Andres Freund <andres@anarazel.de> writes:
> On 2017-08-16 13:40:09 -0400, Tom Lane wrote:
>> I was wondering why the shm_toc code was using BUFFERALIGN and not
>> MAXALIGN, and I now suspect that the answer is "it's an entirely
>> undocumented kluge to make the atomics code not crash on 32-bit
>> machines, so long as nobody puts a pg_atomic_uint64 anywhere except in
>> a shm_toc".
> I don't think there were any atomics in affected code until earlier
> today... And given it didn't work for shm_toc anyway, I'm not quite
> following.
Right, Robert pointed out that it's pre-existing code. My point should
be read as "it's just blind luck that shm_toc is using bigger than
MAXALIGN alignment, or this would never work on 32-bit machines".
>> I'm not sure that that's good enough, and I'm damn sure that it
>> shouldn't be undocumented.
> 8 byte alignment would be good enough, so BUFFERALIGN ought to be
> sufficient. But it'd be nicer to have a separate more descriptive knob.
What I meant by possibly not good enough is that pg_atomic_uint64 used
in other places isn't going to be very safe.
We might be effectively all right as long as we have a coding rule that
pg_atomic_uint64 can only be placed in memory handed out by ShmemAlloc
or shm_toc_allocate, which both have bigger-than-MAXALIGN alignment
practices. But this needs to be documented.
regards, tom lane