, 2014 at 11:50:57AM +0100, Mark Brown wrote:
> > On Thu, Jun 26, 2014 at 04:06:21PM -0500, Andy Gross wrote:
> > 
> > > +	if (xfer->rx_buf) {
> > > +		rx_dma = dma_map_single(controller->dev, xfer->rx_buf,
> > > +			xfer->len, DMA_FROM_DEVICE);
> > 
> > It would be better to use the core DMA mapping code rather than open
> > coding.  This code won't work for vmalloc()ed addresses, or physically
> > non-contiguous addresses unless there's an IOMMU fixing things up.
> 
> Ah, ok.  So I just need a to setup the scatter gather page list and then do a
> dma_map_sg.  I'll resend once I have this in place.

Note that DMA from vmalloc'd memory is non-coherent on some platforms,
even if you use the DMA API.

The only thing that the DMA API guarantees is that the kernel mapping
will be made coherent for DMA purposes.  No other mapping has this
guarantee.

Consider a VIVT cache (like the older ARMs).  For this cache, you need
to find every alias of a physical page and flush it.  The DMA API doesn't
have that information - it can only deal with the kernel's lowmem mapping.

We have introduced a couple of helpers recently to solve the problem of
vmalloc() (since a number of filesystems now do this trick) but the
vmalloc() user has to deal with the problem:

	flush_kernel_vmap_range()
	invalidate_kernel_vmap_range()

See the bottom of Documentation/cachetlb.txt for details.

The long and the short of it is that it's better if vmalloc()'d memory
is avoided where possible.  It's also loads better if subsystems pass
physical references to memory for IO purposes where possible like our
block layer does (iow, struct page + offset, length) rather than
using randomly mapped virtual addresses, where the driver may not
know where the memory has come from.
Andy Gross June 27, 2014, 6:58 p.m. UTC | #4 
On Fri, Jun 27, 2014 at 05:24:11PM +0100, Russell King - ARM Linux wrote:

<snip>

> > > It would be better to use the core DMA mapping code rather than open
> > > coding.  This code won't work for vmalloc()ed addresses, or physically
> > > non-contiguous addresses unless there's an IOMMU fixing things up.
> > 
> > Ah, ok.  So I just need a to setup the scatter gather page list and then do a
> > dma_map_sg.  I'll resend once I have this in place.
> 
> Note that DMA from vmalloc'd memory is non-coherent on some platforms,
> even if you use the DMA API.

Thankfully, that isn't the case on this platform.  As as the QUP controller is
only on Qualcomm chipsets, we don't have to worry about it being used on other
platforms.
Ivan T. Ivanov July 2, 2014, 2:26 p.m. UTC | #5 
Hi Andy,

Just few comments.

On Thu, 2014-06-26 at 16:06 -0500, Andy Gross wrote:
> This patch adds DMA capabilities to the spi-qup driver.  If DMA channels are
> present, the QUP will use DMA instead of block mode for transfers to/from SPI
> peripherals for transactions larger than the length of a block.
> 
> Signed-off-by: Andy Gross <agross@codeaurora.org>
> ---
>  .../devicetree/bindings/spi/qcom,spi-qup.txt       |   10 +
>  drivers/spi/spi-qup.c                              |  361 ++++++++++++++++++--
>  2 files changed, 350 insertions(+), 21 deletions(-)
> 

