[Intel-gfx] [PATCH 2/2] drm/i915/tgl: Skip Wa_1604555607 verification at A0

[Ramalingam C]

On 2019-10-02 at 09:54:12 +0100, Tvrtko Ursulin wrote:
> 
> On 01/10/2019 18:36, Chris Wilson wrote:
> > Quoting Ramalingam C (2019-10-01 18:26:24)
> > > Read of FF_MODE2 register is broken at TGL A0. Hence verification of the
> > > Wa_1604555607 associated to that register needs to be skipped.
> > > 
> > > Signed-off-by: Ramalingam C <ramalingam.c at intel.com>
> > > cc: Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com>
> > > ---
> > >   drivers/gpu/drm/i915/gt/intel_workarounds.c | 22 ++++++++++++++++++++-
> > >   drivers/gpu/drm/i915/i915_drv.h             |  6 ++++++
> > >   2 files changed, 27 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/gt/intel_workarounds.c b/drivers/gpu/drm/i915/gt/intel_workarounds.c
> > > index 4049b876492a..c63d8c3df4d3 100644
> > > --- a/drivers/gpu/drm/i915/gt/intel_workarounds.c
> > > +++ b/drivers/gpu/drm/i915/gt/intel_workarounds.c
> > > @@ -991,10 +991,21 @@ wa_list_apply(struct intel_uncore *uncore, const struct i915_wa_list *wal)
> > >          for (i = 0, wa = wal->list; i < wal->count; i++, wa++) {
> > >                  intel_uncore_rmw_fw(uncore, wa->reg, wa->mask, wa->val);
> > > -               if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM))
> > > +               if (IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)) {
> > > +
> > > +                       /*
> > > +                        * Read of FF_MODE2 is broken on TGL A0.
> > > +                        * Hence skip the corresponding WA verification.
> > > +                        */
> > > +                       if (IS_TGL_REVID(uncore->i915, 0, TGL_REVID_A0) &&
> > > +                           i915_mmio_reg_equal(wa->reg, FF_MODE2) &&
> > > +                           wa->mask == FF_MODE2_TDS_TIMER_MASK)
> > > +                               continue;
> > 
> > This does not scale very well. You will note that wa_verify() already
> > includes a detail on which bits can be read back from HW.
> 
> For the record Chris is referring to wa->read which you can set to zero for
> this particular wa and this should automagically work.
Thanks Tvrtko. We could set that wa->read = 0, only for required steppings.

-Ram
> 
> Then in IGT/gem_workarounds.c you also have existing framework for marking
> registers as write-only. It may need extending to account for steppings and
> so.
> 
> Regards,
> 
> Tvrtko