From: Willy Tarreau <w@1wt.eu>
Date: Sun, 17 Mar 2024 15:55:09 +0000 (+0100)
Subject: MINOR: ring: it's not x86 but all non-ARMv8.1 which needs the read before OR
X-Git-Tag: v3.0-dev6~7
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=39df8c903d65303542a56d0be15f8c6039a75cdd;p=thirdparty%2Fhaproxy.git

MINOR: ring: it's not x86 but all non-ARMv8.1 which needs the read before OR

Archs relying on CAS benefit from a read prior to FETCH_OR, so it's
not just x86 that benefits from this. Let's just change the condition
to only exclude __ARM_FEATURE_ATOMICS which is the only one faster
without.
---

diff --git a/src/ring.c b/src/ring.c
index 1dc4c06eea..7998f09050 100644
--- a/src/ring.c
+++ b/src/ring.c
@@ -280,10 +280,12 @@ ssize_t ring_write(struct ring *ring, size_t maxlen, const struct ist pfx[], siz
 			goto wait_for_flush;
 		__ha_cpu_relax_for_read();
 
-#if defined(__x86_64__)
-		/* x86 prefers a read first */
-		if ((tail_ofs = HA_ATOMIC_LOAD(tail_ptr)) & RING_TAIL_LOCK)
+#if !defined(__ARM_FEATURE_ATOMICS)
+		/* ARMv8.1-a has a true atomic OR and doesn't need the preliminary read */
+		if ((tail_ofs = HA_ATOMIC_LOAD(tail_ptr)) & RING_TAIL_LOCK) {
+			__ha_cpu_relax_for_read();
 			continue;
+		}
 #endif
 		/* OK the queue is locked, let's attempt to get the tail lock */
 		tail_ofs = HA_ATOMIC_FETCH_OR(tail_ptr, RING_TAIL_LOCK);