From: Frederic Weisbecker <frederic@kernel.org>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: LKML <linux-kernel@vger.kernel.org>,
Peter Zijlstra <peterz@infradead.org>,
Paul McKenney <paulmck@kernel.org>,
Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Subject: Re: [patch V2 2/9] irqtime: Make accounting correct on RT
Date: Mon, 7 Dec 2020 02:14:48 +0100 [thread overview]
Message-ID: <20201207011448.GC113660@lothringen> (raw)
In-Reply-To: <87czzm77re.fsf@nanos.tec.linutronix.de>
On Mon, Dec 07, 2020 at 01:57:25AM +0100, Thomas Gleixner wrote:
> On Mon, Dec 07 2020 at 01:23, Frederic Weisbecker wrote:
> >> --- a/kernel/sched/cputime.c
> >> +++ b/kernel/sched/cputime.c
> >> @@ -60,7 +60,7 @@ void irqtime_account_irq(struct task_str
> >> cpu = smp_processor_id();
> >> delta = sched_clock_cpu(cpu) - irqtime->irq_start_time;
> >> irqtime->irq_start_time += delta;
> >> - pc = preempt_count() - offset;
> >> + pc = irq_count() - offset;
> >
> > There are many preempt_count() users all around waiting for similar issues.
> > Wouldn't it be more reasonable to have current->softirq_disable_cnt just saving
> > the softirq count on context switch?
>
> There are not that many and all of them need to be looked at.
>
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index d2003a7d5ab5..6c899c35d6ba 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -3469,6 +3469,10 @@ static inline void prepare_task(struct task_struct *next)
> >
> > static inline void finish_task(struct task_struct *prev)
> > {
> > +#ifdef CONFIG_PREEMPT_RT
> > + prev->softirq_disable_cnt = softirq_count();
> > + __preempt_count_sub(prev->softirq_disable_cnt);
> > +#endif
>
> You fundamentaly break RT with that.
>
> If local_bh_disable() fiddles with the actual preempt_count on RT then
> softirq disabled sections and softirq processing are not longer
> preemtible.
>
> You asked me in the last round of patches to add a proper argument for
> pulling out the softirq count from preempt_count. Here is the revised
> changelog which you agreed with:
>
> "RT requires the softirq processing and local bottomhalf disabled regions to
> be preemptible. Using the normal preempt count based serialization is
> therefore not possible because this implicitely disables preemption.
> ....
> "
>
> Full text in patch 1/9.
>
> According to the above folding of softirq count into the actual preempt
> count cannot work at all.
>
> The current RT approach just works except for the places which look at
> the raw preempt_count and not using the wrappers. Those places are
> restricted to core code and a pretty small number.
>
> Trying to do what you suggest would be a major surgery all over the
> place including a complete trainwreck on the highly optimized
> preempt_enable() --> preempt decision.
I suspected it was more complicated than I imagined :-)
Nevermind.
Thanks.
next prev parent reply other threads:[~2020-12-07 1:15 UTC|newest]
Thread overview: 38+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-04 17:01 [patch V2 0/9] softirq: Make it RT aware Thomas Gleixner
2020-12-04 17:01 ` [patch V2 1/9] softirq: Add RT specific softirq accounting Thomas Gleixner
2020-12-07 13:06 ` Frederic Weisbecker
2020-12-04 17:01 ` [patch V2 2/9] irqtime: Make accounting correct on RT Thomas Gleixner
2020-12-07 0:23 ` Frederic Weisbecker
2020-12-07 0:57 ` Thomas Gleixner
2020-12-07 1:14 ` Frederic Weisbecker [this message]
2020-12-07 13:27 ` Frederic Weisbecker
2020-12-07 14:44 ` Thomas Gleixner
2020-12-04 17:01 ` [patch V2 3/9] softirq: Move various protections into inline helpers Thomas Gleixner
2020-12-07 13:37 ` Frederic Weisbecker
2020-12-04 17:01 ` [patch V2 4/9] softirq: Make softirq control and processing RT aware Thomas Gleixner
2020-12-07 14:16 ` Frederic Weisbecker
2020-12-07 15:08 ` Thomas Gleixner
2020-12-08 0:08 ` Frederic Weisbecker
2020-12-09 10:11 ` Peter Zijlstra
2020-12-09 12:36 ` Thomas Gleixner
2020-12-09 12:42 ` Peter Zijlstra
2020-12-09 13:30 ` Thomas Gleixner
2020-12-09 10:34 ` Peter Zijlstra
2020-12-04 17:01 ` [patch V2 5/9] tick/sched: Prevent false positive softirq pending warnings on RT Thomas Gleixner
2020-12-08 12:23 ` Frederic Weisbecker
2020-12-04 17:01 ` [patch V2 6/9] rcu: Prevent false positive softirq warning " Thomas Gleixner
2020-12-04 17:59 ` Paul E. McKenney
2020-12-04 17:01 ` [patch V2 7/9] softirq: Replace barrier() with cpu_relax() in tasklet_unlock_wait() Thomas Gleixner
2020-12-07 11:39 ` Peter Zijlstra
2020-12-07 15:21 ` Thomas Gleixner
2020-12-04 17:01 ` [patch V2 8/9] tasklets: Use static inlines for stub implementations Thomas Gleixner
2020-12-04 17:02 ` [patch V2 9/9] tasklets: Prevent kill/unlock_wait deadlock on RT Thomas Gleixner
2020-12-07 11:47 ` Peter Zijlstra
2020-12-07 14:00 ` Sebastian Andrzej Siewior
2020-12-07 14:27 ` Peter Zijlstra
2020-12-07 17:55 ` Thomas Gleixner
2020-12-07 15:22 ` Thomas Gleixner
2020-12-07 15:39 ` Sebastian Andrzej Siewior
2020-12-07 17:49 ` Thomas Gleixner
2020-12-07 17:50 ` Thomas Gleixner
2020-12-06 10:05 ` [patch V2 0/9] softirq: Make it RT aware Sebastian Andrzej Siewior
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=20201207011448.GC113660@lothringen \
--to=frederic@kernel.org \
--cc=bigeasy@linutronix.de \
--cc=linux-kernel@vger.kernel.org \
--cc=paulmck@kernel.org \
--cc=peterz@infradead.org \
--cc=tglx@linutronix.de \
/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).