[V4] dmaengine: axi-dmac: Split too large segments

[V4] dmaengine: axi-dmac: Split too large segments

[Ardelean, Alexandru]

From: Lars-Peter Clausen <lars@metafoo.de>

The axi-dmac driver currently rejects transfers with segments that are
larger than what the hardware can handle.

Re-work the driver so that these large segments are split into multiple
segments instead where each segment is smaller or equal to the maximum
segment size.

This allows the driver to handle transfers with segments of arbitrary size.

Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Bogdan Togorean <bogdan.togorean@analog.com>
Signed-off-by: Alexandru Ardelean <alex.ardelean@analog.com>
---

Changelog V1->V2:
* reworked this patch to not use the `sg_nents_for_dma()` helper from
  https://patchwork.kernel.org/patch/9389821/
  After some discussion with Andy Shevchenko and some thought about it, it
  probably does not make sense to have this helper for this driver only,
  since a more generic solution could be implemented via DMAEngine API.
  This generic solution is not yet implemented (as I understood it from the
  discussion).

Changelog V2->V3:
* removed adding `last` field ; patch V2 was adding a `last` field to the
  `struct axi_dmac_sg`, which wasn't being used (neither in later patch);
  also it wasn't properly set

Changelog V3->V4:
* Bogdan found that the `axi_dmac_fill_linear_sg()` function was called
  incorrectly in `axi_dmac_prep_dma_cyclic()` ; it was passing the
  `buf_len` instead of the `period_len` which would then be iterated
  `num_periods`; V4 calls the `axi_dmac_fill_linear_sg()` with `period_len`

 drivers/dma/dma-axi-dmac.c | 81 ++++++++++++++++++++++++++++----------
 1 file changed, 61 insertions(+), 20 deletions(-)

diff --git a/drivers/dma/dma-axi-dmac.c b/drivers/dma/dma-axi-dmac.c
index ffc0adc2f6ce..0fe3a931d8d5 100644
--- a/drivers/dma/dma-axi-dmac.c
+++ b/drivers/dma/dma-axi-dmac.c
@@ -166,7 +166,7 @@ static int axi_dmac_dest_is_mem(struct axi_dmac_chan *chan)
 
 static bool axi_dmac_check_len(struct axi_dmac_chan *chan, unsigned int len)
 {
-	if (len == 0 || len > chan->max_length)
+	if (len == 0)
 		return false;
 	if ((len & chan->align_mask) != 0) /* Not aligned */
 		return false;
@@ -379,6 +379,49 @@ static struct axi_dmac_desc *axi_dmac_alloc_desc(unsigned int num_sgs)
 	return desc;
 }
 
+static struct axi_dmac_sg *axi_dmac_fill_linear_sg(struct axi_dmac_chan *chan,
+	enum dma_transfer_direction direction, dma_addr_t addr,
+	unsigned int num_periods, unsigned int period_len,
+	struct axi_dmac_sg *sg)
+{
+	unsigned int num_segments, i;
+	unsigned int segment_size;
+	unsigned int len;
+
+	/* Split into multiple equally sized segments if necessary */
+	num_segments = DIV_ROUND_UP(period_len, chan->max_length);
+	segment_size = DIV_ROUND_UP(period_len, num_segments);
+	/* Take care of alignment */
+	segment_size = ((segment_size - 1) | chan->align_mask) + 1;
+
+	for (i = 0; i < num_periods; i++) {
+		len = period_len;
+
+		while (len > segment_size) {
+			if (direction == DMA_DEV_TO_MEM)
+				sg->dest_addr = addr;
+			else
+				sg->src_addr = addr;
+			sg->x_len = segment_size;
+			sg->y_len = 1;
+			sg++;
+			addr += segment_size;
+			len -= segment_size;
+		}
+
+		if (direction == DMA_DEV_TO_MEM)
+			sg->dest_addr = addr;
+		else
+			sg->src_addr = addr;
+		sg->x_len = len;
+		sg->y_len = 1;
+		sg++;
+		addr += len;
+	}
+
+	return sg;
+}
+
 static struct dma_async_tx_descriptor *axi_dmac_prep_slave_sg(
 	struct dma_chan *c, struct scatterlist *sgl,
 	unsigned int sg_len, enum dma_transfer_direction direction,
@@ -386,16 +429,24 @@ static struct dma_async_tx_descriptor *axi_dmac_prep_slave_sg(
 {
 	struct axi_dmac_chan *chan = to_axi_dmac_chan(c);
 	struct axi_dmac_desc *desc;
+	struct axi_dmac_sg *dsg;
 	struct scatterlist *sg;
+	unsigned int num_sgs;
 	unsigned int i;
 
 	if (direction != chan->direction)
 		return NULL;
 
