[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250630145034.8JnINEaS@linutronix.de>
Date: Mon, 30 Jun 2025 16:50:34 +0200
From: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Chris Mason <clm@...a.com>, linux-kernel@...r.kernel.org,
Thomas Gleixner <tglx@...utronix.de>
Subject: [PATCH] futex: Temporary disable FUTEX_PRIVATE_HASH
Chris Mason reported a performance regression on big iron. Reports of
this kind were usually reported as part of a micro benchmark but Chris'
test did mimic his real workload. This makes it a real regression.
The root cause is rcuref_get() which is invoked during each futex
operation. If all threads of an application do this simultaneously then
it leads to cache line bouncing and the performance drops.
Disable FUTEX_PRIVATE_HASH entirely for this cycle. The performance
regression will be addressed in the following cycle enabling the option
again.
Reported-by: Chris Mason <clm@...a.com>
Closes: https://lore.kernel.org/all/3ad05298-351e-4d61-9972-ca45a0a50e33@meta.com/
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
---
init/Kconfig | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/init/Kconfig b/init/Kconfig
index af4c2f0854554..666783eb50abd 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1716,9 +1716,13 @@ config FUTEX_PI
depends on FUTEX && RT_MUTEXES
default y
+#
+# marked broken for performance reasons; gives us one more cycle to sort things out.
+#
config FUTEX_PRIVATE_HASH
bool
depends on FUTEX && !BASE_SMALL && MMU
+ depends on BROKEN
default y
config FUTEX_MPOL
--
2.50.0
Powered by blists - more mailing lists