Content-Type: multipart/mixed; boundary="===============5181094557415666432==" MIME-Version: 1.0 From: Frederic Weisbecker To: lkp@lists.01.org Subject: Re: [sched] BUG: kernel boot hang Date: Mon, 16 Feb 2015 19:22:38 +0100 Message-ID: <20150216182235.GA20614@lerouge> In-Reply-To: <20150216150733.GD5029@twins.programming.kicks-ass.net> List-Id: --===============5181094557415666432== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Mon, Feb 16, 2015 at 04:07:33PM +0100, Peter Zijlstra wrote: > On Sun, Feb 15, 2015 at 01:18:07PM -0500, Steven Rostedt wrote: > > On Sun, 15 Feb 2015 16:46:22 +0100 > > Frederic Weisbecker wrote: > > > diff --git a/kernel/sched/core.c b/kernel/sched/core.c > > > index c017a5f..a6d4d6c 100644 > > > --- a/kernel/sched/core.c > > > +++ b/kernel/sched/core.c > > > @@ -2879,7 +2879,7 @@ void __sched schedule_preempt_disabled(void) > > > preempt_disable(); > > > } > > > = > > > -static void preempt_schedule_common(void) > > > +static void __sched notrace preempt_schedule_common(void) > > > { > > > do { > > > preempt_count_add(PREEMPT_ACTIVE); > > = > > Ah, since I added better recursion protection code in function tracer > > this didn't break that. But unfortunately, function graph tracer > > doesn't have that protection. > > = > > If it traces between preempt_schedule() and where it sets > > PREEMPT_ACTIVE, it can indeed go into an infinite recursion. Yeah, > > preempt_schedule_common() should be notrace, at least until we change > > function_graph to have that recursion protection. > > = > > Acked-by: Steven Rostedt > = > = > Can someone shoot me a proper patch with Changelog and such? Sure, I just sent it as: "[PATCH] sched: Fix preempt_schedule_common() trig= gering tracing recursion" Huang Ying hasn't yet reported a Tested-by but I could reproduce (with both= function and function_graph tracers btw) and the patch fixed it. Thanks. --===============5181094557415666432==--