[PATCH v3] drm/amd/display: fix flickering caused by S/G mode

[Christian König]

Am 20.04.23 um 15:44 schrieb Hamza Mahfooz:
> Currently, on a handful of ASICs. We allow the framebuffer for a given
> plane to exist in either VRAM or GTT. However, if the plane's new
> framebuffer is in a different memory domain than it's previous
> framebuffer, flipping between them can cause the screen to flicker. So,
> to fix this, don't perform an immediate flip in the aforementioned case.
>
> Cc: stable at vger.kernel.org
> Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2354
> Fixes: 81d0bcf99009 ("drm/amdgpu: make display pinning more flexible (v2)")
> Signed-off-by: Hamza Mahfooz <hamza.mahfooz at amd.com>
> ---
> v2: make a number of clarifications to the commit message and drop
>      locking.
> v3: use a stronger check
> ---
>   .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c    | 16 ++++++++++++++--
>   1 file changed, 14 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> index dfcb9815b5a8..875111340203 100644
> --- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> +++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c
> @@ -7900,6 +7900,13 @@ static void amdgpu_dm_commit_cursors(struct drm_atomic_state *state)
>   			amdgpu_dm_plane_handle_cursor_update(plane, old_plane_state);
>   }
>   
> +static inline uint32_t get_mem_type(struct drm_framebuffer *fb)
> +{
> +	struct amdgpu_bo *abo = gem_to_amdgpu_bo(fb->obj[0]);
> +
> +	return abo->tbo.resource ? abo->tbo.resource->mem_type : 0;
> +}
> +
>   static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
>   				    struct dc_state *dc_state,
>   				    struct drm_device *dev,
> @@ -7919,6 +7926,7 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
>   			to_dm_crtc_state(drm_atomic_get_old_crtc_state(state, pcrtc));
>   	int planes_count = 0, vpos, hpos;
>   	unsigned long flags;
> +	uint32_t mem_type;
>   	u32 target_vblank, last_flip_vblank;
>   	bool vrr_active = amdgpu_dm_crtc_vrr_active(acrtc_state);
>   	bool cursor_update = false;
> @@ -8040,13 +8048,17 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state,
>   			}
>   		}
>   
> +		mem_type = get_mem_type(old_plane_state->fb);
> +
>   		/*
>   		 * Only allow immediate flips for fast updates that don't
> -		 * change FB pitch, DCC state, rotation or mirroing.
> +		 * change memory domain, FB pitch, DCC state, rotation or
> +		 * mirroring.
>   		 */
>   		bundle->flip_addrs[planes_count].flip_immediate =
>   			crtc->state->async_flip &&
> -			acrtc_state->update_type == UPDATE_TYPE_FAST;
> +			acrtc_state->update_type == UPDATE_TYPE_FAST &&
> +			mem_type && get_mem_type(fb) == mem_type;

I think you can actually drop checking mem_type here.

When the FB has no backing store, we actually don't have anything to 
scan out. In this case flickering is the least of your problems :)

Christian


>   
>   		timestamp_ns = ktime_get_ns();
>   		bundle->flip_addrs[planes_count].flip_timestamp_in_us = div_u64(timestamp_ns, 1000);