[PATCH v5 1/7] dmaengine: xilinx_dma: commonize DMA copy size calculation

This patch removes a bit of duplicated code by introducing a new
function that implements calculations for DMA copy size.

Suggested-by: Vinod Koul <vkoul@kernel.org>
Signed-off-by: Andrea Merello <redacted>
---
Changes in v4:
	- introduce this patch in the patch series
Changes in v5:
	None
---
 drivers/dma/xilinx/xilinx_dma.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index 27b523530c4a..a3aaa0e34cc7 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -952,6 +952,19 @@ static int xilinx_dma_alloc_chan_resources(struct dma_chan *dchan)
 	return 0;
 }
 
+/**
+ * xilinx_dma_calc_copysize - Calculate the amount of data to copy
+ * @size: Total data that needs to be copied
+ * @done: Amount of data that has been already copied
+ *
+ * Return: Amount of data that has to be copied
+ */
+static int xilinx_dma_calc_copysize(int size, int done)
+{
+	return min_t(size_t, size - done,
+		     XILINX_DMA_MAX_TRANS_LEN);
+}
+
 /**
  * xilinx_dma_tx_status - Get DMA transaction status
  * @dchan: DMA channel

@@ -1791,8 +1804,8 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_slave_sg(
 			 * Calculate the maximum number of bytes to transfer,
 			 * making sure it is less than the hw limit
 			 */
-			copy = min_t(size_t, sg_dma_len(sg) - sg_used,
-				     XILINX_DMA_MAX_TRANS_LEN);
+			copy = xilinx_dma_calc_copysize(sg_dma_len(sg),
+							sg_used);
 			hw = &segment->hw;
 
 			/* Fill in the descriptor */

@@ -1896,8 +1909,7 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_dma_cyclic(
 			 * Calculate the maximum number of bytes to transfer,
 			 * making sure it is less than the hw limit
 			 */
-			copy = min_t(size_t, period_len - sg_used,
-				     XILINX_DMA_MAX_TRANS_LEN);
+			copy = xilinx_dma_calc_copysize(period_len, sg_used);
 			hw = &segment->hw;
 			xilinx_axidma_buf(chan, hw, buf_addr, sg_used,
 					  period_len * i);

-- 
2.17.1

Whenever a single or cyclic transaction is prepared, the driver
could eventually split it over several SG descriptors in order
to deal with the HW maximum transfer length.

This could end up in DMA operations starting from a misaligned
address. This seems fatal for the HW if DRE (Data Realignment Engine)
is not enabled.

This patch eventually adjusts the transfer size in order to make sure
all operations start from an aligned address.

Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
---
Changes in v2:
        - don't introduce copy_mask field, rather rely on already-esistent
          copy_align field. Suggested by Radhey Shyam Pandey
        - reword title
Changes in v3:
	- fix bug introduced in v2: wrong copy size when DRE is enabled
	- use implementation suggested by Radhey Shyam Pandey
Changes in v4:
	- rework on the top of 1/6
Changes in v5:
	- fix typo in commit title
	- add hint about "DRE" meaning in commit message
---
 drivers/dma/xilinx/xilinx_dma.c | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index a3aaa0e34cc7..aaa6de8a70e4 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -954,15 +954,28 @@ static int xilinx_dma_alloc_chan_resources(struct dma_chan *dchan)
 
 /**
  * xilinx_dma_calc_copysize - Calculate the amount of data to copy
+ * @chan: Driver specific DMA channel
  * @size: Total data that needs to be copied
  * @done: Amount of data that has been already copied
  *
  * Return: Amount of data that has to be copied
  */
-static int xilinx_dma_calc_copysize(int size, int done)
+static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
+				    int size, int done)
 {
-	return min_t(size_t, size - done,
+	size_t copy = min_t(size_t, size - done,
 		     XILINX_DMA_MAX_TRANS_LEN);
+
+	if ((copy + done < size) &&
+	    chan->xdev->common.copy_align) {
+		/*
+		 * If this is not the last descriptor, make sure
+		 * the next one will be properly aligned
+		 */
+		copy = rounddown(copy,
+				 (1 << chan->xdev->common.copy_align));
+	}
+	return copy;
 }
 
