On Mon, Oct 16, 2017 at 09:36 +0000, Mark Kettenis wrote:
> > Date: Mon, 16 Oct 2017 10:52:09 +0200
> > From: Martin Pieuchot <m...@openbsd.org>
> >
> > As pointed by Mateusz Guzik [0], on x86 the cmpxchg{q,l} used by
> > rw_enter(9) and rw_exit(9) already include an implicit memory
> > barrier, so we can avoids using an explicit expensive one by
> > using the following variants.
> >
> > [0] https://marc.info/?l=openbsd-tech&m=150765959923113&w=2
> >
> > ok?
>
> Is this really safe? The atomic instructions are executed
> conditionally...
>
I've been running without membars on amd64 for about 3 weeks now
and haven't noticed any adverse effects. On my laptop and APU2
serving internet for a hundred of people. It appears to me that
access pattern in rwlock.c allows for that: no reads to be
reordered that matter. But please take this with a grain of salt.
A few people noted that a single membar per packet hugely affects
forwarding performance, so a macro-benchmark can be set up to
quantify the cost.
> > Index: kern/kern_rwlock.c
> > ===================================================================
> > RCS file: /cvs/src/sys/kern/kern_rwlock.c,v
> > retrieving revision 1.31
> > diff -u -p -r1.31 kern_rwlock.c
> > --- kern/kern_rwlock.c 12 Oct 2017 09:19:45 -0000 1.31
> > +++ kern/kern_rwlock.c 16 Oct 2017 08:24:27 -0000
> > @@ -96,7 +96,7 @@ _rw_enter_read(struct rwlock *rwl LOCK_F
> > rw_cas(&rwl->rwl_owner, owner, owner + RWLOCK_READ_INCR)))
> > _rw_enter(rwl, RW_READ LOCK_FL_ARGS);
> > else {
> > - membar_enter();
> > + membar_enter_after_atomic();
> > WITNESS_CHECKORDER(&rwl->rwl_lock_obj, LOP_NEWORDER, file, line,
> > NULL);
> > WITNESS_LOCK(&rwl->rwl_lock_obj, 0, file, line);
> > @@ -112,7 +112,7 @@ _rw_enter_write(struct rwlock *rwl LOCK_
> > RW_PROC(p) | RWLOCK_WRLOCK)))
> > _rw_enter(rwl, RW_WRITE LOCK_FL_ARGS);
> > else {
> > - membar_enter();
> > + membar_enter_after_atomic();
> > WITNESS_CHECKORDER(&rwl->rwl_lock_obj,
> > LOP_EXCLUSIVE | LOP_NEWORDER, file, line, NULL);
> > WITNESS_LOCK(&rwl->rwl_lock_obj, LOP_EXCLUSIVE, file, line);
> > @@ -126,7 +126,7 @@ _rw_exit_read(struct rwlock *rwl LOCK_FL
> >
> > rw_assert_rdlock(rwl);
> >
> > - membar_exit();
> > + membar_exit_before_atomic();
> > if (__predict_false((owner & RWLOCK_WAIT) ||
> > rw_cas(&rwl->rwl_owner, owner, owner - RWLOCK_READ_INCR)))
> > _rw_exit(rwl LOCK_FL_ARGS);
> > @@ -141,7 +141,7 @@ _rw_exit_write(struct rwlock *rwl LOCK_F
> >
> > rw_assert_wrlock(rwl);
> >
> > - membar_exit();
> > + membar_exit_before_atomic();
> > if (__predict_false((owner & RWLOCK_WAIT) ||
> > rw_cas(&rwl->rwl_owner, owner, 0)))
> > _rw_exit(rwl LOCK_FL_ARGS);
> > @@ -261,7 +261,7 @@ retry:
> >
> > if (__predict_false(rw_cas(&rwl->rwl_owner, o, o + inc)))
> > goto retry;
> > - membar_enter();
> > + membar_enter_after_atomic();
> >
> > /*
> > * If old lock had RWLOCK_WAIT and RWLOCK_WRLOCK set, it means we
> > @@ -295,7 +295,7 @@ _rw_exit(struct rwlock *rwl LOCK_FL_VARS
> > WITNESS_UNLOCK(&rwl->rwl_lock_obj, wrlock ? LOP_EXCLUSIVE : 0,
> > file, line);
> >
> > - membar_exit();
> > + membar_exit_before_atomic();
> > do {
> > owner = rwl->rwl_owner;
> > if (wrlock)
> >
> >
>
