From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-8.2 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_HELO_NONE,SPF_PASS, UNPARSEABLE_RELAY,USER_AGENT_SANE_1 autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id E1CFFC33CB1 for ; Tue, 14 Jan 2020 08:54:26 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id A9BCB2075B for ; Tue, 14 Jan 2020 08:54:26 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1728777AbgANIyZ (ORCPT ); Tue, 14 Jan 2020 03:54:25 -0500 Received: from out30-44.freemail.mail.aliyun.com ([115.124.30.44]:58744 "EHLO out30-44.freemail.mail.aliyun.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1725842AbgANIyZ (ORCPT ); Tue, 14 Jan 2020 03:54:25 -0500 X-Alimail-AntiSpam: AC=PASS;BC=-1|-1;BR=01201311R551e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e04426;MF=shile.zhang@linux.alibaba.com;NM=1;PH=DS;RN=5;SR=0;TI=SMTPD_---0TniHf-n_1578992062; Received: from ali-6c96cfdd1403.local(mailfrom:shile.zhang@linux.alibaba.com fp:SMTPD_---0TniHf-n_1578992062) by smtp.aliyun-inc.com(127.0.0.1); Tue, 14 Jan 2020 16:54:23 +0800 Subject: Re: [PATCH RESEND] mm: fix tick_sched timer blocked by pgdat_resize_lock To: Kirill Tkhai , Andrew Morton , Pavel Tatashin Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org References: <20200110082510.172517-2-shile.zhang@linux.alibaba.com> <20200110093053.34777-1-shile.zhang@linux.alibaba.com> <1ee6088c-9e72-8824-3a9a-fc099d196faf@virtuozzo.com> <9420eab3-5e5e-150f-53c9-6cd40bacf859@virtuozzo.com> From: Shile Zhang Message-ID: Date: Tue, 14 Jan 2020 16:54:22 +0800 User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.14; rv:60.0) Gecko/20100101 Thunderbird/60.9.1 MIME-Version: 1.0 In-Reply-To: <9420eab3-5e5e-150f-53c9-6cd40bacf859@virtuozzo.com> Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 8bit Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 2020/1/13 16:11, Kirill Tkhai wrote: > On 13.01.2020 03:54, Shile Zhang wrote: >> >> On 2020/1/10 19:42, Kirill Tkhai wrote: >>> On 10.01.2020 12:30, Shile Zhang wrote: >>>> When 'CONFIG_DEFERRED_STRUCT_PAGE_INIT' is set, 'pgdat_resize_lock' >>>> will be called inside 'pgdatinit' kthread to initialise the deferred >>>> pages with local interrupts disabled. Which is introduced by >>>> commit 3a2d7fa8a3d5 ("mm: disable interrupts while initializing deferred >>>> pages"). >>>> >>>> But 'pgdatinit' kthread is possible be pined on the boot CPU (CPU#0 by >>>> default), especially in small system with NRCPUS <= 2. In this case, the >>>> interrupts are disabled on boot CPU during memory initialising, which >>>> caused the tick_sched timer be blocked, leading to wall clock stuck. >>>> >>>> Fixes: commit 3a2d7fa8a3d5 ("mm: disable interrupts while initializing >>>> deferred pages") >>>> >>>> Signed-off-by: Shile Zhang >>>> --- >>>>   include/linux/memory_hotplug.h | 16 ++++++++++++++-- >>>>   1 file changed, 14 insertions(+), 2 deletions(-) >>>> >>>> diff --git a/include/linux/memory_hotplug.h b/include/linux/memory_hotplug.h >>>> index ba0dca6aac6e..be69a6dc4fee 100644 >>>> --- a/include/linux/memory_hotplug.h >>>> +++ b/include/linux/memory_hotplug.h >>>> @@ -6,6 +6,8 @@ >>>>   #include >>>>   #include >>>>   #include >>>> +#include >>>> +#include >>>>     struct page; >>>>   struct zone; >>>> @@ -282,12 +284,22 @@ static inline bool movable_node_is_enabled(void) >>>>   static inline >>>>   void pgdat_resize_lock(struct pglist_data *pgdat, unsigned long *flags) >>>>   { >>>> -    spin_lock_irqsave(&pgdat->node_size_lock, *flags); >>>> +    /* >>>> +     * Disable local interrupts on boot CPU will stop the tick_sched >>>> +     * timer, which will block jiffies(wall clock) update. >>>> +     */ >>>> +    if (current->cpu != get_boot_cpu_id()) >>>> +        spin_lock_irqsave(&pgdat->node_size_lock, *flags); >>>> +    else >>>> +        spin_lock(&pgdat->node_size_lock); >>>>   } >>>>   static inline >>>>   void pgdat_resize_unlock(struct pglist_data *pgdat, unsigned long *flags) >>>>   { >>>> -    spin_unlock_irqrestore(&pgdat->node_size_lock, *flags); >>>> +    if (current->cpu != get_boot_cpu_id()) >>>> +        spin_unlock_irqrestore(&pgdat->node_size_lock, *flags); >>>> +    else >>>> +        spin_unlock(&pgdat->node_size_lock); >>>>   } >>>>   static inline >>>>   void pgdat_resize_init(struct pglist_data *pgdat) >>> 1)Linux kernel is *preemptible*. Kernel with CONFIG_PREEMPT_RT option even may preempt >>> *kernel* code in the middle of function. When you are executing a code containing >>> pgdat_resize_lock() and pgdat_resize_unlock(), the process may migrate to another cpu >>> between them. >>> >>> bool cpu               another cpu >>> ---------------------------------- >>> pgdat_resize_lock() >>>    spin_lock() >>>    --> migrate to another cpu >>>                        pgdat_resize_unlock() >>>                        spin_unlock_irqrestore() >>> >>> (Yes, in case of CONFIG_PREEMPT_RT, process is preemptible even after spin_lock() call). >>> >>> This looks like a bad helpers, and we should not introduce such the design. >> Hi Kirill, >> >> Thanks for your comments! >> Sorry for I'm not very clear about this lock/unlock, but I encountered this issue >> with "CONFIG_PREEMPT is not set". > The thing is we simply shouldn't introduce such the primitives since the thread > may migrate to another cpu, while you own the lock. This looks like a buggy design. > >>> 2)I think there is no the problem this patch solves. Do we really this statistics? >>> Can't we simple remove print message from deferred_init_memmap() and solve this? >> Sorry for I've not put this issue very clearly. It's *not* just one statistics log >> with wrong time calculate, but the wall clock is stuck. >> So the 'systemd-analyze' command also give a wrong time as I mentioned in the cover >> letter. I don't think is OK just remove the log, it cannot solve the wall clock latency. > Have you tried temporary enabling interrupts in the middle of cycle after a huge enough > memory block is initialized? Something like: > > deferred_init_memmap() > { > while (spfn < epfn) { > nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn); > local_irq_enable(); > local_irq_disable(); > } > } Yes, I'd tried this for issue confirm, before I sent this patch. Likes I mentioned the debug log in cover letter, I also add mdelay between local_irq_enable/disable, this system jiffies is stuck without update. So I think there must be problem to use spin_lock_irqsave in early boot path on boot CPU. > Or, maybe, enable/disable interrupts somewhere inside deferred_init_maxorder(). > >>> Also, you may try to check that sched_clock() gives better results with interrupts >>> disabled (on x86 it uses rdtsc, when it's possible. But it also may fallback to >>> jiffies-based clock in some hardware cases, and they also won't go with interrupts >>> disabled).