It appears that the DRM's vblank timers have always been somewhat buggy in their timestamp calculations. Fix the generic implementation and improve reliability.
Patch 1 returns success/failure to the caller, as expected by DRM. Patch 2 fixes the timestamp calculation to return the time of the first visible scanline in the current vblank phase. Patch 3 switches the hrtimer to absolute values since boot, so that it's easier to work with and we can predict future timestamps reliably. Patches 4 and 5 allow for estimating timestamps even with the timer being disabled. Patches 6 and 7 improve the retrieval of the next vblank timeout's timestamp. Tested under heavy CPU-load on bochs and virtio. No ersors or warnings showed up. There have been reports about vblank timeouts being handled so late, that they trigger DRM's internal error checks. Maybe these fixes and adjustments can help to further avoid hick ups from delayed vblank timers. Thomas Zimmermann (7): drm/vblank: timer: Return success status from get_vblank_timeout drm/vblank: timer: Fix timestamp calculation drm/vblank: timer: Use absolute timer since boot drm/vblank: timer: Reorganize get_vblank_timeout drm/vblank: timer: Estimate vblank timeout if timer is disabled drm/vblank: timer: Verify that expiry time is in the future drm/vblank: timer: Avoid reading the vblank time unnecessarily drivers/gpu/drm/drm_vblank.c | 105 ++++++++++++++++++++-------- drivers/gpu/drm/drm_vblank_helper.c | 4 +- include/drm/drm_vblank.h | 2 +- 3 files changed, 79 insertions(+), 32 deletions(-) base-commit: 4f554688dffcacf48630c14f9fb77a9f60394c1c -- 2.54.0
