Re: [PATCH 09/13] KVM: PPC: Book3S HV: Adapt TLB invalidations to work on POWER9

[Paul Mackerras <paulus@ozlabs.org>]

On Sat, Nov 19, 2016 at 08:57:28AM +1100, Benjamin Herrenschmidt wrote:
> On Fri, 2016-11-18 at 20:11 +0530, Aneesh Kumar K.V wrote:
> > > +      * Work out how many sets the TLB has, for the use of
> > > +      * the TLB invalidation loop in book3s_hv_rmhandlers.S.
> > > +      */
> > > +     if (cpu_has_feature(CPU_FTR_ARCH_300))
> > > +             kvm->arch.tlb_sets = 256;       /* POWER9 */
> > > +     else if (cpu_has_feature(CPU_FTR_ARCH_207S))
> > > +             kvm->arch.tlb_sets = 512;       /* POWER8 */
> > > +     else
> > > +             kvm->arch.tlb_sets = 128;       /* POWER7 */
> > > +
> > 
> > We have 
> > 
> > #define POWER7_TLB_SETS         128     /* # sets in POWER7 TLB */
> > #define POWER8_TLB_SETS         512     /* # sets in POWER8 TLB */
> > #define POWER9_TLB_SETS_HASH    256     /* # sets in POWER9 TLB Hash mode */
> > #define POWER9_TLB_SETS_RADIX   128     /* # sets in POWER9 TLB Radix mode */
> > 
> > May be use that instead of opencoding ?
> 
> Both are bad and are going to kill us for future backward
> compatibility.
> 
> These should be a device-tree property. We can fallback to hard wired
> values if it doesn't exist but we should at least look for one.

Tell me what the property is called and I'll add code to use it. :)
That's the whole reason why I moved this to C code.

> Note: P8 firmwares all have a bug creating a bogus "tlb-sets" property
> in the CPU node, so let's create a new one instead, with 2 entries
> (hash vs. radix) or 2 new ones, one for hash and one for radix (when
> available).

Paul.