On Fri, Aug 25, 2023 at 10:47:12AM -0700, Sean Christopherson <seanjc@xxxxxxxxxx> wrote: > kvm_gmem_error_page() > --------------------- > As pointed out by Vishal[*], guest_memfd()'s error/poison handling is garbage. > KVM needs to unmap, check for poison, and probably also restrict the allowed > mapping size if a partial page is poisoned. > > This item also needs actually testing, e.g. via error injection. Writing a > proper selftest may not be feasible, but at a bare minimum, someone needs to > manually verify an error on a guest_memfd() can get routed all the way into the > guest, e.g. as an #MC on x86. > > This needs an owner. I'm guessing 2-3 weeks? Though I tend to be overly > optimistic when sizing these things... > > [*] https://lore.kernel.org/all/CAGtprH9a2jX-hdww9GPuMrO9noNeXkoqE8oejtVn2vD0AZa3zA@xxxxxxxxxxxxxx I'll look into it. I suppose we can utilize fault injection(Linux kernel or ACPI Error Injection Table). Anyway we will see. -- Isaku Yamahata <isaku.yamahata@xxxxxxxxx>