[Intel-gfx] [PATCH 3/3] drm/i915: Mark FBC B gone if pipe B is gone

[Luca Coelho]

On Thu, 2022-09-22 at 14:57 +0300, Ville Syrjälä wrote:
> On Thu, Sep 22, 2022 at 02:37:35PM +0300, Luca Coelho wrote:
> > On Thu, 2022-09-22 at 12:36 +0300, Ville Syrjälä wrote:
> > > On Thu, Sep 22, 2022 at 11:51:16AM +0300, Jani Nikula wrote:
> > > > On Thu, 22 Sep 2022, Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:
> > > > > On Thu, Sep 22, 2022 at 11:18:55AM +0300, Luca Coelho wrote:
> > > > > > On Fri, 2022-09-16 at 19:52 +0300, Ville Syrjala wrote:
> > > > > > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > > > > 
> > > > > > > If pipe B is fused off we also shouldn't have FBC B.
> > > > > > > 
> > > > > > > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > > > > ---
> > > > > > >  drivers/gpu/drm/i915/intel_device_info.c | 1 +
> > > > > > >  1 file changed, 1 insertion(+)
> > > > > > > 
> > > > > > > diff --git a/drivers/gpu/drm/i915/intel_device_info.c b/drivers/gpu/drm/i915/intel_device_info.c
> > > > > > > index 1434dc33cf49..fbefebc023f1 100644
> > > > > > > --- a/drivers/gpu/drm/i915/intel_device_info.c
> > > > > > > +++ b/drivers/gpu/drm/i915/intel_device_info.c
> > > > > > > @@ -394,6 +394,7 @@ void intel_device_info_runtime_init(struct drm_i915_private *dev_priv)
> > > > > > >  		if (dfsm & SKL_DFSM_PIPE_B_DISABLE) {
> > > > > > >  			runtime->pipe_mask &= ~BIT(PIPE_B);
> > > > > > >  			runtime->cpu_transcoder_mask &= ~BIT(TRANSCODER_B);
> > > > > > > +			runtime->fbc_mask &= ~BIT(INTEL_FBC_B);
> > > > > > >  		}
> > > > > > >  		if (dfsm & SKL_DFSM_PIPE_C_DISABLE) {
> > > > > > >  			runtime->pipe_mask &= ~BIT(PIPE_C);
> > > > > > 
> > > > > > I don't know (yet) what exactly this does, but it makes sense if you
> > > > > > think of consistency: we already do that for PIPE_A.
> > > > > 
> > > > > It's basically saying the entire pipe is fused off, so anything
> > > > > living inside that pipe should also be fused off.
> > > > > 
> > > > > > 
> > > > > > But what about PIPE_C and PIPE_D? Wouldn't it make sense to do the same
> > > > > > thing for them as well?
> > > > > 
> > > > > There is no FBC engine on those pipes (we don't even have
> > > > > the INTEL_FBC_C+ enum values defined), at least for now.
> > > > 
> > > > A future proof way would be to add
> > > > 
> > > > 	runtime->fbc_mask &= runtime->pipe_mask;
> > > 
> > > Dunno if I entirely like the extra assumption that the enums match.
> > > Also would need to make sure we don't accidentally screw up any
> > > old platforms where FBC is not tied to a specific pipe, but I
> > > guess we should never have pipe A fused off on those w/o
> > > the entire display engine fused off as well.
> > 
> > I must say I don't like the idea of making these assumptions across
> > different masks either.
> > 
> > I think that, since you are reading the DFSM register at runtime to
> > check whether those pipes are fused off, you should go all the way and
> > disable everything, including in the fbc_mask for all pipes.  Then you
> > don't need to make any assumptions about whether a pipe has FBC or not.
> > 
> > In short, I think you could add those INTEL_FBC_C+ definitions and
> > force-unset them here too...
> 
> Hmm. I don't see any real problem with adding the FBC C+D
> enum values even if not used by any platform currently.
> Do you want to write that patch?

Sure, I can do it.  I guess it should be done _after_ your patch? Or
should I just add those definitions and you'll rebase your patch? And
there's a third option: I can add the definitions and replace your
patch with one that does this for all PIPEs at once...

Which one do you prefer?

--
Cheers,
Luca.