Re: [patch 02/13] workqueue: Provide work_on_cpu_safe()

From: Peter Zijlstra <peterz@infradead.org>
To: Thomas Gleixner <tglx@linutronix.de>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Ingo Molnar <mingo@kernel.org>,
	Sebastian Siewior <bigeasy@linutronix.de>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	"David S. Miller" <davem@davemloft.net>,
	Fenghua Yu <fenghua.yu@intel.com>,
	Herbert Xu <herbert@gondor.apana.org.au>,
	Lai Jiangshan <jiangshanlai@gmail.com>,
	Len Brown <lenb@kernel.org>,
	Michael Ellerman <mpe@ellerman.id.au>,
	"Rafael J. Wysocki" <rjw@rjwysocki.net>,
	Tejun Heo <tj@kernel.org>, Tony Luck <tony.luck@intel.com>,
	Viresh Kumar <viresh.kumar@linaro.org>
Subject: Re: [patch 02/13] workqueue: Provide work_on_cpu_safe()
Date: Fri, 14 Apr 2017 10:54:59 +0200	[thread overview]
Message-ID: <20170414085459.sp3psluw647ab4lw@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <20170412201042.262610721@linutronix.de>

On Wed, Apr 12, 2017 at 10:07:28PM +0200, Thomas Gleixner wrote:
> +long work_on_cpu_safe(int cpu, long (*fn)(void *), void *arg)
> +{
> +	long ret = -ENODEV;
> +
> +	get_online_cpus();
> +	if (cpu_online(cpu))
> +		ret = work_on_cpu(cpu, fn, arg);
> +	put_online_cpus();
> +	return ret;
> +}

But doesn't workqueue have this lovelt 'feature' where it will unbind
per-cpu work and run it on random CPUs when hotplug happens?

That is, I think you need a flush_work() before put_online_cpus() if you
want to guarantee anything.