From: Peter Sewell <Peter.Sewell@cl.cam.ac.uk>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>,
Torvald Riegel <triegel@redhat.com>,
Will Deacon <will.deacon@arm.com>,
Peter Zijlstra <peterz@infradead.org>,
Ramana Radhakrishnan <Ramana.Radhakrishnan@arm.com>,
David Howells <dhowells@redhat.com>,
"linux-arch@vger.kernel.org" <linux-arch@vger.kernel.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"akpm@linux-foundation.org" <akpm@linux-foundation.org>,
"mingo@kernel.org" <mingo@kernel.org>,
"gcc@gcc.gnu.org" <gcc@gcc.gnu.org>,
Mark Batty <mbatty@cantab.net>
Subject: Re: [RFC][PATCH 0/5] arch: atomic rework
Date: Fri, 21 Feb 2014 19:48:47 +0000 [thread overview]
Message-ID: <CAHWkzRRr5kY==3pZQ1Opb-Vy+2VCCUSaMJQz7Sd=5C--Y=FsxQ@mail.gmail.com> (raw)
In-Reply-To: <CA+55aFz4Lhnx9XzMt_X8SOqmoUdiuW_93SAdV58UiEe3g278CQ@mail.gmail.com>
On 21 February 2014 19:41, Linus Torvalds <torvalds@linux-foundation.org> wrote:
> On Fri, Feb 21, 2014 at 11:16 AM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
>>
>> Why would this be any different, especially since it's easy to
>> understand both for a human and a compiler?
>
> Btw, the actual data path may actually be semantically meaningful even
> at a processor level.
>
> For example, let's look at that gcc bugzilla that got mentioned
> earlier, and let's assume that gcc is fixed to follow the "arithmetic
> is always meaningful, even if it is only syntactic" the letter.
> So we have that gcc bugzilla use-case:
>
> flag ? *(q + flag - flag) : 0;
>
> and let's say that the fixed compiler now generates the code with the
> data dependency that is actually suggested in that bugzilla entry:
>
> and w2, w2, #0
> ldr w0, [x1, w2]
>
> ie the CPU actually sees that address data dependency. Now everything
> is fine, right?
>
> Wrong.
>
> It is actually quite possible that the CPU sees the "and with zero"
> and *breaks the dependencies on the incoming value*.
For reference: the Power and ARM architectures explicitly guarantee
not to do this, the architects are quite clear about it, and we've
tested (some cases) rather thoroughly.
I can't speak about other architectures.
> Modern CPU's literally do things like that. Seriously. Maybe not that
> particular one, but you'll sometimes find that the CPU - int he
> instruction decoding phase (ie very early in the pipeline) notices
> certain patterns that generate constants, and actually drop the data
> dependency on the "incoming" registers.
>
> On x86, generating zero using "xor" on the register with itself is one
> such known sequence.
>
> Can you guarantee that powerpc doesn't do the same for "and r,r,#0"?
> Or what if the compiler generated the much more obvious
>
> sub w2,w2,w2
>
> for that "+flag-flag"? Are you really 100% sure that the CPU won't
> notice that that is just a way to generate a zero, and doesn't depend
> on the incoming values?
>
> Because I'm not. I know CPU designers that do exactly this.
>
> So I would actually and seriously argue that the whole C standard
> attempt to use a syntactic data dependency as a determination of
> whether two things are serialized is wrong, and that you actually
> *want* to have the compiler optimize away false data dependencies.
>
> Because people playing tricks with "+flag-flag" and thinking that that
> somehow generates a data dependency - that's *wrong*. It's not just
> the compiler that decides "that's obviously nonsense, I'll optimize it
> away". The CPU itself can do it.
>
> So my "actual semantic dependency" model is seriously more likely to
> be *correct*. Not just t a compiler level.
>
> Btw, any tricks like that, I would also take a second look at the
> assembler and the linker. Many assemblers do some trivial
> optimizations too.
