diff --git a/[refs] b/[refs]
index 963bfd39b5d9..6c9e0e37a835 100644
--- a/[refs]
+++ b/[refs]
@@ -1,2 +1,2 @@
 ---
-refs/heads/master: 5319c662522db8995ff9276ba9d80549c64b294a
+refs/heads/master: ecc5589f19a52e7e6501fe449047b19087ae11bb
diff --git a/trunk/arch/x86/kvm/mmu.c b/trunk/arch/x86/kvm/mmu.c
index fa3486d64078..dd20b199a7c0 100644
--- a/trunk/arch/x86/kvm/mmu.c
+++ b/trunk/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);