Re: Broken atomics code on PPC with FreeBSD 10.3

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Andres Freund <andres(at)anarazel(dot)de>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: Broken atomics code on PPC with FreeBSD 10.3
Date: 2016-12-30 15:48:22
Views: Raw Message | Whole Thread | Download mbox | Resend email
Lists: pgsql-hackers

Andres Freund <andres(at)anarazel(dot)de> writes:
> On 2016-12-30 00:44:33 -0500, Tom Lane wrote:
>> Perhaps it could be argued that there's a FreeBSD compiler bug here,
>> but what I'm wondering is why configure believes that this:
>> [char lock = 0;
>> __sync_lock_test_and_set(&lock, 1);
>> __sync_lock_release(&lock);])],
>> is going to draw a hard error on platforms without byte-wide atomics.
>> My sense of C semantics is that the best you could hope for is a
>> warning

> Well, in theory these aren't actual functions but intrinsics with
> special behaviour ("by being overloaded so that they work on multiple
> types."). The compiler is supposed to warn and link to an external
> function if a certain operation is not available:

Yeah, I read that. But configure tests don't normally notice warnings.
Therefore, even if the compiler is behaving fully per spec (and I think
FreeBSD's might not be), we are going to think that char and word-wide
operations are interchangeable even when one is an efficient, in-line
operation and the other is inefficiently emulated by a function or
kernel trap. This doesn't really seem good enough, especially for
standard architectures where we know darn well what ought to happen.

There's a reason why we've not converted s_lock.h to rely on compiler
intrinsics everywhere, and this is exactly it: the quality of the
intrinsics implementations are horridly variable.

> So that assumption made in that configure test doesn't seem that
> unreasonable. What assembler do byte-wide atomics generate on that
> platform? Which compiler/version is this?

$ gcc -v
Using built-in specs.
Target: powerpc-undermydesk-freebsd
Configured with: FreeBSD/powerpc system compiler
Thread model: posix
gcc version 4.2.1 20070831 patched [FreeBSD]

I tried "gcc -Wall -O0 -g -S bogus.c" on this input:

test_tas_char(volatile char *ptr)
return __sync_lock_test_and_set(ptr, 1) == 0;

test_tas_int(volatile int *ptr)
return __sync_lock_test_and_set(ptr, 1) == 0;

and got no warnings and the attached output. I'm not very good at reading
PPC assembler, but I think what is happening in the "char" case is that
gcc is trying to emulate a byte-wide operation using a word-wide one,
ie an lwarx/stwcx. loop. Even if that works (and apparently it does not),
we would not want to use it since it implies contention between adjacent
atomic flags.

BTW, I looked around to see why the PPC critters in the buildfarm aren't
failing, and the answer is that they all have compilers that are too old
to have __sync_lock_test_and_set at all, so that the configure probes
just fail outright. But the probes can't tell the difference between
implementations we want to use and implementations we don't.

regards, tom lane

Attachment Content-Type Size
bogus.s text/x-asm 5.5 KB

In response to


Browse pgsql-hackers by date

  From Date Subject
Next Message Erik Rijkers 2016-12-30 15:50:43 Re: Logical Replication WIP
Previous Message Craig Ringer 2016-12-30 15:04:16 Re: Add doc advice about systemd RemoveIPC