Updating the event index has a memory barrier and causes more work
on the other side to actually signal the event. It is unnecessary
if a new buffer has already appeared on the ring, so poll once before
doing the update.
The effect of this on the 0.9 ring implementation is pretty much
invisible, but on the new-style ring it provides a consistent 3%
performance improvement.
Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
---
tools/virtio/ringtest/main.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/tools/virtio/ringtest/main.c b/tools/virtio/ringtest/main.c
index ac1269f87394..19543799c621 100644
--- a/tools/virtio/ringtest/main.c
+++ b/tools/virtio/ringtest/main.c
@@ -147,7 +147,7 @@ static void __attribute__((__flatten__)) run_guest(void)
assert(completed <= bufs);
assert(started <= bufs);
if (do_sleep) {
- if (enable_call())
+ if (used_empty() && enable_call())
wait_for_call();
} else {
poll_used();
@@ -172,7 +172,7 @@ static void __attribute__((__flatten__)) run_host(void)
for (;;) {
if (do_sleep) {
- if (enable_kick())
+ if (avail_empty() && enable_kick())
wait_for_kick();
} else {
poll_avail();
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
