KVM: MMU: optimize set_spte for page sync

[safe/jmp/linux-2.6] / arch / x86 / kvm / mmu.c

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index fa3486d..dd20b19 100644 (file)
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -1593,6 +1593,15 @@ static int set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
 
                spte |= PT_WRITABLE_MASK;
 
+               /*
+                * Optimization: for pte sync, if spte was writable the hash
+                * lookup is unnecessary (and expensive). Write protection
+                * is responsibility of mmu_get_page / kvm_sync_page.
+                * Same reasoning can be applied to dirty page accounting.
+                */
+               if (!can_unsync && is_writeble_pte(*shadow_pte))
+                       goto set_pte;
+
                if (mmu_need_write_protect(vcpu, gfn, can_unsync)) {
                        pgprintk("%s: found shadow page for %lx, marking ro\n",
                                 __func__, gfn);