linux-kernel - Re: [GIT PULL] sched_ext: Initial pull request for v6.11

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 for Android: free password hash cracker in your pocket

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

Message-ID: <ZqmPVUw8htIwkvRb@slm.duckdns.org>
Date: Tue, 30 Jul 2024 15:11:49 -1000
From: Tejun Heo <tj@...nel.org>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Qais Yousef <qyousef@...alina.io>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	linux-kernel@...r.kernel.org, David Vernet <void@...ifault.com>,
	Ingo Molnar <mingo@...hat.com>, Alexei Starovoitov <ast@...nel.org>,
	Vincent Guittot <vincent.guittot@...aro.org>
Subject: Re: [GIT PULL] sched_ext: Initial pull request for v6.11

Hello,

On Tue, Jul 30, 2024 at 11:04:43AM +0200, Peter Zijlstra wrote:
> >  #ifdef CONFIG_NO_HZ_COMMON
> > -static bool sugov_cpu_is_busy(struct sugov_cpu *sg_cpu)
> > +static bool sugov_hold_freq(struct sugov_cpu *sg_cpu)
> >  {
> > -	unsigned long idle_calls = tick_nohz_get_idle_calls_cpu(sg_cpu->cpu);
> > -	bool ret = idle_calls == sg_cpu->saved_idle_calls;
> > +	unsigned long idle_calls;
> > +	bool ret;
> > +
> > +	/*
> > +	 * The heuristics in this function is for the fair class. For SCX, the
> > +	 * performance target comes directly from the BPF scheduler. Let's just
> > +	 * follow it.
> > +	 */
> > +	if (scx_switched_all())
> > +		return false;
> 
> This one does seem really weird. It makes schedutil behave significantly
> different from the BPF pov depending on if you have this partial mode on
> or not.
> 
> So I would really like this to be reconsidered as I agree with Qais,
> things should be consistent.

I replied in the other thread and Vincent raised it too. To reiterate, when
switched_all(), if we want to keep accumulating util signal from the fair
class, we need to keep calling fair's update_blocked_averages() so that the
value can decay. We can but it seems silly to keep calling it to decay it to
zero when we know it's becoming and staying zero.

Thanks.

-- 
tejun