From 573bbbe127608d36e4050fcb19069dc64ba4b2fb Mon Sep 17 00:00:00 2001 From: Willy Tarreau Date: Sun, 17 Mar 2024 10:20:56 +0100 Subject: [PATCH] MEDIUM: ring: improve speed in the queue waiting loop on x86_64 x86_64 doesn't have a native atomic FETCH_OR(), it's implemented using a CAS, which will always cause a write cycle. Here we know we can just wait as long as the lock bit is held so better loop on a load, and only attempt the CAS on success. This requires a tiny ifdef and brings nice benefits. This brings the performance back from 3.33M to 3.75M at 24C48T while doing no change at 3C6T. --- src/ring.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/src/ring.c b/src/ring.c index 74772314a..0393a269b 100644 --- a/src/ring.c +++ b/src/ring.c @@ -281,11 +281,15 @@ ssize_t ring_write(struct ring *ring, size_t maxlen, const struct ist pfx[], siz if (next_cell != &cell) goto wait_for_flush; // FIXME: another thread arrived, we should go to wait now __ha_cpu_relax_for_read(); - - tail_ofs = HA_ATOMIC_FETCH_OR(tail_ptr, RING_TAIL_LOCK); - if (!(tail_ofs & RING_TAIL_LOCK)) - break; - +#if defined(__x86_64__) + /* x86 prefers a read first */ + if (!(HA_ATOMIC_LOAD(tail_ptr) & RING_TAIL_LOCK)) +#endif + { + tail_ofs = HA_ATOMIC_FETCH_OR(tail_ptr, RING_TAIL_LOCK); + if (!(tail_ofs & RING_TAIL_LOCK)) + break; + } __ha_cpu_relax_for_read(); } /* OK the queue is locked, let's attempt to get the tail lock */