linux-kernel - Re: [PATCH V3 1/4] Sched: Scheduler time slice extension

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: <20250502152753.aFffUGVv@linutronix.de>
Date: Fri, 2 May 2025 17:27:53 +0200
From: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To: Peter Zijlstra <peterz@...radead.org>
Cc: Prakash Sangappa <prakash.sangappa@...cle.com>,
	linux-kernel@...r.kernel.org, rostedt@...dmis.org,
	mathieu.desnoyers@...icios.com, tglx@...utronix.de,
	kprateek.nayak@....com
Subject: Re: [PATCH V3 1/4] Sched: Scheduler time slice extension

On 2025-05-02 16:35:44 [+0200], Peter Zijlstra wrote:
> On Fri, May 02, 2025 at 02:34:33PM +0200, Sebastian Andrzej Siewior wrote:
> > On 2025-05-02 01:59:52 [+0000], Prakash Sangappa wrote:
> > > diff --git a/kernel/entry/common.c b/kernel/entry/common.c
> > > index 20154572ede9..b26adccb32df 100644
> > > --- a/kernel/entry/common.c
> > > +++ b/kernel/entry/common.c
> > > @@ -98,8 +99,12 @@ __always_inline unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
> > >  
> > >  		local_irq_enable_exit_to_user(ti_work);
> > >  
> > > -		if (ti_work & (_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY))
> > > -			schedule();
> > > +		if (ti_work & (_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY)) {
> > 
> > I asked to limit this to LAZY only.
> 
> No -- this should work irrespective of the preemption model chosen.

Focusing on LAZY would easily exclude the tasks with priorities.

> It should also have a default time that is shorter than the scheduling
> delay normally observed.
> 
> It should probably also have a PR_WARN on raising the sysctl value.
> 
> I know you worry about RT, but show me a trace where it is a problem.

The default extends the wakeup by 50us. With a 250us period this extends
the wakeup in worst case by 20% of the period. With 1ms it gets just to
5% of the whole period. You basically expect that everything is well
timed and this delay does not hurt anyone.
This delays interrupt driven wakeups (timer, hardware) and not every
wake up as I assumed initially so it is not as bad but still worse than
in has to be.

On top of that sched(7) says:
|   SCHED_FIFO: First in-first out scheduling
|      SCHED_FIFO  can be used only with static priorities higher than 0, which
|      means that when a SCHED_FIFO thread becomes runnable, it will always
|      immediately preempt any currently running SCHED_OTHER, SCHED_BATCH, or
|      SCHED_IDLE thread.  SCHED_FIFO is a simple scheduling algorithm without
|      time slicing.  For threads scheduled under the SCHED_FIFO  policy,  the
…
|   SCHED_DEADLINE: Sporadic task model deadline scheduling
…
|      if any SCHED_DEADLINE thread is runnable, it will preempt any thread
|      scheduled under
…

With this change, this "immediately preempt any currently running" is no
longer true.

While we could continue to argue how much the delay really is, I don't
understand why we can't exclude real time scheduling policy from this.

Sebastian