Hi, On 16 November 2015 at 11:49, Maarten Lankhorst <maarten@xxxxxxxxxxxxxx> wrote: > If an atomic update fails intel_crtc->atomic may have have some values left > from the last atomic check. One example is atomic->wait_for_vblank, > which results in spurious errors in kms_flip. > > Testcase: kms_flip > Reported-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> > Cc: stable@xxxxxxxxxxxxxxx #v4.3 > Signed-off-by: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx> > --- > diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c > index b5f7493213b7..b9539b14fe0d 100644 > --- a/drivers/gpu/drm/i915/intel_display.c > +++ b/drivers/gpu/drm/i915/intel_display.c > @@ -13141,6 +13141,9 @@ static int intel_atomic_check(struct drm_device *dev, > struct intel_crtc_state *pipe_config = > to_intel_crtc_state(crtc_state); > > + memset(&to_intel_crtc(crtc)->atomic, 0, > + sizeof(struct intel_crtc_atomic_commit)); > + > /* Catch I915_MODE_FLAG_INHERITED */ > if (crtc_state->mode.private_flags != crtc->state->mode.private_flags) > crtc_state->mode_changed = true; How about doing it in intel_crtc_duplicate_state instead? Should be a bit more bulletproof/correct against various other failure modes. But I guess this does work, and it is all going away shortly, so: Reviewed-by: Daniel Stone <daniels@xxxxxxxxxxxxx> I think the reason we haven't been bitten by cdclk failures inside intel_atomic_state.atomic, is because intel_atomic_state always starts with a zeroed structure. Cheers, Daniel _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
