linux-kernel - Re: sched: Reenable interrupts in do sched

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] [day] [month] [year] [list]

Message-ID: <20201223120207.csiynjfmzavveesf@e107158-lin>
Date:   Wed, 23 Dec 2020 12:02:07 +0000
From:   Qais Yousef <qais.yousef@....com>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        LKML <linux-kernel@...r.kernel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Ingo Molnar <mingo@...nel.org>,
        Juri Lelli <juri.lelli@...hat.com>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Dietmar Eggemann <dietmar.eggemann@....com>,
        Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>,
        Daniel Bristot de Oliveira <bristot@...hat.com>
Subject: Re: sched: Reenable interrupts in do sched_yield()

On 10/21/20 10:07, Steven Rostedt wrote:
> On Wed, 21 Oct 2020 09:27:22 +0200
> Thomas Gleixner <tglx@...utronix.de> wrote:
> 
> > On Tue, Oct 20 2020 at 16:07, Steven Rostedt wrote:
> > > On Tue, 20 Oct 2020 20:02:55 +0200
> > > Thomas Gleixner <tglx@...utronix.de> wrote:
> > > What I wrote wasn't exactly what I meant. What I meant to have:
> > >
> > > 	/*
> > > 	 * Since we are going to call schedule() anyways, there's
> > > 	 * no need to do the preemption check when the rq_lock is released.
> > > 	 */
> > >
> > > That is, to document why we have the preempt_disable() before the unlock:  
> > 
> > which is pretty obvious, but I let Peter decide on that.
> 
> To us maybe, but I like to have comments that explain why things are done to
> average people. ;-)
> 
> If I go to another kernel developer outside the core kernel, would they know
> why there's a preempt_disable() there?
> 
> 
>  	preempt_disable();
> 	rq_unlock_irq(rq, &rf);
>  	sched_preempt_enable_no_resched();
>  
>  	schedule();
> 
> 
> Not everyone knows that the rq_unlock_irq() would trigger a schedule if an
> interrupt happened as soon as irqs were enabled again and need_resched was
> set.

Sorry a bit late to the party.

Personally, what actually is tripping me off is that rq_unlock_irq() will end
up calling preempt_enable(), and then we do sched_preempt_enable_no_resched().
Was there an earlier preempt_disable() called up in the chain that I couldn't
figure out that's why it's okay to do the 2? Otherwise I see we have imbalanced
preempt_disable/enable.

	preempt_disable()
	rq_unlock_irq()
		__raw_spin_unlock_irq()
			local_irq_enable()
			preempt_enable()	// first preempt_count_dec()
	sched_preempt_enable_no_resched()	// second preempt_count_dec()

Thanks

--
Qais Yousef