<snip>
 
> diff --git a/drivers/spi/spi-qup.c b/drivers/spi/spi-qup.c
> index fc1de86..9b01db5 100644
> --- a/drivers/spi/spi-qup.c
> +++ b/drivers/spi/spi-qup.c
> @@ -22,6 +22,8 @@
>  #include <linux/platform_device.h>
>  #include <linux/pm_runtime.h>
>  #include <linux/spi/spi.h>
> +#include <linux/dmaengine.h>
> +#include <linux/dma-mapping.h>
>  
>  #define QUP_CONFIG			0x0000
>  #define QUP_STATE			0x0004
> @@ -116,6 +118,8 @@
>  
>  #define SPI_NUM_CHIPSELECTS		4
>  
> +#define SPI_MAX_XFER			(SZ_64K - 64)
> +
>  /* high speed mode is when bus rate is greater then 26MHz */
>  #define SPI_HS_MIN_RATE			26000000
>  #define SPI_MAX_RATE			50000000
> @@ -142,6 +146,17 @@ struct spi_qup {
>  	int			w_size;	/* bytes per SPI word */
>  	int			tx_bytes;
>  	int			rx_bytes;
> +
> +	int			use_dma;
> +
> +	struct dma_chan		*rx_chan;
> +	struct dma_slave_config	rx_conf;
> +	struct dma_chan		*tx_chan;
> +	struct dma_slave_config tx_conf;
> +	dma_addr_t		rx_dma;
> +	dma_addr_t		tx_dma;

DMA addresses seems unused.

> +	void			*dummy;

This is not so dummy, probably 'spare'.

> +	atomic_t		dma_outstanding;
>  };

<snip>

>  
>  
> @@ -632,6 +896,56 @@ static int spi_qup_probe(struct platform_device *pdev)
>  	writel_relaxed(SPI_ERROR_CLK_UNDER_RUN | SPI_ERROR_CLK_OVER_RUN,
>  		       base + SPI_ERROR_FLAGS_EN);
>  
> +	/* allocate dma resources, if available */
> +	controller->rx_chan = dma_request_slave_channel(&pdev->dev, "rx");
> +	if (controller->rx_chan) {
> +		controller->tx_chan =
> +			dma_request_slave_channel(&pdev->dev, "tx");
> +
> +		if (!controller->tx_chan) {
> +			dev_err(&pdev->dev, "Failed to allocate dma tx chan");
> +			dma_release_channel(controller->rx_chan);


There is no point to go further with DMA configuration
if there are no channels, right?

> +		}
> +
> +		/* set DMA parameters */
> +		controller->rx_conf.device_fc = 1;
> +		controller->rx_conf.src_addr = res->start + QUP_INPUT_FIFO;
> +		controller->rx_conf.src_maxburst = controller->in_blk_sz;
> +
> +		controller->tx_conf.device_fc = 1;
> +		controller->tx_conf.dst_addr = res->start + QUP_OUTPUT_FIFO;
> +		controller->tx_conf.dst_maxburst = controller->out_blk_sz;
> +


Please, could you share blsp2_bam device node configuration? 
I would like to test these changes.

Regards,
Ivan

--
To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

diff --git a/Documentation/devicetree/bindings/spi/qcom,spi-qup.txt b/Documentation/devicetree/bindings/spi/qcom,spi-qup.txt
index b82a268..4d8977a 100644
--- a/Documentation/devicetree/bindings/spi/qcom,spi-qup.txt
+++ b/Documentation/devicetree/bindings/spi/qcom,spi-qup.txt
@@ -23,6 +23,12 @@  Optional properties:
 - spi-max-frequency: Specifies maximum SPI clock frequency,
                      Units - Hz. Definition as per
                      Documentation/devicetree/bindings/spi/spi-bus.txt
+- dmas :             Two DMA channel specifiers following the convention outlined
+                     in bindings/dma/dma.txt
+- dma-names:         Names for the dma channels, if present. There must be at
+                     least one channel named "tx" for transmit and named "rx" for
+                     receive.
+
 
 SPI slave nodes must be children of the SPI master node and can contain
 properties described in Documentation/devicetree/bindings/spi/spi-bus.txt
@@ -41,6 +47,10 @@  Example:
 		clocks = <&gcc GCC_BLSP2_QUP2_SPI_APPS_CLK>, <&gcc GCC_BLSP2_AHB_CLK>;
 		clock-names = "core", "iface";
 
