On 09/12/2021 07:29, Chen-Yu Tsai wrote:
> DST_QUEUE_OFF_BASE is applied to offset/mem_offset on MMAP capture buffers
> only for the VIDIOC_QUERYBUF ioctl, while the userspace fields (including
> offset/mem_offset) are filled in for VIDIOC_{QUERY,PREPARE,Q,DQ}BUF
> ioctls. This leads to differences in the values presented to userspace.
> If userspace attempts to mmap the capture buffer directly using values
> from DQBUF, it will fail.
>
> Move the code that applies the magic offset into a helper, and call
> that helper from all four ioctl entry points.
>
> Fixes: 7f98639def42 ("V4L/DVB: add memory-to-memory device helper framework for videobuf")
> Fixes: 908a0d7c588e ("[media] v4l: mem2mem: port to videobuf2")
> Signed-off-by: Chen-Yu Tsai <wenst@xxxxxxxxxxxx>
> ---
> This was tested on RK3399 with
>
> gst-launch-1.0 videotestsrc num-buffers=2 ! v4l2jpegenc ! fakesink
>
> and verifying the values using the V4L2 debug messages:
>
> video2: VIDIOC_QUERYBUF: 00:00:00.000000 index=0, type=vid-cap-mplane, request_fd=0, flags=0x00004000, field=any, sequence=0, memory=mmap
> plane 0: bytesused=0, data_offset=0x00000000, offset/userptr=0x40000000, length=2097152
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
> video2: VIDIOC_QUERYBUF: 00:00:00.000000 index=0, type=vid-out-mplane, request_fd=0, flags=0x00004000, field=any, sequence=0, memory=mmap
> plane 0: bytesused=0, data_offset=0x00000000, offset/userptr=0x0, length=153600
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
>
> video2: VIDIOC_QBUF: 00:00:00.000000 index=0, type=vid-cap-mplane, request_fd=0, flags=0x00004003, field=any, sequence=0, memory=mmap
> plane 0: bytesused=2097152, data_offset=0x00000000, offset/userptr=0x40000000, length=2097152
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
> video2: VIDIOC_QBUF: 00:00:00.000000 index=0, type=vid-out-mplane, request_fd=0, flags=0x00004003, field=none, sequence=0, memory=mmap
> plane 0: bytesused=153600, data_offset=0x00000000, offset/userptr=0x0, length=153600
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
>
> video2: VIDIOC_DQBUF: 00:00:00.000000 index=0, type=vid-cap-mplane, request_fd=0, flags=0x00004001, field=none, sequence=0, memory=mmap
> plane 0: bytesused=6324, data_offset=0x00000000, offset/userptr=0x40000000, length=2097152
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
> video2: VIDIOC_DQBUF: 00:00:00.000000 index=0, type=vid-out-mplane, request_fd=0, flags=0x00004001, field=none, sequence=0, memory=mmap
> plane 0: bytesused=153600, data_offset=0x00000000, offset/userptr=0x0, length=153600
> timecode=00:00:00 type=0, flags=0x00000000, frames=0, userbits=0x00000000
>
> Gstreamer doesn't do PREPAREBUF calls, so that path was not verified.
> However the code changes are exactly the same, so I'm quite confident
> about them.
>
> ---
> drivers/media/v4l2-core/v4l2-mem2mem.c | 46 ++++++++++++++++++++------
> 1 file changed, 35 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/media/v4l2-core/v4l2-mem2mem.c b/drivers/media/v4l2-core/v4l2-mem2mem.c
> index e2654b422334..b47f25297c43 100644
> --- a/drivers/media/v4l2-core/v4l2-mem2mem.c
> +++ b/drivers/media/v4l2-core/v4l2-mem2mem.c
> @@ -585,19 +585,14 @@ int v4l2_m2m_reqbufs(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> }
> EXPORT_SYMBOL_GPL(v4l2_m2m_reqbufs);
>
> -int v4l2_m2m_querybuf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> - struct v4l2_buffer *buf)
> +static void v4l2_m2m_adjust_mem_offset(struct vb2_queue *vq,
> + struct v4l2_buffer *buf)
> {
> - struct vb2_queue *vq;
> - int ret = 0;
> - unsigned int i;
> -
> - vq = v4l2_m2m_get_vq(m2m_ctx, buf->type);
> - ret = vb2_querybuf(vq, buf);
> -
> /* Adjust MMAP memory offsets for the CAPTURE queue */
> if (buf->memory == V4L2_MEMORY_MMAP && V4L2_TYPE_IS_CAPTURE(vq->type)) {
> if (V4L2_TYPE_IS_MULTIPLANAR(vq->type)) {
> + unsigned int i;
> +
> for (i = 0; i < buf->length; ++i)
> buf->m.planes[i].m.mem_offset
> += DST_QUEUE_OFF_BASE;
> @@ -605,6 +600,19 @@ int v4l2_m2m_querybuf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> buf->m.offset += DST_QUEUE_OFF_BASE;
> }
> }
> +}
> +
> +int v4l2_m2m_querybuf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> + struct v4l2_buffer *buf)
> +{
> + struct vb2_queue *vq;
> + int ret = 0;
> +
> + vq = v4l2_m2m_get_vq(m2m_ctx, buf->type);
> + ret = vb2_querybuf(vq, buf);
> +
> + /* Adjust MMAP memory offsets for the CAPTURE queue */
> + v4l2_m2m_adjust_mem_offset(vq, buf);
>
> return ret;
> }
> @@ -760,6 +768,10 @@ int v4l2_m2m_qbuf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> }
>
> ret = vb2_qbuf(vq, vdev->v4l2_dev->mdev, buf);
> +
> + /* Adjust MMAP memory offsets for the CAPTURE queue */
> + v4l2_m2m_adjust_mem_offset(vq, buf);
> +
> if (ret)
> return ret;
>
> @@ -784,9 +796,15 @@ int v4l2_m2m_dqbuf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> struct v4l2_buffer *buf)
> {
> struct vb2_queue *vq;
> + int ret;
>
> vq = v4l2_m2m_get_vq(m2m_ctx, buf->type);
> - return vb2_dqbuf(vq, buf, file->f_flags & O_NONBLOCK);
> + ret = vb2_dqbuf(vq, buf, file->f_flags & O_NONBLOCK);
> +
> + /* Adjust MMAP memory offsets for the CAPTURE queue */
> + v4l2_m2m_adjust_mem_offset(vq, buf);
> +
> + return ret;
> }
> EXPORT_SYMBOL_GPL(v4l2_m2m_dqbuf);
>
> @@ -795,9 +813,15 @@ int v4l2_m2m_prepare_buf(struct file *file, struct v4l2_m2m_ctx *m2m_ctx,
> {
> struct video_device *vdev = video_devdata(file);
> struct vb2_queue *vq;
> + int ret;
>
> vq = v4l2_m2m_get_vq(m2m_ctx, buf->type);
> - return vb2_prepare_buf(vq, vdev->v4l2_dev->mdev, buf);
> + ret = vb2_prepare_buf(vq, vdev->v4l2_dev->mdev, buf);
> +
> + /* Adjust MMAP memory offsets for the CAPTURE queue */
> + v4l2_m2m_adjust_mem_offset(vq, buf);
> +
> + return ret;
> }
> EXPORT_SYMBOL_GPL(v4l2_m2m_prepare_buf);
>
>
For all these functions you should only call v4l2_m2m_adjust_mem_offset() if !ret.
If the vb2_* function returned an error, then the offset fields aren't filled in,
so it makes no sense to update them. And besides, the struct isn't copied back to
userspace anyway on error.
Regards,
Hans
