[Intel-gfx] [PATCH 13/16] drm/i915: Do not print DC off mismatch state when DMC firmware in not loaded

[Imre Deak]

On Tue, Oct 16, 2018 at 04:31:37AM +0300, Souza, Jose wrote:
> On Mon, 2018-10-15 at 13:58 +0300, Imre Deak wrote:
> > On Fri, Oct 12, 2018 at 02:52:15PM -0700, José Roberto de Souza
> > wrote:
> > > When DMC firmware is not loaded, it return earlier in
> > > gen9_dc_off_power_well_disable() as it will have no effect without
> > > DMC firmware loaded. But it will cause a mismatch state error when
> > > running intel_power_domains_verify_state(), so skipping this error
> > > in this case.
> > 
> > DC states are disabled when DMC is not loaded and we won't ever
> > enable
> > them, as runtime PM as a whole is disabled. So not sure why you get a
> > mismatch error, but ignoring that by special casing it doesn't seem
> > correct.
> 
> I just found out a bug that was hidden this bug from drm-tip, I just
> sent a separated PR(https://patchwork.freedesktop.org/series/51039/)
> with this patch and the fix please take a look.
> 
> But it was hidden because at every call to
> intel_power_domains_verify_state() something was holding a reference to
> one of DC_OFF domains, keeping it in the on state that means DC is
> disabaled.

Yes, runtime PM is kept disabled if the DMC firmware is not loaded. This
also means we keep all power wells in the INIT domain - including the
DC_OFF power well - enabled.

> If you run drm-tip with the fix without any output all references to
> DC_OFF domain will be release but as DMC firmware was not loaded DC is
> not actualy enabled causing the state mismatch.
> A easy way to reproduce is just run drm-tip with disable_display=1 and
> check the error message in dmesg.
> 
> Although I'm not 100% happy with the way that I'm identifying DC_OFF
> power well, do you have suggestion?
> 
> > 
> > > Signed-off-by: José Roberto de Souza <jose.souza at intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/intel_runtime_pm.c | 13 ++++++++++++-
> > >  1 file changed, 12 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c
> > > b/drivers/gpu/drm/i915/intel_runtime_pm.c
> > > index 8b1c4d0db0af..629091ad8337 100644
> > > --- a/drivers/gpu/drm/i915/intel_runtime_pm.c
> > > +++ b/drivers/gpu/drm/i915/intel_runtime_pm.c
> > > @@ -4014,11 +4014,22 @@ static void
> > > intel_power_domains_verify_state(struct drm_i915_private *dev_priv)
> > >  		enabled = power_well->desc->ops->is_enabled(dev_priv,
> > >  							    power_well)
> > > ;
> > >  		if ((power_well->count || power_well->desc->always_on)
> > > !=
> > > -		    enabled)
> > > +		    enabled) {
> > > +			/* If DMC firmware is not loaded it could cause
> > > a
> > > +			 * mismatch state as we can't disable DC off,
> > > so let's
> > > +			 * do not print any errors in this scenario.
> > > +			 */
> > > +
> > > +			if (!strcmp("DC off", power_well->desc->name)
> > > &&
> > > +			    !dev_priv->csr.dmc_payload)
> > > +				goto skip_state_mismatch_error;
> > > +
> > >  			DRM_ERROR("power well %s state mismatch
> > > (refcount %d/enabled %d)",
> > >  				  power_well->desc->name,
> > >  				  power_well->count, enabled);
> > > +		}
> > >  
> > > +skip_state_mismatch_error:
> > >  		domains_count = 0;
> > >  		for_each_power_domain(domain, power_well->desc-
> > > >domains)
> > >  			domains_count += power_domains-
> > > >domain_use_count[domain];
> > > -- 
> > > 2.19.1
> > > 
> > > _______________________________________________
> > > Intel-gfx mailing list
> > > Intel-gfx at lists.freedesktop.org
> > > https://lists.freedesktop.org/mailman/listinfo/intel-gfx