drm/msm: Avoid dirtyfb stalls on video mode displays (v2)
authorRob Clark <robdclark@chromium.org>
Wed, 23 Feb 2022 19:11:08 +0000 (11:11 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 17 Aug 2022 12:23:28 +0000 (14:23 +0200)
commit340bfec294d5fcc7f735ff7bad093e146198f580
tree40f15e8a3b4181e990f548f6ac55f6e0d6e19e64
parent7404d09455a06f001b0f0d63ec8927ff8c8a9b4d
drm/msm: Avoid dirtyfb stalls on video mode displays (v2)

[ Upstream commit 9e4dde28e9cd34ee13a6b7247f0857fb49fd3f19 ]

Someone on IRC once asked an innocent enough sounding question:  Why
with xf86-video-modesetting is es2gears limited at 120fps.

So I broke out the perfetto tracing mesa MR and took a look.  It turns
out the problem was drm_atomic_helper_dirtyfb(), which would end up
waiting for vblank.. es2gears would rapidly push two frames to Xorg,
which would blit them to screen and in idle hook (I assume) call the
DIRTYFB ioctl.  Which in turn would do an atomic update to flush the
dirty rects, which would stall until the next vblank.  And then the
whole process would repeat.

But this is a bit silly, we only need dirtyfb for command mode DSI
panels.  So track in plane state whether dirtyfb is required, and
track in the fb how many attached planes require dirtyfb so that we
can skip it when not required.  (Note, mdp4 does not actually have
cmd mode support.)

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Link: https://lore.kernel.org/r/20220223191118.881321-1-robdclark@gmail.com
Signed-off-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/gpu/drm/msm/disp/dpu1/dpu_crtc.c
drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c
drivers/gpu/drm/msm/disp/dpu1/dpu_plane.h
drivers/gpu/drm/msm/disp/mdp4/mdp4_plane.c
drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c
drivers/gpu/drm/msm/disp/mdp5/mdp5_kms.h
drivers/gpu/drm/msm/disp/mdp5/mdp5_plane.c
drivers/gpu/drm/msm/msm_atomic.c
drivers/gpu/drm/msm/msm_drv.h
drivers/gpu/drm/msm/msm_fb.c