Content-Type: multipart/mixed; boundary="===============1601536282522514823==" MIME-Version: 1.0 From: Oleg Nesterov To: lkp@lists.01.org Subject: Re: [rfcomm_run] WARNING: CPU: 1 PID: 79 at kernel/sched/core.c:7156 __might_sleep() Date: Thu, 02 Oct 2014 21:18:46 +0200 Message-ID: <20141002191846.GB30606@redhat.com> In-Reply-To: <20141002165739.GC10583@worktop.programming.kicks-ass.net> List-Id: --===============1601536282522514823== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable Ah, somehow I missed this email, I already replied to the previous one. On 10/02, Peter Zijlstra wrote: > > In any case, if we change wait_woken() like the below, then we can > simplify the loops by taking out their signal_pending checks and using > the wait_woken() return value instead. Yes, but let me repeat, > +++ b/kernel/sched/wait.c > @@ -326,8 +326,14 @@ long wait_woken(wait_queue_t *wait, unsi > * woken_wake_function() such that if we observe WQ_FLAG_WOKEN we must > * also observe all state before the wakeup. > */ > - if (!(wait->flags & WQ_FLAG_WOKEN)) > - timeout =3D schedule_timeout(timeout); > + if (!(wait->flags & WQ_FLAG_WOKEN)) { > + if (___wait_is_interruptible(mode)) { ___wait_is_interruptible() is pointless, signal_pending_state() does the same checks. Not to mention it will always return T in this case, note that __builtin_constant_p(state) =3D=3D F. > + if (signal_pending_state(mode, current)) > + timeout =3D -ERESTARTSYS; OK, but unless I missed something this looks overcomplicated. You can simply do this at the start of wait_woken(). Not need to play with current->state, no need to clear WQ_FLAG_WOKEN. Oleg. --===============1601536282522514823==--