[PATCH] videobuf2: call buf_finish() on unprocessed buffers

[PATCH] videobuf2: call buf_finish() on unprocessed buffers

[Jonathan Corbet]

When user space stops streaming, there may be buffers which have been given
to buf_prepare() and which may or may not have been passed to buf_queue().
The videobuf2 core simply takes those buffers back; if buf_prepare() does
work that needs cleaning up (like setting up a DMA mapping), that cleanup
will not happen.

This patch establishes a simple contract with drivers: buffers given to
buf_prepare() will eventually see a buf_finish() call.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>
---
 drivers/media/video/videobuf2-core.c |    8 +++++++-
 1 files changed, 7 insertions(+), 1 deletions(-)

diff --git a/drivers/media/video/videobuf2-core.c b/drivers/media/video/videobuf2-core.c
index 6ba1461..2ba08ab 100644
--- a/drivers/media/video/videobuf2-core.c
+++ b/drivers/media/video/videobuf2-core.c
@@ -1177,6 +1177,7 @@ EXPORT_SYMBOL_GPL(vb2_streamon);
  */
 static void __vb2_queue_cancel(struct vb2_queue *q)
 {
+	struct vb2_buffer *vb;
 	unsigned int i;
 
 	/*
@@ -1188,13 +1189,18 @@ static void __vb2_queue_cancel(struct vb2_queue *q)
 	q->streaming = 0;
 
 	/*
-	 * Remove all buffers from videobuf's list...
+	 * Remove all buffers from videobuf's list...  Give the driver
+	 * a chance to clean them up first, though.
 	 */
+	list_for_each_entry(vb, &q->queued_list, queued_entry)
+		call_qop(q, buf_finish, vb);
 	INIT_LIST_HEAD(&q->queued_list);
 	/*
 	 * ...and done list; userspace will not receive any buffers it
 	 * has not already dequeued before initiating cancel.
 	 */
+	list_for_each_entry(vb, &q->done_list, done_entry)
+		call_qop(q, buf_finish, vb);
 	INIT_LIST_HEAD(&q->done_list);
 	wake_up_all(&q->done_wq);
 
-- 
1.7.6

RE: [PATCH] videobuf2: call buf_finish() on unprocessed buffers

[Marek Szyprowski]

Hello,

On Thursday, July 14, 2011 11:10 PM Jonathan Corbet wrote:

> When user space stops streaming, there may be buffers which have been given
> to buf_prepare() and which may or may not have been passed to buf_queue().
> The videobuf2 core simply takes those buffers back; if buf_prepare() does
> work that needs cleaning up (like setting up a DMA mapping), that cleanup
> will not happen.
> 
> This patch establishes a simple contract with drivers: buffers given to
> buf_prepare() will eventually see a buf_finish() call.
> 
> Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>

> ---
>  drivers/media/video/videobuf2-core.c |    8 +++++++-
>  1 files changed, 7 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/media/video/videobuf2-core.c
> b/drivers/media/video/videobuf2-core.c
> index 6ba1461..2ba08ab 100644
> --- a/drivers/media/video/videobuf2-core.c
> +++ b/drivers/media/video/videobuf2-core.c
> @@ -1177,6 +1177,7 @@ EXPORT_SYMBOL_GPL(vb2_streamon);
>   */
>  static void __vb2_queue_cancel(struct vb2_queue *q)
>  {
> +	struct vb2_buffer *vb;
>  	unsigned int i;
> 
>  	/*
> @@ -1188,13 +1189,18 @@ static void __vb2_queue_cancel(struct vb2_queue *q)
>  	q->streaming = 0;
> 
>  	/*
> -	 * Remove all buffers from videobuf's list...
> +	 * Remove all buffers from videobuf's list...  Give the driver
> +	 * a chance to clean them up first, though.
>  	 */
> +	list_for_each_entry(vb, &q->queued_list, queued_entry)
> +		call_qop(q, buf_finish, vb);
>  	INIT_LIST_HEAD(&q->queued_list);
>  	/*
>  	 * ...and done list; userspace will not receive any buffers it
>  	 * has not already dequeued before initiating cancel.
>  	 */
> +	list_for_each_entry(vb, &q->done_list, done_entry)
> +		call_qop(q, buf_finish, vb);
>  	INIT_LIST_HEAD(&q->done_list);
>  	wake_up_all(&q->done_wq);
> 
> --
> 1.7.6

Best regards
-- 
Marek Szyprowski
Samsung Poland R&D Center

Re: [PATCH] videobuf2: call buf_finish() on unprocessed buffers

[Pawel Osciak]

Hi Jon,
There is a misunderstanding here. This patch will call buf_finish for
some buffers twice. A buffer does not get removed from queued_list
even if it is on the done_list. A buffer gets into the queued_list on
qbuf, gets into the done_list after an operation is finished and is
removed from both lists on dqbuf. This means that you shouldn't call
buf_finish for buffers on both lists, only for all buffers in the
queued_list, as it is a superset of both.
-- 
Best regards,
Pawel Osciak

