On Mon, Jul 25, 2022, David Matlack wrote: > On Mon, Jul 25, 2022 at 4:26 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote: > > The only scenario that jumps to mind is the non-coherent DMA with funky MTRRs > > case. There might be others, but it's been a while since I wrote this... > > > > The MTRRs are per-vCPU (KVM really should just track them as per-VM, but whatever), > > so it's possible that KVM could encounter a fault with a lower fault->req_level > > than a previous fault that set nx_huge_page_disallowed=true (and added the page > > to the possible_nx_huge_pages list because it had a higher req_level). > > But in that case the lower level SP would already have been installed, > so we wouldn't end up calling account_nx_huge_page() and getting to > this point. (account_nx_huge_page() is only called when linking in an > SP.) Hrm, true. I'm 99% certain past me was just maintaining the existing logic in account_huge_nx_page() if (sp->lpage_disallowed) return; Best thing might be to turn that into a WARN as the first patch? > Maybe account_nx_huge_page() needs to be pulled out and called for > every SP on the walk during a fault? Eh, not worth it, the MTRR thing is bogus anyways, e.g. if vCPUs have different MTRR settings and one vCPU allows a huge page but the other does not, KVM will may or may not install a huge page depending on which vCPU faults in the page.
