linux-kernel - Re: [PATCH V7] powercap/drivers/idle_injection: Add an idle injection framework

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20180618103844.t2minxwc7tiu2qgd@vireshk-i7>
Date:   Mon, 18 Jun 2018 16:08:44 +0530
From:   Viresh Kumar <viresh.kumar@...aro.org>
To:     Daniel Lezcano <daniel.lezcano@...aro.org>
Cc:     rjw@...ysocki.net, linux-kernel@...r.kernel.org,
        linux-pm@...r.kernel.org, Eduardo Valentin <edubezval@...il.com>,
        Javi Merino <javi.merino@...nel.org>,
        Leo Yan <leo.yan@...aro.org>,
        Kevin Wangtao <kevin.wangtao@...aro.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Rui Zhang <rui.zhang@...el.com>,
        Daniel Thompson <daniel.thompson@...aro.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Andrea Parri <andrea.parri@...rulasolutions.com>
Subject: Re: [PATCH V7] powercap/drivers/idle_injection: Add an idle
 injection framework

On 18-06-18, 12:35, Daniel Lezcano wrote:
> On 18/06/2018 12:22, Viresh Kumar wrote:
> > On 15-06-18, 11:19, Daniel Lezcano wrote:
> >> +/**
> >> + * idle_injection_stop - stops the idle injections
> >> + * @ii_dev: a pointer to an idle injection_device structure
> >> + *
> >> + * The function stops the idle injection and waits for the threads to
> >> + * complete. If we are in the process of injecting an idle cycle, then
> >> + * this will wait the end of the cycle.
> >> + *
> >> + * When the function returns there is no more idle injection
> >> + * activity. The kthreads are scheduled out and the periodic timer is
> >> + * off.
> >> + */
> >> +void idle_injection_stop(struct idle_injection_device *ii_dev)
> >> +{
> >> +	struct idle_injection_thread *iit;
> >> +	unsigned int cpu;
> >> +
> >> +	pr_debug("Stopping injecting idle cycles on CPUs '%*pbl'\n",
> >> +		 cpumask_pr_args(to_cpumask(ii_dev->cpumask)));
> >> +
> >> +	hrtimer_cancel(&ii_dev->timer);
> >> +
> >> +	/*
> >> +	 * We want the guarantee we have a quescient point where
> >> +	 * parked threads stay in there state while we are stopping
> >> +	 * the idle injection. After exiting the loop, if any CPU is
> >> +	 * plugged in, the 'should_run' boolean being false, the
> >> +	 * smpboot main loop schedules the task out.
> >> +	 */
> >> +	cpu_hotplug_disable();
> >> +
> >> +	for_each_cpu_and(cpu, to_cpumask(ii_dev->cpumask), cpu_online_mask) {
> > 
> > Maybe you should do below for all CPUs in the mask. Is the below usecase
> > possible ?
> > 
> > - CPU0-4 are part of the mask and are all online.
> > - hrtimer fires and sets should_run for all of them to 1.
> 
>     ^^
> hrtimer_cancel gives you the guarantee, the timer is no longer active
> and there is no execution in the timer handler. So the timer can no
> longer fire after hrtimer_cancel() is called (which is a blocking call).

Right but that isn't called yet in my sequence.

> > - Right at this time CPU3 goes offline, so the thread gets parked with
> >   should_run == 1. Is there a reason why this can't happen ?
> > - Now we unregister the stuff and CPU3 again comes online.

It gets called here from unregister/stop.

> > - Because it had should_run as true, we again run the thread and Crash.
> > 
> > makes sense ?

> >> +out_rollback_per_cpu:
> >> +	for_each_cpu(cpu, to_cpumask(ii_dev->cpumask))
> >> +		per_cpu(idle_injection_device, cpu) = NULL;
> > 
> > So if two parts of the kernel call this routine with the same cpumask, then the
> > second call will also overwrite the masks with NULL and return error. That will
> > screw up things a bit here.
> 
> Apparently there is a misunderstanding :)
> 
> https://lkml.org/lkml/2018/5/29/209 (at the end)

Right, your earlier version was doing the right thing :)

-- 
viresh