Re: [PATCH v2] selftests: KVM: avoid failures due to reserved HyperTransport region

[Paolo Bonzini]

On 12/9/21 22:47, Sean Christopherson wrote:
> > +		/*
> > +		 * Otherwise it's at the top of the physical address
> > +		 * space, possibly reduced due to SME by bits 11:6 of
> > +		 * CPUID[0x8000001f].EBX.
> > +		 */
> > +		eax = 0x80000008;
> > +		cpuid(&eax, &ebx, &ecx, &edx);
>
> Should't this check 0x80000000.eax >= 0x80000008 first?  Or do we just accept
> failure if family==0x17 and there's no 0x80000008?  One paranoid option would be
> to use the pre-fam17 value, e.g.
>
>          /* Before family 17h, the HyperTransport area is just below 1T. */
>          ht_gfn = (1 << 28) - num_ht_pages;
>          if (x86_family(eax) < 0x17)
>                  goto out;
>
>          eax = 0x80000000;
>          cpuid(&eax, &ebx, &ecx, &edx);
>          max_ext_leaf = eax;
>
>          /* Use the old, conservative value if MAXPHYADDR isn't enumerated. */
>          if (max_ext_leaf < 0x80000008)
>                  goto out;

Yes, this works for me too.  Though in practice I don't think any 64-bit 
machine ever existed without 0x80000008 (you need it to decide what's a 
canonical address and what isn't), so that would have to be a 32-bit 
fam17h machine.

Paolo

>          /* comment */
>          eax = 0x80000008;
>          cpuid(&eax, &ebx, &ecx, &edx);
>          max_pfn = (1ULL << ((eax & 255) - vm->page_shift)) - 1;
>          if (max_ext_leaf >= 0x8000001f) {
>                  <adjust>
>          }
>          ht_gfn = max_pfn - num_ht_pages;
> out:
>          return min(max_gfn, ht_gfn - 1);
>
> > +             max_pfn = (1ULL << ((eax & 255) - vm->page_shift)) - 1;
> LOL, "& 255", you just couldn't resist, huh?  My version of Rami Code only goes
> up to 15.:-)