On Wed, Feb 22, 2023, Anish Moorthy wrote:
> On Wed, Feb 15, 2023 at 9:24 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
> >
> > On Wed, Feb 15, 2023, Anish Moorthy wrote:
> > > + if (mem_fault_nowait) {
> > > + if (fault->pfn == KVM_PFN_ERR_FAULT) {
> > > + vcpu->run->exit_reason = KVM_EXIT_MEMORY_FAULT;
> > > + vcpu->run->memory_fault.gpa = fault->gfn << PAGE_SHIFT;
> > > + vcpu->run->memory_fault.size = PAGE_SIZE;
> >
> > This belongs in a separate patch, and the exit stuff should be filled in by
> > kvm_handle_error_pfn(). Then this if-statement goes away entirely because the
> > "if (!async)" will always evaluate true in the nowait case.
>
> Hi Sean, what exactly did you want "in a separate patch"?
Separate "exit if fast gup() fails", a.k.a. nowait, from "exit with KVM_EXIT_MEMORY_FAULT
instead of -EFAULT on errors". I.e. don't tie KVM_EXIT_MEMORY_FAULT to the fast
gup() capability (or memslot flag?). That way filing vcpu->run->memory_fault
lands in a separate, largely generic patch, and this patch only introduces the
fast gup() logic.
That probably means adding yet another capability, but capabilities are cheap.