 /**

@@ -1804,7 +1817,7 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_slave_sg(
 			 * Calculate the maximum number of bytes to transfer,
 			 * making sure it is less than the hw limit
 			 */
-			copy = xilinx_dma_calc_copysize(sg_dma_len(sg),
+			copy = xilinx_dma_calc_copysize(chan, sg_dma_len(sg),
 							sg_used);
 			hw = &segment->hw;
 

@@ -1909,7 +1922,8 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_dma_cyclic(
 			 * Calculate the maximum number of bytes to transfer,
 			 * making sure it is less than the hw limit
 			 */
-			copy = xilinx_dma_calc_copysize(period_len, sg_used);
+			copy = xilinx_dma_calc_copysize(chan,
+							period_len, sg_used);
 			hw = &segment->hw;
 			xilinx_axidma_buf(chan, hw, buf_addr, sg_used,
 					  period_len * i);

-- 
2.17.1

On 07-09-18, 08:24, Andrea Merello wrote:

align to preceeding line opening brace please

 {
-	return min_t(size_t, size - done,
+	size_t copy = min_t(size_t, size - done,
 		     XILINX_DMA_MAX_TRANS_LEN);

so we can do this way in patch 1:

        size t copy;

        copy =  min_t(size_t, size - done,
                      XILINX_DMA_MAX_TRANS_LEN);

        return copy;

and then add these here, feels like we are redoing change introduced in
patch 1..

why not keep chan in patch 1 and add only handling in patch 2, seems
less churn to me..

-- 
~Vinod

On Tue, Sep 18, 2018 at 6:21 PM Vinod [off-list ref] wrote:

On 07-09-18, 08:24, Andrea Merello wrote:

quoted

Whenever a single or cyclic transaction is prepared, the driver
could eventually split it over several SG descriptors in order
to deal with the HW maximum transfer length.

This could end up in DMA operations starting from a misaligned
address. This seems fatal for the HW if DRE (Data Realignment Engine)
is not enabled.

This patch eventually adjusts the transfer size in order to make sure
all operations start from an aligned address.

Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
---
Changes in v2:
        - don't introduce copy_mask field, rather rely on already-esistent
          copy_align field. Suggested by Radhey Shyam Pandey
        - reword title
Changes in v3:
      - fix bug introduced in v2: wrong copy size when DRE is enabled
      - use implementation suggested by Radhey Shyam Pandey
Changes in v4:
      - rework on the top of 1/6
Changes in v5:
      - fix typo in commit title
      - add hint about "DRE" meaning in commit message
---
 drivers/dma/xilinx/xilinx_dma.c | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index a3aaa0e34cc7..aaa6de8a70e4 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -954,15 +954,28 @@ static int xilinx_dma_alloc_chan_resources(struct dma_chan *dchan)

 /**
  * xilinx_dma_calc_copysize - Calculate the amount of data to copy
+ * @chan: Driver specific DMA channel
  * @size: Total data that needs to be copied
  * @done: Amount of data that has been already copied
  *
  * Return: Amount of data that has to be copied
  */
-static int xilinx_dma_calc_copysize(int size, int done)
+static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
+                                 int size, int done)

align to preceeding line opening brace please

After applying, I'm seeing it already aligned as you requested; 4 tabs
+ 4 spaces so the 2nd line starts right under the "s" near the opened
brace..
Patch sent using git, so it should pass through without being ruined;
don't know why you see it misaligned :(

quoted

 {
-     return min_t(size_t, size - done,
+     size_t copy = min_t(size_t, size - done,
                   XILINX_DMA_MAX_TRANS_LEN);

so we can do this way in patch 1:

        size t copy;

        copy =  min_t(size_t, size - done,
                      XILINX_DMA_MAX_TRANS_LEN);

        return copy;

and then add these here, feels like we are redoing change introduced in
patch 1..

OK, this sounds good :)

quoted

+     if ((copy + done < size) &&
+         chan->xdev->common.copy_align) {
+             /*
+              * If this is not the last descriptor, make sure
+              * the next one will be properly aligned
+              */
+             copy = rounddown(copy,
+                              (1 << chan->xdev->common.copy_align));
+     }
+     return copy;
 }

 /**

@@ -1804,7 +1817,7 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_slave_sg(
                       * Calculate the maximum number of bytes to transfer,
                       * making sure it is less than the hw limit
                       */
-                     copy = xilinx_dma_calc_copysize(sg_dma_len(sg),
+                     copy = xilinx_dma_calc_copysize(chan, sg_dma_len(sg),

why not keep chan in patch 1 and add only handling in patch 2, seems
less churn to me..

