KVM: MMU: micro-optimize gpte_access

Avoid AND-NOT, most x86 processor lack an instruction for it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

arch/x86/kvm/paging_tmpl.h

@@ -189,8 +189,11 @@ static inline unsigned FNAME(gpte_access)(struct kvm_vcpu *vcpu, u64 gpte)
 		((gpte & VMX_EPT_EXECUTABLE_MASK) ? ACC_EXEC_MASK : 0) |
 		ACC_USER_MASK;
 #else
-	access = (gpte & (PT_WRITABLE_MASK | PT_USER_MASK)) | ACC_EXEC_MASK;
-	access &= ~(gpte >> PT64_NX_SHIFT);
+	BUILD_BUG_ON(ACC_EXEC_MASK != PT_PRESENT_MASK);
+	BUILD_BUG_ON(ACC_EXEC_MASK != 1);
+	access = gpte & (PT_WRITABLE_MASK | PT_USER_MASK | PT_PRESENT_MASK);
+	/* Combine NX with P (which is set here) to get ACC_EXEC_MASK.  */
+	access ^= (gpte >> PT64_NX_SHIFT);
 #endif
 
 	return access;