From: mikeb <mikeb@openbsd.org>
Date: Fri, 5 Aug 2016 18:31:21 +0000 (+0000)
Subject: Switch pending event clearing to an atomic swap operation
X-Git-Url: http://artulab.com/gitweb/?a=commitdiff_plain;h=30494642db82acd378f52a5034c9b671ff98f9f2;p=openbsd

Switch pending event clearing to an atomic swap operation

Rather than performing an atomic bit clearing for every encountered
event bit set we can adjust the code to perform an atomic swap of a
single row of the events array and decrease the amount of expensive
atomic operations.

Same optimization as for Hyper-V.  From FreeBSD.
---

diff --git a/sys/dev/pv/xen.c b/sys/dev/pv/xen.c
index 2ec7c4c7230..f03d7459300 100644
--- a/sys/dev/pv/xen.c
+++ b/sys/dev/pv/xen.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: xen.c,v 1.60 2016/08/03 17:14:41 mikeb Exp $	*/
+/*	$OpenBSD: xen.c,v 1.61 2016/08/05 18:31:21 mikeb Exp $	*/
 
 /*
  * Copyright (c) 2015 Mike Belopuhov
@@ -628,13 +628,14 @@ xen_intr(void)
 	for (row = 0; selector > 0; selector >>= 1, row++) {
 		if ((selector & 1) == 0)
 			continue;
-		pending = sc->sc_ipg->evtchn_pending[row] &
-		    ~(sc->sc_ipg->evtchn_mask[row]);
+		if ((sc->sc_ipg->evtchn_pending[row] &
+		    ~(sc->sc_ipg->evtchn_mask[row])) == 0)
+			continue;
+		pending = atomic_swap_ulong(&sc->sc_ipg->evtchn_pending[row],
+		    0) & ~(sc->sc_ipg->evtchn_mask[row]);
 		for (bit = 0; pending > 0; pending >>= 1, bit++) {
 			if ((pending & 1) == 0)
 				continue;
-			atomic_clearbit_ptr(&sc->sc_ipg->evtchn_pending[row],
-			    bit);
 			port = (row * LONG_BIT) + bit;
 			if ((xi = xen_lookup_intsrc(sc, port)) == NULL) {
 				printf("%s: unhandled interrupt on port %u\n",