Re: [PATCH 3/8] kvm/mmu: rename is_largepage_backed to mapping_level

[Joerg Roedel <joerg.roedel@amd.com>]

On Tue, Jun 23, 2009 at 12:59:33PM -0300, Marcelo Tosatti wrote:
> Hi Joerg,
> 
> On Fri, Jun 19, 2009 at 03:16:24PM +0200, Joerg Roedel wrote:
> >  	gfn = unalias_gfn(kvm, gfn);
> > -	write_count = slot_largepage_idx(gfn,
> > -					 gfn_to_memslot_unaliased(kvm, gfn));
> > -	*write_count += 1;
> > +
> > +	for (i = PT_DIRECTORY_LEVEL;
> > +	     i < PT_PAGE_TABLE_LEVEL + KVM_NR_PAGE_SIZES; ++i) {
> > +		slot          = gfn_to_memslot_unaliased(kvm, gfn);
> 
> Can't you move this call out of the loop?

True. Will do this.

> > @@ -1704,7 +1739,7 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *sptep,
> >  	if ((pte_access & ACC_WRITE_MASK)
> >  	    || (write_fault && !is_write_protection(vcpu) && !user_fault)) {
> >  
> > -		if (largepage && has_wrprotected_page(vcpu->kvm, gfn)) {
> > +		if (largepage && has_wrprotected_page(vcpu->kvm, gfn, 1)) {
> 
> It seems direct_map is missing the large pte overwrite check that
> fetch() contains:
> 
>                 if (is_large_pte(*sptep)) {
>                         rmap_remove(vcpu->kvm, sptep);
>                         __set_spte(sptep, shadow_trap_nonpresent_pte);
>                         kvm_flush_remote_tlbs(vcpu->kvm);
>                 }
> 
> (perhaps its not a possible scenario at the moment, but...).


This function is only called from mmu_set_spte which takes care of this.

Thanks,

	Joerg

-- 
           | Advanced Micro Devices GmbH
 Operating | Karl-Hammerschmidt-Str. 34, 85609 Dornach bei München
 System    | 
 Research  | Geschäftsführer: Thomas M. McCoy, Giuliano Meroni
 Center    | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
           | Registergericht München, HRB Nr. 43632