[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <9f5409d1-82c5-459c-a9cc-416e7320e044@paulmck-laptop>
Date: Tue, 26 Nov 2024 08:17:49 -0800
From: "Paul E. McKenney" <paulmck@...nel.org>
To: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc: kernel test robot <oliver.sang@...el.com>,
Peter Zijlstra <peterz@...radead.org>, oe-lkp@...ts.linux.dev,
lkp@...el.com, linux-kernel@...r.kernel.org, rcu@...r.kernel.org
Subject: Re: [linus:master] [sched, x86] 476e8583ca:
WARNING:at_kernel/rcu/update.c:#torture_sched_setaffinity
On Tue, Nov 26, 2024 at 03:30:24PM +0100, Sebastian Andrzej Siewior wrote:
> On 2024-11-25 22:29:09 [+0800], kernel test robot wrote:
> > Hello,
> Hi,
>
> > by this commit, we see the config has below diff:
> >
> > --- /pkg/linux/x86_64-randconfig-161-20241120/gcc-12/35772d627b55cc7fb4f33bae57c564a25b3121a9/.config 2024-11-22 17:03:32.458344665 +0800
> > +++ /pkg/linux/x86_64-randconfig-161-20241120/gcc-12/476e8583ca16eecec0a3a28b6ee7130f4e369389/.config 2024-11-22 17:02:59.440805587 +0800
> > @@ -121,9 +121,11 @@ CONFIG_BPF_UNPRIV_DEFAULT_OFF=y
> > # end of BPF subsystem
> >
> > CONFIG_PREEMPT_BUILD=y
> > -CONFIG_PREEMPT_NONE=y
> > +CONFIG_ARCH_HAS_PREEMPT_LAZY=y
> > +# CONFIG_PREEMPT_NONE is not set
> > # CONFIG_PREEMPT_VOLUNTARY is not set
> > # CONFIG_PREEMPT is not set
> > +CONFIG_PREEMPT_LAZY=y
> > # CONFIG_PREEMPT_RT is not set
> > CONFIG_PREEMPT_COUNT=y
> > CONFIG_PREEMPTION=y
> >
> …
> > commit: 476e8583ca16eecec0a3a28b6ee7130f4e369389 ("sched, x86: Enable Lazy preemption")
> …
> > runtime: 300s
> > test: cpuhotplug
> > torture_type: trivial
> …
>
> > [ 150.797530][ T445] ------------[ cut here ]------------
> > [ 150.797915][ T445] torture_sched_setaffinity: sched_setaffinity(445) returned -22
> > [ 150.798353][ T445] WARNING: CPU: 0 PID: 445 at kernel/rcu/update.c:535 torture_sched_setaffinity (kernel/rcu/update.c:535 (discriminator 3))
>
> I've been staring at this, and this is actually fine. Your config changes
> from CONFIG_PREEMPT_NONE to CONFIG_PREEMPT_LAZY which implies
> CONFIG_PREEMPTION. The trivial RCU test there does sched_setaffinity()
> while preemption is enabled and CPU-hotplug runs in the background. So
> you get what you expect either by an attempt to move to a CPU which is
> no longer valid or by getting migrated to another CPU in the middle of
> your operation.
>
> This is all fine. You need to update your config file or your test.
Good point, and I suspect that there will be a few rcutorture scenario
adjustments coming along shortly after -rc1 comes out.
Thanx, Paul
Powered by blists - more mailing lists