That's certainly something worth checking.
> Are you sure that "and w2, w2, #0" really ends
> up being encoded as an "and"? Maybe the assembler says "I can do that
> as a "mov w2,#0" instead? Who knows? Even power and ARM have their
> variable-sized encodings (there are some "compressed executable"
> embedded power processors, and there is obviously Thumb2, and many
> assemblers end up trying to use equivalent "small" instructions..
>
> So the whole "fake data dependency" thing is just dangerous on so many levels.
>
> MUCH more dangerous than my "actual real dependency" model.
>
> Linus
next prev parent reply other threads:[~2014-02-21 19:48 UTC|newest]
Thread overview: 299+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-02-06 13:48 [RFC][PATCH 0/5] arch: atomic rework Peter Zijlstra
2014-02-06 13:48 ` [RFC][PATCH 1/5] ia64: Fix up smp_mb__{before,after}_clear_bit Peter Zijlstra
2014-02-06 13:48 ` [RFC][PATCH 2/5] arc,hexagon: Delete asm/barrier.h Peter Zijlstra
2014-02-06 13:48 ` [RFC][PATCH 3/5] arch: s/smp_mb__(before|after)_(atomic|clear)_(dec,inc,bit)/smp_mb__\1/g Peter Zijlstra
2014-02-06 19:12 ` Paul E. McKenney
2014-02-07 9:52 ` Will Deacon
2014-02-06 13:48 ` [RFC][PATCH 4/5] arch: Generic atomic.h cleanup Peter Zijlstra
2014-02-06 17:49 ` Will Deacon
2014-02-06 13:48 ` [RFC][PATCH 5/5] arch: Sanitize atomic_t bitwise ops Peter Zijlstra
2014-02-06 14:43 ` Geert Uytterhoeven
2014-02-06 16:14 ` Peter Zijlstra
2014-02-06 16:53 ` Linus Torvalds
2014-02-06 17:52 ` Peter Zijlstra
2014-02-06 17:56 ` Linus Torvalds
2014-02-06 18:09 ` Peter Zijlstra
2014-02-06 18:25 ` [RFC][PATCH 0/5] arch: atomic rework David Howells
2014-02-06 18:30 ` Peter Zijlstra
2014-02-06 18:42 ` Paul E. McKenney
2014-02-06 18:55 ` Ramana Radhakrishnan
2014-02-06 18:59 ` Will Deacon
2014-02-06 19:27 ` Paul E. McKenney
2014-02-06 21:17 ` Torvald Riegel
2014-02-06 22:11 ` Paul E. McKenney
2014-02-06 23:44 ` Torvald Riegel
2014-02-07 4:20 ` Paul E. McKenney
2014-02-07 7:44 ` Peter Zijlstra
2014-02-07 16:50 ` Paul E. McKenney
2014-02-07 16:55 ` Will Deacon
2014-02-07 17:06 ` Peter Zijlstra
2014-02-07 17:13 ` Will Deacon
2014-02-07 17:20 ` Peter Zijlstra
2014-02-07 18:03 ` Paul E. McKenney
2014-02-07 17:46 ` Joseph S. Myers
2014-02-07 18:43 ` Torvald Riegel
2014-02-07 18:02 ` Paul E. McKenney
2014-02-10 0:27 ` Torvald Riegel
2014-02-10 0:56 ` Linus Torvalds
2014-02-10 1:16 ` Torvald Riegel
2014-02-10 1:24 ` Linus Torvalds
2014-02-10 1:46 ` Torvald Riegel
