Re: [PATCH 3/3] KVM: MMU: consider host cache type in MMIO pfn check

[Haozhong Zhang <haozhong.zhang@intel.com>]

On 10/27/17 10:40 +0200, Ingo Molnar wrote:
> 
> * Haozhong Zhang <haozhong.zhang@intel.com> wrote:
> 
> > By default, KVM treats a reserved page as for MMIO purpose, and maps
> > it to guest with UC memory type.  However, some reserved pages are not
> > for MMIO, such as pages of DAX device (e.g., /dev/daxX.Y). Mapping
> > them with UC memory type will harm the performance.  In order to
> > exclude those cases, we check the host cache mode in addition and only
> > treat UC/UC- pages as MMIO.
> > 
> > Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com>
> > Reported-by: Cuevas Escareno, Ivan D <ivan.d.cuevas.escareno@intel.com>
> > Reported-by: Kumar, Karthik <karthik.kumar@intel.com>
> > ---
> >  arch/x86/kvm/mmu.c | 32 +++++++++++++++++++++++++++++---
> >  1 file changed, 29 insertions(+), 3 deletions(-)
> > 
> > diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> > index 0b481cc9c725..d4c821a6df3d 100644
> > --- a/arch/x86/kvm/mmu.c
> > +++ b/arch/x86/kvm/mmu.c
> > @@ -2707,10 +2707,36 @@ static bool mmu_need_write_protect(struct kvm_vcpu *vcpu, gfn_t gfn,
> >  
> >  static bool kvm_is_mmio_pfn(kvm_pfn_t pfn)
> >  {
> > -	if (pfn_valid(pfn))
> > -		return !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn));
> > +	bool is_mmio = true;
> >  
> > -	return true;
> > +	if (pfn_valid(pfn)) {
> > +		is_mmio = !is_zero_pfn(pfn) && PageReserved(pfn_to_page(pfn));
> > +
> > +		/*
> > +		 * By default, KVM treats a reserved page as for MMIO
> > +		 * purpose, and maps it to guest with UC memory type.
> > +		 * However, some reserved pages are not for MMIO, such
> > +		 * as pages of DAX device (e.g., /dev/daxX.Y). Mapping
> > +		 * them with UC memory type will harm the performance.
> > +		 * In order to exclude those cases, we check the host
> > +		 * cache mode in addition and only treat UC/UC- pages
> > +		 * as MMIO.
> > +		 *
> > +		 * track_pfn_insert() works only when PAT is enabled,
> > +		 * so add pat_enabled() here.
> > +		 */
> > +		if (is_mmio && pat_enabled()) {
> > +			pgprot_t prot;
> > +			enum page_cache_mode cm;
> > +
> > +			track_pfn_insert(NULL, &prot, kvm_pfn_to_pfn(pfn));
> > +			cm = pgprot2cachemode(prot);
> > +			is_mmio = (cm == _PAGE_CACHE_MODE_UC ||
> > +				   cm == _PAGE_CACHE_MODE_UC_MINUS);
> > +		}
> > +	}
> > +
> > +	return is_mmio;
> >  }
> >  
> >  static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
> 
> s/harm the performance
>  /harm performance> 
> 
> But I suspect the rest of the comment should be rewritten too to be more fluid.

I'll refactor the comment.

> 
> Beyond that - I think instead of exposing these low level details a properly named 
> helper function should be put into pat.c instead - and KVM can use that.
>

KVM only needs the memory type information, so lookup_memtype() in
pat.c is probably a better one to be exposed.

Haozhong