Re: [PATCH] Revert mm/vmstat.c: fix vmstat_update() preemption BUG
From: Sebastian Andrzej Siewior
Date: Wed Apr 11 2018 - 10:42:33 EST
On 2018-04-11 07:09:13 [-0700], Tejun Heo wrote:
> Hello,
>
> On Wed, Apr 11, 2018 at 03:56:43PM +0200, Vlastimil Babka wrote:
> > > vmstat_update() is invoked by a kworker on a specific CPU. This worker
> > > it bound to this CPU. The name of the worker was "kworker/1:1" so it
> > > should have been a worker which was bound to CPU1. A worker which can
> > > run on any CPU would have a `u' before the first digit.
> >
> > Oh my, and I have just been assured by Tejun that his cannot happen :)
> > And yet, in the original report [1] I see:
> >
> > CPU: 0 PID: 269 Comm: kworker/1:1 Not tainted
> >
> > So is this perhaps related to the cpu hotplug that [1] mentions? e.g. is
> > the cpu being hotplugged cpu 1, the worker started too early before
> > stuff can be scheduled on the CPU, so it has to run on different than
> > designated CPU?
> >
> > [1] https://marc.info/?l=linux-mm&m=152088260625433&w=2
>
> The report says that it happens when hotplug is attempted. Per-cpu
> doesn't pin the cpu alive, so if the cpu goes down while a work item
> is in flight or a work item is queued while a cpu is offline it'll end
> up executing on some other cpu. So, if a piece of code doesn't want
> that happening, it gotta interlock itself - ie. start queueing when
> the cpu comes online and flush and prevent further queueing when its
> cpu goes down.
I missed that cpuhotplug part while reading it. So in that case, let me
add a CPU-hotplug notifier which cancels that work. After all it is not
need once the CPU is gone.
> Thanks.
>
Sebastian