2014-02-10 2:04 ` Linus Torvalds
2014-02-10 3:21 ` Paul E. McKenney
2014-02-10 3:45 ` Paul E. McKenney
2014-02-10 11:46 ` Peter Zijlstra
2014-02-10 19:09 ` Linus Torvalds
2014-02-11 15:59 ` Paul E. McKenney
2014-02-12 6:06 ` Torvald Riegel
2014-02-12 9:19 ` Peter Zijlstra
2014-02-12 17:42 ` Paul E. McKenney
2014-02-12 18:12 ` Peter Zijlstra
2014-02-17 18:18 ` Paul E. McKenney
2014-02-17 20:39 ` Richard Biener
2014-02-17 22:14 ` Paul E. McKenney
2014-02-17 22:27 ` Torvald Riegel
2014-02-14 5:07 ` Torvald Riegel
2014-02-14 9:50 ` Peter Zijlstra
2014-02-14 19:19 ` Torvald Riegel
2014-02-12 17:39 ` Paul E. McKenney
2014-02-12 5:39 ` Torvald Riegel
2014-02-12 18:07 ` Paul E. McKenney
2014-02-12 20:22 ` Linus Torvalds
2014-02-13 0:23 ` Paul E. McKenney
2014-02-13 20:03 ` Torvald Riegel
2014-02-14 2:01 ` Paul E. McKenney
2014-02-14 4:43 ` Torvald Riegel
2014-02-14 17:29 ` Paul E. McKenney
2014-02-14 19:21 ` Torvald Riegel
2014-02-14 19:50 ` Linus Torvalds
2014-02-14 20:02 ` Linus Torvalds
2014-02-15 2:08 ` Paul E. McKenney
2014-02-15 2:44 ` Linus Torvalds
2014-02-15 2:48 ` Linus Torvalds
2014-02-15 6:35 ` Paul E. McKenney
2014-02-15 6:58 ` Paul E. McKenney
2014-02-15 18:07 ` Torvald Riegel
2014-02-17 18:59 ` Joseph S. Myers
2014-02-17 19:19 ` Will Deacon
2014-02-17 19:41 ` Torvald Riegel
2014-02-17 23:12 ` Joseph S. Myers
2014-02-15 17:45 ` Torvald Riegel
2014-02-15 18:49 ` Linus Torvalds
2014-02-17 19:55 ` Torvald Riegel
2014-02-17 20:18 ` Linus Torvalds
2014-02-17 21:21 ` Torvald Riegel
2014-02-17 22:02 ` Linus Torvalds
2014-02-17 22:25 ` Torvald Riegel
2014-02-17 22:47 ` Linus Torvalds
2014-02-17 23:41 ` Torvald Riegel
2014-02-18 0:18 ` Linus Torvalds
2014-02-18 1:26 ` Paul E. McKenney
2014-02-18 15:38 ` Torvald Riegel
2014-02-18 16:55 ` Paul E. McKenney
2014-02-18 19:57 ` Torvald Riegel
2014-02-17 23:10 ` Alec Teal
2014-02-18 0:05 ` Linus Torvalds
2014-02-18 15:31 ` Torvald Riegel
2014-02-18 16:49 ` Linus Torvalds
2014-02-18 17:16 ` Paul E. McKenney
2014-02-18 18:23 ` Peter Sewell
2014-02-18 19:00 ` Linus Torvalds
2014-02-18 19:42 ` Paul E. McKenney
2014-02-18 21:40 ` Torvald Riegel
2014-02-18 21:52 ` Peter Zijlstra
2014-02-19 9:52 ` Torvald Riegel
2014-02-18 22:58 ` Paul E. McKenney
2014-02-19 10:59 ` Torvald Riegel
2014-02-19 15:14 ` Paul E. McKenney
2014-02-19 17:55 ` Torvald Riegel
2014-02-19 22:12 ` Paul E. McKenney
2014-02-18 21:21 ` Torvald Riegel
2014-02-18 21:40 ` Peter Zijlstra
2014-02-18 21:47 ` Torvald Riegel
