Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230613082012.49615-1-jiahao.os@bytedance.com>
 <20230613082012.49615-2-jiahao.os@bytedance.com> <c06dd33c-bc91-e0a9-096c-d135ce4fedcb@bytedance.com>
In-Reply-To: <c06dd33c-bc91-e0a9-096c-d135ce4fedcb@bytedance.com>
From:   Vincent Guittot <vincent.guittot@linaro.org>
Date:   Tue, 13 Jun 2023 11:13:49 +0200
Message-ID: <CAKfTPtCDyD3SLR38uH07SFaqAeMyvwgD8pCG+DBG4CC1PeYSYg@mail.gmail.com>
Subject: Re: [PATCH v5 1/4] sched/core: Fixed missing rq clock update before
 calling set_rq_offline()
To:     Hao Jia <jiahao.os@bytedance.com>
Cc:     mingo@redhat.com, peterz@infradead.org, mingo@kernel.org,
        juri.lelli@redhat.com, dietmar.eggemann@arm.com,
        rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
        bristot@redhat.com, vschneid@redhat.com,
        mgorman@techsingularity.net, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, 13 Jun 2023 at 10:31, Hao Jia <jiahao.os@bytedance.com> wrote:
>
>
>
> On 2023/6/13 Hao Jia wrote:
> > This is triggered during cpu offline when CONFIG_CPU_FREQ is enabled
> > and cpufreq is set to powersave:
> > ------------[ cut here ]------------
> > rq->clock_update_flags < RQCF_ACT_SKIP
> > WARNING: CPU: 24 PID: 754 at kernel/sched/sched.h:1496
> > enqueue_top_rt_rq+0x139/0x160
> > Call Trace:
> >   <TASK>
> >   ? intel_pstate_update_util+0x3b0/0x3b0
> >   rq_offline_rt+0x1b7/0x250
> >   set_rq_offline.part.120+0x28/0x60
> >   rq_attach_root+0xc4/0xd0
> >   cpu_attach_domain+0x3dc/0x7f0
> >   ? __schedule+0x65e/0x1310
> >   partition_sched_domains_locked+0x2a5/0x3c0
> >   rebuild_sched_domains_locked+0x477/0x830
> >   ? percpu_rwsem_wait+0x140/0x140
> >   rebuild_sched_domains+0x1b/0x30
> >   cpuset_hotplug_workfn+0x2ca/0xc90
> >   ? balance_push+0x56/0x120
> >   ? _raw_spin_unlock+0x15/0x30
> >   ? finish_task_switch+0x98/0x2f0
> >   ? __switch_to+0x116/0x410
> >   ? __schedule+0x65e/0x1310 ? internal_add_timer+0x42/0x60
> >   ? _raw_spin_unlock_irqrestore+0x23/0x40
> >   ? add_timer_on+0xd5/0x130
> >   process_one_work+0x1bc/0x3d0
> >   worker_thread+0x4c/0x380
> >   ? preempt_count_add+0x56/0xa0
> >   ? rescuer_thread+0x310/0x310
> >   kthread+0xe6/0x110
> >   ? kthread_complete_and_exit+0x20/0x20
> >   ret_from_fork+0x1f/0x30
> >
> > More detailed key function call graph:
> > rq_offline_rt()
> >    __disable_runtime()
> >      sched_rt_rq_enqueue()
> >        enqueue_top_rt_rq()
> >          cpufreq_update_util() <-- depends on CONFIG_CPU_FREQ
> >            data->func(data, *rq_clock(rq)*, flags)
> >              intel_pstate_update_util() <-- powersave policy callback function
> >
> > Before calling rq_offline_rt() we need to update the rq clock to avoid
> > using the old rq clock, So we add update_rq_clock() to set_rq_offline()
> > to update rq clock. And we use rq_lock_irqsave()/rq_unlock_irqrestore()
> > to replace raw_spin_rq_lock_irqsave()/raw_spin_rq_unlock_irqrestore()
> > in rq_attach_root() to ensure that rq->clock_update_flags are cleared
> > before updating the rq clock.
> >
> > Steps to reproduce:
> > 1. Enable CONFIG_SMP and CONFIG_CPU_FREQ when compiling the kernel
> > 2. echo 1 > /sys/kernel/debug/clear_warn_once
> > 3. cpupower -c all frequency-set -g powersave
> > 4. Run some rt tasks e.g. Create 5*n rt (100% running) tasks (on a
> >     system with n CPUs)
> > 5. Offline cpu one by one until the warninng is triggered
> >
> > Suggested-by: Ben Segall <bsegall@google.com>
> > Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
>
> Hi Vincent,
>
>    Benjamin Segall suggested adding update_rq_clock() to set_rq_offline()
> instead of calling update_rq_clock() before set_rq_offline() runs.
> The code of patch1 has changed, so I removed your tag "Reviewed-by".
> Please review again.

v5 looks good to me

Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>

>
> Thanks,
> Hao
>
> > ---
> >   kernel/sched/core.c     | 2 +-
> >   kernel/sched/topology.c | 6 +++---
> >   2 files changed, 4 insertions(+), 4 deletions(-)
> >
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index a68d1276bab0..a8be5415daba 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -9548,6 +9548,7 @@ void set_rq_offline(struct rq *rq)
> >       if (rq->online) {
> >               const struct sched_class *class;
> >
> > +             update_rq_clock(rq);
> >               for_each_class(class) {
> >                       if (class->rq_offline)
> >                               class->rq_offline(rq);
> > @@ -9689,7 +9690,6 @@ int sched_cpu_deactivate(unsigned int cpu)
> >
> >       rq_lock_irqsave(rq, &rf);
> >       if (rq->rd) {
> > -             update_rq_clock(rq);
> >               BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
> >               set_rq_offline(rq);
> >       }
> > diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> > index 6682535e37c8..52976eadfee9 100644
> > --- a/kernel/sched/topology.c
> > +++ b/kernel/sched/topology.c
> > @@ -487,9 +487,9 @@ static void free_rootdomain(struct rcu_head *rcu)
> >   void rq_attach_root(struct rq *rq, struct root_domain *rd)
> >   {
> >       struct root_domain *old_rd = NULL;
> > -     unsigned long flags;
> > +     struct rq_flags rf;
> >
> > -     raw_spin_rq_lock_irqsave(rq, flags);
> > +     rq_lock_irqsave(rq, &rf);
> >
> >       if (rq->rd) {
> >               old_rd = rq->rd;
> > @@ -515,7 +515,7 @@ void rq_attach_root(struct rq *rq, struct root_domain *rd)
> >       if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
> >               set_rq_online(rq);
> >
> > -     raw_spin_rq_unlock_irqrestore(rq, flags);
> > +     rq_unlock_irqrestore(rq, &rf);
> >
> >       if (old_rd)
> >               call_rcu(&old_rd->rcu, free_rootdomain);