2021-09-28 03:48:26

by Sergey Senozhatsky

Subject: [PATCH] media: videobuf2: always set buffer vb2 pointer

We need to always link allocated vb2_dc_buf back to vb2_buffer because
we dereference vb2 in prepare() and finish() callbacks.

Signed-off-by: Sergey Senozhatsky <[email protected]>
---
drivers/media/common/videobuf2/videobuf2-dma-contig.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/drivers/media/common/videobuf2/videobuf2-dma-contig.c b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
index b052a4e36961..38767791955d 100644
--- a/drivers/media/common/videobuf2/videobuf2-dma-contig.c
+++ b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
@@ -600,6 +600,7 @@ static void *vb2_dc_get_userptr(struct vb2_buffer *vb, struct device *dev,

buf->dev = dev;
buf->dma_dir = vb->vb2_queue->dma_dir;
+ buf->vb = vb;

offset = lower_32_bits(offset_in_page(vaddr));
vec = vb2_create_framevec(vaddr, size);
@@ -788,6 +789,8 @@ static void *vb2_dc_attach_dmabuf(struct vb2_buffer *vb, struct device *dev,
return ERR_PTR(-ENOMEM);

buf->dev = dev;
+ buf->vb = vb;
+
/* create attachment for the dmabuf with the user device */
dba = dma_buf_attach(dbuf, buf->dev);
if (IS_ERR(dba)) {
--
2.33.0.685.g46640cef36-goog

2021-09-28 06:26:09

by Chen-Yu Tsai

Subject: Re: [PATCH] media: videobuf2: always set buffer vb2 pointer

On Tue, Sep 28, 2021 at 2:16 PM Sergey Senozhatsky
<[email protected]> wrote:
>
> We need to always link allocated vb2_dc_buf back to vb2_buffer because
> we dereference vb2 in prepare() and finish() callbacks.
>
> Signed-off-by: Sergey Senozhatsky <[email protected]>

This fixes the breakage from the "videobuf2: support new noncontiguous DMA
API" series on the RK3399 Scarlet if the ChromeOS patch that changes
min_buffers_needed to 0 [1] is not applied.

Since there are other in-tree drivers that have min_buffers_needed=0,
I would recommend getting some more testing.

ChenYu

[1] https://crrev.com/c/3168489

2021-09-28 07:54:53

by Chen-Yu Tsai

Subject: Re: [PATCH] media: videobuf2: always set buffer vb2 pointer

On Tue, Sep 28, 2021 at 2:24 PM Chen-Yu Tsai <[email protected]> wrote:
>
> On Tue, Sep 28, 2021 at 2:16 PM Sergey Senozhatsky
> <[email protected]> wrote:
> >
> > We need to always link allocated vb2_dc_buf back to vb2_buffer because
> > we dereference vb2 in prepare() and finish() callbacks.
> >
> > Signed-off-by: Sergey Senozhatsky <[email protected]>
>
> This fixes the breakage from the "videobuf2: support new noncontiguous DMA
> API" series on the RK3399 Scarlet if the ChromeOS patch that changes
> min_buffers_needed to 0 [1] is not applied.
>
> Since there are other in-tree drivers that have min_buffers_needed=0,
> I would recommend getting some more testing.

Seems I had a stale kernel when testing the min_buffers_needed=0 case.

Everythings works now.

Tested-by: Chen-Yu Tsai <[email protected]>

2021-10-05 09:59:24

by Tomasz Figa

Subject: Re: [PATCH] media: videobuf2: always set buffer vb2 pointer

On Tue, Sep 28, 2021 at 12:46 PM Sergey Senozhatsky
<[email protected]> wrote:
>
> We need to always link allocated vb2_dc_buf back to vb2_buffer because
> we dereference vb2 in prepare() and finish() callbacks.
>
> Signed-off-by: Sergey Senozhatsky <[email protected]>
> ---
> drivers/media/common/videobuf2/videobuf2-dma-contig.c | 3 +++
> 1 file changed, 3 insertions(+)
>

Acked-by: Tomasz Figa <[email protected]>

Best regards,
Tomasz

> diff --git a/drivers/media/common/videobuf2/videobuf2-dma-contig.c b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> index b052a4e36961..38767791955d 100644
> --- a/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> +++ b/drivers/media/common/videobuf2/videobuf2-dma-contig.c
> @@ -600,6 +600,7 @@ static void *vb2_dc_get_userptr(struct vb2_buffer *vb, struct device *dev,
>
> buf->dev = dev;
> buf->dma_dir = vb->vb2_queue->dma_dir;
> + buf->vb = vb;
>
> offset = lower_32_bits(offset_in_page(vaddr));
> vec = vb2_create_framevec(vaddr, size);
> @@ -788,6 +789,8 @@ static void *vb2_dc_attach_dmabuf(struct vb2_buffer *vb, struct device *dev,
> return ERR_PTR(-ENOMEM);
>
> buf->dev = dev;
> + buf->vb = vb;
> +
> /* create attachment for the dmabuf with the user device */
> dba = dma_buf_attach(dbuf, buf->dev);
> if (IS_ERR(dba)) {
> --
> 2.33.0.685.g46640cef36-goog
>

2021-10-12 02:04:37

by Sergey Senozhatsky

Subject: Re: [PATCH] media: videobuf2: always set buffer vb2 pointer

On (21/10/05 18:57), Tomasz Figa wrote:
>
> On Tue, Sep 28, 2021 at 12:46 PM Sergey Senozhatsky
> <[email protected]> wrote:
> >
> > We need to always link allocated vb2_dc_buf back to vb2_buffer because
> > we dereference vb2 in prepare() and finish() callbacks.
> >
> > Signed-off-by: Sergey Senozhatsky <[email protected]>
> > ---
> > drivers/media/common/videobuf2/videobuf2-dma-contig.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
>
> Acked-by: Tomasz Figa <[email protected]>

Hans, can you please pick up this patch?

2021-10-12 07:54:52

by Hans Verkuil

Subject: Re: [PATCH] media: videobuf2: always set buffer vb2 pointer

On 12/10/2021 03:57, Sergey Senozhatsky wrote:
> On (21/10/05 18:57), Tomasz Figa wrote:
>>
>> On Tue, Sep 28, 2021 at 12:46 PM Sergey Senozhatsky
>> <[email protected]> wrote:
>>>
>>> We need to always link allocated vb2_dc_buf back to vb2_buffer because
>>> we dereference vb2 in prepare() and finish() callbacks.
>>>
>>> Signed-off-by: Sergey Senozhatsky <[email protected]>
>>> ---
>>> drivers/media/common/videobuf2/videobuf2-dma-contig.c | 3 +++
>>> 1 file changed, 3 insertions(+)
>>>
>>
>> Acked-by: Tomasz Figa <[email protected]>
>
> Hans, can you please pick up this patch?
>

Yes, it's already part of my upcoming pull request.

Hans