2014-02-19 15:23 ` David Lang
2014-02-19 18:11 ` Torvald Riegel
2014-02-18 21:47 ` Peter Zijlstra
2014-02-19 11:07 ` Torvald Riegel
2014-02-19 11:42 ` Peter Zijlstra
2014-02-18 22:14 ` Linus Torvalds
2014-02-19 14:40 ` Torvald Riegel
2014-02-19 19:49 ` Linus Torvalds
2014-02-18 3:00 ` Paul E. McKenney
2014-02-18 3:24 ` Linus Torvalds
2014-02-18 3:42 ` Linus Torvalds
2014-02-18 5:22 ` Paul E. McKenney
2014-02-18 16:17 ` Torvald Riegel
2014-02-18 17:44 ` Linus Torvalds
2014-02-18 19:40 ` Paul E. McKenney
2014-02-18 19:47 ` Torvald Riegel
2014-02-20 0:53 ` Linus Torvalds
2014-02-20 4:01 ` Paul E. McKenney
2014-02-20 4:43 ` Linus Torvalds
2014-02-20 8:30 ` Paul E. McKenney
2014-02-20 9:20 ` Paul E. McKenney
2014-02-20 17:01 ` Linus Torvalds
2014-02-20 18:11 ` Paul E. McKenney
2014-02-20 18:32 ` Linus Torvalds
2014-02-20 18:53 ` Torvald Riegel
2014-02-20 19:09 ` Linus Torvalds
2014-02-22 18:53 ` Torvald Riegel
2014-02-22 21:53 ` Linus Torvalds
2014-02-23 0:39 ` Paul E. McKenney
2014-02-23 3:50 ` Linus Torvalds
2014-02-23 6:34 ` Paul E. McKenney
2014-02-23 19:31 ` Linus Torvalds
2014-02-24 1:16 ` Paul E. McKenney
2014-02-24 1:35 ` Linus Torvalds
2014-02-24 4:59 ` Paul E. McKenney
2014-02-24 5:25 ` Linus Torvalds
2014-02-24 15:57 ` Linus Torvalds
2014-02-24 16:27 ` Richard Biener
2014-02-24 16:37 ` Linus Torvalds
2014-02-24 16:40 ` Linus Torvalds
2014-02-24 16:55 ` Michael Matz
2014-02-24 17:28 ` Paul E. McKenney
2014-02-24 17:57 ` Paul E. McKenney
2014-02-26 17:39 ` Torvald Riegel
2014-02-24 17:38 ` Linus Torvalds
2014-02-24 18:12 ` Paul E. McKenney
2014-02-26 17:34 ` Torvald Riegel
2014-02-24 17:21 ` Paul E. McKenney
2014-02-24 18:14 ` Linus Torvalds
2014-02-24 18:53 ` Paul E. McKenney
2014-02-24 19:54 ` Linus Torvalds
2014-02-24 22:37 ` Paul E. McKenney
2014-02-24 23:35 ` Linus Torvalds
2014-02-25 6:00 ` Paul E. McKenney
2014-02-26 1:47 ` Linus Torvalds
2014-02-26 5:12 ` Paul E. McKenney
2014-02-25 6:05 ` Linus Torvalds
2014-02-26 0:15 ` Paul E. McKenney
2014-02-26 3:32 ` Jeff Law
2014-02-26 5:23 ` Paul E. McKenney
2014-02-27 15:37 ` Torvald Riegel
2014-02-27 17:01 ` Linus Torvalds
2014-02-27 19:06 ` Paul E. McKenney
2014-02-27 19:47 ` Linus Torvalds
2014-02-27 20:53 ` Paul E. McKenney
2014-03-01 0:50 ` Paul E. McKenney
2014-03-01 10:06 ` Peter Sewell
2014-03-01 14:03 ` Paul E. McKenney
2014-03-02 10:05 ` Peter Sewell
2014-03-02 23:20 ` Paul E. McKenney
2014-03-02 23:44 ` Peter Sewell
2014-03-03 4:25 ` Paul E. McKenney
