Merge tag 'dmaengine-4.3-rc1' of git://git.infradead.org/users/vkoul/slave-dma

Pull dmaengine updates from Vinod Koul:
 "This time we have aded a new capability for scatter-gathered memset
  using dmaengine APIs.  This is supported in xdmac & hdmac drivers

  We have added support for reusing descriptors for examples like video
  buffers etc.  Driver will follow

  The behaviour of descriptor ack has been clarified and documented

  New devices added are:
   - dma controller in sun[457]i SoCs
   - lpc18xx dmamux
   - ZTE ZX296702 dma controller
   - Analog Devices AXI-DMAC DMA controller
   - eDMA support for dma-crossbar
   - imx6sx support in imx-sdma driver
   - imx-sdma device to device support

  Other:
   - jz4780 fixes
   - ioatdma large refactor and cleanup for removal of ioat v1 and v2
     which is deprecated and fixes
   - ACPI support in X-Gene DMA engine driver
   - ipu irq fixes
   - mvxor fixes
   - minor fixes spread thru drivers"

[ The Kconfig and Makefile entries got re-sorted alphabetically, and I
  handled the conflict with the new Intel integrated IDMA driver by
  slightly mis-sorting it on purpose: "IDMA64" got sorted after "IMX" in
  order to keep the Intel entries together.  I think it might be a good
  idea to just rename the IDMA64 config entry to INTEL_IDMA64 to make
  the sorting be a true sort, not this mismash.

  Also, this merge disables the COMPILE_TEST for the sun4i DMA
  controller, because it does not compile cleanly at all.     - Linus ]

* tag 'dmaengine-4.3-rc1' of git://git.infradead.org/users/vkoul/slave-dma: (89 commits)
  dmaengine: ioatdma: add Broadwell EP ioatdma PCI dev IDs
  dmaengine :ipu: change ipu_irq_handler() to remove compile warning
  dmaengine: ioatdma: Fix variable array length
  dmaengine: ioatdma: fix sparse "error" with prep lock
  dmaengine: hdmac: Add memset capabilities
  dmaengine: sort the sh Makefile
  dmaengine: sort the sh Kconfig
  dmaengine: sort the dw Kconfig
  dmaengine: sort the Kconfig
  dmaengine: sort the makefile
  drivers/dma: make mv_xor.c driver explicitly non-modular
  dmaengine: Add support for the Analog Devices AXI-DMAC DMA controller
  devicetree: Add bindings documentation for Analog Devices AXI-DMAC
  dmaengine: xgene-dma: Fix the lock to allow client for further submission of requests
  dmaengine: ioatdma: fix coccinelle warning
  dmaengine: ioatdma: fix zero day warning on incompatible pointer type
  dmaengine: tegra-apb: Simplify locking for device using global pause
  dmaengine: tegra-apb: Remove unnecessary return statements and variables
  dmaengine: tegra-apb: Avoid unnecessary channel base address calculation
  dmaengine: tegra-apb: Remove unused variables
  ...

1 files changed, 172 insertions, 11 deletions

diff --git a/drivers/dma/at_xdmac.c b/drivers/dma/at_xdmac.c
index 40afa2a16cfc..a165b4bfd330 100644
--- a/drivers/dma/at_xdmac.c
+++ b/drivers/dma/at_xdmac.c
@@ -625,12 +625,12 @@ at_xdmac_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		       unsigned int sg_len, enum dma_transfer_direction direction,
 		       unsigned long flags, void *context)
 {
-	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
-	struct at_xdmac_desc	*first = NULL, *prev = NULL;
-	struct scatterlist	*sg;
-	int			i;
-	unsigned int		xfer_size = 0;
-	unsigned long		irqflags;
+	struct at_xdmac_chan		*atchan = to_at_xdmac_chan(chan);
+	struct at_xdmac_desc		*first = NULL, *prev = NULL;
+	struct scatterlist		*sg;
+	int				i;
+	unsigned int			xfer_size = 0;
+	unsigned long			irqflags;
 	struct dma_async_tx_descriptor	*ret = NULL;
 
 	if (!sgl)
@@ -797,10 +797,7 @@ at_xdmac_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t buf_addr,
 		list_add_tail(&desc->desc_node, &first->descs_list);
 	}
 
