[Intel-gfx] [PATCH] drm/i915/dsi: Silence atomic update failure with DSI panel

[Mika Kahola]

On Thu, 2017-09-07 at 13:29 +0200, Maarten Lankhorst wrote:
> Op 05-09-17 om 15:35 schreef Mika Kahola:
> > 
> > It appears that we cannot trust scanline counters when MIPI/DSI
> > display is
> > connected. In CI system this appears as flickering errors that
> > randomly
> > appear in test cases. To avoid this flickering, let's just silence
> > atomic
> > update failure in case with DSI panel.
> > 
> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102403
> > Signed-off-by: Mika Kahola <mika.kahola at intel.com>
> > ---
> >  drivers/gpu/drm/i915/intel_sprite.c | 32 +++++++++++++++++------
> > ---------
> >  1 file changed, 17 insertions(+), 15 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/intel_sprite.c
> > b/drivers/gpu/drm/i915/intel_sprite.c
> > index b0d6e3e..8511072 100644
> > --- a/drivers/gpu/drm/i915/intel_sprite.c
> > +++ b/drivers/gpu/drm/i915/intel_sprite.c
> > @@ -205,23 +205,25 @@ void intel_pipe_update_end(struct
> > intel_crtc_state *new_crtc_state)
> >  	if (intel_vgpu_active(dev_priv))
> >  		return;
> >  
> > -	if (crtc->debug.start_vbl_count &&
> > -	    crtc->debug.start_vbl_count != end_vbl_count) {
> > -		DRM_ERROR("Atomic update failure on pipe %c
> > (start=%u end=%u) time %lld us, min %d, max %d, scanline start %d,
> > end %d\n",
> > -			  pipe_name(pipe), crtc-
> > >debug.start_vbl_count,
> > -			  end_vbl_count,
> > -			  ktime_us_delta(end_vbl_time, crtc-
> > >debug.start_vbl_time),
> > -			  crtc->debug.min_vbl, crtc-
> > >debug.max_vbl,
> > -			  crtc->debug.scanline_start,
> > scanline_end);
> > -	}
> > +	if (!intel_crtc_has_type(new_crtc_state,
> > INTEL_OUTPUT_DSI)) {
> > +		if (crtc->debug.start_vbl_count &&
> > +		    crtc->debug.start_vbl_count != end_vbl_count)
> > {
> > +			DRM_ERROR("Atomic update failure on pipe
> > %c (start=%u end=%u) time %lld us, min %d, max %d, scanline start
> > %d, end %d\n",
> > +				  pipe_name(pipe), crtc-
> > >debug.start_vbl_count,
> > +				  end_vbl_count,
> > +				  ktime_us_delta(end_vbl_time,
> > crtc->debug.start_vbl_time),
> > +				  crtc->debug.min_vbl, crtc-
> > >debug.max_vbl,
> > +				  crtc->debug.scanline_start,
> > scanline_end);
> > +		}
> >  #ifdef CONFIG_DRM_I915_DEBUG_VBLANK_EVADE
> > -	else if (ktime_us_delta(end_vbl_time, crtc-
> > >debug.start_vbl_time) >
> > -		 VBLANK_EVASION_TIME_US)
> > -		DRM_WARN("Atomic update on pipe (%c) took %lld us,
> > max time under evasion is %u us\n",
> > -			 pipe_name(pipe),
> > -			 ktime_us_delta(end_vbl_time, crtc-
> > >debug.start_vbl_time),
> > -			 VBLANK_EVASION_TIME_US);
> > +		else if (ktime_us_delta(end_vbl_time, crtc-
> > >debug.start_vbl_time) >
> > +			 VBLANK_EVASION_TIME_US)
> > +			DRM_WARN("Atomic update on pipe (%c) took
> > %lld us, max time under evasion is %u us\n",
> > +				 pipe_name(pipe),
> > +				 ktime_us_delta(end_vbl_time,
> > crtc->debug.start_vbl_time),
> > +				 VBLANK_EVASION_TIME_US);
> >  #endif
> > +	}
> >  }
> >  
> >  static void
> I don't think this goes far enough. We should stop claiming accurate
> vblanks when MIPI/DSI is used.
> intel_get_crtc_scanline will currently spin for 100 us to see if we
> can move from scanline offset = 0,
> this means that we add an additional 100 us wait for MIPI/DSI always.
> 
Definitely there's room for optimization here. We could get rid of
those delays.

> i915_get_crtc_scanoutpos should return false as well.
> 
> Does this mean need_vlv_dsi_wa in intel_pipe_update_start is now a
> noop? Should we perhaps only apply this
> for gen9+ MIPI/DSI?
We should limit this to gen9+ MIPI/DSI. The BSpec says that MIPI DSI is
not supported from Broxton onwards.
> 
> ~Maarten
> 
-- 
Mika Kahola - Intel OTC