Indeed this was something I was unsure about.. I ended up in feeling
better not to add introduce a function that takes an unused (yet)
argument, but I can change this of course :)

--
~Vinod

On 28-09-18, 09:11, Andrea Merello wrote:

On Tue, Sep 18, 2018 at 6:21 PM Vinod [off-list ref] wrote:

quoted

quoted

@@ -1804,7 +1817,7 @@ static struct dma_async_tx_descriptor *xilinx_dma_prep_slave_sg(
                       * Calculate the maximum number of bytes to transfer,
                       * making sure it is less than the hw limit
                       */
-                     copy = xilinx_dma_calc_copysize(sg_dma_len(sg),
+                     copy = xilinx_dma_calc_copysize(chan, sg_dma_len(sg),

why not keep chan in patch 1 and add only handling in patch 2, seems
less churn to me..

Indeed this was something I was unsure about.. I ended up in feeling
better not to add introduce a function that takes an unused (yet)
argument, but I can change this of course :)

IMO It is fine to add a user in subsequent patch in a series. Not fine to
add something and not use in "that" series :)

-- 
~Vinod

The AXIDMA and CDMA HW can be either direct-access or scatter-gather
version. These are SW incompatible.

The driver can handle both versions: a DT property was used to
tell the driver whether to assume the HW is in scatter-gather mode.

This patch makes the driver to autodetect this information. The DT
property is not required anymore.

No changes for VDMA.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
---
Changes in v2:
	- autodetect only in !VDMA case
Changes in v3:
	- cc DT maintainers/ML
Changes in v4:
	- fix typos in commit message
Changes in v5:
	None
---
 drivers/dma/xilinx/xilinx_dma.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index b17f24e4ec35..78d0f2f8225e 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -86,6 +86,7 @@
 #define XILINX_DMA_DMASR_DMA_DEC_ERR		BIT(6)
 #define XILINX_DMA_DMASR_DMA_SLAVE_ERR		BIT(5)
 #define XILINX_DMA_DMASR_DMA_INT_ERR		BIT(4)
+#define XILINX_DMA_DMASR_SG_MASK		BIT(3)
 #define XILINX_DMA_DMASR_IDLE			BIT(1)
 #define XILINX_DMA_DMASR_HALTED		BIT(0)
 #define XILINX_DMA_DMASR_DELAY_MASK		GENMASK(31, 24)

@@ -407,7 +408,6 @@ struct xilinx_dma_config {
  * @dev: Device Structure
  * @common: DMA device structure
  * @chan: Driver specific DMA channel
- * @has_sg: Specifies whether Scatter-Gather is present or not
  * @mcdma: Specifies whether Multi-Channel is present or not
  * @flush_on_fsync: Flush on frame sync
  * @ext_addr: Indicates 64 bit addressing is supported by dma device

@@ -427,7 +427,6 @@ struct xilinx_dma_device {
 	struct device *dev;
 	struct dma_device common;
 	struct xilinx_dma_chan *chan[XILINX_DMA_MAX_CHANS_PER_DEVICE];
-	bool has_sg;
 	bool mcdma;
 	u32 flush_on_fsync;
 	bool ext_addr;

@@ -2400,7 +2399,6 @@ static int xilinx_dma_chan_probe(struct xilinx_dma_device *xdev,
 
 	chan->dev = xdev->dev;
 	chan->xdev = xdev;
-	chan->has_sg = xdev->has_sg;
 	chan->desc_pendingcount = 0x0;
 	chan->ext_addr = xdev->ext_addr;
 	/* This variable ensures that descriptors are not

@@ -2493,6 +2491,15 @@ static int xilinx_dma_chan_probe(struct xilinx_dma_device *xdev,
 		chan->stop_transfer = xilinx_dma_stop_transfer;
 	}
 
+	/* check if SG is enabled (only for AXIDMA and CDMA) */
+	if (xdev->dma_config->dmatype != XDMA_TYPE_VDMA) {
+		if (dma_ctrl_read(chan, XILINX_DMA_REG_DMASR) &
+		    XILINX_DMA_DMASR_SG_MASK)
+			chan->has_sg = true;
+		dev_dbg(chan->dev, "ch %d: SG %s\n", chan->id,
+			chan->has_sg ? "enabled" : "disabled");
+	}
+
 	/* Initialize the tasklet */
 	tasklet_init(&chan->tasklet, xilinx_dma_do_tasklet,
 			(unsigned long)chan);

