[PATCH v3] drm/i915/panelreplay: Panel replay workaround with VRR

[Ville Syrjälä]

On Wed, Feb 21, 2024 at 11:08:18PM +0200, Ville Syrjälä wrote:
> On Wed, Feb 21, 2024 at 03:58:48PM -0500, Rodrigo Vivi wrote:
> > On Wed, Feb 21, 2024 at 08:19:35PM +0000, Manna, Animesh wrote:
> > > 
> > > 
> > > > -----Original Message-----
> > > > From: Vivi, Rodrigo <rodrigo.vivi at intel.com>
> > > > Sent: Tuesday, February 20, 2024 11:12 PM
> > > > To: Manna, Animesh <animesh.manna at intel.com>
> > > > Cc: intel-gfx at lists.freedesktop.org; ville.syrjala at linux.intel.com; Hogander,
> > > > Jouni <jouni.hogander at intel.com>; Murthy, Arun R
> > > > <arun.r.murthy at intel.com>
> > > > Subject: Re: [PATCH v3] drm/i915/panelreplay: Panel replay workaround with
> > > > VRR
> > > > 
> > > > On Tue, Feb 20, 2024 at 07:49:19PM +0530, Animesh Manna wrote:
> > > > > Panel Replay VSC SDP not getting sent when VRR is enabled and W1 and
> > > > > W2 are 0. So Program Set Context Latency in
> > > > TRANS_SET_CONTEXT_LATENCY
> > > > > register to at least a value of 1.
> > > > >
> > > > > HSD: 14015406119
> > > > 
> > > > Unnecessary mark since the wa_name already is a pointer to the HSD.
> > > > 
> > > > >
> > > > > v1: Initial version.
> > > > > v2: Update timings stored in adjusted_mode struct. [Ville]
> > > > > v3: Add WA in compute_config(). [Ville]
> > > > >
> > > > > Signed-off-by: Animesh Manna <animesh.manna at intel.com>
> > > > > ---
> > > > >  drivers/gpu/drm/i915/display/intel_dp.c | 12 ++++++++++++
> > > > >  1 file changed, 12 insertions(+)
> > > > >
> > > > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > b/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > index 217196196e50..eb0fa513cd0f 100644
> > > > > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > @@ -2948,6 +2948,18 @@ intel_dp_compute_config(struct intel_encoder
> > > > *encoder,
> > > > >  	intel_dp_compute_vsc_sdp(intel_dp, pipe_config, conn_state);
> > > > >  	intel_dp_compute_hdr_metadata_infoframe_sdp(intel_dp,
> > > > pipe_config,
> > > > > conn_state);
> > > > >
> > > > > +	/*
> > > > > +	 * WA: HSD-14015406119
> > > > 
> > > > this is not the right one. You should use the lineage one and then mark the
> > > > platforms.
> > > > 
> > > > /* wa_14015401596: xe_lpd, xe_hpd */
> > > > 
> > > > or perhaps
> > > > 
> > > > /* wa_14015401596: display versions: 13, 14 */
> > > > 
> > > > and also add a check for the display version with it.
> > > 
> > > Sure.
> > > 
> > > > 
> > > > > +	 * Program Set Context Latency in TRANS_SET_CONTEXT_LATENCY
> > > > register
> > > > > +	 * to at least a value of 1 when Panel Replay is enabled with VRR.
> > > > > +	 * Value for TRANS_SET_CONTEXT_LATENCY is calculated by
> > > > substracting
> > > > > +	 * crtc_vdisplay from crtc_vblank_start, so incrementing
> > > > crtc_vblank_start
> > > > > +	 * by 1 if both are equal.
> > > > > +	 */
> > > > > +	if (pipe_config->vrr.enable && pipe_config->has_panel_replay &&
> > > > > +	    adjusted_mode->crtc_vblank_start == adjusted_mode-
> > > > >crtc_vdisplay)
> > > > > +		adjusted_mode->crtc_vblank_start += 1;
> > > > 
> > > > why to mess with the vblank start instead of going to
> > > > intel_set_transcoder_timings() and change directly what is getting written to
> > > > the register when the register gets written?
> > > 
> > > I have done in previous version of this patch. But as per review feedback, added now here.
> > > https://patchwork.freedesktop.org/series/129632/#rev1
> > > https://patchwork.freedesktop.org/series/129632/#rev2
> > >  
> > > > 
> > > > In case the answer is becasue by then we don't have the vrr.enable or
> > > > something like that, then we should consider move around when we set that
> > > > register?
> > > 
> > > This was not acceptable in earlier versions. As per feedback instead of atomic-commit need to add in compute-config phase.  
> > > 
> > > > 
> > > > or perhaps create a specific flag? one extra variable, 3 less comment lines...
> > > 
> > > The above comment is not clear to me, can you please elaborate more here.
> > 
> > something like:
> > 
> > @intel_dp_compute_config()
> > 
> > +if (pipe_config->vrr.enable && pipe_config->has_panel_replay &&
> > +	adjusted_mode->crtc_vblank_start == adjusted_mode-crtc_vdisplay)
> > +	pipe_config->mode_flags = I915_MODE_FLAG_MIN_TRANS_CONTEXT_LATENCY_1
> > 
> > then
> > @intel_set_transcoder_timings()
> > +u32 context_latency;
> > 
> > +if (crtc_state->mode_flags & I915_MODE_FLAG_MIN_TRANS_CONTEXT_LATENCY_1)
> > +   context_latency = 1;
> > +else
> > +   crtc_vblank_start - crtc_vdisplay;
> > 
> > -crtc_vblank_start - crtc_vdisplay);
> > +context_latency);
> > 
> > Ville, thoughts?
> 
> I think what we need is a intel_crtc_vblank_delay() or somesuch thing
> that accounts for all the things (eg. this w/a, dsb execution latency
> when we start to use dsb for double buffered registers, etc). 
> And it should probably be called from some central place so that
> it works for all output types. intel_crtc_compute_config() comes to
> mind, but I guess we need to first ponder whether there are bits
> of code being executed prior to that that would need to know the
> actual vblank_start...

PSR2 itself might have a chicken vs. egg issue here. Wehether we
can enable it not depends on the length of the vblank. Whether
that means the full vblank or the pipe's shrunken vblank I don't
know. Someone needs to figure that out.

-- 
Ville Syrjälä
Intel