From: "Paul E. McKenney" <paulmck@kernel.org>
To: Alan Stern <stern@rowland.harvard.edu>
Cc: Andrea Parri <parri.andrea@gmail.com>,
Jonas Oberhauser <jonas.oberhauser@huawei.com>,
Peter Zijlstra <peterz@infradead.org>, will <will@kernel.org>,
"boqun.feng" <boqun.feng@gmail.com>, npiggin <npiggin@gmail.com>,
dhowells <dhowells@redhat.com>, "j.alglave" <j.alglave@ucl.ac.uk>,
"luc.maranget" <luc.maranget@inria.fr>, akiyks <akiyks@gmail.com>,
dlustig <dlustig@nvidia.com>, joel <joel@joelfernandes.org>,
urezki <urezki@gmail.com>,
quic_neeraju <quic_neeraju@quicinc.com>,
frederic <frederic@kernel.org>,
Kernel development list <linux-kernel@vger.kernel.org>
Subject: Re: Internal vs. external barriers (was: Re: Interesting LKMM litmus test)
Date: Fri, 20 Jan 2023 11:20:32 -0800 [thread overview]
Message-ID: <20230120192032.GR2948950@paulmck-ThinkPad-P17-Gen-1> (raw)
In-Reply-To: <Y8rff3HJ2o9wUyGT@rowland.harvard.edu>
On Fri, Jan 20, 2023 at 01:37:51PM -0500, Alan Stern wrote:
> On Fri, Jan 20, 2023 at 09:58:04AM -0800, Paul E. McKenney wrote:
> > On Fri, Jan 20, 2023 at 11:01:03AM -0500, Alan Stern wrote:
> > > On Wed, Jan 18, 2023 at 04:02:14PM -0800, Paul E. McKenney wrote:
> > > > There are pairs of per-CPU counters. One pair (->srcu_lock_count[])
> > > > counts the number of srcu_down_read() operations that took place on
> > > > that CPU and another pair (->srcu_unlock_count[]) counts the number
> > > > of srcu_down_read() operations that took place on that CPU. There is
> > > > an ->srcu_idx that selects which of the ->srcu_lock_count[] elements
> > > > should be incremented by srcu_down_read(). Of course, srcu_down_read()
> > > > returns the value of ->srcu_idx that it used so that the matching
> > > > srcu_up_read() will use that same index when incrementing its CPU's
> > > > ->srcu_unlock_count[].
> > > >
> > > > Grace periods go something like this:
> > > >
> > > > 1. Sum up the ->srcu_unlock_count[!ssp->srcu_idx] counters.
> > > >
> > > > 2. smp_mb().
> > > >
> > > > 3. Sum up the ->srcu_unlock_count[!ssp->srcu_idx] counters.
> > >
> > > Presumably you meant to write "lock" here, not "unlock".
> >
> > You are quite right, and apologies for my confusion.
> >
> > > > 4. If the sums are not equal, retry from #1.
> > > >
> > > > 5. smp_mb().
> > > >
> > > > 6. WRITE_ONCE(ssp->srcu_idx, !ssp->srcu_idx);
> > > >
> > > > 7. smp_mb().
> > > >
> > > > 8. Same loop as #1-4.
> > > >
> > > > So similar to r/w semaphores, but with two separate distributed counts.
> > > > This means that the number of readers need not go to zero at any given
> > > > point in time, consistent with the need to wait only on old readers.
> > >
> > > Reasoning from first principles, I deduce the following:
> > >
> > > You didn't describe exactly how srcu_down_read() and srcu_up_read()
> > > work. Evidently the unlock increment in srcu_up_read() should have
> > > release semantics, to prevent accesses from escaping out the end of the
> > > critical section. But the lock increment in srcu_down_read() has to be
> > > stronger than an acquire; to prevent accesses in the critical section
> > > escaping out the start, the increment has to be followed by smp_mb().