@@ -2631,7 +2638,6 @@ static int xilinx_dma_probe(struct platform_device *pdev)
 		return PTR_ERR(xdev->regs);
 
 	/* Retrieve the DMA engine properties from the device tree */
-	xdev->has_sg = of_property_read_bool(node, "xlnx,include-sg");
 	xdev->max_buffer_len = GENMASK(XILINX_DMA_MAX_TRANS_LEN_MAX - 1, 0);
 
 	if (xdev->dma_config->dmatype == XDMA_TYPE_AXIDMA) {

-- 
2.17.1

xilinx_vdma_start_transfer() is used only for VDMA IP, still it contains
conditional code on has_sg variable. has_sg is set only whenever the HW
does support SG mode, that is never true for VDMA IP.

This patch drops the never-taken branches.

Signed-off-by: Andrea Merello <redacted>
---
Changes in V4: introduced this patch in series
Changes in v5:
	None
---
 drivers/dma/xilinx/xilinx_dma.c | 84 +++++++++++++--------------------
 1 file changed, 32 insertions(+), 52 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index 78d0f2f8225e..07ceadef0a00 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -1093,6 +1093,8 @@ static void xilinx_vdma_start_transfer(struct xilinx_dma_chan *chan)
 	struct xilinx_dma_tx_descriptor *desc, *tail_desc;
 	u32 reg, j;
 	struct xilinx_vdma_tx_segment *tail_segment;
+	struct xilinx_vdma_tx_segment *segment, *last = NULL;
+	int i = 0;
 
 	/* This function was invoked with lock held */
 	if (chan->err)

@@ -1112,14 +1114,6 @@ static void xilinx_vdma_start_transfer(struct xilinx_dma_chan *chan)
 	tail_segment = list_last_entry(&tail_desc->segments,
 				       struct xilinx_vdma_tx_segment, node);
 
-	/*
-	 * If hardware is idle, then all descriptors on the running lists are
-	 * done, start new transfers
-	 */
-	if (chan->has_sg)
-		dma_ctrl_write(chan, XILINX_DMA_REG_CURDESC,
-				desc->async_tx.phys);
-
 	/* Configure the hardware using info in the config structure */
 	reg = dma_ctrl_read(chan, XILINX_DMA_REG_DMACR);
 

@@ -1128,15 +1122,11 @@ static void xilinx_vdma_start_transfer(struct xilinx_dma_chan *chan)
 	else
 		reg &= ~XILINX_DMA_DMACR_FRAMECNT_EN;
 
-	/*
-	 * With SG, start with circular mode, so that BDs can be fetched.
-	 * In direct register mode, if not parking, enable circular mode
-	 */
-	if (chan->has_sg || !config->park)
-		reg |= XILINX_DMA_DMACR_CIRC_EN;
-
+	/* If not parking, enable circular mode */
 	if (config->park)
 		reg &= ~XILINX_DMA_DMACR_CIRC_EN;
+	else
+		reg |= XILINX_DMA_DMACR_CIRC_EN;
 
 	dma_ctrl_write(chan, XILINX_DMA_REG_DMACR, reg);
 

@@ -1158,48 +1148,38 @@ static void xilinx_vdma_start_transfer(struct xilinx_dma_chan *chan)
 		return;
 
 	/* Start the transfer */
-	if (chan->has_sg) {
-		dma_ctrl_write(chan, XILINX_DMA_REG_TAILDESC,
-				tail_segment->phys);
-		list_splice_tail_init(&chan->pending_list, &chan->active_list);
-		chan->desc_pendingcount = 0;
-	} else {
-		struct xilinx_vdma_tx_segment *segment, *last = NULL;
-		int i = 0;
-
-		if (chan->desc_submitcount < chan->num_frms)
-			i = chan->desc_submitcount;
-
-		list_for_each_entry(segment, &desc->segments, node) {
-			if (chan->ext_addr)
-				vdma_desc_write_64(chan,
-					XILINX_VDMA_REG_START_ADDRESS_64(i++),
-					segment->hw.buf_addr,
-					segment->hw.buf_addr_msb);
-			else
-				vdma_desc_write(chan,
+	if (chan->desc_submitcount < chan->num_frms)
+		i = chan->desc_submitcount;
+
+	list_for_each_entry(segment, &desc->segments, node) {
+		if (chan->ext_addr)
+			vdma_desc_write_64(chan,
+				   XILINX_VDMA_REG_START_ADDRESS_64(i++),
+				   segment->hw.buf_addr,
+				   segment->hw.buf_addr_msb);
+		else
+			vdma_desc_write(chan,
 					XILINX_VDMA_REG_START_ADDRESS(i++),
 					segment->hw.buf_addr);
 
