From: Linus Torvalds <torvalds@linux-foundation.org>
To: Chris Snook <csnook@redhat.com>
Cc: David Miller <davem@davemloft.net>,
piggin@cyberone.com.au, satyam@infradead.org,
herbert@gondor.apana.org.au, paulus@samba.org, clameter@sgi.com,
ilpo.jarvinen@helsinki.fi, paulmck@linux.vnet.ibm.com,
stefanr@s5r6.in-berlin.de, linux-kernel@vger.kernel.org,
linux-arch@vger.kernel.org, netdev@vger.kernel.org,
akpm@linux-foundation.org, ak@suse.de, heiko.carstens@de.ibm.com,
schwidefsky@de.ibm.com, wensong@linux-vs.org, horms@verge.net.au,
wjiang@resilience.com, cfriesen@nortel.com, zlynx@acm.org,
rpjday@mindspring.com, jesper.juhl@gmail.com,
segher@kernel.crashing.org
Subject: Re: [PATCH 0/24] make atomic_read() behave consistently across all architectures
Date: Tue, 21 Aug 2007 09:43:52 -0700 (PDT) [thread overview]
Message-ID: <alpine.LFD.0.999.0708210922290.30176@woody.linux-foundation.org> (raw)
In-Reply-To: <46CAED8B.9030006@redhat.com>
On Tue, 21 Aug 2007, Chris Snook wrote:
>
> Moore's law is definitely working against us here. Register counts, pipeline
> depths, core counts, and clock multipliers are all increasing in the long run.
> At some point in the future, barrier() will be universally regarded as a
> hammer too big for most purposes.
Note that "barrier()" is purely a compiler barrier. It has zero impact on
the CPU pipeline itself, and also has zero impact on anything that gcc
knows isn't visible in memory (ie local variables that don't have their
address taken), so barrier() really is pretty cheap.
Now, it's possible that gcc messes up in some circumstances, and that the
memory clobber will cause gcc to also do things like flush local registers
unnecessarily to their stack slots, but quite frankly, if that happens,
it's a gcc problem, and I also have to say that I've not seen that myself.
So in a very real sense, "barrier()" will just make sure that there is a
stronger sequence point for the compiler where things are stable. In most
cases it has absolutely zero performance impact - apart from the
-intended- impact of making sure that the compiler doesn't re-order or
cache stuff around it.
And sure, we could make it more finegrained, and also introduce a
per-variable barrier, but the fact is, people _already_ have problems with
thinking about these kinds of things, and adding new abstraction issues
with subtle semantics is the last thing we want.
So I really think you'd want to show a real example of real code that
actually gets noticeably slower or bigger.
In removing "volatile", we have shown that. It may not have made a big
difference on powerpc, but it makes a real difference on x86 - and more
importantly, it removes something that people clearly don't know how it
works, and incorrectly expect to just fix bugs.
[ There are *other* barriers - the ones that actually add memory barriers
to the CPU - that really can be quite expensive. The good news is that
the expense is going down rather than up: both Intel and AMD are not
only removing the need for some of them (ie "smp_rmb()" will become a
compiler-only barrier), but we're _also_ seeing the whole "pipeline
flush" approach go away, and be replaced by the CPU itself actually
being better - so even the actual CPU pipeline barriers are getting
cheaper, not more expensive. ]
For example, did anybody even _test_ how expensive "barrier()" is? Just
as a lark, I did
#undef barrier
#define barrier() do { } while (0)
in kernel/sched.c (which only has three of them in it, but hey, that's
more than most files), and there were _zero_ code generation downsides.
One instruction was moved (and a few line numbers changed), so it wasn't
like the assembly language was identical, but the point is, barrier()
simply doesn't have the same kinds of downsides that "volatile" has.
(That may not be true on other architectures or in other source files, of
course. This *does* depend on code generation details. But anybody who
thinks that "barrier()" is fundamentally expensive is simply incorrect. It
is *fundamnetally* a no-op).
