[Intel-gfx] [PATCH] drm/i915/vlv: assert and de-assert sideband reset on resume

[Jesse Barnes]

On Fri, 11 Apr 2014 21:06:31 +0300
Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:

> On Fri, Apr 11, 2014 at 10:34:19AM -0700, Jesse Barnes wrote:
> > On Fri, 11 Apr 2014 19:16:32 +0200
> > Daniel Vetter <daniel at ffwll.ch> wrote:
> > 
> > > On Fri, Apr 11, 2014 at 10:00:16AM -0700, Jesse Barnes wrote:
> > > > This is a bit like the CMN reset de-assert we do in DPIO_CTL, except
> > > > that it resets the whole common lane section of the PHY.  This is
> > > > required on machines where the BIOS doesn't do this for us on resume to
> > > > properly re-calibrate and get the PHY ready to transmit data.
> > > > 
> > > > Without this patch, such machines won't resume correctly much of the time,
> > > > with the symptom being a 'port ready' timeout and/or a link training
> > > > failure.
> > > > 
> > > > I'm open to better suggestions on how to do the power well toggle, with
> > > > the existing code it looks like I'd have to walk through a bunch of
> > > > power domains looking for a match, then call a generic function which
> > > > will warn.  I'd prefer to just expose the specific domains directly for
> > > > low level platform code like this.
> > > > 
> > > > Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
> > > > ---
> > > >  drivers/gpu/drm/i915/intel_pm.c     |    4 ++--
> > > >  drivers/gpu/drm/i915/intel_uncore.c |   19 +++++++++++++++++++
> > > >  2 files changed, 21 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> > > > index fa00185..3afd0bc 100644
> > > > --- a/drivers/gpu/drm/i915/intel_pm.c
> > > > +++ b/drivers/gpu/drm/i915/intel_pm.c
> > > > @@ -5454,8 +5454,8 @@ static bool i9xx_always_on_power_well_enabled(struct drm_i915_private *dev_priv,
> > > >  	return true;
> > > >  }
> > > >  
> > > > -static void vlv_set_power_well(struct drm_i915_private *dev_priv,
> > > > -			       struct i915_power_well *power_well, bool enable)
> > > > +void vlv_set_power_well(struct drm_i915_private *dev_priv,
> > > > +			struct i915_power_well *power_well, bool enable)
> > > >  {
> > > >  	enum punit_power_well power_well_id = power_well->data;
> > > >  	u32 mask;
> > > > diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
> > > > index 2a72bab..f1abd2d 100644
> > > > --- a/drivers/gpu/drm/i915/intel_uncore.c
> > > > +++ b/drivers/gpu/drm/i915/intel_uncore.c
> > > > @@ -363,6 +363,9 @@ static void intel_uncore_forcewake_reset(struct drm_device *dev, bool restore)
> > > >  	spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
> > > >  }
> > > >  
> > > > +void vlv_set_power_well(struct drm_i915_private *dev_priv,
> > > > +			struct i915_power_well *power_well, bool enable);
> > > > +
> > > >  void intel_uncore_early_sanitize(struct drm_device *dev)
> > > >  {
> > > >  	struct drm_i915_private *dev_priv = dev->dev_private;
> > > > @@ -381,6 +384,22 @@ void intel_uncore_early_sanitize(struct drm_device *dev)
> > > >  		DRM_INFO("Found %zuMB of eLLC\n", dev_priv->ellc_size);
> > > >  	}
> > > >  
> > > > +	/*
> > > > +	 * From VLV2A0_DP_eDP_HDMI_DPIO_driver_vbios_notes_11.docx:
> > > > +	 * Need to assert and de-assert PHY SB reset by gating the common
> > > > +	 * lane power, then un-gating it.
> > > > +	 * Simply ungating isn't enough to reset the PHY enough to get
> > > > +	 * ports and lanes running.
> > > > +	 */
> > > > +	if (IS_VALLEYVIEW(dev)) {
> > > > +		struct i915_power_well cmn_well = {
> > > > +			.data = PUNIT_POWER_WELL_DPIO_CMN_BC
> > > > +		};
> > > > +
> > > > +		vlv_set_power_well(dev_priv, &cmn_well, false);
> > > > +		vlv_set_power_well(dev_priv, &cmn_well, true);
> > > > +	}
> > > 
> > > Relationship with intel_reset_dpio? Should we move this bit of code over
> > > there? I'm lost in this maze of kick-me-harder patches for byt dpio ...
> > 
> > That happens too late.  This will clobber register state, whereas the
> > DPIO reset just resets the interface between the phy and the display.
> 
> As a clarification to the cmnreset thing, we never actually assert
> that signal, we just deassert it. The idea being that it should be
> asserted by default when things get powered on. But I wonder if we
> should assert it before suspending anyway.

And maybe do a write of 0 then 1 on resume too.  That's what Windows
does afaik.

> Oh and I think if we power gate the cmnlane we would need to
> assert/deassert cmnreset around it. In some CHV doc I see a note
> that side reset must be deasserted before cmnreset. The timing
> diagrams in VLV docs seem to have that order as well. So unless
> there's some internal logic which hold cmnreset asserted for the
> required time, we should do it by hand.

Yeah would be good to do that to be on the safe side.

> Oh and there's another intersting looking note:
> 
> "NOTE1 : Common lane reset must not be de-asserted until REFCLK to PLL is
>  enabled by i_pll*refclkbufen and the clock is running and stable"
> 
> I guess we managed to follow that by accident since we enable the
> refclock for DPLLB for the hotplug workaround. But perhaps we should
> enable the refclk for all PLLs just to be sure.

Yeah that shouldn't hurt.  In talking with the PHY guys, the cmnreset
de-assert will cause the PHY to re-calibrate, and any PLL settings
won't take effect until that's complete.  They will however be latched
& pended into the display, so it's safe to write them before, but
they'll take longer to lock the first time.

-- 
Jesse Barnes, Intel Open Source Technology Center