> >
> > You got it! There is some work going on to see if srcu_read_lock()'s
> > smp_mb() can be weakened to pure release, but we will see.
>
> That doesn't make sense. Release ordering in srcu_read_lock() would
> only prevent accesses from leaking _in_ to the critical section. It
> would do nothing to prevent accesses from leaking _out_.
Yes, I should have said srcu_read_unlock(). I do seem to be having
lock/unlock difficulties. :-/
We could remove the smp_mb() from srcu_read_lock(), but at the expense
of a round of IPIs from the grace-period code, along with interactions
with things like the CPU-hotplug code paths. I am not proposing
doing that, for one thing, one of the attractions of SRCU is its fast
and disturbance-free grace period when there are no readers in flight.
It is possible, though: Tasks Trace RCU does just this, IPIs, CPU hotplug,
and all.
There are other ways to do this, but the ones I know of would restrict
the contexts in which srcu_read_lock() and srcu_read_unlock() can be
executed, for example, in the context of offline CPUs.
> > > The smp_mb() fences in steps 5 and 7 appear to be completely
> > > unnecessary.
> >
> > For correctness, agreed. Their purpose is instead forward progress.
>
> It's hard to say whether they would be effective at that. smp_mb()
> forces the processor to wait until some time when previous writes have
> become visible to all CPUs. But if you don't wait for that potentially
> excessively long delay, you may be able to continue and be lucky enough
> to find that all previous writes have already become visible to all the
> CPUs that matter.
>
> As far as I know, smp_mb() doesn't expedite the process of making
> previous writes visible. However, I am very far from being an expert
> on system architecture design.
As you noticed, without the step-7 smp_mb(), a potentially large number
of invocations of srcu_read_lock() could use the old index value, that
is, the inces value that is to be counted in step 8. Then the step-8
phase of the grace period could unncessarily wait on those readers.
Similarly, without the step-5 smp_mb() and without the control
dependencies extending from the loads feeding into step 4's sum,
srcu_read_lock() and srcu_read_unlock() on other CPUs might prematurely
use the new index, which could force the step 1-4 phase of the grace
period to unnecessarily wait on those readers.
> > One can argue that step 5 is redundant due to control dependency, but
> > control dependencies are fragile, and as you say below, this code is
> > nowhere near a fastpath.
>
> Also, control dependencies do not contribute to forward progress.
I might be mistaken, and you can argue that the risk is small, but without
that ordering, step 4 could see unintended increments that could force
unnecessary retries of steps 1-3.
> > > Provided an smp_mb() is added at the very start and end of the grace
> > > period, the memory barrier in step 2 and its copy in step 8 can be
> > > demoted to smp_rmb().
> >
> > This might need to be smp_mb() to allow srcu_read_unlock() to be
> > demoted to release ordering. Work in progress.
>
> I thought srcu_read_unlock() already _is_ a release operation. The
> smp_mb() fence mentioned earlier needs to be in srcu_read_lock(), not
> unlock(). And there's no way that one can be demoted.
Agreed, my mistake earlier. The smp_mb() in srcu_read_lock() must
remaain smp_mb().
> srcu_read_unlock() does not need a full smp_mb().
That is quite possible, and that is what we are looking into. And testing
thus far agrees with you. But the grace-period ordering constraints
are quite severe, so this requires careful checking and severe testing.
> > > These changes would be small optimizations at best, and you may consider
> > > them unimportant in view of the fact that grace periods often last quite
> > > a long time.
> >
> > Agreed, keeping it simple and obvious is important on this code, which
> > is nowhere near a fastpath. The case of srcu_read_unlock() is another
> > thing altogether.
>
> Unfortunately, the full fence in srcu_read_lock() is unavoidable without
> very major changes to the algorithm -- probably a complete redesign.
> Without it, a read inside the critical section could be executed before
> the store part of the increment, which could lead synchronize_srcu() to
> believe that the critical section had not yet started when in fact it
> had.
