Re: [PATCH v2 1/8] drm/dumb-buffers: Fix drm_mode_create_dumb() for bpp < 8

[Geert Uytterhoeven]

Hi Thomas,

On Thu, Aug 31, 2023 at 9:40 AM Thomas Zimmermann <tzimmermann@xxxxxxx> wrote:
> Am 24.08.23 um 17:08 schrieb Geert Uytterhoeven:
> > drm_mode_create_dumb() calculates the number of characters per pixel
> > from the number of bits per pixel by rounding up, which is not correct
> > as the actual value of cpp may be non-integer.  While we do not need to
> > care here about complex formats like YUV, bpp < 8 is a valid use case.
> >
> >    - The overflow check for the buffer width is not correct if bpp < 8.
> >      However, it doesn't hurt, as widths larger than U32_MAX / 8 should
> >      not happen for real anyway.  Add a comment to clarify.
> >    - Calculating the stride from the number of characters per pixel is
> >      not correct.  Fix this by calculating it from the number of bits per
> >      pixel instead.
> >
> > Signed-off-by: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> > Reviewed-by: Javier Martinez Canillas <javierm@xxxxxxxxxx>
> > Tested-by: Javier Martinez Canillas <javierm@xxxxxxxxxx>
> > ---
> > v2:
> >    - Add Reviewed-by, Tested-by.
> > ---
> >   drivers/gpu/drm/drm_dumb_buffers.c | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/drm_dumb_buffers.c b/drivers/gpu/drm/drm_dumb_buffers.c
> > index 70032bba1c97e787..21a04c32a5e3d785 100644
> > --- a/drivers/gpu/drm/drm_dumb_buffers.c
> > +++ b/drivers/gpu/drm/drm_dumb_buffers.c
> > @@ -71,10 +71,11 @@ int drm_mode_create_dumb(struct drm_device *dev,
> >       /* overflow checks for 32bit size calculations */
> >       if (args->bpp > U32_MAX - 8)
> >               return -EINVAL;
> > +     /* Incorrect (especially if bpp < 8), but doesn't hurt much */
> >       cpp = DIV_ROUND_UP(args->bpp, 8);
> >       if (cpp > U32_MAX / args->width)
> >               return -EINVAL;
>
> At first, I was confused by this.

Me too ;-)

> So I'd really prefer to outright
> delete this code. As you say, it's incorrect and doesn't add anything.
> The concept of cpp is somewhat wrong anyway.
>
> > -     stride = cpp * args->width;
> > +     stride = DIV_ROUND_UP(args->bpp * args->width, 8);
>
> Do we need an overflow check for (bpp * width < U32_MAX) ?

I think I thought it would be covered by the above, but on more thought,
that is indeed not true.

So perhaps drop the cpp code above, and test bpp instead:

    -        cpp = DIV_ROUND_UP(args->bpp, 8);
    -        if (cpp > U32_MAX / args->width)
    +        if (args->bpp > U32_MAX / args->width)
                     return -EINVAL;

That would preclude a very wide buffer with height 1, though.

BTW, in v1 I also had this question:

    Why is drm_mode_create_dumb.size __u64?  The test for "args->height >
    U32_MAX / stride" rejects all sizes not fitting in __u32 anyway.

Gr{oetje,eeting}s,

                        Geert


--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds