Re: [RT] BUG in sched/cpupri.c

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 21.12.21 17:45, John Keeping wrote:
> On Tue, 21 Dec 2021 16:11:34 +0000
> Valentin Schneider <valentin.schneider@arm.com> wrote:
> 
>> On 20/12/21 18:35, Dietmar Eggemann wrote:

[...]

>> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
>> index fd7c4f972aaf..7d61ceec1a3b 100644
>> --- a/kernel/sched/deadline.c
>> +++ b/kernel/sched/deadline.c
>> @@ -2467,10 +2467,13 @@ static void switched_from_dl(struct rq *rq, struct task_struct *p)
>>  	 * this is the right place to try to pull some other one
>>  	 * from an overloaded CPU, if any.
>>  	 */
>> -	if (!task_on_rq_queued(p) || rq->dl.dl_nr_running)
>> +	if (!task_on_rq_queued(p))
>>  		return;
>>  
>> -	deadline_queue_pull_task(rq);
>> +	if (!rq->dl.dl_nr_running)
>> +		deadline_queue_pull_task(rq);
>> +	else if (task_current(rq, p) && (p->sched_class < &dl_sched_class))
>> +		resched_curr(rq);
>>  }
>>  
>>  /*
>> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
>> index ef8228d19382..1ea2567612fb 100644
>> --- a/kernel/sched/rt.c
>> +++ b/kernel/sched/rt.c
>> @@ -2322,10 +2322,13 @@ static void switched_from_rt(struct rq *rq, struct task_struct *p)
>>  	 * we may need to handle the pulling of RT tasks
>>  	 * now.
>>  	 */
>> -	if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
>> +	if (!task_on_rq_queued(p))
>>  		return;
>>  
>> -	rt_queue_pull_task(rq);
>> +	if (!rq->rt.rt_nr_running)
>> +		rt_queue_pull_task(rq);
>> +	else if (task_current(rq, p) && (p->sched_class < &rt_sched_class))
>> +		resched_curr(rq);

switched_from_rt() -> rt_queue_pull_task(, pull_rt_task)
  pull_rt_task()->tell_cpu_to_push()->irq_work_queue_on(&rq->rd->rto_push_work,)
    rto_push_irq_work_func() -> push_rt_task(rq, true)

seems to be the only way with pull=true.

In my tests, rq->rt.rt_nr_running seems to be 0 when it happens.

[   22.288537] CPU3 switched_to_rt: p=[ksoftirqd/3 35]
[   22.288554] rt_mutex_setprio: CPU3 p=[ksoftirqd/3 35] pi_task=[rcu_preempt 11] queued=1 running=0 prio=98 oldprio=120
[   22.288636] CPU3 switched_from_rt: p=[ksoftirqd/3 35] rq->rt.rt_nr_running=0
                                                         ^^^^^^^^^^^^^^^^^^^^^^ 
[   22.288649] rt_mutex_setprio: CPU3 p=[ksoftirqd/3 35] queued=1 running=1 prio=120 oldprio=98
[   22.288681] CPU3 push_rt_task: next_task=[rcu_preempt 11] migr_dis=1 rq->curr=[ksoftirqd/3 35] pull=1
                                                             ^^^^^^^^^^                           ^^^^^^ 
[   22.288698] CPU: 3 PID: 35 Comm: ksoftirqd/3 Not tainted 5.15.10-rt24-dirty #36
[   22.288711] Hardware name: ARM Juno development board (r0) (DT)
[   22.288718] Call trace:
[   22.288722]  dump_backtrace+0x0/0x1ac
[   22.288747]  show_stack+0x1c/0x70
[   22.288763]  dump_stack_lvl+0x68/0x84
[   22.288777]  dump_stack+0x1c/0x38
[   22.288788]  push_rt_task.part.0+0x364/0x370
[   22.288805]  rto_push_irq_work_func+0x180/0x190
[   22.288821]  irq_work_single+0x34/0xa0
[   22.288836]  flush_smp_call_function_queue+0x138/0x244
[   22.288852]  generic_smp_call_function_single_interrupt+0x18/0x24
[   22.288867]  ipi_handler+0xb0/0x15c
...

What about slightly changing the layout in switched_from_rt() (only lightly tested):


@@ -2322,7 +2338,15 @@ static void switched_from_rt(struct rq *rq, struct task_struct *p)
         * we may need to handle the pulling of RT tasks
         * now.
         */
-       if (!task_on_rq_queued(p) || rq->rt.rt_nr_running)
+       if (!task_on_rq_queued(p))
+               return;
+
+       if (task_current(rq, p) && (p->sched_class < &rt_sched_class)) {
+               resched_curr(rq);
+               return;
+       }
+
+       if (rq->rt.rt_nr_running)
                return;
 
        rt_queue_pull_task(rq);