Linus
next prev parent reply other threads:[~2007-08-21 16:51 UTC|newest]
Thread overview: 311+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-08-09 13:14 [PATCH 0/24] make atomic_read() behave consistently across all architectures Chris Snook
2007-08-09 12:41 ` Arnd Bergmann
2007-08-09 14:29 ` Chris Snook
2007-08-09 15:30 ` Arnd Bergmann
2007-08-14 22:31 ` Christoph Lameter
2007-08-14 22:45 ` Chris Snook
2007-08-14 22:51 ` Christoph Lameter
2007-08-14 23:08 ` Satyam Sharma
2007-08-14 23:04 ` Chris Snook
2007-08-14 23:14 ` Christoph Lameter
2007-08-15 6:49 ` Herbert Xu
2007-08-15 8:18 ` Heiko Carstens
2007-08-15 13:53 ` Stefan Richter
2007-08-15 14:35 ` Satyam Sharma
2007-08-15 14:52 ` Herbert Xu
2007-08-15 16:09 ` Stefan Richter
2007-08-15 16:27 ` Paul E. McKenney
2007-08-15 17:13 ` Satyam Sharma
2007-08-15 18:31 ` Segher Boessenkool
2007-08-15 18:57 ` Paul E. McKenney
2007-08-15 19:54 ` Satyam Sharma
2007-08-15 20:17 ` Paul E. McKenney
2007-08-15 20:52 ` Segher Boessenkool
2007-08-15 22:42 ` Paul E. McKenney
2007-08-15 20:47 ` Segher Boessenkool
2007-08-16 0:36 ` Satyam Sharma
2007-08-16 0:32 ` your mail Herbert Xu
2007-08-16 0:58 ` [PATCH 0/24] make atomic_read() behave consistently across all architectures Satyam Sharma
2007-08-16 0:51 ` Herbert Xu
2007-08-16 1:18 ` Satyam Sharma
2007-08-16 1:38 ` Segher Boessenkool
2007-08-15 21:05 ` [PATCH 0/24] make atomic_read() behave consistently across all architectures Segher Boessenkool
2007-08-15 22:44 ` Paul E. McKenney
2007-08-16 1:23 ` Segher Boessenkool
2007-08-16 2:22 ` Paul E. McKenney
2007-08-15 19:58 ` Stefan Richter
2007-08-16 0:39 ` [PATCH] i386: Fix a couple busy loops in mach_wakecpu.h:wait_for_init_deassert() Satyam Sharma
2007-08-24 11:59 ` Denys Vlasenko
2007-08-24 12:07 ` Andi Kleen
2007-08-24 12:12 ` Kenn Humborg
2007-08-24 14:25 ` Denys Vlasenko
2007-08-24 17:34 ` Linus Torvalds
2007-08-24 13:30 ` Satyam Sharma
2007-08-24 17:06 ` Christoph Lameter
2007-08-24 20:26 ` Denys Vlasenko
2007-08-24 20:34 ` Chris Snook
2007-08-24 16:19 ` Luck, Tony
2007-08-15 16:13 ` [PATCH 0/24] make atomic_read() behave consistently across all architectures Chris Snook
2007-08-15 23:40 ` Herbert Xu
2007-08-15 23:51 ` Paul E. McKenney
2007-08-16 1:30 ` Segher Boessenkool
2007-08-16 2:30 ` Paul E. McKenney
2007-08-16 19:33 ` Segher Boessenkool
2007-08-16 1:26 ` Segher Boessenkool
2007-08-16 2:23 ` Nick Piggin
2007-08-16 19:32 ` Segher Boessenkool
2007-08-17 2:19 ` Nick Piggin
2007-08-17 3:16 ` Paul Mackerras
2007-08-17 3:32 ` Nick Piggin
2007-08-17 3:50 ` Linus Torvalds
2007-08-17 23:59 ` Paul E. McKenney
2007-08-18 0:09 ` Herbert Xu
2007-08-18 1:08 ` Paul E. McKenney
2007-08-18 1:24 ` Christoph Lameter
2007-08-18 1:41 ` Satyam Sharma
2007-08-18 4:13 ` Linus Torvalds
2007-08-18 13:36 ` Satyam Sharma
2007-08-18 21:54 ` Paul E. McKenney
2007-08-18 22:41 ` Linus Torvalds
2007-08-18 23:19 ` Paul E. McKenney
2007-08-24 12:19 ` Denys Vlasenko
2007-08-24 17:19 ` Linus Torvalds
2007-08-18 21:56 ` Paul E. McKenney
2007-08-20 13:31 ` Chris Snook
2007-08-20 22:04 ` Segher Boessenkool