-			last = segment;
-		}
-
-		if (!last)
-			return;
+		last = segment;
+	}
 
-		/* HW expects these parameters to be same for one transaction */
-		vdma_desc_write(chan, XILINX_DMA_REG_HSIZE, last->hw.hsize);
-		vdma_desc_write(chan, XILINX_DMA_REG_FRMDLY_STRIDE,
-				last->hw.stride);
-		vdma_desc_write(chan, XILINX_DMA_REG_VSIZE, last->hw.vsize);
+	if (!last)
+		return;
 
-		chan->desc_submitcount++;
-		chan->desc_pendingcount--;
-		list_del(&desc->node);
-		list_add_tail(&desc->node, &chan->active_list);
-		if (chan->desc_submitcount == chan->num_frms)
-			chan->desc_submitcount = 0;
-	}
+	/* HW expects these parameters to be same for one transaction */
+	vdma_desc_write(chan, XILINX_DMA_REG_HSIZE, last->hw.hsize);
+	vdma_desc_write(chan, XILINX_DMA_REG_FRMDLY_STRIDE,
+			last->hw.stride);
+	vdma_desc_write(chan, XILINX_DMA_REG_VSIZE, last->hw.vsize);
+
+	chan->desc_submitcount++;
+	chan->desc_pendingcount--;
+	list_del(&desc->node);
+	list_add_tail(&desc->node, &chan->active_list);
+	if (chan->desc_submitcount == chan->num_frms)
+		chan->desc_submitcount = 0;
 
 	chan->idle = false;
 }

-- 
2.17.1

The width of the "length register" cannot be autodetected, and it is now
specified with a DT property. Add documentation for it.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
---
Changes in v2:
	- change property name
	- property is now optional
	- cc DT maintainer
Changes in v3:
	- reword
	- cc DT maintainerS and ML
Changes in v4:
	- specify the unit, the valid range and the default value
Changes in v5:
	- commit message trivial fix
	- fix spaces before tab
---
 Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt b/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt
index a2b8bfaec43c..5df4eac7300c 100644
--- a/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt
+++ b/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt

@@ -41,6 +41,10 @@ Optional properties:
 - xlnx,include-sg: Tells configured for Scatter-mode in
 	the hardware.
 Optional properties for AXI DMA:
+- xlnx,sg-length-width: Should be set to the width in bits of the length
+	register as configured in h/w. Takes values {8...26}. If the property
+	is missing or invalid then the default value 23 is used. This is the
+	maximum value that is supported by all IP versions.
 - xlnx,mcdma: Tells whether configured for multi-channel mode in the hardware.
 Optional properties for VDMA:
 - xlnx,flush-fsync: Tells which channel to Flush on Frame sync.

-- 
2.17.1

On Fri,  7 Sep 2018 08:24:58 +0200, Andrea Merello wrote:

The width of the "length register" cannot be autodetected, and it is now
specified with a DT property. Add documentation for it.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
---
Changes in v2:
	- change property name
	- property is now optional
	- cc DT maintainer
Changes in v3:
	- reword
	- cc DT maintainerS and ML
Changes in v4:
	- specify the unit, the valid range and the default value
Changes in v5:
	- commit message trivial fix
	- fix spaces before tab
---
 Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt | 4 ++++
 1 file changed, 4 insertions(+)

Reviewed-by: Rob Herring <robh@kernel.org>

From: Radhey Shyam Pandey <redacted>

AXI-DMA IP supports configurable (c_sg_length_width) buffer length
register width, hence read buffer length (xlnx,sg-length-width) DT
property and ensure that driver doesn't program buffer length
exceeding the supported limit. For VDMA and CDMA there is no change.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Signed-off-by: Radhey Shyam Pandey <redacted>
Signed-off-by: Michal Simek <redacted>
Signed-off-by: Andrea Merello <redacted> [rebase, reword]
---
Changes in v2:
        - drop original patch and replace with the one in Xilinx tree
Changes in v3:
	- cc DT maintainers/ML
Changes in v4:
	- upper bound for the property should be 26, not 23
	- add warn for width > 23 as per xilinx original patch
	- rework due to changes introduced in 1/6
Changes in v5:
	None
---
 drivers/dma/xilinx/xilinx_dma.c | 36 +++++++++++++++++++++++++--------
 1 file changed, 28 insertions(+), 8 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index aaa6de8a70e4..b17f24e4ec35 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -158,7 +158,9 @@
 #define XILINX_DMA_REG_BTT		0x28
 