On Thu, Jul 14, 2011 at 14:09, Jonathan Corbet <corbet@lwn.net> wrote:
> When user space stops streaming, there may be buffers which have been given
> to buf_prepare() and which may or may not have been passed to buf_queue().
> The videobuf2 core simply takes those buffers back; if buf_prepare() does
> work that needs cleaning up (like setting up a DMA mapping), that cleanup
> will not happen.
>
> This patch establishes a simple contract with drivers: buffers given to
> buf_prepare() will eventually see a buf_finish() call.
>
> Signed-off-by: Jonathan Corbet <corbet@lwn.net>
> ---
>  drivers/media/video/videobuf2-core.c |    8 +++++++-
>  1 files changed, 7 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/media/video/videobuf2-core.c b/drivers/media/video/videobuf2-core.c
> index 6ba1461..2ba08ab 100644
> --- a/drivers/media/video/videobuf2-core.c
> +++ b/drivers/media/video/videobuf2-core.c
> @@ -1177,6 +1177,7 @@ EXPORT_SYMBOL_GPL(vb2_streamon);
>  */
>  static void __vb2_queue_cancel(struct vb2_queue *q)
>  {
> +       struct vb2_buffer *vb;
>        unsigned int i;
>
>        /*
> @@ -1188,13 +1189,18 @@ static void __vb2_queue_cancel(struct vb2_queue *q)
>        q->streaming = 0;
>
>        /*
> -        * Remove all buffers from videobuf's list...
> +        * Remove all buffers from videobuf's list...  Give the driver
> +        * a chance to clean them up first, though.
>         */
> +       list_for_each_entry(vb, &q->queued_list, queued_entry)
> +               call_qop(q, buf_finish, vb);
>        INIT_LIST_HEAD(&q->queued_list);
>        /*
>         * ...and done list; userspace will not receive any buffers it
>         * has not already dequeued before initiating cancel.
>         */
> +       list_for_each_entry(vb, &q->done_list, done_entry)
> +               call_qop(q, buf_finish, vb);
>        INIT_LIST_HEAD(&q->done_list);
>        wake_up_all(&q->done_wq);
>
> --
> 1.7.6
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-media" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

Re: [PATCH] videobuf2: call buf_finish() on unprocessed buffers

[Mauro Carvalho Chehab]

Em 24-07-2011 19:10, Pawel Osciak escreveu:
> Hi Jon,
> There is a misunderstanding here. This patch will call buf_finish for
> some buffers twice. A buffer does not get removed from queued_list
> even if it is on the done_list. A buffer gets into the queued_list on
> qbuf, gets into the done_list after an operation is finished and is
> removed from both lists on dqbuf. This means that you shouldn't call
> buf_finish for buffers on both lists, only for all buffers in the
> queued_list, as it is a superset of both.

Is there any update on this?

Thanks,
Mauro

> -- Best regards, Pawel Osciak On Thu, Jul 14, 2011 at 14:09, Jonathan Corbet <corbet@lwn.net> wrote:
>> > When user space stops streaming, there may be buffers which have been given
>> > to buf_prepare() and which may or may not have been passed to buf_queue().
>> > The videobuf2 core simply takes those buffers back; if buf_prepare() does
>> > work that needs cleaning up (like setting up a DMA mapping), that cleanup
>> > will not happen.
>> >
>> > This patch establishes a simple contract with drivers: buffers given to
>> > buf_prepare() will eventually see a buf_finish() call.
>> >
>> > Signed-off-by: Jonathan Corbet <corbet@lwn.net>
>> > ---
>> >  drivers/media/video/videobuf2-core.c |    8 +++++++-
>> >  1 files changed, 7 insertions(+), 1 deletions(-)
>> >
>> > diff --git a/drivers/media/video/videobuf2-core.c b/drivers/media/video/videobuf2-core.c
>> > index 6ba1461..2ba08ab 100644
>> > --- a/drivers/media/video/videobuf2-core.c
>> > +++ b/drivers/media/video/videobuf2-core.c
>> > @@ -1177,6 +1177,7 @@ EXPORT_SYMBOL_GPL(vb2_streamon);
>> >  */
>> >  static void __vb2_queue_cancel(struct vb2_queue *q)
>> >  {
>> > +       struct vb2_buffer *vb;
>> >        unsigned int i;
>> >
>> >        /*
>> > @@ -1188,13 +1189,18 @@ static void __vb2_queue_cancel(struct vb2_queue *q)
>> >        q->streaming = 0;
>> >
>> >        /*
>> > -        * Remove all buffers from videobuf's list...
>> > +        * Remove all buffers from videobuf's list...  Give the driver
>> > +        * a chance to clean them up first, though.
>> >         */
>> > +       list_for_each_entry(vb, &q->queued_list, queued_entry)
>> > +               call_qop(q, buf_finish, vb);
>> >        INIT_LIST_HEAD(&q->queued_list);
>> >        /*
>> >         * ...and done list; userspace will not receive any buffers it
>> >         * has not already dequeued before initiating cancel.
>> >         */
>> > +       list_for_each_entry(vb, &q->done_list, done_entry)
>> > +               call_qop(q, buf_finish, vb);
>> >        INIT_LIST_HEAD(&q->done_list);
>> >        wake_up_all(&q->done_wq);
>> >
>> > --
>> > 1.7.6
>> >
>> > --
>> > To unsubscribe from this list: send the line "unsubscribe linux-media" in
>> > the body of a message to majordomo@vger.kernel.org
>> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> >
> --
> To unsubscribe from this list: send the line "unsubscribe linux-media" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>