linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Oleg Nesterov <oleg@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	Steven Rostedt <rostedt@goodmis.org>
Cc: Sasha Levin <sasha.levin@oracle.com>,
	Ingo Molnar <mingo@kernel.org>, Peter Anvin <hpa@zytor.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Andy Lutomirski <luto@amacapital.net>,
	Denys Vlasenko <dvlasenk@redhat.com>,
	Thomas Gleixner <tglx@linutronix.de>,
	Chuck Ebbert <cebbert.lkml@gmail.com>
Subject: [PATCH 0/1] stop the unbound recursion in preempt_schedule_context()
Date: Sun, 5 Oct 2014 22:23:00 +0200	[thread overview]
Message-ID: <20141005202300.GA27962@redhat.com> (raw)
In-Reply-To: <20141004003300.GA6297@redhat.com>

On 10/04, Oleg Nesterov wrote:
>
> On 10/03, Linus Torvalds wrote:
> >
> > On Fri, Oct 3, 2014 at 5:01 PM, Linus Torvalds
> > <torvalds@linux-foundation.org> wrote:
> > >
> > > The real fix would appear to be to use
> > > "preempt_enable_no_resched_notrace()", which your patch did, but
> > > without the loop.
> >
> > Actually, the real fix would be to not be stupid, and just make the
> > code do something like
> >
> > >        if (likely(!preemptible()))
> > >                return;
> > >
> > >         __preempt_count_add(PREEMPT_ACTIVE);
> > >         prev_ctx = exception_enter();
> > >
> > >         __schedule();
> > >
> > >         exception_exit(prev_ctx);
> > >         __preempt_count_sub(PREEMPT_ACTIVE);
> >
> > and *not* enable preemption around the scheduling at all.

Yes, I think you are right. And I hate to admit that I didn't think about
this simplification. The only complication is that we should move this
function from context_tracking.c to sched/core.c.

However, I still think we need the need_resched() loop, we can't do this
only once.

And in fact the simplest fix could just turn it into

	local_irq_disable();
	preempt_schedule_irq();
	local_irq_enable();

> Again, it is too late for me... Most probably I am wrong, but somehow
> it seems to me that the real fix should try to kill preempt_schedule_context()
> altogether and teach preempt_schedule() to play well with CONTEXT_TRACKING.

Yes, the very fact that

	preempt_enable() != trace_preempt_on() + preempt_enable_notrace()

looks simply wrong imo. And preempt_enable_notrace() has users outside of
the tracing code which can run in IN_USER state. For example, why should
__perf_sw_event() worry about context_tracking.state?

OTOH, if the caller of preempt_enable_notrace() actually needs to take care
of potential IN_USER state, then it probably has other problems. And indeed,
ftrace_ops_control_func() has to check rcu_is_watching() anyway.

IOW, imho 29bb9e5a75 "tracing/context-tracking: Add preempt_schedule_context()
for tracing" was not a right solution. Perhaps the tracing functions can be
changed to switch to IN_KERNEL mode, or at least perhaps we can add the
special preempt_disable_ftrace() helper. But this needs another discussion.

Either way, I do think that preempt_schedule_context() in its current form
must die.

Oleg.


  reply	other threads:[~2014-10-05 20:26 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-09-21 18:41 [PATCH v3 0/2] x86: reimplement ___preempt_schedule*() using THUNK helpers Oleg Nesterov
2014-09-21 18:41 ` [PATCH v3 1/2] " Oleg Nesterov
2014-09-24 15:02   ` [tip:x86/asm] x86: Speed up ___preempt_schedule*() by " tip-bot for Oleg Nesterov
2014-10-03  4:50     ` Sasha Levin
2014-10-03 13:39       ` Chuck Ebbert
2014-10-03 21:41         ` Oleg Nesterov
2014-10-03 21:56           ` Andy Lutomirski
2014-10-03 23:48             ` Linus Torvalds
2014-10-03 23:51             ` Oleg Nesterov
2014-10-03 22:48           ` Chuck Ebbert
2014-10-03 22:53             ` Andy Lutomirski
2014-10-03 23:13               ` H. Peter Anvin
2014-10-03 23:37             ` Oleg Nesterov
2014-10-03 21:16       ` Oleg Nesterov
2014-10-03 23:26       ` Oleg Nesterov
2014-10-04  0:01         ` Linus Torvalds
2014-10-04  0:11           ` Linus Torvalds
2014-10-04  0:33             ` Oleg Nesterov
2014-10-05 20:23               ` Oleg Nesterov [this message]
2014-10-05 20:23                 ` [PATCH 1/1] stop the unbound recursion in preempt_schedule_context() Oleg Nesterov
2014-10-28 11:03                   ` [tip:sched/core] sched: " tip-bot for Oleg Nesterov
2014-10-05 23:53                 ` [PATCH 0/1] " Oleg Nesterov
2014-10-04  0:19           ` [tip:x86/asm] x86: Speed up ___preempt_schedule*() by using THUNK helpers Oleg Nesterov
2014-09-21 18:42 ` [PATCH v3 2/2] x86, lib/Makefile: remove the unnecessary "+= thunk_64.o" Oleg Nesterov
2014-09-24 15:02   ` [tip:x86/asm] x86/lib/Makefile: Remove " tip-bot for Oleg Nesterov

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=20141005202300.GA27962@redhat.com \
    --to=oleg@redhat.com \
    --cc=cebbert.lkml@gmail.com \
    --cc=dvlasenk@redhat.com \
    --cc=fweisbec@gmail.com \
    --cc=hpa@zytor.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=luto@amacapital.net \
    --cc=mingo@kernel.org \
    --cc=peterz@infradead.org \
    --cc=rostedt@goodmis.org \
    --cc=sasha.levin@oracle.com \
    --cc=tglx@linutronix.de \
    --cc=torvalds@linux-foundation.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).