Similar to the previous patch, the atomic xchg in sock_error() results in
extra pipeline flushes due to the need to perform full synchronization as
a memory barrier. Avoid calling sock_error() in a couple of places where
it is safe to do a racy version of the test. Combined with the refcounting
patch, this improves LMbench's bw_unix results from ~346K/s to ~570K/s on
my test system.
-ben
Signed-off-by: Benjamin LaHaise <[EMAIL PROTECTED]>
diff --git a/net/core/datagram.c b/net/core/datagram.c
index 1bcfef5..1c4af6f 100644
--- a/net/core/datagram.c
+++ b/net/core/datagram.c
@@ -149,11 +149,15 @@ struct sk_buff *skb_recv_datagram(struct
long timeo;
/*
* Caller is allowed not to check sk->sk_err before skb_recv_datagram()
+ * because of this, we allow ourselves to do the fast (racy) test first.
*/
- int error = sock_error(sk);
+ int error;
- if (error)
- goto no_packet;
+ if (sk->sk_err) {
+ error = sock_error(sk);
+ if (error)
+ goto no_packet;
+ }
timeo = sock_rcvtimeo(sk, noblock);
diff --git a/net/core/sock.c b/net/core/sock.c
index 13cc3be..463576b 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -950,9 +950,14 @@ static struct sk_buff *sock_alloc_send_p
timeo = sock_sndtimeo(sk, noblock);
while (1) {
- err = sock_error(sk);
- if (err != 0)
- goto failure;
+ /* Avoid atomic xchg operation -- the pending error will
+ * be tested again, so the racy check is okay.
+ */
+ if (sk->sk_err) {
+ err = sock_error(sk);
+ if (err != 0)
+ goto failure;
+ }
err = -EPIPE;
if (sk->sk_shutdown & SEND_SHUTDOWN)
--
"You know, I've seen some crystals do some pretty trippy shit, man."
Don't Email: <[EMAIL PROTECTED]>.
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
