On Thu, 2023-02-09 at 17:55 +0800, Chao Gao wrote: > On Thu, Feb 09, 2023 at 10:40:15AM +0800, Robert Hoo wrote: > > --- a/arch/x86/kvm/mmu/mmu.c > > +++ b/arch/x86/kvm/mmu/mmu.c > > @@ -3698,8 +3698,11 @@ static int mmu_alloc_shadow_roots(struct > > kvm_vcpu *vcpu) > > gfn_t root_gfn, root_pgd; > > int quadrant, i, r; > > hpa_t root; > > - > > The blank line should be kept. OK > > > +#ifdef CONFIG_X86_64 > > + root_pgd = mmu->get_guest_pgd(vcpu) & ~(X86_CR3_LAM_U48 | > > X86_CR3_LAM_U57); > > +#else > > root_pgd = mmu->get_guest_pgd(vcpu); > > +#endif > > Why are other call sites of mmu->get_guest_pgd() not changed? Emm, the other 3 are FNAME(walk_addr_generic)() kvm_arch_setup_async_pf() kvm_arch_async_page_ready In former version, I clear CR3.LAM bits for guest_pgd inside mmu- >get_guest_pgd(). I think this is generic. Perhaps I should still do it in that way. Let's wait for other's comments on this. Thanks for pointing out. > And what's > the value of the #ifdef? LAM is only available in 64 bit mode. > > > root_gfn = root_pgd >> PAGE_SHIFT; > > > > if (mmu_check_root(vcpu, root_gfn)) > > -- > > 2.31.1 > >
