Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 19 Aug 2014 14:24:16 -0500
From:      Alan Cox <alan.l.cox@gmail.com>
To:        Mateusz Guzik <mjguzik@gmail.com>
Cc:        Konstantin Belousov <kostikbel@gmail.com>, Johan Schuijt <johan@transip.nl>, "freebsd-arch@freebsd.org" <freebsd-arch@freebsd.org>
Subject:   Re: [PATCH 1/2] Implement simple sequence counters with memory barriers.
Message-ID:  <CAJUyCcPA7ZDNbwyfx3fT7mq3SE7M-mL5he=eXZ8bY3z-xUCJ-g@mail.gmail.com>
In-Reply-To: <20140817012646.GA21025@dft-labs.eu>
References:  <1408064112-573-1-git-send-email-mjguzik@gmail.com> <1408064112-573-2-git-send-email-mjguzik@gmail.com> <20140816093811.GX2737@kib.kiev.ua> <20140816185406.GD2737@kib.kiev.ua> <20140817012646.GA21025@dft-labs.eu>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, Aug 16, 2014 at 8:26 PM, Mateusz Guzik <mjguzik@gmail.com> wrote:

> On Sat, Aug 16, 2014 at 09:54:06PM +0300, Konstantin Belousov wrote:
> > On Sat, Aug 16, 2014 at 12:38:11PM +0300, Konstantin Belousov wrote:
> > > On Fri, Aug 15, 2014 at 02:55:11AM +0200, Mateusz Guzik wrote:
> > > > ---
> > > >  sys/sys/seq.h | 126
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > > >  1 file changed, 126 insertions(+)
> > > >  create mode 100644 sys/sys/seq.h
> > > >
> > > > diff --git a/sys/sys/seq.h b/sys/sys/seq.h
> > > > new file mode 100644
> > > > index 0000000..0971aef
> > > > --- /dev/null
> > > > +++ b/sys/sys/seq.h
> [..]
> > > > +#ifndef _SYS_SEQ_H_
> > > > +#define _SYS_SEQ_H_
> > > > +
> > > > +#ifdef _KERNEL
> > > > +
> > > > +/*
> > > > + * Typical usage:
> > > > + *
> > > > + * writers:
> > > > + *       lock_exclusive(&obj->lock);
> > > > + *       seq_write_begin(&obj->seq);
> > > > + *       .....
> > > > + *       seq_write_end(&obj->seq);
> > > > + *       unlock_exclusive(&obj->unlock);
> > > > + *
> > > > + * readers:
> > > > + *       obj_t lobj;
> > > > + *       seq_t seq;
> > > > + *
> > > > + *       for (;;) {
> > > > + *               seq =3D seq_read(&gobj->seq);
> > > > + *               lobj =3D gobj;
> > > > + *               if (seq_consistent(&gobj->seq, seq))
> > > > + *                       break;
> > > > + *               cpu_spinwait();
> > > > + *       }
> > > > + *       foo(lobj);
> > > > + */
> > > > +
> > > > +typedef uint32_t seq_t;
> > > > +
> > > > +/* A hack to get MPASS macro */
> > > > +#include <sys/systm.h>
> > > > +#include <sys/lock.h>
> > > > +
> > > > +#include <machine/cpu.h>
> > > > +
> > > > +static __inline bool
> > > > +seq_in_modify(seq_t seqp)
> > > > +{
> > > > +
> > > > + return (seqp & 1);
> > > > +}
> > > > +
> > > > +static __inline void
> > > > +seq_write_begin(seq_t *seqp)
> > > > +{
> > > > +
> > > > + MPASS(!seq_in_modify(*seqp));
> > > > + (*seqp)++;
> > > > + wmb();
> > > This probably ought to be written as atomic_add_rel_int(seqp, 1);
> > Alan Cox rightfully pointed out that better expression is
> > v =3D *seqp + 1;
> > atomic_store_rel_int(seqp, v);
> > which also takes care of TSO on x86.
> >
>
> Well, my memory-barrier-and-so-on-fu is rather weak.
>
> I had another look at the issue. At least on amd64, it looks like only
> compiler barrier is required for both reads and writes.
>
> According to AMD64 Architecture Programmer=E2=80=99s Manual Volume 2: Sys=
tem
> Programming, 7.2 Multiprocessor Memory Access Ordering states:
>
> "Loads do not pass previous loads (loads are not reordered). Stores do
> not pass previous stores (stores are not reordered)"
>
> Since the code modifying stuff only performs a series of writes and we
> expect exclusive writers, I find it applicable to this scenario.
>
> I checked linux sources and generated assembly, they indeed issue only
> a compiler barrier on amd64 (and for intel processors as well).
>
> atomic_store_rel_int on amd64 seems fine in this regard, but the only
> function for loads issues lock cmpxhchg which kills performance
> (median 55693659 -> 12789232 ops in a microbenchmark) for no gain.
>
> Additionally release and acquire semantics seems to be a stronger than
> needed guarantee.
>
>

This statement left me puzzled and got me to look at our x86 atomic.h for
the first time in years.  It appears that our implementation of
atomic_load_acq_int() on x86 is, umm ..., unconventional.  That is, it is
enforcing a constraint that simple acquire loads don't normally enforce.
For example, the C11 stdatomic.h simple acquire load doesn't enforce this
constraint.  Moreover, our own implementation of atomic_load_acq_int() on
ia64, where the mapping from atomic_load_acq_int() to machine instructions
is straightforward, doesn't enforce this constraint either.

Give us a chance to sort this out before you do anything further.  As
Kostik said, but in different words, we've always written our
machine-independent layer code using acquires and releases to express the
required ordering constraints and not {r,w}mb() primitives.



> As far as sequence counters go, we should be able to get away with
> making the following:
> - all relevant reads are performed between given points
> - all relevant writes are performed between given points
>
> As such, I propose introducing another atomic_* function variants
> (or stealing smp_{w,r,}mb idea from linux) which provide just that.
>
> So for amd64 reading guarantee and writing guarantee could be provided
> in the same way with a compiler barrier.
>
> > > Same note for all other linux-style barriers.  In fact, on x86
> > > wmb() is sfence and it serves no useful purpose in seq_write*.
> > >
> > > Overall, it feels too alien and linux-ish for my taste.
> > > Since we have sequence bound to some lock anyway, could we introduce
> > > some sort of generation-aware locks variants, which extend existing
> > > locks, and where lock/unlock bump generation number ?
> > Still, merging it to the guts of lock implementation is right
> > approach, IMO.
> >
>
> Current usage would be along with filedesc (sx) lock. The lock protects
> writes to entire fd table (and lock holders can block in malloc), while
> each file descriptor has its own counter. Also areas covered by seq are
> short and cannot block.
>
> As such, I don't really see any way to merge the lock with the counter.
>
> I agree it would be useful, provided area protected by the lock would be
> the same as the one protected by the counter. If this code hits the tree
> and one day turns out someone needs such functionality, there should not
> be any problems (apart from time effort) in implementing this.
>
> --
> Mateusz Guzik <mjguzik gmail.com>
> _______________________________________________
> freebsd-arch@freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-arch
> To unsubscribe, send any mail to "freebsd-arch-unsubscribe@freebsd.org"
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJUyCcPA7ZDNbwyfx3fT7mq3SE7M-mL5he=eXZ8bY3z-xUCJ-g>