ACK/Cmnt [PATCH 2/2] Revert "drm/amd/amdgpu: fix refcount leak"

Krzysztof Kozlowski krzysztof.kozlowski at canonical.com
Thu Jul 29 07:45:17 UTC 2021


On 28/07/2021 17:02, Stefan Bader wrote:
> This reverts commit 6c0065eb69de90d50186c04f51ffe889e5bbe037. The 4.19.y
> series which this patch comes from has a previous change which is not in
> bionic which stores GEM info in the framebuffer. Suspecting that due to
> this we actually have no refcount leak in our kernel.
> 

The get() happens via:
amdgpufb_create() -> amdgpu_display_gem_fb_init() ->
amdgpu_display_framebuffer_init().

The backport is a fix for 37ac3dc00da0 ("drm/amdgpu: Use device specific
BO size & stride check."), which we do not have. So obviously it's bogus
in our context and the cause of memory corruption.

However maybe we should consider backporting both - 37ac3dc00da0 and fix
reverted here... but that's different story :)

Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski at canonical.com>


Best regards,
Krzysztof



More information about the kernel-team mailing list