[Intel-gfx] [PATCH] drm/i915: Unset legacy_cursor_update early in intel_atomic_commit, v2.

[Ville Syrjälä]

On Tue, Sep 19, 2017 at 11:06:52AM +0200, Maarten Lankhorst wrote:
> Op 18-09-17 om 17:03 schreef Ville Syrjälä:
> > On Mon, Sep 18, 2017 at 12:12:50PM +0200, Maarten Lankhorst wrote:
> >> Commit b44d5c0c105a ("drm/i915: Always wait for flip_done, v2.") removed
> >> the call to wait_for_vblanks and replaced it with flip_done.
> >>
> >> Unfortunately legacy_cursor_update was unset too late, and the
> >> replacement call drm_atomic_helper_wait_for_flip_done() was
> >> a noop. Make sure that its unset before setup_commit() is
> >> called to fix this issue.
> >>
> >> Changes since v1:
> >> - Force vblank wait for watermarks not yet converted to atomic too. (Ville)
> >> - Use for_each_new_intel_crtc_in_state. (Ville)
> >>
> >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
> >> Fixes: b44d5c0c105a ("drm/i915: Always wait for flip_done, v2.")
> >> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102675
> >> Testcase: kms_cursor_crc
> >> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> >> Cc: Jani Nikula <jani.nikula at linux.intel.com>
> >> Reported-by: Marta Löfstedt <marta.lofstedt at intel.com>
> >> Cc: Marta Löfstedt <marta.lofstedt at intel.com>
> >> Tested-by: Marta Löfstedt <marta.lofstedt at intel.com>
> >> Cc: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >> ---
> >>  drivers/gpu/drm/i915/intel_display.c | 45 +++++++++++++++++++++---------------
> >>  1 file changed, 26 insertions(+), 19 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> >> index 8599e425abb1..8d051256da1e 100644
> >> --- a/drivers/gpu/drm/i915/intel_display.c
> >> +++ b/drivers/gpu/drm/i915/intel_display.c
> >> @@ -12517,21 +12517,10 @@ static int intel_atomic_commit(struct drm_device *dev,
> >>  	struct drm_i915_private *dev_priv = to_i915(dev);
> >>  	int ret = 0;
> >>  
> >> -	ret = drm_atomic_helper_setup_commit(state, nonblock);
> >> -	if (ret)
> >> -		return ret;
> >> -
> >>  	drm_atomic_state_get(state);
> >>  	i915_sw_fence_init(&intel_state->commit_ready,
> >>  			   intel_atomic_commit_ready);
> >>  
> >> -	ret = intel_atomic_prepare_commit(dev, state);
> >> -	if (ret) {
> >> -		DRM_DEBUG_ATOMIC("Preparing state failed with %i\n", ret);
> >> -		i915_sw_fence_commit(&intel_state->commit_ready);
> >> -		return ret;
> >> -	}
> >> -
> >>  	/*
> >>  	 * The intel_legacy_cursor_update() fast path takes care
> >>  	 * of avoiding the vblank waits for simple cursor
> >> @@ -12540,19 +12529,37 @@ static int intel_atomic_commit(struct drm_device *dev,
> >>  	 * updates happen during the correct frames. Gen9+ have
> >>  	 * double buffered watermarks and so shouldn't need this.
> >>  	 *
> >> -	 * Do this after drm_atomic_helper_setup_commit() and
> >> -	 * intel_atomic_prepare_commit() because we still want
> >> -	 * to skip the flip and fb cleanup waits. Although that
> >> -	 * does risk yanking the mapping from under the display
> >> -	 * engine.
> >> +	 * Unset state->legacy_cursor_update before the call to
> >> +	 * drm_atomic_helper_setup_commit() because otherwise
> >> +	 * drm_atomic_helper_wait_for_flip_done() is a noop and
> >> +	 * we get FIFO underruns because we didn't wait
> >> +	 * for vblank.
> >>  	 *
> >>  	 * FIXME doing watermarks and fb cleanup from a vblank worker
> >>  	 * (assuming we had any) would solve these problems.
> >>  	 */
> >> -	if (INTEL_GEN(dev_priv) < 9)
> >> -		state->legacy_cursor_update = false;
> >> +	if (INTEL_GEN(dev_priv) < 9 && state->legacy_cursor_update) {
> >> +		struct intel_crtc_state *new_crtc_state;
> >> +		struct intel_crtc *crtc;
> >> +		int i;
> >> +
> >> +		for_each_new_intel_crtc_in_state(intel_state, crtc, new_crtc_state, i)
> >> +			if (new_crtc_state->wm.need_postvbl_update ||
> >> +			    new_crtc_state->update_wm_post)
> >> +				state->legacy_cursor_update = false;
> > Hmm. I guess that's better. But I still don't see why you want to change
> > this bit of code in this patch. AFAICS it's got nothing to do with the fix
> > itself, and instead it's just trying to optimize some cursor updates
> > that were kicked over to the slow path. Or am I missing something?
> 
> We accidentally removed the vblank wait for the slowpath, but I don't think we should reintroduce the vblank except where we need it..

IMO any regression fix should ideally get us back exactly where we were.

-- 
Ville Syrjälä
Intel OTC