[Intel-gfx] [PATCH 05/20] drm/i915: Save all GT WAs and apply them at a later time

[Chris Wilson]

Quoting Joonas Lahtinen (2017-10-31 14:14:52)
> On Mon, 2017-10-30 at 13:17 -0700, Oscar Mateo wrote:
> > By doing this, we can dump these workarounds in debugfs for validation (which,
> > at the moment, we are only able to do for the contexts WAs).
> > 
> > v2:
> >   - Wrong macro used for MMIO set bit masked
> >   - Improved naming
> >   - Rebased
> > 
> > v3:
> >   - GT instead of MMIO (Chris, Mika)
> >   - Leave L3_PRIO_CREDITS_MASK for a separate patch
> >   - Rebased
> > 
> > v4: Carry the init_early nomenclature over (Chris)
> > 
> > Signed-off-by: Oscar Mateo <oscar.mateo at intel.com>
> > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> > Cc: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
> 
> This and the following patch are still a no-go and won't be merged. The
> required changes for the series to be accepted (to make it more
> declarative) were clearly described previously. If there are further
> questions, we should discuss those instead wasting time looking at
> respins that do not address the input given.

I would like draw everyone's attention to

https://bugs.freedesktop.org/show_bug.cgi?id=103549

As much as I don't like gem_workarounds for its incestrous relationship
with the kernel it purports to be testing, that bug is exactly the type
of regression it prevents. It could not find this regression because it
requires us to be very formal in our w/a handling, i.e. we had not
declared the w/a for it to check; such formality being sought after here.

Whatever the design outcome, a good test plan is essential.
-Chris