 /* AXI DMA Specific Masks/Bit fields */
-#define XILINX_DMA_MAX_TRANS_LEN	GENMASK(22, 0)
+#define XILINX_DMA_MAX_TRANS_LEN_MIN	8
+#define XILINX_DMA_MAX_TRANS_LEN_MAX	23
+#define XILINX_DMA_V2_MAX_TRANS_LEN_MAX	26
 #define XILINX_DMA_CR_COALESCE_MAX	GENMASK(23, 16)
 #define XILINX_DMA_CR_CYCLIC_BD_EN_MASK	BIT(4)
 #define XILINX_DMA_CR_COALESCE_SHIFT	16

@@ -418,6 +420,7 @@ struct xilinx_dma_config {
  * @rxs_clk: DMA s2mm stream clock
  * @nr_channels: Number of channels DMA device supports
  * @chan_id: DMA channel identifier
+ * @max_buffer_len: Max buffer length
  */
 struct xilinx_dma_device {
 	void __iomem *regs;

@@ -437,6 +440,7 @@ struct xilinx_dma_device {
 	struct clk *rxs_clk;
 	u32 nr_channels;
 	u32 chan_id;
+	u32 max_buffer_len;
 };
 
 /* Macros */

@@ -964,7 +968,7 @@ static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
 				    int size, int done)
 {
 	size_t copy = min_t(size_t, size - done,
-		     XILINX_DMA_MAX_TRANS_LEN);
+			    chan->xdev->max_buffer_len);
 
 	if ((copy + done < size) &&
 	    chan->xdev->common.copy_align) {

@@ -1011,7 +1015,7 @@ static enum dma_status xilinx_dma_tx_status(struct dma_chan *dchan,
 			list_for_each_entry(segment, &desc->segments, node) {
 				hw = &segment->hw;
 				residue += (hw->control - hw->status) &
-					   XILINX_DMA_MAX_TRANS_LEN;
+					   chan->xdev->max_buffer_len;
 			}
 		}
 		spin_unlock_irqrestore(&chan->lock, flags);

@@ -1263,7 +1267,7 @@ static void xilinx_cdma_start_transfer(struct xilinx_dma_chan *chan)
 
 		/* Start the transfer */
 		dma_ctrl_write(chan, XILINX_DMA_REG_BTT,
-				hw->control & XILINX_DMA_MAX_TRANS_LEN);
+				hw->control & chan->xdev->max_buffer_len);
 	}
 
 	list_splice_tail_init(&chan->pending_list, &chan->active_list);

@@ -1366,7 +1370,7 @@ static void xilinx_dma_start_transfer(struct xilinx_dma_chan *chan)
 
 		/* Start the transfer */
 		dma_ctrl_write(chan, XILINX_DMA_REG_BTT,
-			       hw->control & XILINX_DMA_MAX_TRANS_LEN);
+			       hw->control & chan->xdev->max_buffer_len);
 	}
 
 	list_splice_tail_init(&chan->pending_list, &chan->active_list);

@@ -1727,7 +1731,7 @@ xilinx_cdma_prep_memcpy(struct dma_chan *dchan, dma_addr_t dma_dst,
 	struct xilinx_cdma_tx_segment *segment;
 	struct xilinx_cdma_desc_hw *hw;
 
-	if (!len || len > XILINX_DMA_MAX_TRANS_LEN)
+	if (!len || len > chan->xdev->max_buffer_len)
 		return NULL;
 
 	desc = xilinx_dma_alloc_tx_descriptor(chan);

@@ -2596,7 +2600,7 @@ static int xilinx_dma_probe(struct platform_device *pdev)
 	struct xilinx_dma_device *xdev;
 	struct device_node *child, *np = pdev->dev.of_node;
 	struct resource *io;
-	u32 num_frames, addr_width;
+	u32 num_frames, addr_width, len_width;
 	int i, err;
 