2014-03-03 20:44 ` Torvald Riegel
2014-03-04 22:11 ` Peter Sewell
2014-03-05 17:15 ` Torvald Riegel
2014-03-05 18:37 ` Peter Sewell
2014-03-03 18:55 ` Torvald Riegel
2014-03-03 19:20 ` Paul E. McKenney
2014-03-03 20:46 ` Torvald Riegel
2014-03-04 19:00 ` Paul E. McKenney
2014-03-04 21:35 ` Paul E. McKenney
2014-03-05 16:54 ` Torvald Riegel
2014-03-05 18:15 ` Paul E. McKenney
2014-03-07 18:33 ` Torvald Riegel
2014-03-07 19:11 ` Paul E. McKenney
2014-03-05 16:26 ` Torvald Riegel
2014-03-05 18:01 ` Paul E. McKenney
2014-03-07 17:45 ` Torvald Riegel
2014-03-07 19:02 ` Paul E. McKenney
2014-03-03 18:59 ` Torvald Riegel
2014-03-03 15:36 ` Torvald Riegel
2014-02-27 17:50 ` Paul E. McKenney
2014-02-27 19:22 ` Paul E. McKenney
2014-02-28 1:02 ` Paul E. McKenney
2014-03-03 19:01 ` Torvald Riegel
2014-02-20 18:56 ` Paul E. McKenney
2014-02-20 19:45 ` Linus Torvalds
2014-02-20 22:10 ` Paul E. McKenney
2014-02-20 22:52 ` Linus Torvalds
2014-02-21 18:35 ` Michael Matz
2014-02-21 19:13 ` Paul E. McKenney
2014-02-21 22:10 ` Joseph S. Myers
2014-02-21 22:37 ` Paul E. McKenney
2014-02-26 13:09 ` Torvald Riegel
2014-02-26 18:43 ` Joseph S. Myers
2014-02-27 0:52 ` Torvald Riegel
2014-02-24 13:55 ` Michael Matz
2014-02-24 17:40 ` Paul E. McKenney
2014-02-26 13:04 ` Torvald Riegel
2014-02-26 18:27 ` Paul E. McKenney
2014-02-20 18:44 ` Torvald Riegel
2014-02-20 18:56 ` Paul E. McKenney
2014-02-20 18:23 ` Torvald Riegel
[not found] ` <CAHWkzRQZ8+gOGMFNyTKjFNzpUv6d_J1G9KL0x_iCa=YCgvEojQ@mail.gmail.com>
2014-02-21 19:16 ` Linus Torvalds
2014-02-21 19:41 ` Linus Torvalds
2014-02-21 19:48 ` Peter Sewell [this message]
[not found] ` <CAHWkzRSO82jU-9dtTEjHaW2FeLcEqdZXxp5Q8cmVTTT9uhZQYw@mail.gmail.com>
2014-02-21 20:22 ` Linus Torvalds
[not found] ` <CAHWkzRRxqhH+DnuQHu9bM4ywGBen3oqtT8W4Xqt1CFAHy2WQRg@mail.gmail.com>
2014-02-21 19:24 ` Paul E. McKenney
2014-02-20 17:54 ` Torvald Riegel
2014-02-20 18:11 ` Paul E. McKenney
2014-02-20 17:49 ` Torvald Riegel
2014-02-20 18:25 ` Linus Torvalds
2014-02-20 19:02 ` Linus Torvalds
2014-02-20 19:06 ` Linus Torvalds
2014-02-20 17:26 ` Torvald Riegel
2014-02-20 18:18 ` Paul E. McKenney
2014-02-22 18:30 ` Torvald Riegel
2014-02-22 20:17 ` Paul E. McKenney
2014-02-20 17:14 ` Torvald Riegel
2014-02-20 17:34 ` Linus Torvalds
2014-02-20 18:12 ` Torvald Riegel
2014-02-20 18:26 ` Paul E. McKenney
2014-02-18 5:01 ` Paul E. McKenney
2014-02-18 15:56 ` Torvald Riegel
2014-02-18 16:51 ` Paul E. McKenney
