On 2014-06-26 14:13:07 -0700, Tom Lane wrote:
> Andres Freund <andres@2ndquadrant.com> writes:
> > I think we should rework things so that we fall back to
> > pg_write_barrier(), (*((volatile slock_t *) (lock)) = 0) instead of what
> > we have right now.
>
> Surely it had better be a read barrier as well?
I don't immediately see why it has to be read barrier? Hoisting a load
from after the release into the locked area of code should be safe? Note
that 'bad' reads can only happen for variables which aren't protected by
the spinlock since the S_LOCK needs to have acquire semantics and no
other process can modify protected variables concurrently.
The important thing is that all modifications that have been done inside
the spinlock are visible to other backends and that no writes are moved
outside the protected are.
> And S_LOCK the same?
It better be a read barrier, yes. I haven't checked yet, but I assume
that pretty much all TAS/tas implementation already guarantee that. I
think if not we'd seen problems. Well, at least on platforms that
receive testing under concurrent circumstances :/
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services