2007-08-20 22:48 ` Russell King
2007-08-20 23:02 ` Segher Boessenkool
2007-08-21 0:05 ` Paul E. McKenney
2007-08-21 7:08 ` Russell King
2007-08-21 7:05 ` Russell King
2007-08-21 9:33 ` Paul Mackerras
2007-08-21 11:37 ` Andi Kleen
2007-08-21 14:48 ` Segher Boessenkool
2007-08-21 16:16 ` Paul E. McKenney
2007-08-21 22:51 ` Valdis.Kletnieks
2007-08-22 0:50 ` Paul E. McKenney
2007-08-22 21:38 ` Adrian Bunk
2007-08-21 14:39 ` Segher Boessenkool
2007-08-17 3:42 ` Linus Torvalds
2007-08-17 5:18 ` Paul E. McKenney
2007-08-17 5:56 ` Satyam Sharma
2007-08-17 7:26 ` Nick Piggin
2007-08-17 8:47 ` Satyam Sharma
2007-08-17 9:15 ` Nick Piggin
2007-08-17 10:12 ` Satyam Sharma
2007-08-17 12:14 ` Nick Piggin
2007-08-17 13:05 ` Satyam Sharma
2007-08-17 9:48 ` Paul Mackerras
2007-08-17 10:23 ` Satyam Sharma
2007-08-17 22:49 ` Segher Boessenkool
2007-08-17 23:51 ` Satyam Sharma
2007-08-17 23:55 ` Segher Boessenkool
2007-08-17 6:42 ` Geert Uytterhoeven
2007-08-17 8:52 ` Andi Kleen
2007-08-17 10:08 ` Satyam Sharma
2007-08-17 22:29 ` Segher Boessenkool
2007-08-17 17:37 ` Segher Boessenkool
2007-08-14 23:26 ` Paul E. McKenney
2007-08-15 10:35 ` Stefan Richter
2007-08-15 12:04 ` Herbert Xu
2007-08-15 12:31 ` Satyam Sharma
2007-08-15 13:08 ` Stefan Richter
2007-08-15 13:11 ` Stefan Richter
2007-08-15 13:47 ` Satyam Sharma
2007-08-15 14:25 ` Paul E. McKenney
2007-08-15 15:33 ` Herbert Xu
2007-08-15 16:08 ` Paul E. McKenney
2007-08-15 17:18 ` Satyam Sharma
2007-08-15 17:33 ` Paul E. McKenney
2007-08-15 18:05 ` Satyam Sharma
2007-08-15 17:55 ` Satyam Sharma
2007-08-15 19:07 ` Paul E. McKenney
2007-08-15 21:07 ` Segher Boessenkool
2007-08-15 20:58 ` Segher Boessenkool
2007-08-15 18:19 ` David Howells
2007-08-15 18:45 ` Paul E. McKenney
2007-08-15 23:41 ` Herbert Xu
2007-08-15 23:53 ` Paul E. McKenney
2007-08-16 0:12 ` Herbert Xu
2007-08-16 0:23 ` Paul E. McKenney
2007-08-16 0:30 ` Herbert Xu
2007-08-16 0:49 ` Paul E. McKenney
2007-08-16 0:53 ` Herbert Xu
2007-08-16 1:14 ` Paul E. McKenney
2007-08-15 18:31 ` Segher Boessenkool
2007-08-15 19:40 ` Satyam Sharma
2007-08-15 20:42 ` Segher Boessenkool
2007-08-16 1:23 ` Satyam Sharma
2007-08-15 23:22 ` Paul Mackerras
2007-08-16 0:26 ` Christoph Lameter
2007-08-16 0:34 ` Paul Mackerras
2007-08-16 0:40 ` Christoph Lameter
2007-08-16 0:39 ` Paul E. McKenney
2007-08-16 0:42 ` Christoph Lameter
2007-08-16 0:53 ` Paul E. McKenney
2007-08-16 0:59 ` Christoph Lameter
2007-08-16 1:14 ` Paul E. McKenney
2007-08-16 1:41 ` Christoph Lameter
2007-08-16 2:15 ` Satyam Sharma
2007-08-16 2:08 ` Herbert Xu
2007-08-16 2:18 ` Christoph Lameter
2007-08-16 3:23 ` Paul Mackerras
2007-08-16 3:33 ` Herbert Xu
2007-08-16 3:48 ` Paul Mackerras
2007-08-16 4:03 ` Herbert Xu
2007-08-16 4:34 ` Paul Mackerras
2007-08-16 5:37 ` Herbert Xu
2007-08-16 6:00 ` Paul Mackerras
2007-08-16 18:50 ` Christoph Lameter
2007-08-16 18:48 ` Christoph Lameter
2007-08-16 19:44 ` Segher Boessenkool
2007-08-16 2:18 ` Chris Friesen
2007-08-16 2:32 ` Paul E. McKenney
2007-08-16 1:51 ` Paul Mackerras
2007-08-16 2:00 ` Herbert Xu
