On Fri, Dec 01, 2017 at 04:15:37PM +0100, Paolo Bonzini wrote: > On 30/11/2017 19:05, Radim Krčmář wrote: > > Does roughly what kvm_mmu_notifier_invalidate_page did before. > > > > I am not certain why this would be needed. It might mean that we have > > another bug with start/end or just that I missed something. > > I don't think this is needed, because we don't have shared page tables. > My understanding is that without shared page tables, you can assume that > all page modifications go through invalidate_range_start/end. With > shared page tables, there are additional TLB flushes to take care of, > which require invalidate_range. Agreed, invalidate_range only is ever needed if you the secondary MMU (i.e. KVM) shares the same pagetables of the primary MMU in the host. Only in such case we need a special secondary MMU invalidate in the tlb gather before the page is freed because there's no way to block the secondary MMU from walking the host pagetables in invalidate_range_start. In KVM case the secondary MMU always go through the shadow pagetables, so all shadow pagetable invalidates can happen in invalidate_range_start and patch 2/2 is not needed here. Note that the host kernel could have always decided to call invalidate_range_start/end and never to call invalidate_page even before invalidate_page was removed. So the problem in practice could only be noticed after the removal of invalidate_page of course, but in more theoretical terms 1/2 is actually fixing a longstanding bug. The removal of invalidate_page made the lack of kvm_arch_mmu_notifier_invalidate_page call in invalidate_range_start more apparent. Thanks, Andrea -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
