[Intel-gfx] [RFCv5 2/2] drm/i915: Introduce private PAT management

[Wang, Zhi A]

I see. 

For 1) I can fix it in the next RFC.
For 2) I can find some VPG guys to ask for the details.

Thanks,
Zhi.

-----Original Message-----
From: Chris Wilson [mailto:chris at chris-wilson.co.uk] 
Sent: Tuesday, August 29, 2017 9:14 PM
To: Wang, Zhi A <zhi.a.wang at intel.com>; Joonas Lahtinen <joonas.lahtinen at linux.intel.com>; intel-gfx at lists.freedesktop.org; intel-gvt-dev at lists.freedesktop.org
Cc: zhenyuw at linux.intel.com; Widawsky, Benjamin <benjamin.widawsky at intel.com>; Vivi, Rodrigo <rodrigo.vivi at intel.com>
Subject: RE: [RFCv5 2/2] drm/i915: Introduce private PAT management

Quoting Wang, Zhi A (2017-08-29 18:54:51)
> Another finding during the re-factoring are:
> 
> a)It looks like that the PPAT_CACHE_INDEX on BDW/SKL is mapped to:
> GEN8_PPAT_WB | GEN8_PPAT_LLCELLC | GEN8_PPAT_AGE(0);
> 
> But the PPAT_CACHE_INDEX on CNL is mapped to GEN8_PPAT_LLCELLC | 
> GEN8_PPAT_AGE(0);
> 
> GEN8_PPAT_WB is missing here and by default the cache attribute is UC.
> 
> Is this set intentionally?

That sounds like a nasty little bug.

> b) Looks like all the ages of PPAT in windows driver is AGE(3) because of some performance gains, is there any reason that i915 has to set it to AGE(0)?

Nope, it's never been rigorously tested. On occasion, we've swapped it around (at least for the older gen) and never found a significant difference; I haven't even heard if anyone has tried such experiments on gen8+. Off the top of my head, the age should only matter when you have PTE with different ages (unless there's some automatic clock algorithm tracking the age on each page in a shadow, the challenge being then when you decide to refresh the age from the PTE.) -Chris