Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 27 Jun 2006 09:31:02 -0700
From:      "Kip Macy" <kip.macy@gmail.com>
To:        "John Baldwin" <jhb@freebsd.org>
Cc:        kris@freebsd.org, Perforce Change Reviews <perforce@freebsd.org>, Kip Macy <kmacy@freebsd.org>
Subject:   Re: PERFORCE change 100121 for review
Message-ID:  <b1fa29170606270931p6fbbbf98vcfcad8dc50987024@mail.gmail.com>
In-Reply-To: <200606271127.13228.jhb@freebsd.org>
References:  <200606270742.k5R7gCfT001310@repoman.freebsd.org> <200606271127.13228.jhb@freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help
Point me at your tree. I do :-).

                     -Kip


On 6/27/06, John Baldwin <jhb@freebsd.org> wrote:
> On Tuesday 27 June 2006 03:42, Kip Macy wrote:
> > http://perforce.freebsd.org/chv.cgi?CH=100121
> >
> > Change 100121 by kmacy@kmacy_storage:sun4v_work_sleepq on 2006/06/27
> 07:41:14
> >
> >       avoid pointless turnstile contention if mutex owner is running
>
> Yes, I did this already in jhb_lock for both mutexes and rwlocks at BSDCan.
> Kris said my patch panic'd for him at BSDCan but I don't have an SMP machine
> I can test it on now to debug it. :(
>
> >
> > Affected files ...
> >
> > .. //depot/projects/kmacy_sun4v/src/sys/kern/kern_mutex.c#20 edit
> >
> > Differences ...
> >
> > ==== //depot/projects/kmacy_sun4v/src/sys/kern/kern_mutex.c#20 (text+ko)
> ====
> >
> > @@ -176,7 +176,7 @@
> >  static int mutex_prof_maxrecords = MPROF_HASH_SIZE;
> >  SYSCTL_INT(_debug_mutex_prof, OID_AUTO, maxrecords, CTLFLAG_RD,
> >      &mutex_prof_maxrecords, 0, "Maximum number of profiling records");
> > -int mutex_prof_rejected;
> > +int mutex_prof_rejected = 0;
> >  SYSCTL_INT(_debug_mutex_prof, OID_AUTO, rejected, CTLFLAG_RD,
> >      &mutex_prof_rejected, 0, "Number of rejected profiling records");
> >  static int mutex_prof_hashsize = MPROF_HASH_SIZE;
> > @@ -418,6 +418,16 @@
> >
> >       while (!_obtain_lock(m, tid)) {
> >               lock_profile_obtain_lock_failed(&m->mtx_object, &contested);
> > +#if defined(SMP) && !defined(NO_ADAPTIVE_MUTEXES)
> > +             /*
> > +              * If the current owner of the lock is executing on another
> > +              * CPU, spin instead of blocking.
> > +              */
> > +             for (owner = mtx_owner(m); owner && TD_IS_RUNNING(owner); owner =
> mtx_owner(m))
> > +                     cpu_spinwait();
> > +             if (mtx_unowned(m))
> > +                     continue;
> > +#endif       /* SMP && !NO_ADAPTIVE_MUTEXES */
> >               turnstile_lock(&m->mtx_object);
> >               v = m->mtx_lock;
> >
> > @@ -425,7 +435,7 @@
> >                * Check if the lock has been released while spinning for
> >                * the turnstile chain lock.
> >                */
> > -             if (v == MTX_UNOWNED) {
> > +             if (mtx_unowned(m)) {
> >                       turnstile_release(&m->mtx_object);
> >                       cpu_spinwait();
> >                       continue;
> > @@ -837,6 +847,8 @@
> >       mtx_validate(m);
> >  #endif
> >
> > +
> > +
> >       /* Determine lock class and lock flags. */
> >       if (opts & MTX_SPIN)
> >               class = &lock_class_mtx_spin;
> >
>
> --
> John Baldwin
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b1fa29170606270931p6fbbbf98vcfcad8dc50987024>