2014-02-17 20:23 ` Paul E. McKenney
2014-02-17 21:05 ` Torvald Riegel
2014-02-15 17:30 ` Torvald Riegel
2014-02-15 19:15 ` Linus Torvalds
2014-02-17 22:09 ` Torvald Riegel
2014-02-17 22:32 ` Linus Torvalds
2014-02-17 23:17 ` Torvald Riegel
2014-02-18 0:09 ` Linus Torvalds
2014-02-18 15:46 ` Torvald Riegel
2014-02-10 11:48 ` Peter Zijlstra
2014-02-10 11:49 ` Will Deacon
2014-02-10 12:05 ` Peter Zijlstra
2014-02-10 15:04 ` Paul E. McKenney
2014-02-10 16:22 ` Will Deacon
2014-02-07 18:44 ` Torvald Riegel
2014-02-10 0:06 ` Torvald Riegel
2014-02-10 3:51 ` Paul E. McKenney
2014-02-12 5:13 ` Torvald Riegel
2014-02-12 18:26 ` Paul E. McKenney
2014-02-06 21:09 ` Torvald Riegel
2014-02-06 21:55 ` Paul E. McKenney
2014-02-06 22:58 ` Torvald Riegel
2014-02-07 4:06 ` Paul E. McKenney
2014-02-07 9:13 ` Torvald Riegel
2014-02-07 16:44 ` Paul E. McKenney
2014-02-06 22:13 ` Joseph S. Myers
2014-02-06 23:25 ` Torvald Riegel
2014-02-06 23:33 ` Joseph S. Myers
2014-02-07 12:01 ` Will Deacon
2014-02-07 16:47 ` Paul E. McKenney
2014-02-06 19:21 ` Linus Torvalds
[not found] ` <52F93B7C.2090304@tilera.com>
[not found] ` <20140210205719.GY5002@laptop.programming.kicks-ass.net>
2014-02-10 21:08 ` Chris Metcalf
2014-02-10 21:14 ` Peter Zijlstra
2014-02-18 12:12 Peter Sewell
2014-02-18 12:53 ` Peter Zijlstra
2014-02-18 16:08 ` Peter Sewell
2014-02-18 14:56 ` Paul E. McKenney
2014-02-18 15:16 ` Mark Batty
2014-02-18 17:17 ` Paul E. McKenney
2014-02-18 15:33 ` Peter Sewell
2014-02-18 16:47 ` Paul E. McKenney
2014-02-18 17:38 ` Linus Torvalds
2014-02-18 18:21 ` Peter Sewell
2014-02-18 18:49 ` Linus Torvalds
2014-02-18 19:47 ` Paul E. McKenney
2014-02-18 20:46 ` Torvald Riegel
2014-02-18 20:43 ` Torvald Riegel
2014-02-18 21:29 ` Paul E. McKenney
2014-02-18 23:48 ` Peter Sewell
2014-02-19 9:46 ` Torvald Riegel
2014-02-26 3:06 George Spelvin
2014-02-26 5:22 ` Paul E. McKenney
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='CAHWkzRRr5kY==3pZQ1Opb-Vy+2VCCUSaMJQz7Sd=5C--Y=FsxQ@mail.gmail.com' \
--to=peter.sewell@cl.cam.ac.uk \
--cc=Ramana.Radhakrishnan@arm.com \
--cc=akpm@linux-foundation.org \
--cc=dhowells@redhat.com \
--cc=gcc@gcc.gnu.org \
--cc=linux-arch@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=mbatty@cantab.net \
--cc=mingo@kernel.org \
--cc=paulmck@linux.vnet.ibm.com \
--cc=peterz@infradead.org \
--cc=torvalds@linux-foundation.org \
--cc=triegel@redhat.com \
--cc=will.deacon@arm.com \
/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).