Re: [PATCH] Revert mm/vmstat.c: fix vmstat_update() preemption BUG

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To: Steven Rostedt <rostedt@goodmis.org>
Cc: Tejun Heo <htejun@gmail.com>, Vlastimil Babka <vbabka@suse.cz>,
	linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	tglx@linutronix.de, "Steven J . Hill" <steven.hill@cavium.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Christoph Lameter <cl@linux.com>
Subject: Re: [PATCH] Revert mm/vmstat.c: fix vmstat_update() preemption BUG
Date: Wed, 27 Jun 2018 14:47:06 +0200	[thread overview]
Message-ID: <20180627124706.g33vlaku7xa4veqq@linutronix.de> (raw)
In-Reply-To: <20180627123657.2hb7ow4szjyhg5aj@home.goodmis.org>

On 2018-06-27 08:36:57 [-0400], Steven Rostedt wrote:
> On Wed, Apr 11, 2018 at 09:07:30PM +0200, Sebastian Andrzej Siewior wrote:
> > 
> > This already happens:
> > - vmstat_shepherd() does get_online_cpus() and within this block it does
> >   queue_delayed_work_on(). So this has to wait until cpuhotplug
> >   completed before it can schedule something and then it won't schedule
> >   anything on the "off" CPU.
> 
> But can't we have something like this happen: ?
> 
> 	CPU0			CPU1			CPU2
> 	----			----			----
>  get_online_cpus()
>  queue_work(vmstat_update, cpu1)
>     wakeup(kworker/1)
> 			     High prio task running
>  put_online_cpus()
>  						     Shutdown CPU 1
> 			     migrate kworker/1
>  schedule kworker/1
>  (smp_processor_id() != 1)

I don't get CPU1 doing in here. kworker/1 should not be migrated to
begin with. The work item should be done before CPU2 shutdowns CPU1 (or
0) because vmstat_cpu_down_prep() cancels the work while the CPUs goes
down so it should not be migrated. Also, the work is enqueued while
holding the CPU HP lock.

Sebastian