 	/* Allocate and initialize the DMA engine structure */

@@ -2628,8 +2632,24 @@ static int xilinx_dma_probe(struct platform_device *pdev)
 
 	/* Retrieve the DMA engine properties from the device tree */
 	xdev->has_sg = of_property_read_bool(node, "xlnx,include-sg");
-	if (xdev->dma_config->dmatype == XDMA_TYPE_AXIDMA)
+	xdev->max_buffer_len = GENMASK(XILINX_DMA_MAX_TRANS_LEN_MAX - 1, 0);
+
+	if (xdev->dma_config->dmatype == XDMA_TYPE_AXIDMA) {
 		xdev->mcdma = of_property_read_bool(node, "xlnx,mcdma");
+		if (!of_property_read_u32(node, "xlnx,sg-length-width",
+					  &len_width)) {
+			if (len_width < XILINX_DMA_MAX_TRANS_LEN_MIN ||
+			    len_width > XILINX_DMA_V2_MAX_TRANS_LEN_MAX) {
+				dev_warn(xdev->dev,
+					 "invalid xlnx,sg-length-width property value. Using default width\n");
+			} else {
+				if (len_width > XILINX_DMA_MAX_TRANS_LEN_MAX)
+					dev_warn(xdev->dev, "Please ensure that IP supports buffer length > 23 bits\n");
+				xdev->max_buffer_len =
+					GENMASK(len_width - 1, 0);
+			}
+		}
+	}
 
 	if (xdev->dma_config->dmatype == XDMA_TYPE_VDMA) {
 		err = of_property_read_u32(node, "xlnx,num-fstores",

-- 
2.17.1

On 07-09-18, 08:24, Andrea Merello wrote:

hmm why not add max_buffer_len in patch 1 again, and then use default
len as XILINX_DMA_MAX_TRANS_LEN and add multiple lengths here :)

- 
~Vinod

On Tue, Sep 18, 2018 at 6:25 PM Vinod [off-list ref] wrote:

On 07-09-18, 08:24, Andrea Merello wrote:

quoted

From: Radhey Shyam Pandey <redacted>

AXI-DMA IP supports configurable (c_sg_length_width) buffer length
register width, hence read buffer length (xlnx,sg-length-width) DT
property and ensure that driver doesn't program buffer length
exceeding the supported limit. For VDMA and CDMA there is no change.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Signed-off-by: Radhey Shyam Pandey <redacted>
Signed-off-by: Michal Simek <redacted>
Signed-off-by: Andrea Merello <redacted> [rebase, reword]
---
Changes in v2:
        - drop original patch and replace with the one in Xilinx tree
Changes in v3:
      - cc DT maintainers/ML
Changes in v4:
      - upper bound for the property should be 26, not 23
      - add warn for width > 23 as per xilinx original patch
      - rework due to changes introduced in 1/6
Changes in v5:
      None
---
 drivers/dma/xilinx/xilinx_dma.c | 36 +++++++++++++++++++++++++--------
 1 file changed, 28 insertions(+), 8 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dma.c b/drivers/dma/xilinx/xilinx_dma.c
index aaa6de8a70e4..b17f24e4ec35 100644
--- a/drivers/dma/xilinx/xilinx_dma.c
+++ b/drivers/dma/xilinx/xilinx_dma.c

@@ -158,7 +158,9 @@
 #define XILINX_DMA_REG_BTT           0x28

 /* AXI DMA Specific Masks/Bit fields */
-#define XILINX_DMA_MAX_TRANS_LEN     GENMASK(22, 0)
+#define XILINX_DMA_MAX_TRANS_LEN_MIN 8
+#define XILINX_DMA_MAX_TRANS_LEN_MAX 23
+#define XILINX_DMA_V2_MAX_TRANS_LEN_MAX      26
 #define XILINX_DMA_CR_COALESCE_MAX   GENMASK(23, 16)
 #define XILINX_DMA_CR_CYCLIC_BD_EN_MASK      BIT(4)
 #define XILINX_DMA_CR_COALESCE_SHIFT 16

@@ -418,6 +420,7 @@ struct xilinx_dma_config {
  * @rxs_clk: DMA s2mm stream clock
  * @nr_channels: Number of channels DMA device supports
  * @chan_id: DMA channel identifier
+ * @max_buffer_len: Max buffer length
  */
 struct xilinx_dma_device {
      void __iomem *regs;