+		dmas = <&blsp2_bam 2>,
+			<&blsp2_bam 3>;
+		dma-names = "rx", "tx";
+
 		pinctrl-names = "default";
 		pinctrl-0 = <&spi8_default>;
 
diff --git a/drivers/spi/spi-qup.c b/drivers/spi/spi-qup.c
index fc1de86..9b01db5 100644
--- a/drivers/spi/spi-qup.c
+++ b/drivers/spi/spi-qup.c
@@ -22,6 +22,8 @@ 
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
 #include <linux/spi/spi.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
 
 #define QUP_CONFIG			0x0000
 #define QUP_STATE			0x0004
@@ -116,6 +118,8 @@ 
 
 #define SPI_NUM_CHIPSELECTS		4
 
+#define SPI_MAX_XFER			(SZ_64K - 64)
+
 /* high speed mode is when bus rate is greater then 26MHz */
 #define SPI_HS_MIN_RATE			26000000
 #define SPI_MAX_RATE			50000000
@@ -142,6 +146,17 @@  struct spi_qup {
 	int			w_size;	/* bytes per SPI word */
 	int			tx_bytes;
 	int			rx_bytes;
+
+	int			use_dma;
+
+	struct dma_chan		*rx_chan;
+	struct dma_slave_config	rx_conf;
+	struct dma_chan		*tx_chan;
+	struct dma_slave_config tx_conf;
+	dma_addr_t		rx_dma;
+	dma_addr_t		tx_dma;
+	void			*dummy;
+	atomic_t		dma_outstanding;
 };
 
 
@@ -265,6 +280,221 @@  static void spi_qup_fifo_write(struct spi_qup *controller,
 	}
 }
 
