| From: | Andres Freund <andres(at)anarazel(dot)de> |
|---|---|
| To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
| Cc: | Antonin Houska <ah(at)cybertec(dot)at>, pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Non-volatile variables used for spinlock manipulation |
| Date: | 2015-09-04 14:14:26 |
| Message-ID: | 20150904141426.GA5516@awork2.anarazel.de |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On 2015-09-04 09:42:20 -0400, Tom Lane wrote:
> Antonin Houska <ah(at)cybertec(dot)at> writes:
> > Since SpinLockAcquire() / SpinLockRelease() macros usually reference variables
> > declared as volatile, I wonder if the following changes should be applied.
>
> We've been making changes to remove that requirement, so I think that at
> least in 9.5/HEAD this isn't an issue anymore.
And even before that the pointer passed to SpinLockAcquire/Release()
didn't have to be volatile. The important thing was that read/writes
that are protected by the spinlock were only done via volatile
variables, to prevent them from being moved outside the critical
section. All the tas()/unlock section already force the pointer to the
lock to be volatile anyway.
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Alexander Korotkov | 2015-09-04 14:26:36 | Re: WIP: Rework access method interface |
| Previous Message | Tom Lane | 2015-09-04 13:52:16 | Re: Fwd: Core dump with nested CREATE TEMP TABLE |