@@ -437,6 +440,7 @@ struct xilinx_dma_device {
      struct clk *rxs_clk;
      u32 nr_channels;
      u32 chan_id;
+     u32 max_buffer_len;
 };

 /* Macros */

@@ -964,7 +968,7 @@ static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
                                  int size, int done)
 {
      size_t copy = min_t(size_t, size - done,
-                  XILINX_DMA_MAX_TRANS_LEN);
+                         chan->xdev->max_buffer_len);

hmm why not add max_buffer_len in patch 1 again, and then use default
len as XILINX_DMA_MAX_TRANS_LEN and add multiple lengths here :)

Sorry, I'm not getting your point. Could you please elaborate the "add
multiple lengths here" thing ?

-
~Vinod

On 28-09-18, 08:53, Andrea Merello wrote:

On Tue, Sep 18, 2018 at 6:25 PM Vinod [off-list ref] wrote:

quoted

quoted

@@ -964,7 +968,7 @@ static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
                                  int size, int done)
 {
      size_t copy = min_t(size_t, size - done,
-                  XILINX_DMA_MAX_TRANS_LEN);
+                         chan->xdev->max_buffer_len);

hmm why not add max_buffer_len in patch 1 again, and then use default
len as XILINX_DMA_MAX_TRANS_LEN and add multiple lengths here :)

Sorry, I'm not getting your point. Could you please elaborate the "add
multiple lengths here" thing ?

IIRC (sorry been travelling and vacation), add
chan->xdev->max_buffer_len in patch 1 and initialize it to
XILINX_DMA_MAX_TRANS_LEN. Then in subsequent patches update the length.

-- 
~Vinod

On Tue, Oct 2, 2018 at 4:56 PM Vinod [off-list ref] wrote:

On 28-09-18, 08:53, Andrea Merello wrote:

quoted

On Tue, Sep 18, 2018 at 6:25 PM Vinod [off-list ref] wrote:

quoted

quoted

quoted

@@ -964,7 +968,7 @@ static int xilinx_dma_calc_copysize(struct xilinx_dma_chan *chan,
                                  int size, int done)
 {
      size_t copy = min_t(size_t, size - done,
-                  XILINX_DMA_MAX_TRANS_LEN);
+                         chan->xdev->max_buffer_len);

hmm why not add max_buffer_len in patch 1 again, and then use default
len as XILINX_DMA_MAX_TRANS_LEN and add multiple lengths here :)

Sorry, I'm not getting your point. Could you please elaborate the "add
multiple lengths here" thing ?

IIRC (sorry been travelling and vacation), add
chan->xdev->max_buffer_len in patch 1 and initialize it to
XILINX_DMA_MAX_TRANS_LEN. Then in subsequent patches update the length.

Ah ok. IMO introducing max_buffer_len seems more related to what 4/7
does (actually getting the max transfer len from DT,  thus it is not
constant anymore) rather than to what 1/7 does (commonizing the
calculation of transfer len as it is).. This is why I've introduced it
in 4/7..

.. But if you prefer this way, I'll change this :) .. Maybe we can
change 1/7 commit message so that this change looks less off-topic..
But I have not found a very good title yet.. Something like "Prepare
for DMA copy size calculation rework" ?

--
~Vinod

This property is not needed anymore, because the driver now autodetects it.
Delete references in documentation.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: devicetree at vger.kernel.org
Cc: Radhey Shyam Pandey <redacted>
Signed-off-by: Andrea Merello <redacted>
Reviewed-by: Radhey Shyam Pandey <redacted>
Reviewed-by: Rob Herring <robh@kernel.org>
---
Changes in v2:
	- cc DT maintainer
Changes in v3:
	- cc DT maintainerS/ML
Changes in v4:
	None
Changes in v5:
	None
---
 Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt b/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt
index 5df4eac7300c..6303ce7fcc3d 100644
--- a/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt
+++ b/Documentation/devicetree/bindings/dma/xilinx/xilinx_dma.txt

@@ -37,9 +37,6 @@ Required properties:
 Required properties for VDMA:
 - xlnx,num-fstores: Should be the number of framebuffers as configured in h/w.
 
-Optional properties:
-- xlnx,include-sg: Tells configured for Scatter-mode in
-	the hardware.
 Optional properties for AXI DMA:
 - xlnx,sg-length-width: Should be set to the width in bits of the length
 	register as configured in h/w. Takes values {8...26}. If the property

-- 
2.17.1