From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753987AbbE1NMJ (ORCPT ); Thu, 28 May 2015 09:12:09 -0400 Received: from cantor2.suse.de ([195.135.220.15]:47625 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752172AbbE1NMF (ORCPT ); Thu, 28 May 2015 09:12:05 -0400 Date: Thu, 28 May 2015 15:12:02 +0200 From: Petr Mladek To: Andrew Morton Cc: Frederic Weisbecker , Steven Rostedt , Dave Anderson , "Paul E. McKenney" , Kay Sievers , Jiri Kosina , Michal Hocko , Jan Kara , linux-kernel@vger.kernel.org, Wang Long , peifeiyue@huawei.com, dzickus@redhat.com, morgan.wang@huawei.com, sasha.levin@oracle.com, Peter Zijlstra Subject: Re: [PATCH 04/10] printk: Merge and flush NMI buffer predictably via IRQ work Message-ID: <20150528131202.GE3135@pathway.suse.cz> References: <1432557993-20458-1-git-send-email-pmladek@suse.cz> <1432557993-20458-5-git-send-email-pmladek@suse.cz> <20150527161423.9ca10028e4edf4ecbd28a533@linux-foundation.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20150527161423.9ca10028e4edf4ecbd28a533@linux-foundation.org> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed 2015-05-27 16:14:23, Andrew Morton wrote: > On Mon, 25 May 2015 14:46:27 +0200 Petr Mladek wrote: > > > It might take ages until users see messages from NMI context. They cannot > > be flushed to the console because the operation involves taking and > > releasing a bunch of locks. Everything gets fixed by the followup printk > > in normal context but it is not predictable. > > > > The same problem has printk_sched() and this patch reuses the existing > > solution. > > > > There is no special printk() variant for NMI context. Hence the IRQ work > > need to get queued from vprintk_emit(). > > > > ... > > > > --- a/kernel/printk/printk.c > > +++ b/kernel/printk/printk.c > > @@ -1554,9 +1554,6 @@ int printk_deferred(const char *fmt, ...) > > va_start(args, fmt); > > r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args); > > va_end(args); > > - > > - __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT); > > - irq_work_queue(this_cpu_ptr(&wake_up_klogd_work)); > > preempt_enable(); > > > > return r; > > @@ -1880,7 +1877,10 @@ asmlinkage int vprintk_emit(int facility, int level, > > * If called from the scheduler or NMI context, we can not get console > > * without a possible deadlock. > > */ > > - if (!in_sched && !in_nmi()) { > > + if (in_sched || in_nmi()) { > > + __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT); > > + irq_work_queue(this_cpu_ptr(&wake_up_klogd_work)); > > + } else { > > This looks hairy. Why irq_work_queue() OK to call from NMI? IMHO, irq_work_queue() was primary created for exactly this job. I mean to move some work from NMI into a more relaxed context. See the initial commit e360adbe29241a01 ("irq_work: Add generic hardirq context callbacks"). In each case, the code seems to be well prepared for this. Especially, there is a big effort to manipulate work->flags lock-less way. The only exception is irq_work_queue_on() but it seems to be related to the particular arch_send_call_function_single_ipi(cpu) call that is not NMI safe. > arch/arm64/kernel/smp.c uses smp_cross_call() which might use NMI! smp_cross_call() is defined by set_smp_cross_call(). I see only three functions that are assigned this way: drivers/irqchip/irq-armada-370-xp.c: set_smp_cross_call(armada_mpic_send_doorbell); drivers/irqchip/irq-gic-v3.c: set_smp_cross_call(gic_raise_softirq); drivers/irqchip/irq-gic.c: set_smp_cross_call(gic_raise_softirq); drivers/irqchip/irq-hip04.c: set_smp_cross_call(hip04_raise_softirq); They all seems to work with soft IRQ. > Presumably it'll call directly if the target CPU==this_cpu but I didn't > run around and audit everything. IMHO, this is the case of smp_call_function_single() that has the function as a parameter. But irq_work_queue() always puts the function into the work list and sends interrupt. Of course, it is possible that I have missed something. I hope that Peter or Frederic could confirm my observation. Best Regards, Petr