[Intel-gfx] [PATCH] drm/i915: Return IDI Hash Mask setup for gen9+

[Rodrigo Vivi]

On Thu, Sep 14, 2017 at 07:44:29PM +0000, Ville Syrjälä wrote:
> On Thu, Sep 14, 2017 at 12:18:41PM -0700, Rodrigo Vivi wrote:
> > Spec teels:
> > "It is required for GFX Driver to set [19:16] to 1 when
> > eDRAM configuration is enabled."
> > 
> > This basically reverts commit 666fbcf5c21d ("drm/i915: Don't
> > program eLLC IDI hash mask for gen9+")
> > 
> > This requirement hasn't changed since HSW up to CNL.
> > But that commit was created before EDRAM was properly
> > organized with commit 3accaf7e734d ("drm/i915: Store
> > and use edram capabilities")
> > 
> > Cc: Mika Kuoppala <mika.kuoppala at intel.com>
> > Cc: Matthew Auld <matthew.auld at intel.com>
> > Cc: Chris Wilson <chris at chris-wilson.co.uk>
> > Signed-off-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_gem.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > index f445587c1a4b..25b59dbb29b0 100644
> > --- a/drivers/gpu/drm/i915/i915_gem.c
> > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > @@ -4714,7 +4714,7 @@ int i915_gem_init_hw(struct drm_i915_private *dev_priv)
> >  	/* Double layer security blanket, see i915_gem_init() */
> >  	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
> >  
> > -	if (HAS_EDRAM(dev_priv) && INTEL_GEN(dev_priv) < 9)
> > +	if (HAS_EDRAM(dev_priv))
> >  		I915_WRITE(HSW_IDICR, I915_READ(HSW_IDICR) | IDIHASHMSK(0xf));
> 
> Hmm. For me the spec seems to be saying
> "Note: It is required for GFX Driver to set [19:16] to 1 when eDRAM
>  configuration is enabled.
>  For Skylake, S/W is not needed to program this register as eDRAM is a
>  memory side cache."

Oh I had missed the SKL specoific version... was looking to an older and a newer...

> 
> But for CNL I only see the first sentence. Does that mean they moved
> the eLLC back to the ring side of the system agent, or just someone
> forgot to update this for CNL? The fact that eLLC moved to the memory
> side of the system agent on SKL isn't really mentioned anywhere else
> in the spec either AFAICS. Seems like a rather big elephant to overlook.
> Might be nice to get some clarification into the spec for this stuff.
> 
> Oh and I wonder what we should be doing with bits [21:20]?

good question...

also if it is whole 0xf or only 1 to the whole field.

> 
> >  
> >  	if (IS_HASWELL(dev_priv))
> > -- 
> > 2.13.5
> > 
> > _______________________________________________
> > Intel-gfx mailing list
> > Intel-gfx at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> 
> -- 
> Ville Syrjälä
> Intel OTC