drivers/media/v4l2-core/v4l2-h264.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
In v4l2_h264_build_b_ref_lists(), the B0/B1 list equality
check passes the entry count builder->num_valid to memcmp()
instead of a byte size. Since struct v4l2_h264_reference is
two bytes (fields and index), only half of each list is
compared, so distinct lists can be wrongly treated as equal
and trigger an incorrect swap(b1_reflist[0], b1_reflist[1]).
Change the memcmp() size argument to builder->num_valid *
sizeof(*b1_reflist) so that the full byte length of both
reference lists is compared.
Fixes: 624922a2739b ("media: v4l2-core: Add helpers to build the H264 P/B0/B1 reflists")
Signed-off-by: Haotian Zhang <vulab@iscas.ac.cn>
---
drivers/media/v4l2-core/v4l2-h264.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
index c00197d095e7..1b1421caf488 100644
--- a/drivers/media/v4l2-core/v4l2-h264.c
+++ b/drivers/media/v4l2-core/v4l2-h264.c
@@ -440,7 +440,8 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
}
if (builder->num_valid > 1 &&
- !memcmp(b1_reflist, b0_reflist, builder->num_valid))
+ !memcmp(b1_reflist, b0_reflist,
+ builder->num_valid * sizeof(*b1_reflist)))
swap(b1_reflist[0], b1_reflist[1]);
print_ref_list_b(builder, b0_reflist, 0);
--
2.43.0
Le vendredi 28 août 2026 à 10:17 +0800, Haotian Zhang a écrit :
> In v4l2_h264_build_b_ref_lists(), the B0/B1 list equality
> check passes the entry count builder->num_valid to memcmp()
> instead of a byte size. Since struct v4l2_h264_reference is
> two bytes (fields and index), only half of each list is
> compared, so distinct lists can be wrongly treated as equal
> and trigger an incorrect swap(b1_reflist[0], b1_reflist[1]).
>
> Change the memcmp() size argument to builder->num_valid *
> sizeof(*b1_reflist) so that the full byte length of both
> reference lists is compared.
>
> Fixes: 624922a2739b ("media: v4l2-core: Add helpers to build the H264 P/B0/B1 reflists")
> Signed-off-by: Haotian Zhang <vulab@iscas.ac.cn>
> ---
> drivers/media/v4l2-core/v4l2-h264.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
> index c00197d095e7..1b1421caf488 100644
> --- a/drivers/media/v4l2-core/v4l2-h264.c
> +++ b/drivers/media/v4l2-core/v4l2-h264.c
> @@ -440,7 +440,8 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
> }
>
> if (builder->num_valid > 1 &&
> - !memcmp(b1_reflist, b0_reflist, builder->num_valid))
> + !memcmp(b1_reflist, b0_reflist,
> + builder->num_valid * sizeof(*b1_reflist)))
One last comment, I think this would be more consistent with the memcpy() calls:
+ sizeof(b1_reflist[0]) * builder->num_valid))
Nicolas
> swap(b1_reflist[0], b1_reflist[1]);
>
> print_ref_list_b(builder, b0_reflist, 0);
Le vendredi 28 août 2026 à 10:17 +0800, Haotian Zhang a écrit :
> In v4l2_h264_build_b_ref_lists(), the B0/B1 list equality
> check passes the entry count builder->num_valid to memcmp()
> instead of a byte size. Since struct v4l2_h264_reference is
> two bytes (fields and index), only half of each list is
> compared, so distinct lists can be wrongly treated as equal
> and trigger an incorrect swap(b1_reflist[0], b1_reflist[1]).
>
> Change the memcmp() size argument to builder->num_valid *
> sizeof(*b1_reflist) so that the full byte length of both
> reference lists is compared.
>
> Fixes: 624922a2739b ("media: v4l2-core: Add helpers to build the H264 P/B0/B1 reflists")
This is the wrong commit. It was fine until I broke it 4 years ago it seems.
Boris don't do that kind of silly mistake :-S
commit 2e2c3d6c0ef88ffac0d6b5079ee88cf8408f5f3b
Author: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Date: Fri May 13 22:29:03 2022 +0200
media: h264: Use v4l2_h264_reference for reflist
> Signed-off-by: Haotian Zhang <vulab@iscas.ac.cn>
> ---
> drivers/media/v4l2-core/v4l2-h264.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
> index c00197d095e7..1b1421caf488 100644
> --- a/drivers/media/v4l2-core/v4l2-h264.c
> +++ b/drivers/media/v4l2-core/v4l2-h264.c
> @@ -440,7 +440,8 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
> }
>
> if (builder->num_valid > 1 &&
> - !memcmp(b1_reflist, b0_reflist, builder->num_valid))
> + !memcmp(b1_reflist, b0_reflist,
> + builder->num_valid * sizeof(*b1_reflist)))
> swap(b1_reflist[0], b1_reflist[1]);
And of course this swap need to be replaced.
Nicolas
>
> print_ref_list_b(builder, b0_reflist, 0);
Le vendredi 28 août 2026 à 14:31 -0400, Nicolas Dufresne a écrit :
> Le vendredi 28 août 2026 à 10:17 +0800, Haotian Zhang a écrit :
> > In v4l2_h264_build_b_ref_lists(), the B0/B1 list equality
> > check passes the entry count builder->num_valid to memcmp()
> > instead of a byte size. Since struct v4l2_h264_reference is
> > two bytes (fields and index), only half of each list is
> > compared, so distinct lists can be wrongly treated as equal
> > and trigger an incorrect swap(b1_reflist[0], b1_reflist[1]).
> >
> > Change the memcmp() size argument to builder->num_valid *
> > sizeof(*b1_reflist) so that the full byte length of both
> > reference lists is compared.
> >
> > Fixes: 624922a2739b ("media: v4l2-core: Add helpers to build the H264 P/B0/B1 reflists")
>
> This is the wrong commit. It was fine until I broke it 4 years ago it seems.
> Boris don't do that kind of silly mistake :-S
>
> commit 2e2c3d6c0ef88ffac0d6b5079ee88cf8408f5f3b
> Author: Nicolas Dufresne <nicolas.dufresne@collabora.com>
> Date: Fri May 13 22:29:03 2022 +0200
>
> media: h264: Use v4l2_h264_reference for reflist
>
>
> > Signed-off-by: Haotian Zhang <vulab@iscas.ac.cn>
> > ---
> > drivers/media/v4l2-core/v4l2-h264.c | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/media/v4l2-core/v4l2-h264.c b/drivers/media/v4l2-core/v4l2-h264.c
> > index c00197d095e7..1b1421caf488 100644
> > --- a/drivers/media/v4l2-core/v4l2-h264.c
> > +++ b/drivers/media/v4l2-core/v4l2-h264.c
> > @@ -440,7 +440,8 @@ v4l2_h264_build_b_ref_lists(const struct v4l2_h264_reflist_builder *builder,
> > }
> >
> > if (builder->num_valid > 1 &&
> > - !memcmp(b1_reflist, b0_reflist, builder->num_valid))
> > + !memcmp(b1_reflist, b0_reflist,
> > + builder->num_valid * sizeof(*b1_reflist)))
> > swap(b1_reflist[0], b1_reflist[1]);
>
> And of course this swap need to be replaced.
Ignore this one, I should have read the implementation of the swap macro ... its
working fine for this case:
do { typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
>
> Nicolas
>
> >
> > print_ref_list_b(builder, b0_reflist, 0);
© 2016 - 2026 Red Hat, Inc.