I actually did type "srcu_read_unlock()" correctly in this case. ;-)
But yes, removing the smp_mb() from srcu_read_lock() is not in the cards.
On the other hand, doing so for srcu_read_unlock() just might be both
doable and worthwhile.
Thanx, Paul
next prev parent reply other threads:[~2023-01-20 19:20 UTC|newest]
Thread overview: 161+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20220921173109.GA1214281@paulmck-ThinkPad-P17-Gen-1>
[not found] ` <YytfFiMT2Xsdwowf@rowland.harvard.edu>
[not found] ` <YywXuzZ/922LHfjI@hirez.programming.kicks-ass.net>
[not found] ` <114ECED5-FED1-4361-94F7-8D9BC02449B7>
[not found] ` <YzSAnclenTz7KQyt@rowland.harvard.edu>
[not found] ` <f763bd5ff835458d8750b61da47fe316@huawei.com>
2023-01-03 18:56 ` Alan Stern
2023-01-04 15:37 ` Andrea Parri
2023-01-04 20:58 ` Alan Stern
[not found] ` <ee186bc17a5e48298a5373f688496dce@huawei.com>
2023-01-05 17:32 ` Paul E. McKenney
[not found] ` <bea712c82e6346f8973399a5711ff78a@huawei.com>
2023-01-11 15:06 ` Alan Stern
[not found] ` <768ffe7edc7f4ddfacd5b0a8e844ed84@huawei.com>
2023-01-11 17:01 ` Alan Stern
[not found] ` <07579baee4b84532a76ea8b0b33052bb@huawei.com>
2023-01-12 21:57 ` Paul E. McKenney
2023-01-13 16:38 ` Alan Stern
2023-01-13 19:54 ` Paul E. McKenney
[not found] ` <06a8aef7eb8d46bca34521a80880dae3@huawei.com>
2023-01-14 17:42 ` Paul E. McKenney
[not found] ` <e51c82a113484b6bb62354a49376f248@huawei.com>
2023-01-14 16:42 ` Alan Stern
2023-01-17 17:48 ` Jonas Oberhauser
2023-01-17 21:19 ` Alan Stern
2023-01-18 11:25 ` Jonas Oberhauser
2023-01-19 2:28 ` Alan Stern
2023-01-19 11:22 ` Jonas Oberhauser
2023-01-19 16:41 ` Alan Stern
2023-01-19 18:43 ` Paul E. McKenney
2023-01-23 16:16 ` Jonas Oberhauser
2023-01-23 19:58 ` Alan Stern
2023-01-23 20:06 ` Jonas Oberhauser
2023-01-23 20:41 ` Alan Stern
2023-01-24 13:21 ` Jonas Oberhauser
2023-01-24 15:54 ` Jonas Oberhauser
2023-01-24 17:22 ` Alan Stern
[not found] ` <4c1abc7733794519ad7c5153ae8b58f9@huawei.com>
2023-01-13 16:28 ` Alan Stern
2023-01-13 20:07 ` Paul E. McKenney
2023-01-13 20:32 ` Paul E. McKenney
2023-01-14 17:40 ` Alan Stern
2023-01-14 17:48 ` Paul E. McKenney
[not found] ` <136d019d8c8049f6b737627df830e66f@huawei.com>
2023-01-14 17:53 ` Paul E. McKenney
2023-01-14 18:15 ` Paul E. McKenney
2023-01-14 19:58 ` Alan Stern
2023-01-15 5:19 ` Paul E. McKenney
2023-01-14 20:19 ` Alan Stern
2023-01-15 5:15 ` Paul E. McKenney
2023-01-15 16:23 ` Alan Stern
2023-01-15 18:10 ` Paul E. McKenney