2007-08-16 2:05 ` Paul Mackerras
2007-08-16 2:11 ` Herbert Xu
2007-08-16 2:35 ` Paul E. McKenney
2007-08-16 3:15 ` Paul Mackerras
2007-08-16 3:43 ` Herbert Xu
2007-08-16 2:15 ` Christoph Lameter
2007-08-16 2:17 ` Christoph Lameter
2007-08-16 2:33 ` Satyam Sharma
2007-08-16 3:01 ` Satyam Sharma
2007-08-16 4:11 ` Paul Mackerras
2007-08-16 5:39 ` Herbert Xu
2007-08-16 6:56 ` Paul Mackerras
2007-08-16 7:09 ` Herbert Xu
2007-08-16 8:06 ` Stefan Richter
2007-08-16 8:10 ` Herbert Xu
2007-08-16 9:54 ` Stefan Richter
2007-08-16 10:31 ` Stefan Richter
2007-08-16 10:42 ` Herbert Xu
2007-08-16 16:34 ` Paul E. McKenney
2007-08-16 23:59 ` Herbert Xu
2007-08-17 1:01 ` Paul E. McKenney
2007-08-17 7:39 ` Satyam Sharma
2007-08-17 14:31 ` Paul E. McKenney
2007-08-17 18:31 ` Satyam Sharma
2007-08-17 18:56 ` Paul E. McKenney
2007-08-17 3:15 ` Nick Piggin
2007-08-17 4:02 ` Paul Mackerras
2007-08-17 4:39 ` Nick Piggin
2007-08-17 7:25 ` Stefan Richter
2007-08-17 8:06 ` Nick Piggin
2007-08-17 8:58 ` Satyam Sharma
2007-08-17 9:15 ` Nick Piggin
2007-08-17 10:03 ` Satyam Sharma
2007-08-17 11:50 ` Nick Piggin
2007-08-17 12:50 ` Satyam Sharma
2007-08-17 12:56 ` Nick Piggin
2007-08-18 2:15 ` Satyam Sharma
2007-08-17 10:48 ` Stefan Richter
2007-08-17 10:58 ` Stefan Richter
2007-08-18 14:35 ` LDD3 pitfalls (was Re: [PATCH 0/24] make atomic_read() behave consistently across all architectures) Stefan Richter
2007-08-20 13:28 ` Chris Snook
2007-08-17 22:14 ` [PATCH 0/24] make atomic_read() behave consistently across all architectures Segher Boessenkool
2007-08-17 5:04 ` Paul Mackerras
2007-08-16 10:35 ` Herbert Xu
2007-08-16 19:48 ` Chris Snook
2007-08-17 0:02 ` Herbert Xu
2007-08-17 2:04 ` Chris Snook
2007-08-17 2:13 ` Herbert Xu
2007-08-17 2:31 ` Nick Piggin
2007-08-17 5:09 ` Paul Mackerras
2007-08-17 5:32 ` Herbert Xu
2007-08-17 5:41 ` Paul Mackerras
2007-08-17 8:28 ` Satyam Sharma
2007-08-16 14:48 ` Ilpo Järvinen
2007-08-16 16:19 ` Stefan Richter
2007-08-16 19:55 ` Chris Snook
2007-08-16 20:20 ` Christoph Lameter
2007-08-17 1:02 ` Paul E. McKenney
2007-08-17 1:28 ` Herbert Xu
2007-08-17 5:07 ` Paul E. McKenney
2007-08-17 2:16 ` Paul Mackerras
2007-08-17 3:03 ` Linus Torvalds
2007-08-17 3:43 ` Paul Mackerras
2007-08-17 3:53 ` Herbert Xu
2007-08-17 6:26 ` Satyam Sharma
2007-08-17 8:38 ` Nick Piggin
2007-08-17 9:14 ` Satyam Sharma
2007-08-17 9:31 ` Nick Piggin
2007-08-17 10:55 ` Satyam Sharma
2007-08-17 12:39 ` Nick Piggin
2007-08-17 13:36 ` Satyam Sharma
2007-08-17 16:48 ` Linus Torvalds
2007-08-17 18:50 ` Chris Friesen
2007-08-17 18:54 ` Arjan van de Ven
2007-08-17 19:49 ` Paul E. McKenney
2007-08-17 19:49 ` Arjan van de Ven
2007-08-17 20:12 ` Paul E. McKenney
2007-08-17 19:08 ` Linus Torvalds
2007-08-20 13:15 ` Chris Snook
2007-08-20 13:32 ` Herbert Xu
2007-08-20 13:38 ` Chris Snook
2007-08-20 22:07 ` Segher Boessenkool
2007-08-21 5:46 ` Linus Torvalds
2007-08-21 7:04 ` David Miller
2007-08-21 13:50 ` Chris Snook
2007-08-21 14:59 ` Segher Boessenkool
2007-08-21 16:31 ` Satyam Sharma
2007-08-21 16:43 ` Linus Torvalds [this message]
2007-09-09 18:02 ` Denys Vlasenko
