From: "Paul E. McKenney" <paulmck@kernel.org>
To: Joel Fernandes <joel@joelfernandes.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
Scott Wood <swood@redhat.com>,
linux-rt-users@vger.kernel.org, linux-kernel@vger.kernel.org,
Thomas Gleixner <tglx@linutronix.de>,
Peter Zijlstra <peterz@infradead.org>,
Juri Lelli <juri.lelli@redhat.com>,
Clark Williams <williams@redhat.com>
Subject: Re: [PATCH RT v2 2/3] sched: migrate_enable: Use sleeping_lock to indicate involuntary sleep
Date: Tue, 27 Aug 2019 08:58:13 -0700 [thread overview]
Message-ID: <20190827155813.GG26530@linux.ibm.com> (raw)
In-Reply-To: <20190827130853.GB132568@google.com>
On Tue, Aug 27, 2019 at 09:08:53AM -0400, Joel Fernandes wrote:
> On Tue, Aug 27, 2019 at 11:23:33AM +0200, Sebastian Andrzej Siewior wrote:
> [snip]
> > > However, if this was instead an rcu_read_lock() critical section within
> > > a PREEMPT=y kernel, then if a schedule() occured within stop_one_task(),
> > > RCU would consider that critical section to be preempted. This means
> > > that any RCU grace period that is blocked by this RCU read-side critical
> > > section would remain blocked until stop_one_cpu() resumed, returned,
> > > and so on until the matching rcu_read_unlock() was reached. In other
> > > words, RCU would consider that RCU read-side critical section to span
> > > the call to stop_one_cpu() even if stop_one_cpu() invoked schedule().
> >
> > Isn't that my example from above and what we do in RT? My understanding
> > is that this is the reason why we need BOOST on RT otherwise the RCU
> > critical section could remain blocked for some time.
>
> Not just for boost, it is needed to block the grace period itself on
> PREEMPT=y. On PREEMPT=y, if rcu_note_context_switch() happens in middle of a
> rcu_read_lock() reader section, then the task is added to a blocked list
> (rcu_preempt_ctxt_queue). Then just after that, the CPU reports a QS state
> (rcu_qs()) as you can see in the PREEMPT=y implementation of
> rcu_note_context_switch(). Even though the CPU has reported a QS, the grace
> period will not end because the preempted (or block as could be in -rt) task
> is still blocking the grace period. This is fundamental to the function of
> Preemptible-RCU where there is the concept of tasks blocking a grace period,
> not just CPUs.
>
> I think what Paul is trying to explain AIUI (Paul please let me know if I
> missed something):
>
> (1) Anyone calling rcu_note_context_switch() and expecting it to respect
> RCU-readers that are readers as a result of interrupt disabled regions, have
> incorrect expectations. So calling rcu_note_context_switch() has to be done
> carefully.
>
> (2) Disabling interrupts is "generally" implied as an RCU-sched flavor
> reader. However, invoking rcu_note_context_switch() from a disabled interrupt
> region is *required* for rcu_note_context_switch() to work correctly.
>
> (3) On PREEMPT=y kernels, invoking rcu_note_context_switch() from an
> interrupt disabled region does not mean that that the task will be added to a
> blocked list (unless it is also in an RCU-preempt reader) so
> rcu_note_context_switch() may immediately report a quiescent state and
> nothing blockings the grace period.
> So callers of rcu_note_context_switch() must be aware of this behavior.
>
> (4) On PREEMPT=n, unlike PREEMPT=y, there is no blocked list handling and so
> nothing will block the grace period once rcu_note_context_switch() is called.
> So any path calling rcu_note_context_switch() on a PREEMPT=n kernel, in the
> middle of something that is expected to be an RCU reader would be really bad
> from an RCU view point.
>
> Probably, we should add this all to documentation somewhere.
I think that Sebastian understands this and was using the example of RCU
priority boosting to confirm his understanding. But documentation would
be good. Extremely difficult to keep current, but good. I believe that
the requirements document does cover this.
Thanx, Paul
> thanks!