2023-01-15 20:46 ` Alan Stern
2023-01-16 4:23 ` Paul E. McKenney
2023-01-16 18:11 ` Alan Stern
2023-01-16 19:06 ` Paul E. McKenney
2023-01-16 19:20 ` Alan Stern
2023-01-16 22:13 ` Paul E. McKenney
2023-01-17 11:46 ` Andrea Parri
2023-01-17 15:14 ` Paul E. McKenney
2023-01-17 15:56 ` Alan Stern
2023-01-17 17:43 ` Paul E. McKenney
2023-01-17 18:27 ` Jonas Oberhauser
2023-01-17 18:55 ` Paul E. McKenney
2023-01-17 20:20 ` Jonas Oberhauser
2023-01-17 20:15 ` Alan Stern
2023-01-18 3:50 ` Paul E. McKenney
2023-01-18 16:50 ` Alan Stern
2023-01-18 19:42 ` Jonas Oberhauser
2023-01-18 20:19 ` Paul E. McKenney
2023-01-18 20:30 ` Jonas Oberhauser
2023-01-18 21:12 ` Paul E. McKenney
2023-01-18 21:24 ` Jonas Oberhauser
2023-01-19 0:11 ` Paul E. McKenney
2023-01-19 13:39 ` Jonas Oberhauser
2023-01-19 18:41 ` Paul E. McKenney
2023-01-19 19:51 ` Alan Stern
2023-01-19 21:53 ` Paul E. McKenney
2023-01-19 22:04 ` Alan Stern
2023-01-19 23:03 ` Paul E. McKenney
2023-01-20 9:43 ` Jonas Oberhauser
2023-01-20 15:39 ` Paul E. McKenney
2023-01-20 20:46 ` Jonas Oberhauser
2023-01-20 21:37 ` Paul E. McKenney
2023-01-20 22:36 ` Jonas Oberhauser
2023-01-20 23:19 ` Paul E. McKenney
2023-01-21 0:03 ` Jonas Oberhauser
2023-01-21 0:34 ` Paul E. McKenney
2023-01-20 3:55 ` Paul E. McKenney
2023-01-20 9:20 ` Jonas Oberhauser
2023-01-20 12:34 ` Jonas Oberhauser
2023-01-20 12:51 ` Jonas Oberhauser
2023-01-20 15:32 ` Paul E. McKenney
2023-01-20 20:56 ` Jonas Oberhauser
2023-01-20 21:40 ` Paul E. McKenney
2023-01-20 16:14 ` Alan Stern
2023-01-20 17:30 ` Paul E. McKenney
2023-01-20 18:15 ` Alan Stern
2023-01-20 18:59 ` Paul E. McKenney
2023-01-20 10:13 ` Jonas Oberhauser
2023-01-20 15:47 ` Paul E. McKenney
2023-01-20 22:21 ` Jonas Oberhauser
2023-01-20 16:18 ` Alan Stern
2023-01-20 21:41 ` Jonas Oberhauser
2023-01-21 4:38 ` Paul E. McKenney
2023-01-21 17:36 ` Alan Stern
2023-01-21 18:40 ` Paul E. McKenney
2023-01-21 19:56 ` Alan Stern
2023-01-21 20:10 ` Paul E. McKenney
2023-01-21 21:03 ` Alan Stern
2023-01-21 23:49 ` Paul E. McKenney
2023-01-23 11:48 ` Jonas Oberhauser
2023-01-23 15:55 ` Alan Stern
2023-01-23 19:40 ` Jonas Oberhauser
2023-01-23 20:34 ` Alan Stern
2023-01-18 20:06 ` Paul E. McKenney
2023-01-18 20:54 ` Alan Stern
2023-01-18 21:05 ` Jonas Oberhauser
2023-01-19 0:02 ` Paul E. McKenney
2023-01-19 2:19 ` Alan Stern
2023-01-19 11:23 ` Paul E. McKenney
2023-01-20 16:01 ` Alan Stern
2023-01-20 17:58 ` Paul E. McKenney
2023-01-20 18:37 ` Alan Stern
