[PATCHv3] media: vb2: always set buffer cache sync hints
diff mbox series

Message ID 20201129031545.557586-1-sergey.senozhatsky@gmail.com
State New, archived
Headers show
Series
  • [PATCHv3] media: vb2: always set buffer cache sync hints
Related show

Commit Message

Sergey Senozhatsky Nov. 29, 2020, 3:15 a.m. UTC
We need to always set ->need_cache_sync_on_prepare and
->need_cache_sync_on_finish when we initialize vb2 buffer.

Currently these flags are set/adjusted only in V4L2's
vb2_queue_or_prepare_buf(), which means that for the code
paths that don't use V4L2 vb2 will always tell videobuf2
core to skip ->prepare() and ->finish() cache syncs/flushes.

This is a quick solution that should do the trick. The
proper fix, however, is much more complicated and requires
a rather big videobuf2 refactoring - we need to move cache
sync/flush decision making out of core videobuf2 to the
allocators.

Fixes: f5f5fa73fbfb ("media: videobuf2: handle V4L2 buffer cache flags")
Reported-by: Tomasz Figa <tfiga@chromium.org>
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
---
 drivers/media/common/videobuf2/videobuf2-core.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

v3: Improved code comment and dropped queue allow_cache_hints check (Tomasz)
v2: Added a comment and set cache sync flags only for specific buffers (Hans)

Comments

Hans Verkuil Nov. 30, 2020, 10:02 a.m. UTC | #1
On 29/11/2020 04:15, Sergey Senozhatsky wrote:
> We need to always set ->need_cache_sync_on_prepare and
> ->need_cache_sync_on_finish when we initialize vb2 buffer.
> 
> Currently these flags are set/adjusted only in V4L2's
> vb2_queue_or_prepare_buf(), which means that for the code
> paths that don't use V4L2 vb2 will always tell videobuf2
> core to skip ->prepare() and ->finish() cache syncs/flushes.
> 
> This is a quick solution that should do the trick. The
> proper fix, however, is much more complicated and requires
> a rather big videobuf2 refactoring - we need to move cache
> sync/flush decision making out of core videobuf2 to the
> allocators.
> 
> Fixes: f5f5fa73fbfb ("media: videobuf2: handle V4L2 buffer cache flags")
> Reported-by: Tomasz Figa <tfiga@chromium.org>
> Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
> ---
>  drivers/media/common/videobuf2/videobuf2-core.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> v3: Improved code comment and dropped queue allow_cache_hints check (Tomasz)
> v2: Added a comment and set cache sync flags only for specific buffers (Hans)
> 
> diff --git a/drivers/media/common/videobuf2/videobuf2-core.c b/drivers/media/common/videobuf2/videobuf2-core.c
> index 5499013cf82e..3f11fc5b5d9a 100644
> --- a/drivers/media/common/videobuf2/videobuf2-core.c
> +++ b/drivers/media/common/videobuf2/videobuf2-core.c
> @@ -414,6 +414,20 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
>  		vb->index = q->num_buffers + buffer;
>  		vb->type = q->type;
>  		vb->memory = memory;
> +		/*
> +		 * A workaround fix. We need to set these flags here so that
> +		 * videobuf2 core will call ->prepare()/->finish() cache
> +		 * sync/flush on vb2 buffers when appropriate. Otherwise, for
> +		 * backends that don't rely on V4L2 (perhaps dvb) these flags
> +		 * will always be false and, hence, videobuf2 core will skip
> +		 * cache sync/flush operations. However, we can avoid explicit
> +		 * ->prepare() and ->finish() cache sync for DMABUF buffers,
> +		 * because DMA exporter takes care of it.
> +		 */
> +		if (q->memory != VB2_MEMORY_DMABUF) {
> +			vb->need_cache_sync_on_prepare = 1;
> +			vb->need_cache_sync_on_finish = 1;
> +		}

Is this a work-around fix? Isn't this just a bug fix? It seems reasonable
to always set this when allocating buffers for a queue. And for v4l2 these
values can be changed if supported by the driver (allow_cache_hints is set).

So the comment would be:

	/*
	 * We need to set these flags here so that the videobuf2 core
	 * will call ->prepare()/->finish() cache sync/flush on vb2
	 * buffers when appropriate. However, we can avoid explicit
	 * ->prepare() and ->finish() cache sync for DMABUF buffers,
	 * because DMA exporter takes care of it.
	 */

The commit message would need to be modified as well.