-	desc = axi_dmac_alloc_desc(sg_len);
+	num_sgs = 0;
+	for_each_sg(sgl, sg, sg_len, i)
+		num_sgs += DIV_ROUND_UP(sg_dma_len(sg), chan->max_length);
+
+	desc = axi_dmac_alloc_desc(num_sgs);
 	if (!desc)
 		return NULL;
 
+	dsg = desc->sg;
+
 	for_each_sg(sgl, sg, sg_len, i) {
 		if (!axi_dmac_check_addr(chan, sg_dma_address(sg)) ||
 		    !axi_dmac_check_len(chan, sg_dma_len(sg))) {
@@ -403,12 +454,8 @@ static struct dma_async_tx_descriptor *axi_dmac_prep_slave_sg(
 			return NULL;
 		}
 
-		if (direction == DMA_DEV_TO_MEM)
-			desc->sg[i].dest_addr = sg_dma_address(sg);
-		else
-			desc->sg[i].src_addr = sg_dma_address(sg);
-		desc->sg[i].x_len = sg_dma_len(sg);
-		desc->sg[i].y_len = 1;
+		dsg = axi_dmac_fill_linear_sg(chan, direction, sg_dma_address(sg), 1,
+			sg_dma_len(sg), dsg);
 	}
 
 	desc->cyclic = false;
@@ -423,7 +470,7 @@ static struct dma_async_tx_descriptor *axi_dmac_prep_dma_cyclic(
 {
 	struct axi_dmac_chan *chan = to_axi_dmac_chan(c);
 	struct axi_dmac_desc *desc;
-	unsigned int num_periods, i;
+	unsigned int num_periods, num_segments;
 
 	if (direction != chan->direction)
 		return NULL;
@@ -436,20 +483,14 @@ static struct dma_async_tx_descriptor *axi_dmac_prep_dma_cyclic(
 		return NULL;
 
 	num_periods = buf_len / period_len;
+	num_segments = DIV_ROUND_UP(period_len, chan->max_length);
 
-	desc = axi_dmac_alloc_desc(num_periods);
+	desc = axi_dmac_alloc_desc(num_periods * num_segments);
 	if (!desc)
 		return NULL;
 
-	for (i = 0; i < num_periods; i++) {
-		if (direction == DMA_DEV_TO_MEM)
-			desc->sg[i].dest_addr = buf_addr;
-		else
-			desc->sg[i].src_addr = buf_addr;
-		desc->sg[i].x_len = period_len;
-		desc->sg[i].y_len = 1;
-		buf_addr += period_len;
-	}
+	axi_dmac_fill_linear_sg(chan, direction, buf_addr, num_periods,
+		period_len, desc->sg);
 
 	desc->cyclic = true;
 
@@ -647,7 +688,7 @@ static int axi_dmac_probe(struct platform_device *pdev)
 	of_node_put(of_channels);
 
 	pdev->dev.dma_parms = &dmac->dma_parms;
-	dma_set_max_seg_size(&pdev->dev, dmac->chan.max_length);
+	dma_set_max_seg_size(&pdev->dev, UINT_MAX);
 
 	dma_dev = &dmac->dma_dev;
 	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);

[V4] dmaengine: axi-dmac: Split too large segments

[Vinod Koul]

On 08-03-19, 15:02, Alexandru Ardelean wrote:
> From: Lars-Peter Clausen <lars@metafoo.de>
> 
> The axi-dmac driver currently rejects transfers with segments that are
> larger than what the hardware can handle.
> 
> Re-work the driver so that these large segments are split into multiple
> segments instead where each segment is smaller or equal to the maximum
> segment size.
> 
> This allows the driver to handle transfers with segments of arbitrary size.

Applied, thanks