From mboxrd@z Thu Jan 1 00:00:00 1970 From: Oleg Nesterov Subject: Re: [regression] boot failure on alpha, bisected Date: Sun, 7 Oct 2012 19:33:36 +0200 Message-ID: <20121007173336.GA14804@redhat.com> References: <20121006204736.GA1830@ds20.borg.net> <20121007165534.GA8024@redhat.com> <20121007170850.GJ2616@ZenIV.linux.org.uk> Mime-Version: 1.0 Return-path: Content-Disposition: inline In-Reply-To: <20121007170850.GJ2616@ZenIV.linux.org.uk> Sender: linux-alpha-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Al Viro Cc: dl8bcu@dl8bcu.de, peterz@infradead.org, mingo@kernel.org, linux-kernel@vger.kernel.org, linux-alpha@vger.kernel.org, Richard Henderson , Ivan Kokshaysky , Matt Turner On 10/07, Al Viro wrote: > > On Sun, Oct 07, 2012 at 06:55:34PM +0200, Oleg Nesterov wrote: > > > Hmm. I know nothing about arch/alpha and I can't understand its entry.S. > > But _it seems_ to me that do_notify_resume() is called with irqs disabled. > > If this is true, then imho arch/alpha should be fixed. > > > > Before this commit task_work_run() enabled irqs, but this was the "side > > effect" of spin_lock_irq/spin_unlock_irq, we should not rely on this. > > Um... There's a bunch of architectures that are in the same situation. > grep for do_notify_resume() and you'll see... And every do_notify_resume() should be changed anyway, do_signal() and tracehook_notify_resume() should be re-ordered. > It needs to be dealt with sanely, and actually have patches for alpha > going in that direction, but breaking a bunch of architectures is not a good > thing, obviously. So you've bought yourself a major PITA for coming > weeks... So perhaps the patch below until they are fixed? --- x/kernel/task_work.c +++ x/kernel/task_work.c @@ -51,6 +51,9 @@ void task_work_run(void) struct task_struct *task = current; struct callback_head *work, *head, *next; + if (WARN_ONCE(irqs_disabled(), "notify_resume() with irqs_disabled")) + local_irq_enable(); + for (;;) { /* * work->func() can do task_work_add(), do not set