Horton

Re: global barrier & atomics in signal handlers (Re: Atomic operations within spinlocks)

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Robert Haas <robertmhaas@gmail.com>
Cc: Andres Freund <andres@anarazel.de>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2020-06-18T16:29:40Z
Lists: pgsql-hackers 
Robert Haas <robertmhaas@gmail.com> writes:
> On Thu, Jun 18, 2020 at 11:59 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Sure, but wouldn't making the SpinLockAcquire layer into static inlines be
>> sufficient to address that point, with no need to touch s_lock.h at all?

> I mean, wouldn't you then end up with a bunch of 1-line functions
> where you can step into the function but not through whatever
> individual things it does?

Not following your point.  The s_lock.h implementations tend to be either
simple C statements ("*lock = 0") or asm blocks; if you feel a need to
step through them you're going to be resorting to "si" anyway.

I think the main usefulness of doing anything here would be (a) separating
the spinlock infrastructure from callers and (b) ensuring that we have a
declared argument type, and single-evaluation semantics, for the spinlock
function parameters.  Both of those are adequately addressed by fixing
spin.h, IMO anyway.

			regards, tom lane

Commits

  1. Convert SpinLock* macros to static inline functions.

  2. Clean up includes of s_lock.h.

  3. Fix deadlock danger when atomic ops are done under spinlock.

  4. Add basic spinlock tests to regression tests.

  5. spinlock emulation: Fix bug when more than INT_MAX spinlocks are initialized.

  6. Avoid potential spinlock in a signal handler as part of global barriers.

  7. Make pg_stat_wal_receiver consistent with the WAL receiver's shmem info

  8. Improve 64bit atomics support.