+static void qup_dma_callback(void *data)
+{
+	struct spi_qup *controller = data;
+
+	if (atomic_dec_and_test(&controller->dma_outstanding))
+		complete(&controller->done);
+}
+
+static int spi_qup_do_dma(struct spi_qup *controller, struct spi_transfer *xfer)
+{
+	struct dma_async_tx_descriptor *rxd, *txd;
+	dma_cookie_t rx_cookie, tx_cookie;
+	u32 xfer_len, rx_align = 0, tx_align = 0, n_words;
+	struct scatterlist tx_sg[2], rx_sg[2];
+	int ret = 0;
+	u32 bytes_to_xfer = xfer->len;
+	u32 offset = 0;
+	u32 rx_nents = 0, tx_nents = 0;
+	dma_addr_t rx_dma = 0, tx_dma = 0, rx_dummy_dma = 0, tx_dummy_dma = 0;
+
+
+	if (xfer->rx_buf) {
+		rx_dma = dma_map_single(controller->dev, xfer->rx_buf,
+			xfer->len, DMA_FROM_DEVICE);
+
+		if (dma_mapping_error(controller->dev, rx_dma)) {
+			ret = -ENOMEM;
+			return ret;
+		}
+
+		/* check to see if we need dummy buffer for leftover bytes */
+		rx_align = xfer->len % controller->in_blk_sz;
+		if (rx_align) {
+			rx_dummy_dma = dma_map_single(controller->dev,
+				controller->dummy, controller->in_fifo_sz,
+				DMA_FROM_DEVICE);
+
+			if (dma_mapping_error(controller->dev, rx_dummy_dma)) {
+				ret = -ENOMEM;
+				goto err_map_rx_dummy;
+			}
+		}
+	}
+
+	if (xfer->tx_buf) {
+		tx_dma = dma_map_single(controller->dev,
+			(void *)xfer->tx_buf, xfer->len, DMA_TO_DEVICE);
+
+		if (dma_mapping_error(controller->dev, tx_dma)) {
+			ret = -ENOMEM;
+			goto err_map_tx;
+		}
+
+		/* check to see if we need dummy buffer for leftover bytes */
+		tx_align = xfer->len % controller->out_blk_sz;
+		if (tx_align) {
+			memcpy(controller->dummy + SZ_1K,
+				xfer->tx_buf + xfer->len - tx_align,
+				tx_align);
+			memset(controller->dummy + SZ_1K + tx_align, 0,
+				controller->out_blk_sz - tx_align);
+
+			tx_dummy_dma = dma_map_single(controller->dev,
+				controller->dummy + SZ_1K,
+				controller->out_blk_sz, DMA_TO_DEVICE);
+
+			if (dma_mapping_error(controller->dev, tx_dummy_dma)) {
+				ret = -ENOMEM;
+				goto err_map_tx_dummy;
+			}
+		}
+	}
+
+	atomic_set(&controller->dma_outstanding, 0);
+
+	while (bytes_to_xfer > 0) {
+		xfer_len = min_t(u32, bytes_to_xfer, SPI_MAX_XFER);
+		n_words = DIV_ROUND_UP(xfer_len, controller->w_size);
+
+		/* write out current word count to controller */
+		writel_relaxed(n_words, controller->base + QUP_MX_INPUT_CNT);
+		writel_relaxed(n_words, controller->base + QUP_MX_OUTPUT_CNT);
+
+		reinit_completion(&controller->done);
+
+		if (xfer->tx_buf) {
+			/* recalc align for each transaction */
+			tx_align = xfer_len % controller->out_blk_sz;
+
+			if (tx_align)
+				tx_nents = 2;
+			else
+				tx_nents = 1;
+
+			/* initialize scatterlists */
+			sg_init_table(tx_sg, tx_nents);
+			sg_dma_len(&tx_sg[0]) = xfer_len - tx_align;
+			sg_dma_address(&tx_sg[0]) = tx_dma + offset;
+
+			/* account for non block size transfer */
+			if (tx_align) {
+				sg_dma_len(&tx_sg[1]) = controller->out_blk_sz;
+				sg_dma_address(&tx_sg[1]) = tx_dummy_dma;
+			}
+
+			txd = dmaengine_prep_slave_sg(controller->tx_chan,
+					tx_sg, tx_nents, DMA_MEM_TO_DEV, 0);
+			if (!txd) {
+				ret = -ENOMEM;
+				goto err_unmap;
+			}
+
+			atomic_inc(&controller->dma_outstanding);
+
+			txd->callback = qup_dma_callback;
+			txd->callback_param = controller;
+
+			tx_cookie = dmaengine_submit(txd);
+
+			dma_async_issue_pending(controller->tx_chan);
+		}
+
+		if (xfer->rx_buf) {
+			/* recalc align for each transaction */
+			rx_align = xfer_len % controller->in_blk_sz;
+
+			if (rx_align)
+				rx_nents = 2;
+			else
+				rx_nents = 1;
+
+			/* initialize scatterlists */
+			sg_init_table(rx_sg, rx_nents);
+			sg_dma_address(&rx_sg[0]) = rx_dma + offset;
+			sg_dma_len(&rx_sg[0]) = xfer_len - rx_align;
+
+			/* account for non block size transfer */
+			if (rx_align) {
+				sg_dma_len(&rx_sg[1]) = controller->in_blk_sz;
+				sg_dma_address(&rx_sg[1]) = rx_dummy_dma;
+			}
+
+			rxd = dmaengine_prep_slave_sg(controller->rx_chan,
+					rx_sg, rx_nents, DMA_DEV_TO_MEM, 0);
+			if (!rxd) {
+				ret = -ENOMEM;
+				goto err_unmap;
+			}
+
+			atomic_inc(&controller->dma_outstanding);
+
+			rxd->callback = qup_dma_callback;
+			rxd->callback_param = controller;
+
+			rx_cookie = dmaengine_submit(rxd);
+
+			dma_async_issue_pending(controller->rx_chan);
+		}
+
+		if (spi_qup_set_state(controller, QUP_STATE_RUN)) {
+			dev_warn(controller->dev, "cannot set EXECUTE state\n");
+			goto err_unmap;
+		}
+
+		if (!wait_for_completion_timeout(&controller->done,
+			msecs_to_jiffies(1000))) {
+			ret = -ETIMEDOUT;
+
+			/* clear out all the DMA transactions */
+			if (xfer->tx_buf)
+				dmaengine_terminate_all(controller->tx_chan);
+			if (xfer->rx_buf)
+				dmaengine_terminate_all(controller->rx_chan);
+
+			goto err_unmap;
+		}
+
+		if (rx_align)
+			memcpy(xfer->rx_buf + offset + xfer->len - rx_align,
+				controller->dummy, rx_align);
+
+		/* adjust remaining bytes to transfer */
+		bytes_to_xfer -= xfer_len;
+		offset += xfer_len;
+
+
+		/* reset mini-core state so we can program next transaction */
+		if (spi_qup_set_state(controller, QUP_STATE_RESET)) {
+			dev_err(controller->dev, "cannot set RESET state\n");
+			goto err_unmap;
+		}
+	}
+
+	ret = 0;
+
+err_unmap:
+	if (tx_align)
+		dma_unmap_single(controller->dev, tx_dummy_dma,
+			controller->out_fifo_sz, DMA_TO_DEVICE);
+err_map_tx_dummy:
+	if (xfer->tx_buf)
+		dma_unmap_single(controller->dev, tx_dma, xfer->len,
+			DMA_TO_DEVICE);
+err_map_tx:
+	if (rx_align)
+		dma_unmap_single(controller->dev, rx_dummy_dma,
+			controller->in_fifo_sz, DMA_FROM_DEVICE);
+err_map_rx_dummy:
+	if (xfer->rx_buf)
+		dma_unmap_single(controller->dev, rx_dma, xfer->len,
+			DMA_FROM_DEVICE);
+
+	return ret;
+}
+
 static irqreturn_t spi_qup_qup_irq(int irq, void *dev_id)
 {
 	struct spi_qup *controller = dev_id;
@@ -314,11 +544,13 @@  static irqreturn_t spi_qup_qup_irq(int irq, void *dev_id)
 		error = -EIO;
 	}
 