2007-09-09 18:18 ` Arjan van de Ven
2007-09-10 10:56 ` Denys Vlasenko
2007-09-10 11:15 ` Herbert Xu
2007-09-10 12:22 ` Kyle Moffett
2007-09-10 13:38 ` Denys Vlasenko
2007-09-10 14:16 ` Denys Vlasenko
2007-09-10 15:09 ` Linus Torvalds
2007-09-10 16:46 ` Denys Vlasenko
2007-09-10 19:59 ` Kyle Moffett
2007-09-10 18:59 ` Christoph Lameter
2007-09-10 23:19 ` [PATCH] Document non-semantics of atomic_read() and atomic_set() Chris Snook
2007-09-10 23:44 ` Paul E. McKenney
2007-09-11 19:35 ` Christoph Lameter
2007-09-10 14:51 ` [PATCH 0/24] make atomic_read() behave consistently across all architectures Arjan van de Ven
2007-09-10 14:38 ` Denys Vlasenko
2007-09-10 17:02 ` Arjan van de Ven
2007-08-17 11:08 ` Stefan Richter
2007-08-17 22:09 ` Segher Boessenkool
2007-08-17 17:41 ` Segher Boessenkool
2007-08-17 18:38 ` Satyam Sharma
2007-08-17 23:17 ` Segher Boessenkool
2007-08-17 23:55 ` Satyam Sharma
2007-08-18 0:04 ` Segher Boessenkool
2007-08-18 1:56 ` Satyam Sharma
2007-08-18 2:15 ` Segher Boessenkool
2007-08-18 3:33 ` Satyam Sharma
2007-08-18 5:18 ` Segher Boessenkool
2007-08-18 13:20 ` Satyam Sharma
2007-09-10 18:59 ` Christoph Lameter
2007-09-10 20:54 ` Paul E. McKenney
2007-09-10 21:36 ` Christoph Lameter
2007-09-10 21:50 ` Paul E. McKenney
2007-09-11 2:27 ` Segher Boessenkool
2007-08-16 21:08 ` Luck, Tony
2007-08-16 19:55 ` Chris Snook
2007-08-16 18:54 ` Christoph Lameter
2007-08-16 20:07 ` Paul E. McKenney
2007-08-16 3:05 ` Paul Mackerras
2007-08-16 19:39 ` Segher Boessenkool
2007-08-16 2:07 ` Segher Boessenkool
2007-08-24 12:50 ` Denys Vlasenko
2007-08-24 17:15 ` Christoph Lameter
2007-08-24 20:21 ` Denys Vlasenko
2007-08-16 3:37 ` Bill Fink
2007-08-16 5:20 ` Satyam Sharma
2007-08-16 5:57 ` Satyam Sharma
2007-08-16 9:25 ` Satyam Sharma
2007-08-16 21:00 ` Segher Boessenkool
2007-08-17 4:32 ` Satyam Sharma
2007-08-17 22:38 ` Segher Boessenkool
2007-08-18 14:42 ` Satyam Sharma
2007-08-16 20:50 ` Segher Boessenkool
2007-08-16 22:40 ` David Schwartz
2007-08-17 4:36 ` Satyam Sharma
2007-08-17 4:24 ` Satyam Sharma
2007-08-17 22:34 ` Segher Boessenkool
2007-08-15 19:59 ` Christoph Lameter
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=alpine.LFD.0.999.0708210922290.30176@woody.linux-foundation.org \
--to=torvalds@linux-foundation.org \
--cc=ak@suse.de \
--cc=akpm@linux-foundation.org \
--cc=cfriesen@nortel.com \
--cc=clameter@sgi.com \
--cc=csnook@redhat.com \
--cc=davem@davemloft.net \
--cc=heiko.carstens@de.ibm.com \
--cc=herbert@gondor.apana.org.au \
--cc=horms@verge.net.au \
--cc=ilpo.jarvinen@helsinki.fi \
--cc=jesper.juhl@gmail.com \
--cc=linux-arch@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=netdev@vger.kernel.org \
--cc=paulmck@linux.vnet.ibm.com \
--cc=paulus@samba.org \
--cc=piggin@cyberone.com.au \
--cc=rpjday@mindspring.com \
--cc=satyam@infradead.org \
--cc=schwidefsky@de.ibm.com \
--cc=segher@kernel.crashing.org \
--cc=stefanr@s5r6.in-berlin.de \
--cc=wensong@linux-vs.org \
--cc=wjiang@resilience.com \
--cc=zlynx@acm.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).