-	prev->lld.mbr_nda = first->tx_dma_desc.phys;
-	dev_dbg(chan2dev(chan),
-		"%s: chain lld: prev=0x%p, mbr_nda=%pad\n",
-		__func__, prev, &prev->lld.mbr_nda);
+	at_xdmac_queue_desc(chan, prev, first);
 	first->tx_dma_desc.flags = flags;
 	first->xfer_size = buf_len;
 	first->direction = direction;
@@ -1135,7 +1132,7 @@ static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan,
 	 * SAMA5D4x), so we can use the same interface for source and dest,
 	 * that solves the fact we don't know the direction.
 	 */
-	u32			chan_cc = AT_XDMAC_CC_DAM_INCREMENTED_AM
+	u32			chan_cc = AT_XDMAC_CC_DAM_UBS_AM
 					| AT_XDMAC_CC_SAM_INCREMENTED_AM
 					| AT_XDMAC_CC_DIF(0)
 					| AT_XDMAC_CC_SIF(0)
@@ -1203,6 +1200,168 @@ at_xdmac_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,
 	return &desc->tx_dma_desc;
 }
 
+static struct dma_async_tx_descriptor *
+at_xdmac_prep_dma_memset_sg(struct dma_chan *chan, struct scatterlist *sgl,
+			    unsigned int sg_len, int value,
+			    unsigned long flags)
+{
+	struct at_xdmac_chan	*atchan = to_at_xdmac_chan(chan);
+	struct at_xdmac_desc	*desc, *pdesc = NULL,
+				*ppdesc = NULL, *first = NULL;
+	struct scatterlist	*sg, *psg = NULL, *ppsg = NULL;
+	size_t			stride = 0, pstride = 0, len = 0;
+	int			i;
+
+	if (!sgl)
+		return NULL;
+
+	dev_dbg(chan2dev(chan), "%s: sg_len=%d, value=0x%x, flags=0x%lx\n",
+		__func__, sg_len, value, flags);
+
+	/* Prepare descriptors. */
+	for_each_sg(sgl, sg, sg_len, i) {
+		dev_dbg(chan2dev(chan), "%s: dest=0x%08x, len=%d, pattern=0x%x, flags=0x%lx\n",
+			__func__, sg_dma_address(sg), sg_dma_len(sg),
+			value, flags);
+		desc = at_xdmac_memset_create_desc(chan, atchan,
+						   sg_dma_address(sg),
+						   sg_dma_len(sg),
+						   value);
+		if (!desc && first)
+			list_splice_init(&first->descs_list,
+					 &atchan->free_descs_list);
+
+		if (!first)
+			first = desc;
+
+		/* Update our strides */
+		pstride = stride;
+		if (psg)
+			stride = sg_dma_address(sg) -
+				(sg_dma_address(psg) + sg_dma_len(psg));
+
+		/*
+		 * The scatterlist API gives us only the address and
+		 * length of each elements.
+		 *
+		 * Unfortunately, we don't have the stride, which we
+		 * will need to compute.
+		 *
+		 * That make us end up in a situation like this one:
+		 *    len    stride    len    stride    len
+		 * +-------+        +-------+        +-------+
+		 * |  N-2  |        |  N-1  |        |   N   |
+		 * +-------+        +-------+        +-------+
+		 *
+		 * We need all these three elements (N-2, N-1 and N)
+		 * to actually take the decision on whether we need to
+		 * queue N-1 or reuse N-2.
+		 *
+		 * We will only consider N if it is the last element.
+		 */
+		if (ppdesc && pdesc) {
+			if ((stride == pstride) &&
+			    (sg_dma_len(ppsg) == sg_dma_len(psg))) {
+				dev_dbg(chan2dev(chan),
+					"%s: desc 0x%p can be merged with desc 0x%p\n",
+					__func__, pdesc, ppdesc);
+
+				/*
+				 * Increment the block count of the
+				 * N-2 descriptor
+				 */
+				at_xdmac_increment_block_count(chan, ppdesc);
+				ppdesc->lld.mbr_dus = stride;
+
+				/*
+				 * Put back the N-1 descriptor in the
