On 2026-11-03 16:13, James Lin wrote:
> From: James Lin <[email protected]>
>
> amdgpu_get_vblank_counter_kms() is DRM's hardware ->get_vblank_counter
> callback, so drm_update_vblank_count() stores its return value into
> vblank->last and derives diff = cur_vblank - vblank->last.
>
> Commit f2bc71d16e73 ("drm/amdgpu: serialize vblank counter reads
> against GPU reset") made the callback fall back to
> drm_crtc_vblank_count() when the reset semaphore is held. That returns
> DRM's software vblank count, which is a different numeric domain than
> the hardware counter. It poisons vblank->last, so a later update hits
> diff == 0 while cur_vblank != vblank->last and trips
> drm_WARN_ON_ONCE() in drm_update_vblank_count(). Reset still succeeds,
> but the WARN fails the mode1_reset dmesg check.
>
> Return the cached hardware value vblank->last via drm_crtc_vblank_crtc()
> instead, so the core sees "no new vblanks" and keeps its accounting
> consistent. vblank->last is protected by drm_device.vbl_lock, already
> held across this callback.
>
> Fixes: f2bc71d16e73 ("drm/amdgpu: serialize vblank counter reads against GPU
> reset")
> Signed-off-by: James Lin <[email protected]>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 13 +++++++++++--
> 1 file changed, 11 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> index 14a72a406d37..365b4fc86a9e 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> @@ -1699,9 +1699,18 @@ u32 amdgpu_get_vblank_counter_kms(struct drm_crtc
> *crtc)
> DRM_ERROR("Invalid crtc %u\n", pipe);
> return -EINVAL;
> }
> - if (!down_read_trylock(&adev->reset_domain->sem))
> - return drm_crtc_vblank_count(crtc);
>
> + /*
> + * A GPU reset is in progress and the display hardware cannot be read
> + * safely. Return the hardware counter value DRM already cached rather
> + * than the software vblank count: the core feeds this callback's
> + * result back into vblank->last, and a software count lives in a
> + * different numeric domain, which later trips
> + * drm_WARN_ON_ONCE(cur_vblank != vblank->last) in
> + * drm_update_vblank_count().
> + */
> + if (!down_read_trylock(&adev->reset_domain->sem))
> + return drm_crtc_vblank_crtc(crtc)->last;
Reviewed-by: Leo Li <[email protected]>
Thanks!
>
> /* The hw increments its frame counter at start of vsync, not at start
> * of vblank, as is required by DRM core vblank counter handling.
