On Fri, Nov 03, 2017 at 01:27:55PM +0000, Chris Wilson wrote:
> Quoting Ville Syrjala (2017-11-03 13:03:37)
> > From: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
> >
> > Apparently setting up a bunch of GT registers before we've properly
> > initialized the rest of the GT hardware leads to these setting being
> > lost. So looks like I broke HSW with commit b7048ea12fbb ("drm/i915:
> > Do .init_clock_gating() earlier to avoid it clobbering watermarks")
> > by doing init_clock_gating() too early. This should actually affect
> > other platforms as well, but apparently not to such a great degree.
> >
> > What I was ultimately after in that commit was to move the
> > ilk_init_lp_watermarks() call earlier. So let's undo the damage and
> > move init_clock_gating() back to where it was, and call
> > ilk_init_lp_watermarks() just before the watermark state readout.
> >
> > This highlights how fragile and messed up our init order really is.
> > I wonder why we even initialize the display before gem. The opposite
> > order would make much more sense to me...
>
> Indeed this will cause some fun for me momentarily as I will have to
> move init_clock_gating() to i915_gem_init(). We can only wish for that
> magic wand to be waved sooner.
>
> Does that make sense, or will I have to start carving up
> init_clock_gating()?
i915_gem_init() should be fine as far as the display is concerned
at least, albeit a bit unexpected. Do we need to do that already in
this patch, or a followup?
>
> > diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> > index 07118c0b69d3..352a6739ed70 100644
> > --- a/drivers/gpu/drm/i915/intel_pm.c
> > +++ b/drivers/gpu/drm/i915/intel_pm.c
> > @@ -5754,12 +5754,30 @@ void vlv_wm_sanitize(struct drm_i915_private *dev_priv)
> > mutex_unlock(&dev_priv->wm.wm_mutex);
> > }
> >
> > +/*
> > + * FIXME should probably kill this and improve
> > + * the real watermark readout/sanitation instead
> > + */
> > +static void ilk_init_lp_watermarks(struct drm_i915_private *dev_priv)
> > +{
> > + I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
> > + I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
> > + I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
> > +
> > + /*
> > + * Don't touch WM1S_LP_EN here.
> > + * Doing so could cause underruns.
> > + */
> > +}
> > +
> > void ilk_wm_get_hw_state(struct drm_device *dev)
> > {
> > struct drm_i915_private *dev_priv = to_i915(dev);
> > struct ilk_wm_values *hw = &dev_priv->wm.hw;
> > struct drm_crtc *crtc;
> >
> > + ilk_init_lp_watermarks(dev_priv);
>
> Wasn't expecting this, but the rest lgtm. Could you explain you decision
> to put it here in a bit more detail?
The original problem was that this guy turned off the LP1+ watermarks
after we'd already done the state readout in ilk_wm_get_hw_state(). So
the state we had read out no longer matched the hardware state.
To keep the software and hardware states in sync we just need to make
sure ilk_init_lp_watermarks() is called before the readout. And the
obvious thing to do then is to call it immediately before to the
readout.
--
Ville Syrjälä
Intel OTC