2023-01-20 19:20 ` Paul E. McKenney [this message]
2023-01-20 20:36 ` Alan Stern
2023-01-20 21:20 ` Paul E. McKenney
2023-01-22 20:32 ` Alan Stern
2023-01-23 20:16 ` Paul E. McKenney
2023-01-24 2:18 ` Alan Stern
2023-01-24 4:06 ` Paul E. McKenney
2023-01-24 11:09 ` Andrea Parri
2023-01-24 14:54 ` Paul E. McKenney
2023-01-24 15:11 ` Jonas Oberhauser
2023-01-24 16:22 ` Paul E. McKenney
2023-01-24 16:39 ` Jonas Oberhauser
2023-01-24 17:26 ` Paul E. McKenney
2023-01-24 19:30 ` Jonas Oberhauser
2023-01-24 22:15 ` Paul E. McKenney
2023-01-24 22:35 ` Alan Stern
2023-01-24 22:54 ` Paul E. McKenney
2023-01-25 1:54 ` Alan Stern
2023-01-25 2:20 ` Paul E. McKenney
2023-01-25 13:10 ` Jonas Oberhauser
2023-01-25 15:05 ` Paul E. McKenney
2023-01-25 15:34 ` Alan Stern
2023-01-25 17:18 ` Paul E. McKenney
2023-01-25 17:42 ` Jonas Oberhauser
2023-01-25 19:08 ` Alan Stern
2023-01-25 19:46 ` Paul E. McKenney
2023-01-25 20:36 ` Andrea Parri
2023-01-25 21:10 ` Jonas Oberhauser
2023-01-25 21:23 ` Paul E. McKenney
2023-01-25 20:46 ` Alan Stern
2023-01-25 21:38 ` Paul E. McKenney
2023-01-25 23:33 ` Paul E. McKenney
2023-01-26 1:45 ` Alan Stern
2023-01-26 1:53 ` Paul E. McKenney
2023-01-26 12:17 ` Jonas Oberhauser
2023-01-26 18:48 ` Paul E. McKenney
2023-01-27 15:03 ` Jonas Oberhauser
2023-01-27 16:50 ` Paul E. McKenney
2023-01-27 16:54 ` Paul E. McKenney
2023-01-18 19:57 ` Jonas Oberhauser
2023-01-18 21:06 ` Paul E. McKenney
2023-01-18 2:15 ` Alan Stern
2023-01-18 5:17 ` Paul E. McKenney
2023-01-18 16:03 ` Alan Stern
2023-01-18 16:59 ` Boqun Feng
2023-01-18 17:08 ` Alan Stern
2023-01-18 17:41 ` Paul E. McKenney
2023-01-19 19:07 ` Paul E. McKenney
2023-01-14 16:55 ` Alan Stern
2023-01-14 17:35 ` Paul E. McKenney
[not found] ` <17078dd97cb6480f9c51e27058af3197@huawei.com>
2023-01-14 17:27 ` Alan Stern
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230120192032.GR2948950@paulmck-ThinkPad-P17-Gen-1 \
--to=paulmck@kernel.org \
--cc=akiyks@gmail.com \
--cc=boqun.feng@gmail.com \
--cc=dhowells@redhat.com \
--cc=dlustig@nvidia.com \
--cc=frederic@kernel.org \
--cc=j.alglave@ucl.ac.uk \
--cc=joel@joelfernandes.org \
--cc=jonas.oberhauser@huawei.com \
--cc=linux-kernel@vger.kernel.org \
--cc=luc.maranget@inria.fr \
--cc=npiggin@gmail.com \
--cc=parri.andrea@gmail.com \
--cc=peterz@infradead.org \
--cc=quic_neeraju@quicinc.com \
--cc=stern@rowland.harvard.edu \
--cc=urezki@gmail.com \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®