Regards,

	Hans

>  		for (plane = 0; plane < num_planes; ++plane) {
>  			vb->planes[plane].length = plane_sizes[plane];
>  			vb->planes[plane].min_length = plane_sizes[plane];
>
Sergey Senozhatsky Nov. 30, 2020, 10:44 a.m. UTC | #2
On (20/11/30 11:02), Hans Verkuil wrote:
> > @@ -414,6 +414,20 @@ static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
> >  		vb->index = q->num_buffers + buffer;
> >  		vb->type = q->type;
> >  		vb->memory = memory;
> > +		/*
> > +		 * A workaround fix. We need to set these flags here so that
> > +		 * videobuf2 core will call ->prepare()/->finish() cache
> > +		 * sync/flush on vb2 buffers when appropriate. Otherwise, for
> > +		 * backends that don't rely on V4L2 (perhaps dvb) these flags
> > +		 * will always be false and, hence, videobuf2 core will skip
> > +		 * cache sync/flush operations. However, we can avoid explicit
> > +		 * ->prepare() and ->finish() cache sync for DMABUF buffers,
> > +		 * because DMA exporter takes care of it.
> > +		 */
> > +		if (q->memory != VB2_MEMORY_DMABUF) {
> > +			vb->need_cache_sync_on_prepare = 1;
> > +			vb->need_cache_sync_on_finish = 1;
> > +		}
> 
> Is this a work-around fix? Isn't this just a bug fix? It seems reasonable
> to always set this when allocating buffers for a queue. And for v4l2 these
> values can be changed if supported by the driver (allow_cache_hints is set).

Yes, it's a bug fix. It's a workaround bug fix in a sense that a proper
bug fix is to rename these flags and invert them (and patch all the
functions that set and test them). A sneak preview of what's in my tree:

---

diff --git a/include/media/videobuf2-core.h b/include/media/videobuf2-core.h
index bd78f6c7f342..3b79042d9d5c 100644
--- a/include/media/videobuf2-core.h
+++ b/include/media/videobuf2-core.h
@@ -264,10 +264,10 @@ struct vb2_buffer {
         *                      after the 'buf_finish' op is called.
         * copied_timestamp:    the timestamp of this capture buffer was copied
         *                      from an output buffer.
-        * need_cache_sync_on_prepare: when set buffer's ->prepare() function
-        *                      performs cache sync/invalidation.
-        * need_cache_sync_on_finish: when set buffer's ->finish() function
-        *                      performs cache sync/invalidation.
+        * no_cache_sync_on_prepare: when set buffer's ->prepare() function
+        *                      skips cache sync/invalidation.
+        * no_cache_sync_on_finish: when set buffer's ->finish() function
+        *                      skips cache sync/invalidation.
         * queued_entry:        entry on the queued buffers list, which holds
         *                      all buffers queued from userspace
         * done_entry:          entry on the list that stores all buffers ready
@@ -278,8 +278,8 @@ struct vb2_buffer {
        unsigned int            synced:1;
        unsigned int            prepared:1;
        unsigned int            copied_timestamp:1;
-       unsigned int            need_cache_sync_on_prepare:1;
-       unsigned int            need_cache_sync_on_finish:1;
+       unsigned int            no_cache_sync_on_prepare:1;
+       unsigned int            no_cache_sync_on_finish:1;
 
        struct vb2_plane        planes[VB2_MAX_PLANES];
        struct list_head        queued_entry;

---

and it's much bigger than a 3-liner ("workaround") fix.

> So the comment would be:
> 
> 	/*
> 	 * We need to set these flags here so that the videobuf2 core
> 	 * will call ->prepare()/->finish() cache sync/flush on vb2
> 	 * buffers when appropriate. However, we can avoid explicit
> 	 * ->prepare() and ->finish() cache sync for DMABUF buffers,
> 	 * because DMA exporter takes care of it.
> 	 */
> 
> The commit message would need to be modified as well.

OK.

	-ss

Patch
diff mbox series

diff --git a/drivers/media/common/videobuf2/videobuf2-core.c b/drivers/media/common/videobuf2/videobuf2-core.c
index 5499013cf82e..3f11fc5b5d9a 100644
--- a/drivers/media/common/videobuf2/videobuf2-core.c
+++ b/drivers/media/common/videobuf2/videobuf2-core.c
@@ -414,6 +414,20 @@  static int __vb2_queue_alloc(struct vb2_queue *q, enum vb2_memory memory,
 		vb->index = q->num_buffers + buffer;
 		vb->type = q->type;
 		vb->memory = memory;
+		/*
+		 * A workaround fix. We need to set these flags here so that
+		 * videobuf2 core will call ->prepare()/->finish() cache
+		 * sync/flush on vb2 buffers when appropriate. Otherwise, for
+		 * backends that don't rely on V4L2 (perhaps dvb) these flags
+		 * will always be false and, hence, videobuf2 core will skip
+		 * cache sync/flush operations. However, we can avoid explicit
+		 * ->prepare() and ->finish() cache sync for DMABUF buffers,
+		 * because DMA exporter takes care of it.
+		 */
+		if (q->memory != VB2_MEMORY_DMABUF) {
+			vb->need_cache_sync_on_prepare = 1;
+			vb->need_cache_sync_on_finish = 1;
+		}
 		for (plane = 0; plane < num_planes; ++plane) {
 			vb->planes[plane].length = plane_sizes[plane];
 			vb->planes[plane].min_length = plane_sizes[plane];