From: "Paul E. McKenney" <paulmck@kernel.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
Peter Zijlstra <peterz@infradead.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
the arch/x86 maintainers <x86@kernel.org>
Subject: Re: [GIT pull] locking/urgent for v5.10-rc6
Date: Sun, 29 Nov 2020 16:29:18 -0800 [thread overview]
Message-ID: <20201130002918.GD1437@paulmck-ThinkPad-P72> (raw)
In-Reply-To: <CAHk-=wi3o-wwFVbAXb7YZZViDBsZ_yMVqyOAEZsx5qcskLsOcg@mail.gmail.com>
On Sun, Nov 29, 2020 at 11:31:41AM -0800, Linus Torvalds wrote:
> On Sun, Nov 29, 2020 at 5:38 AM Thomas Gleixner <tglx@linutronix.de> wrote:
> >
> > Yet two more places which invoke tracing from RCU disabled regions in the
> > idle path. Similar to the entry path the low level idle functions have to
> > be non-instrumentable.
>
> This really seems less than optimal.
>
> In particular, lookie here:
>
> > @@ -94,9 +94,35 @@ void __cpuidle default_idle_call(void)
> >
> > trace_cpu_idle(1, smp_processor_id());
> > stop_critical_timings();
> > +
> > + /*
> > + * arch_cpu_idle() is supposed to enable IRQs, however
> > + * we can't do that because of RCU and tracing.
> > + *
> > + * Trace IRQs enable here, then switch off RCU, and have
> > + * arch_cpu_idle() use raw_local_irq_enable(). Note that
> > + * rcu_idle_enter() relies on lockdep IRQ state, so switch that
> > + * last -- this is very similar to the entry code.
> > + */
> > + trace_hardirqs_on_prepare();
> > + lockdep_hardirqs_on_prepare(_THIS_IP_);
> > rcu_idle_enter();
> > + lockdep_hardirqs_on(_THIS_IP_);
> > +
> > arch_cpu_idle();
> > +
> > + /*
> > + * OK, so IRQs are enabled here, but RCU needs them disabled to
> > + * turn itself back on.. funny thing is that disabling IRQs
> > + * will cause tracing, which needs RCU. Jump through hoops to
> > + * make it 'work'.
> > + */
> > + raw_local_irq_disable();
> > + lockdep_hardirqs_off(_THIS_IP_);
> > rcu_idle_exit();
> > + lockdep_hardirqs_on(_THIS_IP_);
> > + raw_local_irq_enable();
> > +
> > start_critical_timings();
> > trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
> > }
>
> And look at what the code generation for the idle exit path is when
> lockdep isn't even on.
>
> It's *literally*
>
> cli
> call rcu_idle_exit
> sti
>
> and guess what rcu_idle_exit does?
>
> Yeah, that one does "pushf; cli; call rcu_eqs_exit; popf".
>
> So here we are, in the somewhat critical "an interrupt woke us up"
> section, and we're doing just ridiculously stupid things.
>
> I've pulled this, because it solves a problem, but there's a deeper
> problem here in how all this is done.
>
> The idle path is actually quite important. I can point to real loads
> where this is a big part of the CPU profile, because you end up having
> lots of "go to sleep for very short times, because the thing we were
> waiting for takes almost no time at all".
This is because of the noinline implied by the noinstr on rcu_eqs_exit().
If I replace that with inline, it does get inlined. Except that, if
I remember correctly, making that change messes up the tooling that
enforces the no-instrumentation regions.
I -think- that a combination of instrumentation_end() and s/noinstr/inline/
might work, but I will need to consult with the experts on this.
Thanx, Paul
next prev parent reply other threads:[~2020-11-30 0:30 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-29 13:37 [GIT pull] irq/urgent for v5.10-rc6 Thomas Gleixner
2020-11-29 13:38 ` [GIT pull] locking/urgent " Thomas Gleixner
2020-11-29 19:31 ` Linus Torvalds
2020-11-30 0:29 ` Paul E. McKenney [this message]
2020-11-30 7:56 ` Peter Zijlstra
2020-11-30 8:23 ` Sven Schnelle
2020-11-30 12:31 ` Sven Schnelle
2020-11-30 12:52 ` Peter Zijlstra
2020-11-30 13:03 ` Peter Zijlstra
2020-11-30 18:04 ` Linus Torvalds
2020-11-30 19:31 ` Christian Borntraeger
2020-12-01 8:07 ` Peter Zijlstra
2020-12-01 11:07 ` Peter Zijlstra
2020-12-01 14:46 ` Paul E. McKenney
2020-12-01 14:55 ` Peter Zijlstra
2020-12-01 16:53 ` Paul E. McKenney
2020-12-01 18:15 ` Peter Zijlstra
2020-12-01 18:48 ` Peter Zijlstra
2020-12-01 18:57 ` Mark Rutland
2020-12-01 19:14 ` Peter Zijlstra
2020-12-01 19:18 ` Heiko Carstens
2020-12-02 9:21 ` Peter Zijlstra
2020-12-02 10:56 ` Heiko Carstens
2020-12-02 11:16 ` Mark Rutland
2020-12-02 13:46 ` Heiko Carstens
2020-12-02 12:27 ` Peter Zijlstra
2020-12-01 7:48 ` Sven Schnelle
2020-12-02 7:54 ` Heiko Carstens
2020-12-02 9:38 ` Peter Zijlstra
2020-12-02 10:52 ` Heiko Carstens
2020-11-30 8:36 ` Peter Zijlstra
2020-11-30 17:55 ` Linus Torvalds
2020-12-01 7:39 ` Peter Zijlstra
2020-12-01 7:56 ` Peter Zijlstra
2020-12-01 19:45 ` Linus Torvalds
2020-12-02 7:53 ` Peter Zijlstra
2020-11-29 20:06 ` pr-tracker-bot
2020-11-29 20:06 ` [GIT pull] irq/urgent " pr-tracker-bot
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=20201130002918.GD1437@paulmck-ThinkPad-P72 \
--to=paulmck@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=peterz@infradead.org \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
--cc=x86@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;
as well as URLs for NNTP newsgroup(s).