Re: [bug report] drm/amdgpu/amdgpu_connectors: remove amdgpu_connector_free_edid

["Joshua Peisach" <jpeisach@ubuntu.com>]

On Sat Apr 11, 2026 at 1:30 AM EDT, SRINIVASAN SHANMUGAM wrote:

> I went through the code path, and the warning looks valid:
>
> In amdgpu_connector_dvi_detect(), we do:
> drm_edid_free(amdgpu_connector->edid);
> After that, we call:
> amdgpu_connector_get_edid(connector);
> But inside amdgpu_connector_get_edid():
> It immediately returns if amdgpu_connector->edid is non-NULL
> Since we did not set amdgpu_connector->edid = NULL after freeing:
> The pointer is still non-NULL (but already freed)
> So amdgpu_connector_get_edid() becomes a no-op
> No new EDID is read
> Then later we do:
> drm_edid_is_digital(amdgpu_connector->edid);
> At this point:
> amdgpu_connector->edid still points to freed memory
> So this becomes a real use-after-free
>
> So the issue is not just the removal of amdgpu_connector_free_edid(),
> but that we lost the behavior of clearing the cached EDID pointer after free.
>
> Because of this, the EDID cache logic breaks.
>
> About reverting:
>
> Reverting the commit would fix it indirectly
> But I think a minimal fix is better:
> Set amdgpu_connector->edid = NULL after drm_edid_free()
>
> This keeps the current design intact and fixes the bug cleanly.
>

I like it - less is more :) and it also makes it clear that the
pointer is set to NULL, instead of being hidden behind a function.

-Josh