From: will.deacon@arm.com (Will Deacon)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH 13/18] arm64: cmpxchg: avoid memory barrier on comparison failure
Date: Mon, 13 Jul 2015 12:22:27 +0100 [thread overview]
Message-ID: <20150713112226.GC2632@arm.com> (raw)
In-Reply-To: <20150713102848.GX19282@twins.programming.kicks-ass.net>
On Mon, Jul 13, 2015 at 11:28:48AM +0100, Peter Zijlstra wrote:
> On Mon, Jul 13, 2015 at 10:25:14AM +0100, Will Deacon wrote:
> > cmpxchg doesn't require memory barrier semantics when the value
> > comparison fails, so make the barrier conditional on success.
>
> So this isn't actually a documented condition.
There's *something* in Documentation/atomic_ops.txt (literally, the last
sentence in the file) but it's terrible at best:
"Note that this also means that for the case where the counter
is not dropping to zero, there are no memory ordering
requirements."
[you probably want to see the example for some context]
> I would very much like a fwe extra words on this, that you've indeed
> audited cmpxchg() users and preferably even a Documentation/ update to
> match.
Happy to update the docs. In terms of code audit, I couldn't find any
cmpxchg users that do something along the lines of "if the comparison
fails, don't loop, and instead do something to an independent address,
without barrier semantics that must be observed after the failed CAS":
- Most (as in, it's hard to find other cases) users just loop until
success, so there's no issue there.
- One use-case with work on the failure path is stats update (e.g.
drivers/net/ethernet/intel/ixgbe/ixgbe.h), but barrier semantics
aren't required here anyway.
- Another use-case is where you optimistically try a cmpxchg, then
fall back on a lock if you fail (e.g. slub and cmpxchg_double).
- Some other archs appear to do the same trick (alpha and powerpc).
So I'm confident with this change, but agree that a Docs update would
be beneficial. Something like below, or do you want some additional text,
too?
Will
--->8
diff --git a/Documentation/memory-barriers.txt b/Documentation/memory-barriers.txt
index 13feb697271f..18fc860df1be 100644
--- a/Documentation/memory-barriers.txt
+++ b/Documentation/memory-barriers.txt
@@ -2383,9 +2383,7 @@ about the state (old or new) implies an SMP-conditional general memory barrier
explicit lock operations, described later). These include:
xchg();
- cmpxchg();
atomic_xchg(); atomic_long_xchg();
- atomic_cmpxchg(); atomic_long_cmpxchg();
atomic_inc_return(); atomic_long_inc_return();
atomic_dec_return(); atomic_long_dec_return();
atomic_add_return(); atomic_long_add_return();
@@ -2398,7 +2396,9 @@ explicit lock operations, described later). These include:
test_and_clear_bit();
test_and_change_bit();
- /* when succeeds (returns 1) */
+ /* when succeeds */
+ cmpxchg();
+ atomic_cmpxchg(); atomic_long_cmpxchg();
atomic_add_unless(); atomic_long_add_unless();
These are used for such things as implementing ACQUIRE-class and RELEASE-class
next prev parent reply other threads:[~2015-07-13 11:22 UTC|newest]
Thread overview: 35+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-07-13 9:25 [PATCH 00/18] arm64: support for 8.1 LSE atomic instructions Will Deacon
2015-07-13 9:25 ` [PATCH 01/18] arm64: cpufeature.h: add missing #include of kernel.h Will Deacon
2015-07-13 9:25 ` [PATCH 02/18] arm64: atomics: move ll/sc atomics into separate header file Will Deacon
2015-07-13 9:25 ` [PATCH 03/18] arm64: elf: advertise 8.1 atomic instructions as new hwcap Will Deacon
2015-07-17 13:48 ` Catalin Marinas
2015-07-17 13:57 ` Russell King - ARM Linux
2015-07-13 9:25 ` [PATCH 04/18] arm64: alternatives: add cpu feature for lse atomics Will Deacon
2015-07-13 9:25 ` [PATCH 05/18] arm64: introduce CONFIG_ARM64_LSE_ATOMICS as fallback to ll/sc atomics Will Deacon
2015-07-17 16:32 ` Catalin Marinas
2015-07-17 17:25 ` Will Deacon
2015-07-13 9:25 ` [PATCH 06/18] arm64: atomics: patch in lse instructions when supported by the CPU Will Deacon
2015-07-13 9:25 ` [PATCH 07/18] arm64: locks: " Will Deacon
2015-07-21 16:53 ` Catalin Marinas
2015-07-21 17:29 ` Will Deacon
2015-07-23 13:39 ` Will Deacon
2015-07-23 14:14 ` Catalin Marinas
2015-07-13 9:25 ` [PATCH 08/18] arm64: bitops: " Will Deacon
2015-07-13 9:25 ` [PATCH 09/18] arm64: xchg: " Will Deacon
2015-07-13 9:25 ` [PATCH 10/18] arm64: cmpxchg: " Will Deacon
2015-07-13 9:25 ` [PATCH 11/18] arm64: cmpxchg_dbl: " Will Deacon
2015-07-13 9:25 ` [PATCH 12/18] arm64: cmpxchg: avoid "cc" clobber in ll/sc routines Will Deacon
2015-07-21 17:16 ` Catalin Marinas
2015-07-21 17:32 ` Will Deacon
2015-07-13 9:25 ` [PATCH 13/18] arm64: cmpxchg: avoid memory barrier on comparison failure Will Deacon
2015-07-13 10:28 ` Peter Zijlstra
2015-07-13 11:22 ` Will Deacon [this message]
2015-07-13 13:39 ` Peter Zijlstra
2015-07-13 14:52 ` Will Deacon
2015-07-13 15:32 ` Peter Zijlstra
2015-07-13 15:58 ` Will Deacon
2015-07-13 9:25 ` [PATCH 14/18] arm64: atomics: tidy up common atomic{,64}_* macros Will Deacon
2015-07-13 9:25 ` [PATCH 15/18] arm64: atomics: prefetch the destination word for write prior to stxr Will Deacon
2015-07-13 9:25 ` [PATCH 16/18] arm64: atomics: implement atomic{, 64}_cmpxchg using cmpxchg Will Deacon
2015-07-13 9:25 ` [PATCH 17/18] arm64: atomic64_dec_if_positive: fix incorrect branch condition Will Deacon
2015-07-13 9:25 ` [PATCH 18/18] arm64: kconfig: select HAVE_CMPXCHG_LOCAL Will Deacon
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20150713112226.GC2632@arm.com \
--to=will.deacon@arm.com \
--cc=linux-arm-kernel@lists.infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).