Simplify xdma_xfer_stop(). Stop the dma engine and clear its status
register unconditionally - just do what its name states. This change
also allows to call it without grabbing a lock, which minimizes
the total time spent with a spinlock held.

Delete the currently processed vd.node from the vc.desc_issued list
prior to passing it to vchan_terminate_vdesc(). In case there's more
than one descriptor pending on vc.desc_issued list, calling
vchan_terminate_desc() results in losing the link between
vc.desc_issued list head and the second descriptor on the list. Doing so
results in resources leakege, as vchan_dma_desc_free_list() won't be
able to properly free memory resources attached to descriptors,
resulting in dma_pool_destroy() failure.

Don't call vchan_dma_desc_free_list() from within xdma_terminate_all().
Move all terminated descriptors to the vc.desc_terminated list instead.
This allows to postpone freeing memory resources associated with
descriptors until the call to vchan_synchronize(), which is called from
xdma_synchronize() callback. This is the right way to do it -
xdma_terminate_all() should return as soon as possible, while freeing
resources (that may be time consuming in case of large number of
descriptors) can be done safely later.

Fixes: 290bb5d2d1e2
("dmaengine: xilinx: xdma: Add terminate_all/synchronize callbacks")

Signed-off-by: Jan Kuliga <jankul@alatek.krakow.pl>
---
 drivers/dma/xilinx/xdma.c | 32 ++++++++++++++++----------------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/drivers/dma/xilinx/xdma.c b/drivers/dma/xilinx/xdma.c
index 1bce48e5d86c..521ba2a653b6 100644
--- a/drivers/dma/xilinx/xdma.c
+++ b/drivers/dma/xilinx/xdma.c
@@ -379,20 +379,20 @@ static int xdma_xfer_start(struct xdma_chan *xchan)
  */
 static int xdma_xfer_stop(struct xdma_chan *xchan)
 {
-	struct virt_dma_desc *vd = vchan_next_desc(&xchan->vchan);
-	struct xdma_device *xdev = xchan->xdev_hdl;
 	int ret;
-
-	if (!vd || !xchan->busy)
-		return -EINVAL;
+	u32 val;
+	struct xdma_device *xdev = xchan->xdev_hdl;

 	/* clear run stop bit to prevent any further auto-triggering */
 	ret = regmap_write(xdev->rmap, xchan->base + XDMA_CHAN_CONTROL_W1C,
-			   CHAN_CTRL_RUN_STOP);
+							CHAN_CTRL_RUN_STOP);
 	if (ret)
 		return ret;

-	xchan->busy = false;
+	/* Clear the channel status register */
+	ret = regmap_read(xdev->rmap, xchan->base + XDMA_CHAN_STATUS_RC, &val);
+	if (ret)
+		return ret;

 	return 0;
 }
@@ -505,25 +505,25 @@ static void xdma_issue_pending(struct dma_chan *chan)
 static int xdma_terminate_all(struct dma_chan *chan)
 {
 	struct xdma_chan *xdma_chan = to_xdma_chan(chan);
-	struct xdma_desc *desc = NULL;
 	struct virt_dma_desc *vd;
 	unsigned long flags;
 	LIST_HEAD(head);

-	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
 	xdma_xfer_stop(xdma_chan);

+	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
+
+	xdma_chan->busy = false;
 	vd = vchan_next_desc(&xdma_chan->vchan);
-	if (vd)
-		desc = to_xdma_desc(vd);
-	if (desc) {
-		dma_cookie_complete(&desc->vdesc.tx);
-		vchan_terminate_vdesc(&desc->vdesc);
+	if (vd) {
+		list_del(&vd->node);
+		dma_cookie_complete(&vd->tx);
+		vchan_terminate_vdesc(vd);
 	}
-
 	vchan_get_all_descriptors(&xdma_chan->vchan, &head);
+	list_splice_tail(&head, &xdma_chan->vchan.desc_terminated);
+
 	spin_unlock_irqrestore(&xdma_chan->vchan.lock, flags);
-	vchan_dma_desc_free_list(&xdma_chan->vchan, &head);