-	if (opflags & QUP_OP_IN_SERVICE_FLAG)
-		spi_qup_fifo_read(controller, xfer);
+	if (!controller->use_dma) {
+		if (opflags & QUP_OP_IN_SERVICE_FLAG)
+			spi_qup_fifo_read(controller, xfer);
 
-	if (opflags & QUP_OP_OUT_SERVICE_FLAG)
-		spi_qup_fifo_write(controller, xfer);
+		if (opflags & QUP_OP_OUT_SERVICE_FLAG)
+			spi_qup_fifo_write(controller, xfer);
+	}
 
 	spin_lock_irqsave(&controller->lock, flags);
 	controller->error = error;
@@ -338,6 +570,8 @@  static int spi_qup_io_config(struct spi_device *spi, struct spi_transfer *xfer)
 	struct spi_qup *controller = spi_master_get_devdata(spi->master);
 	u32 config, iomode, mode;
 	int ret, n_words, w_size;
+	size_t dma_align = dma_get_cache_alignment();
+	u32 dma_available = 0;
 
 	if (spi->mode & SPI_LOOP && xfer->len > controller->in_fifo_sz) {
 		dev_err(controller->dev, "too big size for loopback %d > %d\n",
@@ -366,6 +600,11 @@  static int spi_qup_io_config(struct spi_device *spi, struct spi_transfer *xfer)
 	n_words = xfer->len / w_size;
 	controller->w_size = w_size;
 
