Re: [Intel-gfx] [PATCH v4] drm/i915/display: Dual refresh rate fastset fixes with VRR fastset

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Thu, Aug 24, 2023 at 02:27:49PM +0300, Jani Nikula wrote:
> On Fri, 18 Aug 2023, Manasi Navare <navaremanasi@chromium.org> wrote:
> > Dual refresh rate (DRR) fastset seamlessly lets refresh rate
> > throttle without needing a full modeset.
> > However with the recent VRR fastset patches that got merged this
> > logic was broken. This is broken because now with VRR fastset
> > VRR parameters are calculated by default at the nominal refresh rate say 120Hz.
> > Now when DRR throttle happens to switch refresh rate to 60Hz, crtc clock
> > changes and this throws a mismatch in VRR parameters and fastset logic
> > for DRR gets thrown off and full modeset is indicated.
> >
> > This patch fixes this by ignoring the pipe mismatch for VRR parameters
> > in the case of DRR and when VRR is not enabled. With this fix, DRR
> > will still throttle seamlessly as long as VRR is not enabled.
> >
> > This will still need a full modeset for both DRR and VRR operating together
> > during the refresh rate throttle and then enabling VRR since now VRR
> > parameters need to be recomputed with new crtc clock and written to HW.
> >
> > This DRR + VRR fastset in conjunction needs more work in the driver and
> > will be fixed in later patches.
> >
> > v3:
> > Compute new VRR params whenever there is fastset and its non DRRS.
> > This will ensure it computes while switching to a fixed RR (Mitul)
> >
> > v2:
> > Check for pipe config mismatch in crtc clock whenever VRR is enabled
> >
> > Fixes: 1af1d18825d3 ("drm/i915/vrr: Allow VRR to be toggled during fastsets")
> 
> How could this have broken fastsets, when this made it possible to do
> vrr enable/disable fastsets to begin with? I was hoping to get a
> regressing commit to make this easier to reason.
> 
> > Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/9154
> > Cc: Drew Davenport <ddavenport@chromium.org>
> > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > Cc: Sean Paul <seanpaul@chromium.org>
> > Cc: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
> > Signed-off-by: Manasi Navare <navaremanasi@chromium.org>
> > ---
> >  drivers/gpu/drm/i915/display/intel_display.c | 14 ++++++++------
> >  1 file changed, 8 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > index 763ab569d8f3..2cf3b22adaf7 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > @@ -5352,7 +5352,7 @@ intel_pipe_config_compare(const struct intel_crtc_state *current_config,
> >  	if (IS_G4X(dev_priv) || DISPLAY_VER(dev_priv) >= 5)
> >  		PIPE_CONF_CHECK_I(pipe_bpp);
> >  
> > -	if (!fastset || !pipe_config->seamless_m_n) {
> > +	if (!fastset || !pipe_config->seamless_m_n || pipe_config->vrr.enable) {
> >  		PIPE_CONF_CHECK_I(hw.pipe_mode.crtc_clock);
> >  		PIPE_CONF_CHECK_I(hw.adjusted_mode.crtc_clock);
> >  	}
> > @@ -5387,11 +5387,13 @@ intel_pipe_config_compare(const struct intel_crtc_state *current_config,
> >  
> >  	if (!fastset)
> >  		PIPE_CONF_CHECK_BOOL(vrr.enable);
> > -	PIPE_CONF_CHECK_I(vrr.vmin);
> > -	PIPE_CONF_CHECK_I(vrr.vmax);
> > -	PIPE_CONF_CHECK_I(vrr.flipline);
> > -	PIPE_CONF_CHECK_I(vrr.pipeline_full);
> > -	PIPE_CONF_CHECK_I(vrr.guardband);
> > +	if ((fastset && !pipe_config->seamless_m_n) || pipe_config->vrr.enable) {
> 
> I just don't get the conditions here and above. For example, why
> wouldn't we check the parameters e.g. on full modeset that disables vrr?
> 
> I think we'll need a matrix of the features, which of them can be
> combined together, which are mutually exclusive, and which are expected
> to be fastsets.

I wouldn't expect a panel to support both DRRS and VRR. Pick one and
stick to it.

I haven't thought through all the implications of changing all the VRR
parameters live, so fastsets doing that are currently not possible. I
have a branch for LRR (which is essentially manual VRR) but that was
writtent before the VRR fastset support, so needs a full redesign now.
Until then I don't think we can do anything.

-- 
Ville Syrjälä
Intel