 	return 0;
 }
--
2.34.1

On 08-12-23, 14:49, Jan Kuliga wrote:
> Simplify xdma_xfer_stop(). Stop the dma engine and clear its status
> register unconditionally - just do what its name states. This change
> also allows to call it without grabbing a lock, which minimizes
> the total time spent with a spinlock held.
> 
> Delete the currently processed vd.node from the vc.desc_issued list
> prior to passing it to vchan_terminate_vdesc(). In case there's more
> than one descriptor pending on vc.desc_issued list, calling
> vchan_terminate_desc() results in losing the link between
> vc.desc_issued list head and the second descriptor on the list. Doing so
> results in resources leakege, as vchan_dma_desc_free_list() won't be
> able to properly free memory resources attached to descriptors,
> resulting in dma_pool_destroy() failure.
> 
> Don't call vchan_dma_desc_free_list() from within xdma_terminate_all().
> Move all terminated descriptors to the vc.desc_terminated list instead.
> This allows to postpone freeing memory resources associated with
> descriptors until the call to vchan_synchronize(), which is called from
> xdma_synchronize() callback. This is the right way to do it -
> xdma_terminate_all() should return as soon as possible, while freeing
> resources (that may be time consuming in case of large number of
> descriptors) can be done safely later.
> 
> Fixes: 290bb5d2d1e2
> ("dmaengine: xilinx: xdma: Add terminate_all/synchronize callbacks")
> 
> Signed-off-by: Jan Kuliga <jankul@alatek.krakow.pl>
> ---
>  drivers/dma/xilinx/xdma.c | 32 ++++++++++++++++----------------
>  1 file changed, 16 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/dma/xilinx/xdma.c b/drivers/dma/xilinx/xdma.c
> index 1bce48e5d86c..521ba2a653b6 100644
> --- a/drivers/dma/xilinx/xdma.c
> +++ b/drivers/dma/xilinx/xdma.c
> @@ -379,20 +379,20 @@ static int xdma_xfer_start(struct xdma_chan *xchan)
>   */
>  static int xdma_xfer_stop(struct xdma_chan *xchan)
>  {
> -	struct virt_dma_desc *vd = vchan_next_desc(&xchan->vchan);
> -	struct xdma_device *xdev = xchan->xdev_hdl;
>  	int ret;
> -
> -	if (!vd || !xchan->busy)
> -		return -EINVAL;
> +	u32 val;
> +	struct xdma_device *xdev = xchan->xdev_hdl;
> 
>  	/* clear run stop bit to prevent any further auto-triggering */
>  	ret = regmap_write(xdev->rmap, xchan->base + XDMA_CHAN_CONTROL_W1C,
> -			   CHAN_CTRL_RUN_STOP);
> +							CHAN_CTRL_RUN_STOP);

Why this change, checkpatch would tell you this is not expected
alignment (run with strict)

>  	if (ret)
>  		return ret;
> 
> -	xchan->busy = false;
> +	/* Clear the channel status register */
> +	ret = regmap_read(xdev->rmap, xchan->base + XDMA_CHAN_STATUS_RC, &val);
> +	if (ret)
> +		return ret;
> 
>  	return 0;
>  }
> @@ -505,25 +505,25 @@ static void xdma_issue_pending(struct dma_chan *chan)
>  static int xdma_terminate_all(struct dma_chan *chan)
>  {
>  	struct xdma_chan *xdma_chan = to_xdma_chan(chan);
> -	struct xdma_desc *desc = NULL;
>  	struct virt_dma_desc *vd;
>  	unsigned long flags;
>  	LIST_HEAD(head);
> 
> -	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
>  	xdma_xfer_stop(xdma_chan);
> 
> +	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
> +
> +	xdma_chan->busy = false;
>  	vd = vchan_next_desc(&xdma_chan->vchan);
> -	if (vd)
> -		desc = to_xdma_desc(vd);
> -	if (desc) {
> -		dma_cookie_complete(&desc->vdesc.tx);
> -		vchan_terminate_vdesc(&desc->vdesc);
> +	if (vd) {
> +		list_del(&vd->node);
> +		dma_cookie_complete(&vd->tx);
> +		vchan_terminate_vdesc(vd);
>  	}
> -
>  	vchan_get_all_descriptors(&xdma_chan->vchan, &head);
> +	list_splice_tail(&head, &xdma_chan->vchan.desc_terminated);
> +
>  	spin_unlock_irqrestore(&xdma_chan->vchan.lock, flags);
> -	vchan_dma_desc_free_list(&xdma_chan->vchan, &head);
> 
>  	return 0;
>  }
> --
> 2.34.1

-- 
~Vinod

Hi Vinod,

Thanks for reviewing my patchset.

