[Intel-gfx] [PATCH v2] drm/i915: Fix MST disable sequence

[Ville Syrjälä]

On Fri, Jan 10, 2020 at 06:40:38PM +0000, Souza, Jose wrote:
> On Wed, 2020-01-08 at 18:20 +0200, Ville Syrjälä wrote:
> > On Wed, Jan 08, 2020 at 04:09:31PM +0000, Souza, Jose wrote:
> > > On Wed, 2020-01-08 at 16:45 +0200, Ville Syrjala wrote:
> > > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > 
> > > > When moving the pipe disable & co. function calls from
> > > > haswell_crtc_disable() into the encoder .post_disable() hooks I
> > > > neglected to account for the MST vs. DDI interactions properly.
> > > > This now leads us to call these functions two times for the last
> > > > MST stream (once from the MST code and a second time from the DDI
> > > > code). The calls from the DDI code should only be done for SST
> > > > and not MST. Add the proper check for that.
> > > 
> > > Oohh I forgot that too.
> > > 
> > > > This results in an MCE on ICL. My vague theory is that we turn
> > > > off
> > > > the transcoder clock from the MST code and then we proceed to
> > > > touch
> > > > something in the DDI code which still depends on that clock
> > > > causing
> > > > the hardware to become upset. Though I can't really explain why
> > > > Stan's hack of omitting the pipe disable in the MST code would
> > > > avoid
> > > > the MCE since we should still be turning off the transcoder
> > > > clock.
> > > > But maybe there's something magic in the hw that keeps the clock
> > > > on
> > > > as long as the pipe is on. Or maybe the clock isn't the problem
> > > > and
> > > > we now touch something in the DDI disable code that really does
> > > > need
> > > > the pipe to be still enabled.
> > > > 
> > > > v2: Rebase to latest drm-tip
> > > > 
> > > > Cc: José Roberto de Souza <jose.souza at intel.com>
> > > > Cc: Manasi Navare <manasi.d.navare at intel.com>
> > > > Reported-by: Stanislav Lisovskiy <stanislav.lisovskiy at intel.com>
> > > > Closes: https://gitlab.freedesktop.org/drm/intel/issues/901
> > > > Fixes: 773b4b54351c ("drm/i915: Move stuff from
> > > > haswell_crtc_disable() into encoder .post_disable()")
> > > > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > ---
> > > >  drivers/gpu/drm/i915/display/intel_ddi.c | 22 ++++++++++++----
> > > > ------
> > > >  1 file changed, 12 insertions(+), 10 deletions(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > b/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > index 07acd0daca25..6e0a75d1e6ca 100644
> > > > --- a/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > +++ b/drivers/gpu/drm/i915/display/intel_ddi.c
> > > > @@ -3897,21 +3897,23 @@ static void intel_ddi_post_disable(struct
> > > > intel_encoder *encoder,
> > > >  	enum phy phy = intel_port_to_phy(dev_priv, encoder->port);
> > > >  	bool is_tc_port = intel_phy_is_tc(dev_priv, phy);
> > > >  
> > > > -	intel_crtc_vblank_off(old_crtc_state);
> > > > +	if (!intel_crtc_has_type(old_crtc_state, INTEL_OUTPUT_DP_MST))
> > > > {
> > > > +		intel_crtc_vblank_off(old_crtc_state);
> > > >  
> > > > -	intel_disable_pipe(old_crtc_state);
> > > > +		intel_disable_pipe(old_crtc_state);
> > > >  
> > > > -	if (INTEL_GEN(dev_priv) >= 11)
> > > > -		icl_disable_transcoder_port_sync(old_crtc_state);
> > > > +		if (INTEL_GEN(dev_priv) >= 11)
> > > > +			icl_disable_transcoder_port_sync(old_crtc_state
> > > > );
> > > >  
> > > > -	intel_ddi_disable_transcoder_func(old_crtc_state);
> > > > +		intel_ddi_disable_transcoder_func(old_crtc_state);
> > > >  
> > > > -	intel_dsc_disable(old_crtc_state);
> > > > +		intel_dsc_disable(old_crtc_state);
> > > >  
> > > > -	if (INTEL_GEN(dev_priv) >= 9)
> > > > -		skl_scaler_disable(old_crtc_state);
> > > > -	else
> > > > -		ilk_pfit_disable(old_crtc_state);
> > > > +		if (INTEL_GEN(dev_priv) >= 9)
> > > > +			skl_scaler_disable(old_crtc_state);
> > > > +		else
> > > > +			ilk_pfit_disable(old_crtc_state);
> > > > +	}
> > > 
> > > Other option would be replace
> > > intel_dig_port->base.post_disable(&intel_dig_port->base,
> > > old_crtc_state, NULL);
> > > in intel_mst_post_disable_dp() by:
> > > 
> > > 
> > > intel_ddi_post_disable_dp(encoder, old_crtc_state, old_conn_state);
> > > 
> > > if (INTEL_GEN(dev_priv) >= 11)
> > > 	icl_unmap_plls_to_ports(encoder);
> > > 
> > > if (intel_crtc_has_dp_encoder(old_crtc_state) || is_tc_port)
> > > 	intel_display_power_put_unchecked(dev_priv,
> > > intel_ddi_main_link_aux_domain(dig_port));
> > > 
> > > if (is_tc_port)
> > > 	intel_tc_port_put_link(dig_port);
> > 
> > Yeah, the current way is a bit of a mess. We probably want to think
> > of
> > ways to make it less sucky.
> 
> Can I go forward and implement the above and undoing this patch?

I'm thinking we should start looking at the enable side too so that
we can come up with some kind of scheme that isn't totally out of
sync for enable vs. disable.

-- 
Ville Syrjälä
Intel