+				 * free descriptor list
+				 */
+				list_add_tail(&pdesc->desc_node,
+					      &atchan->free_descs_list);
+
+				/*
+				 * Make our N-1 descriptor pointer
+				 * point to the N-2 since they were
+				 * actually merged.
+				 */
+				pdesc = ppdesc;
+
+			/*
+			 * Rule out the case where we don't have
+			 * pstride computed yet (our second sg
+			 * element)
+			 *
+			 * We also want to catch the case where there
+			 * would be a negative stride,
+			 */
+			} else if (pstride ||
+				   sg_dma_address(sg) < sg_dma_address(psg)) {
+				/*
+				 * Queue the N-1 descriptor after the
+				 * N-2
+				 */
+				at_xdmac_queue_desc(chan, ppdesc, pdesc);
+
+				/*
+				 * Add the N-1 descriptor to the list
+				 * of the descriptors used for this
+				 * transfer
+				 */
+				list_add_tail(&desc->desc_node,
+					      &first->descs_list);
+				dev_dbg(chan2dev(chan),
+					"%s: add desc 0x%p to descs_list 0x%p\n",
+					__func__, desc, first);
+			}
+		}
+
+		/*
+		 * If we are the last element, just see if we have the
+		 * same size than the previous element.
+		 *
+		 * If so, we can merge it with the previous descriptor
+		 * since we don't care about the stride anymore.
+		 */
+		if ((i == (sg_len - 1)) &&
+		    sg_dma_len(ppsg) == sg_dma_len(psg)) {
+			dev_dbg(chan2dev(chan),
+				"%s: desc 0x%p can be merged with desc 0x%p\n",
+				__func__, desc, pdesc);
+
+			/*
+			 * Increment the block count of the N-1
+			 * descriptor
+			 */
+			at_xdmac_increment_block_count(chan, pdesc);
+			pdesc->lld.mbr_dus = stride;
+
+			/*
+			 * Put back the N descriptor in the free
+			 * descriptor list
+			 */
+			list_add_tail(&desc->desc_node,
+				      &atchan->free_descs_list);
+		}
+
+		/* Update our descriptors */
+		ppdesc = pdesc;
+		pdesc = desc;
+
+		/* Update our scatter pointers */
+		ppsg = psg;
+		psg = sg;
+
+		len += sg_dma_len(sg);
+	}
+
+	first->tx_dma_desc.cookie = -EBUSY;
+	first->tx_dma_desc.flags = flags;
+	first->xfer_size = len;
+
+	return &first->tx_dma_desc;
+}
+
 static enum dma_status
 at_xdmac_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
 		struct dma_tx_state *txstate)
@@ -1736,6 +1895,7 @@ static int at_xdmac_probe(struct platform_device *pdev)
 	dma_cap_set(DMA_INTERLEAVE, atxdmac->dma.cap_mask);
 	dma_cap_set(DMA_MEMCPY, atxdmac->dma.cap_mask);
 	dma_cap_set(DMA_MEMSET, atxdmac->dma.cap_mask);
+	dma_cap_set(DMA_MEMSET_SG, atxdmac->dma.cap_mask);
 	dma_cap_set(DMA_SLAVE, atxdmac->dma.cap_mask);
 	/*
 	 * Without DMA_PRIVATE the driver is not able to allocate more than
@@ -1751,6 +1911,7 @@ static int at_xdmac_probe(struct platform_device *pdev)
 	atxdmac->dma.device_prep_interleaved_dma	= at_xdmac_prep_interleaved;
 	atxdmac->dma.device_prep_dma_memcpy		= at_xdmac_prep_dma_memcpy;
 	atxdmac->dma.device_prep_dma_memset		= at_xdmac_prep_dma_memset;
+	atxdmac->dma.device_prep_dma_memset_sg		= at_xdmac_prep_dma_memset_sg;
 	atxdmac->dma.device_prep_slave_sg		= at_xdmac_prep_slave_sg;
 	atxdmac->dma.device_config			= at_xdmac_device_config;
 	atxdmac->dma.device_pause			= at_xdmac_device_pause;