>
> - Joel
>
>
> > > On the other hand, within a PREEMPT=n kernel, the call to schedule()
> > > would split even an rcu_read_lock() critical section. Which is why I
> > > asked earlier if sleeping_lock_inc() and sleeping_lock_dec() are no-ops
> > > in !PREEMPT_RT_BASE kernels. We would after all want the usual lockdep
> > > complaints in that case.
> >
> > sleeping_lock_inc() +dec() is only RT specific. It is part of RT's
> > spin_lock() implementation and used by RCU (rcu_note_context_switch())
> > to not complain if invoked within a critical section.
> >
> > > Does that help, or am I missing the point?
> > >
> > > Thanx, Paul
> > Sebastian
next prev parent reply other threads:[~2019-08-27 15:58 UTC|newest]
Thread overview: 43+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-21 23:19 [PATCH RT v2 0/3] RCU fixes Scott Wood
2019-08-21 23:19 ` [PATCH RT v2 1/3] rcu: Acquire RCU lock when disabling BHs Scott Wood
2019-08-21 23:33 ` Paul E. McKenney
2019-08-22 13:39 ` Joel Fernandes
2019-08-22 15:27 ` Paul E. McKenney
2019-08-23 1:50 ` Joel Fernandes
2019-08-23 2:11 ` Paul E. McKenney
2019-08-23 3:23 ` Scott Wood
2019-08-23 12:30 ` Paul E. McKenney
2019-08-23 16:17 ` Sebastian Andrzej Siewior
2019-08-23 19:46 ` Scott Wood
2019-08-26 15:59 ` Sebastian Andrzej Siewior
2019-08-26 23:21 ` Scott Wood
2019-08-23 2:36 ` Scott Wood
2019-08-23 2:54 ` Paul E. McKenney
2019-08-21 23:19 ` [PATCH RT v2 2/3] sched: migrate_enable: Use sleeping_lock to indicate involuntary sleep Scott Wood
2019-08-21 23:35 ` Paul E. McKenney
2019-08-23 1:21 ` Scott Wood
2019-08-23 16:20 ` Sebastian Andrzej Siewior
2019-08-23 19:28 ` Scott Wood
2019-08-24 3:10 ` Joel Fernandes
2019-08-26 15:25 ` Sebastian Andrzej Siewior
2019-08-26 16:29 ` Paul E. McKenney
2019-08-26 17:49 ` Scott Wood
2019-08-26 18:12 ` Paul E. McKenney
2019-08-27 9:23 ` Sebastian Andrzej Siewior
2019-08-27 13:08 ` Joel Fernandes
2019-08-27 15:58 ` Paul E. McKenney [this message]
2019-08-27 16:06 ` Joel Fernandes
2019-08-27 15:53 ` Paul E. McKenney
2019-08-28 9:27 ` Sebastian Andrzej Siewior
2019-08-28 12:54 ` Paul E. McKenney
2019-08-28 13:14 ` Sebastian Andrzej Siewior
2019-08-28 13:59 ` Joel Fernandes
2019-08-28 15:51 ` Paul E. McKenney
2019-08-28 15:50 ` Paul E. McKenney
2019-08-21 23:19 ` [PATCH RT v2 3/3] rcu: Disable use_softirq on PREEMPT_RT Scott Wood
2019-08-21 23:40 ` Paul E. McKenney
2019-08-23 16:32 ` Sebastian Andrzej Siewior
2019-08-22 13:59 ` Joel Fernandes
2019-08-22 15:29 ` Paul E. McKenney
2019-08-22 19:31 ` Scott Wood
2019-08-23 0:52 ` Joel Fernandes
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=20190827155813.GG26530@linux.ibm.com \
--to=paulmck@kernel.org \
--cc=bigeasy@linutronix.de \
--cc=joel@joelfernandes.org \
--cc=juri.lelli@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rt-users@vger.kernel.org \
--cc=peterz@infradead.org \
--cc=swood@redhat.com \
--cc=tglx@linutronix.de \
--cc=williams@redhat.com \
/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;
as well as URLs for NNTP newsgroup(s).