On Fri, 05 Sep 2025, Zilin Guan <[email protected]> wrote:
> The drm_edid_alloc() function uses kmemdup() to copy raw EDID data,
> which can be loaded from firmware. A problem exists because the firmware
> subsystem uses vfree() to release its data buffer. This implies the
> buffer may be allocated with vmalloc() and can thus be larger than
> kmalloc() typically supports. Since kmemdup() uses kmalloc() internally,
> attempting to duplicate a large, vmalloc'd EDID buffer can lead to an
> allocation failure.

Real world EDIDs are usually under 1 kB and if the user provides a much
bigger EDID via the firmware loader it's okay to fail at the allocation
time.

BR,
Jani.

>
> To fix this mismatch and robustly handle EDID data of any size, this
> patch replaces kmemdup() with kvmemdup(). The kvmemdup() function is
> designed for this scenario, as it can safely handle a vmalloc'd source
> and choose an appropriate allocator for the destination. The corresponding
> free calls are therefore updated to kvfree().
>
> Signed-off-by: Zilin Guan <[email protected]>
> ---
>  drivers/gpu/drm/drm_edid.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c
> index e2e85345aa9a..97142bfc45ad 100644
> --- a/drivers/gpu/drm/drm_edid.c
> +++ b/drivers/gpu/drm/drm_edid.c
> @@ -2530,13 +2530,13 @@ const struct drm_edid *drm_edid_alloc(const void 
> *edid, size_t size)
>       if (!edid || !size || size < EDID_LENGTH)
>               return NULL;
>  
> -     edid = kmemdup(edid, size, GFP_KERNEL);
> +     edid = kvmemdup(edid, size, GFP_KERNEL);
>       if (!edid)
>               return NULL;
>  
>       drm_edid = _drm_edid_alloc(edid, size);
>       if (!drm_edid)
> -             kfree(edid);
> +             kvfree(edid);
>  
>       return drm_edid;
>  }
> @@ -2568,7 +2568,7 @@ void drm_edid_free(const struct drm_edid *drm_edid)
>       if (!drm_edid)
>               return;
>  
> -     kfree(drm_edid->edid);
> +     kvfree(drm_edid->edid);
>       kfree(drm_edid);
>  }
>  EXPORT_SYMBOL(drm_edid_free);

-- 
Jani Nikula, Intel

Reply via email to