On 11.12.2023 11:54, Vinod Koul wrote:
> On 08-12-23, 14:49, Jan Kuliga wrote:
>> Simplify xdma_xfer_stop(). Stop the dma engine and clear its status
>> register unconditionally - just do what its name states. This change
>> also allows to call it without grabbing a lock, which minimizes
>> the total time spent with a spinlock held.
>>
>> Delete the currently processed vd.node from the vc.desc_issued list
>> prior to passing it to vchan_terminate_vdesc(). In case there's more
>> than one descriptor pending on vc.desc_issued list, calling
>> vchan_terminate_desc() results in losing the link between
>> vc.desc_issued list head and the second descriptor on the list. Doing so
>> results in resources leakege, as vchan_dma_desc_free_list() won't be
>> able to properly free memory resources attached to descriptors,
>> resulting in dma_pool_destroy() failure.
>>
>> Don't call vchan_dma_desc_free_list() from within xdma_terminate_all().
>> Move all terminated descriptors to the vc.desc_terminated list instead.
>> This allows to postpone freeing memory resources associated with
>> descriptors until the call to vchan_synchronize(), which is called from
>> xdma_synchronize() callback. This is the right way to do it -
>> xdma_terminate_all() should return as soon as possible, while freeing
>> resources (that may be time consuming in case of large number of
>> descriptors) can be done safely later.
>>
>> Fixes: 290bb5d2d1e2
>> ("dmaengine: xilinx: xdma: Add terminate_all/synchronize callbacks")
>>
>> Signed-off-by: Jan Kuliga <jankul@alatek.krakow.pl>
>> ---
>>  drivers/dma/xilinx/xdma.c | 32 ++++++++++++++++----------------
>>  1 file changed, 16 insertions(+), 16 deletions(-)
>>
>> diff --git a/drivers/dma/xilinx/xdma.c b/drivers/dma/xilinx/xdma.c
>> index 1bce48e5d86c..521ba2a653b6 100644
>> --- a/drivers/dma/xilinx/xdma.c
>> +++ b/drivers/dma/xilinx/xdma.c
>> @@ -379,20 +379,20 @@ static int xdma_xfer_start(struct xdma_chan *xchan)
>>   */
>>  static int xdma_xfer_stop(struct xdma_chan *xchan)
>>  {
>> -	struct virt_dma_desc *vd = vchan_next_desc(&xchan->vchan);
>> -	struct xdma_device *xdev = xchan->xdev_hdl;
>>  	int ret;
>> -
>> -	if (!vd || !xchan->busy)
>> -		return -EINVAL;
>> +	u32 val;
>> +	struct xdma_device *xdev = xchan->xdev_hdl;
>>
>>  	/* clear run stop bit to prevent any further auto-triggering */
>>  	ret = regmap_write(xdev->rmap, xchan->base + XDMA_CHAN_CONTROL_W1C,
>> -			   CHAN_CTRL_RUN_STOP);
>> +							CHAN_CTRL_RUN_STOP);
> 
> Why this change, checkpatch would tell you this is not expected
> alignment (run with strict)
Actually, it does not. I've run it like this:
$LINUX_DIR/scripts/checkpatch.pl --strict -g <commit-id>

and it produced no output related to this line. Anyway, I've already prepared v5 patchset, that conforms to your hint:
Message-Id: 20231218113904.9071-1-jankul@alatek.krakow.pl

> 
>>  	i.f (ret)
>>  		return ret;
>>
>> -	xchan->busy = false;
>> +	/* Clear the channel status register */
>> +	ret = regmap_read(xdev->rmap, xchan->base + XDMA_CHAN_STATUS_RC, &val);
>> +	if (ret)
>> +		return ret;
>>
>>  	return 0;
>>  }
>> @@ -505,25 +505,25 @@ static void xdma_issue_pending(struct dma_chan *chan)
>>  static int xdma_terminate_all(struct dma_chan *chan)
>>  {
>>  	struct xdma_chan *xdma_chan = to_xdma_chan(chan);
>> -	struct xdma_desc *desc = NULL;
>>  	struct virt_dma_desc *vd;
>>  	unsigned long flags;
>>  	LIST_HEAD(head);
>>
>> -	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
>>  	xdma_xfer_stop(xdma_chan);
>>
>> +	spin_lock_irqsave(&xdma_chan->vchan.lock, flags);
>> +
>> +	xdma_chan->busy = false;
>>  	vd = vchan_next_desc(&xdma_chan->vchan);
>> -	if (vd)
>> -		desc = to_xdma_desc(vd);
>> -	if (desc) {
>> -		dma_cookie_complete(&desc->vdesc.tx);
>> -		vchan_terminate_vdesc(&desc->vdesc);
>> +	if (vd) {
>> +		list_del(&vd->node);
>> +		dma_cookie_complete(&vd->tx);
>> +		vchan_terminate_vdesc(vd);
>>  	}
>> -
>>  	vchan_get_all_descriptors(&xdma_chan->vchan, &head);
>> +	list_splice_tail(&head, &xdma_chan->vchan.desc_terminated);
>> +
>>  	spin_unlock_irqrestore(&xdma_chan->vchan.lock, flags);
>> -	vchan_dma_desc_free_list(&xdma_chan->vchan, &head);
>>
>>  	return 0;
>>  }
>> --
>> 2.34.1
> 

Thanks,
Jan