From: Marc Zyngier marc.zyngier@arm.com
When triggering a CoW, we unmap the RO page via an MMU notifier (invalidate_range_start), and then populate the new PTE using another one (change_pte). In the meantime, we'll have copied the old page into the new one.
The problem is that the data for the new page is sitting in the cache, and should the guest have an uncached mapping to that page (or its MMU off), following accesses will bypass the cache.
In a way, this is similar to what happens on a translation fault: We need to clean the page to the PoC before mapping it. So let's just do that.
This fixes a KVM unit test regression observed on a HiSilicon platform, and subsequently reproduced on Seattle.
Fixes: a9c0e12ebee5 ("KVM: arm/arm64: Only clean the dcache on translation fault") Cc: stable@vger.kernel.org # v4.16+ Reported-by: Mike Galbraith efault@gmx.de Signed-off-by: Marc Zyngier marc.zyngier@arm.com Signed-off-by: Christoffer Dall christoffer.dall@arm.com --- virt/kvm/arm/mmu.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c index 91aaf73b00df..111a660be3be 100644 --- a/virt/kvm/arm/mmu.c +++ b/virt/kvm/arm/mmu.c @@ -1860,13 +1860,20 @@ static int kvm_set_spte_handler(struct kvm *kvm, gpa_t gpa, u64 size, void *data void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte) { unsigned long end = hva + PAGE_SIZE; + kvm_pfn_t pfn = pte_pfn(pte); pte_t stage2_pte;
if (!kvm->arch.pgd) return;
trace_kvm_set_spte_hva(hva); - stage2_pte = pfn_pte(pte_pfn(pte), PAGE_S2); + + /* + * We've moved a page around, probably through CoW, so let's treat it + * just like a translation fault and clean the cache to the PoC. + */ + clean_dcache_guest_page(pfn, PAGE_SIZE); + stage2_pte = pfn_pte(pfn, PAGE_S2); handle_hva_to_gpa(kvm, hva, end, &kvm_set_spte_handler, &stage2_pte); }