+	if (controller->rx_chan &&
+		IS_ALIGNED((size_t)xfer->tx_buf, dma_align) &&
+		IS_ALIGNED((size_t)xfer->rx_buf, dma_align))
+		dma_available = 1;
+
 	if (n_words <= (controller->in_fifo_sz / sizeof(u32))) {
 		mode = QUP_IO_M_MODE_FIFO;
 		writel_relaxed(n_words, controller->base + QUP_MX_READ_CNT);
@@ -373,19 +612,31 @@  static int spi_qup_io_config(struct spi_device *spi, struct spi_transfer *xfer)
 		/* must be zero for FIFO */
 		writel_relaxed(0, controller->base + QUP_MX_INPUT_CNT);
 		writel_relaxed(0, controller->base + QUP_MX_OUTPUT_CNT);
-	} else {
+		controller->use_dma = 0;
+	} else if (!dma_available) {
 		mode = QUP_IO_M_MODE_BLOCK;
 		writel_relaxed(n_words, controller->base + QUP_MX_INPUT_CNT);
 		writel_relaxed(n_words, controller->base + QUP_MX_OUTPUT_CNT);
 		/* must be zero for BLOCK and BAM */
 		writel_relaxed(0, controller->base + QUP_MX_READ_CNT);
 		writel_relaxed(0, controller->base + QUP_MX_WRITE_CNT);
+		controller->use_dma = 0;
+	} else {
+		mode = QUP_IO_M_MODE_DMOV;
+		writel_relaxed(0, controller->base + QUP_MX_READ_CNT);
+		writel_relaxed(0, controller->base + QUP_MX_WRITE_CNT);
+		controller->use_dma = 1;
 	}
 
 	iomode = readl_relaxed(controller->base + QUP_IO_M_MODES);
 	/* Set input and output transfer mode */
 	iomode &= ~(QUP_IO_M_INPUT_MODE_MASK | QUP_IO_M_OUTPUT_MODE_MASK);
-	iomode &= ~(QUP_IO_M_PACK_EN | QUP_IO_M_UNPACK_EN);
+
+	if (!controller->use_dma)
+		iomode &= ~(QUP_IO_M_PACK_EN | QUP_IO_M_UNPACK_EN);
+	else
+		iomode |= QUP_IO_M_PACK_EN | QUP_IO_M_UNPACK_EN;
+
 	iomode |= (mode << QUP_IO_M_OUTPUT_MODE_MASK_SHIFT);
 	iomode |= (mode << QUP_IO_M_INPUT_MODE_MASK_SHIFT);
 
@@ -418,6 +669,14 @@  static int spi_qup_io_config(struct spi_device *spi, struct spi_transfer *xfer)
 	config &= ~(QUP_CONFIG_NO_INPUT | QUP_CONFIG_NO_OUTPUT | QUP_CONFIG_N);
 	config |= xfer->bits_per_word - 1;
 	config |= QUP_CONFIG_SPI_MODE;
+
+	if (controller->use_dma) {
+		if (!xfer->tx_buf)
+			config |= QUP_CONFIG_NO_OUTPUT;
+		if (!xfer->rx_buf)
+			config |= QUP_CONFIG_NO_INPUT;
+	}
+
 	writel_relaxed(config, controller->base + QUP_CONFIG);
 
 	writel_relaxed(0, controller->base + QUP_OPERATIONAL_MASK);
@@ -474,25 +733,29 @@  static int spi_qup_transfer_one(struct spi_master *master,
 	controller->tx_bytes = 0;
 	spin_unlock_irqrestore(&controller->lock, flags);
 
-	if (spi_qup_set_state(controller, QUP_STATE_RUN)) {
-		dev_warn(controller->dev, "cannot set RUN state\n");
-		goto exit;
-	}
+	if (controller->use_dma) {
+		ret = spi_qup_do_dma(controller, xfer);
+	} else {
+		if (spi_qup_set_state(controller, QUP_STATE_RUN)) {
+			dev_warn(controller->dev, "cannot set RUN state\n");
+			goto exit;
+		}
 
-	if (spi_qup_set_state(controller, QUP_STATE_PAUSE)) {
-		dev_warn(controller->dev, "cannot set PAUSE state\n");
-		goto exit;
-	}
+		if (spi_qup_set_state(controller, QUP_STATE_PAUSE)) {
+			dev_warn(controller->dev, "cannot set PAUSE state\n");
+			goto exit;
+		}
 
-	spi_qup_fifo_write(controller, xfer);
+		spi_qup_fifo_write(controller, xfer);
 
-	if (spi_qup_set_state(controller, QUP_STATE_RUN)) {
-		dev_warn(controller->dev, "cannot set EXECUTE state\n");
-		goto exit;
-	}
+		if (spi_qup_set_state(controller, QUP_STATE_RUN)) {
+			dev_warn(controller->dev, "cannot set EXECUTE state\n");
+			goto exit;
+		}
 
-	if (!wait_for_completion_timeout(&controller->done, timeout))
-		ret = -ETIMEDOUT;
+		if (!wait_for_completion_timeout(&controller->done, timeout))
+			ret = -ETIMEDOUT;
+	}
 exit:
 	spi_qup_set_state(controller, QUP_STATE_RESET);
 	spin_lock_irqsave(&controller->lock, flags);
@@ -580,6 +843,7 @@  static int spi_qup_probe(struct platform_device *pdev)
 	master->transfer_one = spi_qup_transfer_one;
 	master->dev.of_node = pdev->dev.of_node;
 	master->auto_runtime_pm = true;
+	master->dma_alignment = dma_get_cache_alignment();
 
 	platform_set_drvdata(pdev, master);
 
@@ -632,6 +896,56 @@  static int spi_qup_probe(struct platform_device *pdev)
 	writel_relaxed(SPI_ERROR_CLK_UNDER_RUN | SPI_ERROR_CLK_OVER_RUN,
 		       base + SPI_ERROR_FLAGS_EN);
 
+	/* allocate dma resources, if available */
+	controller->rx_chan = dma_request_slave_channel(&pdev->dev, "rx");
+	if (controller->rx_chan) {
+		controller->tx_chan =
+			dma_request_slave_channel(&pdev->dev, "tx");
+
+		if (!controller->tx_chan) {
+			dev_err(&pdev->dev, "Failed to allocate dma tx chan");
+			dma_release_channel(controller->rx_chan);
+		}
+
+		/* set DMA parameters */
+		controller->rx_conf.device_fc = 1;
+		controller->rx_conf.src_addr = res->start + QUP_INPUT_FIFO;
+		controller->rx_conf.src_maxburst = controller->in_blk_sz;
+
+		controller->tx_conf.device_fc = 1;
+		controller->tx_conf.dst_addr = res->start + QUP_OUTPUT_FIFO;
+		controller->tx_conf.dst_maxburst = controller->out_blk_sz;
+
+		if (dmaengine_slave_config(controller->rx_chan,
+				&controller->rx_conf)) {
+			dev_err(&pdev->dev, "failed to configure RX channel\n");
+
+			dma_release_channel(controller->rx_chan);
+			dma_release_channel(controller->tx_chan);
+			controller->tx_chan = NULL;
+			controller->rx_chan = NULL;
+		} else if (dmaengine_slave_config(controller->tx_chan,
+				&controller->tx_conf)) {
+			dev_err(&pdev->dev, "failed to configure TX channel\n");
+
+			dma_release_channel(controller->rx_chan);
+			dma_release_channel(controller->tx_chan);
+			controller->tx_chan = NULL;
+			controller->rx_chan = NULL;
+		}
+
+		controller->dummy = devm_kmalloc(controller->dev, PAGE_SIZE,
+			GFP_KERNEL);
+
+		if (!controller->dummy) {
+			dma_release_channel(controller->rx_chan);
+			dma_release_channel(controller->tx_chan);
+			controller->tx_chan = NULL;
+			controller->rx_chan = NULL;
+		}
+	}
+
+
 	writel_relaxed(0, base + SPI_CONFIG);
 	writel_relaxed(SPI_IO_C_NO_TRI_STATE, base + SPI_IO_CONTROL);
 
@@ -741,6 +1055,11 @@  static int spi_qup_remove(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	if (controller->rx_chan)
+		dma_release_channel(controller->rx_chan);
+	if (controller->tx_chan)
+		dma_release_channel(controller->tx_chan);
+
 	clk_disable_unprepare(controller->cclk);
 	clk_disable_unprepare(controller->iclk);