diff mbox series

[v5,2/4] sched/core: Avoid double calling update_rq_clock() in __balance_push_cpu_stop()

Message ID 20230613082012.49615-3-jiahao.os@bytedance.com
State New
Headers
Series Fix some warnings about rq clock |

Commit Message

Hao Jia June 13, 2023, 8:20 a.m. UTC 
  The WARN_DOUBLE_CLOCK warning is triggered during cpu offline.
------------[ cut here ]------------
rq->clock_update_flags & RQCF_UPDATED
WARNING: CPU: 17 PID: 138 at kernel/sched/core.c:741
update_rq_clock+0xaf/0x180
Call Trace:
 <TASK>
 __balance_push_cpu_stop+0x146/0x180
 ? migration_cpu_stop+0x2a0/0x2a0
 cpu_stopper_thread+0xa3/0x140
 smpboot_thread_fn+0x14f/0x210
 ? sort_range+0x20/0x20
 kthread+0xe6/0x110
 ? kthread_complete_and_exit+0x20/0x20
 ret_from_fork+0x1f/0x30

To avoid this warning, we remove update_rq_clock() from
the __migrate_task() function. And in order to avoid
missing rq clock update, add update_rq_clock() call before
migration_cpu_stop() calls __migrate_task().

Signed-off-by: Hao Jia <jiahao.os@bytedance.com>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/core.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

Comments

Peter Zijlstra June 15, 2023, 12:39 p.m. UTC | #1 
On Tue, Jun 13, 2023 at 04:20:10PM +0800, Hao Jia wrote:
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index a8be5415daba..1eca36299d8b 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -2398,7 +2398,6 @@ static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
>  	if (!is_cpu_allowed(p, dest_cpu))
>  		return rq;
>  
> -	update_rq_clock(rq);
>  	rq = move_queued_task(rq, rf, p, dest_cpu);
>  
>  	return rq;
> @@ -2456,10 +2455,12 @@ static int migration_cpu_stop(void *data)
>  				goto out;
>  		}
>  
> -		if (task_on_rq_queued(p))
> +		if (task_on_rq_queued(p)) {
> +			update_rq_clock(rq);
>  			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
> -		else
> +		} else {
>  			p->wake_cpu = arg->dest_cpu;
> +		}
>  
>  		/*
>  		 * XXX __migrate_task() can fail, at which point we might end

So now you've got update_rq_clock() in both callers, why not remove it
from __balance_push_cpu_stop() ?

Afaict nothing actually needs it before __migrate_task().

---
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -9435,8 +9435,6 @@ static int __balance_push_cpu_stop(void
 	raw_spin_lock_irq(&p->pi_lock);
 	rq_lock(rq, &rf);
 
-	update_rq_clock(rq);
-
 	if (task_rq(p) == rq && task_on_rq_queued(p)) {
 		cpu = select_fallback_rq(rq->cpu, p);
 		rq = __migrate_task(rq, &rf, p, cpu);
Hao Jia June 16, 2023, 7:56 a.m. UTC | #2 
On 2023/6/15 Peter Zijlstra wrote:
> On Tue, Jun 13, 2023 at 04:20:10PM +0800, Hao Jia wrote:
>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>> index a8be5415daba..1eca36299d8b 100644
>> --- a/kernel/sched/core.c
>> +++ b/kernel/sched/core.c
>> @@ -2398,7 +2398,6 @@ static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
>>   	if (!is_cpu_allowed(p, dest_cpu))
>>   		return rq;
>>   
>> -	update_rq_clock(rq);
>>   	rq = move_queued_task(rq, rf, p, dest_cpu);
>>   
>>   	return rq;
>> @@ -2456,10 +2455,12 @@ static int migration_cpu_stop(void *data)
>>   				goto out;
>>   		}
>>   
>> -		if (task_on_rq_queued(p))
>> +		if (task_on_rq_queued(p)) {
>> +			update_rq_clock(rq);
>>   			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
>> -		else
>> +		} else {
>>   			p->wake_cpu = arg->dest_cpu;
>> +		}
>>   
>>   		/*
>>   		 * XXX __migrate_task() can fail, at which point we might end
> 
> So now you've got update_rq_clock() in both callers, why not remove it
> from __balance_push_cpu_stop() ?
> 
> Afaict nothing actually needs it before __migrate_task().
> 

Thanks for your review.
I'm afraid not, the rq clock also needs to be updated before
select_fallback_rq() is called.


> ---
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -9435,8 +9435,6 @@ static int __balance_push_cpu_stop(void
>   	raw_spin_lock_irq(&p->pi_lock);
>   	rq_lock(rq, &rf);
>   
> -	update_rq_clock(rq);
> -
>   	if (task_rq(p) == rq && task_on_rq_queued(p)) {
>   		cpu = select_fallback_rq(rq->cpu, p);
>   		rq = __migrate_task(rq, &rf, p, cpu);

If we just remove update_rq_clock() from __balance_push_cpu_stop(), we 
will get this warning.


[ 1260.960166] rq->clock_update_flags < RQCF_ACT_SKIP
[ 1260.960170] WARNING: CPU: 25 PID: 196 at kernel/sched/sched.h:1496 
update_curr+0xf6/0x1f0

[ 1260.960318] Call Trace:
[ 1260.960320]  <TASK>
[ 1260.960323]  ? show_regs+0x5b/0x60
[ 1260.960327]  ? __warn+0x89/0x140
[ 1260.960331]  ? update_curr+0xf6/0x1f0
[ 1260.960334]  ? report_bug+0x1b7/0x1e0
[ 1260.960338]  ? __wake_up_klogd.part.25+0x5a/0x80
[ 1260.960342]  ? handle_bug+0x45/0x80
[ 1260.960346]  ? exc_invalid_op+0x18/0x70
[ 1260.960348]  ? asm_exc_invalid_op+0x1b/0x20
[ 1260.960354]  ? update_curr+0xf6/0x1f0
[ 1260.960356]  ? update_curr+0xf6/0x1f0
[ 1260.960359]  dequeue_entity+0x3b/0x410
[ 1260.960361]  dequeue_task_fair+0xc7/0x3c0
[ 1260.960363]  dequeue_task+0x30/0xf0
[ 1260.960365]  __do_set_cpus_allowed+0x94/0x130
[ 1260.960366]  do_set_cpus_allowed+0x38/0x60
[ 1260.960368]  cpuset_cpus_allowed_fallback+0x70/0x80
[ 1260.960372]  select_fallback_rq+0x20f/0x250            <----
[ 1260.960374]  __balance_push_cpu_stop+0x13f/0x1a0
[ 1260.960377]  ? migration_cpu_stop+0x2b0/0x2b0
[ 1260.960379]  cpu_stopper_thread+0xaf/0x140
[ 1260.960382]  smpboot_thread_fn+0x158/0x220
[ 1260.960387]  ? sort_range+0x30/0x30
[ 1260.960390]  kthread+0xfe/0x130
[ 1260.960392]  ? kthread_complete_and_exit+0x20/0x20
[ 1260.960394]  ret_from_fork+0x1f/0x30
[ 1260.960399]  </TASK>

Thanks,
Hao
Peter Zijlstra June 16, 2023, 9:18 a.m. UTC | #3 
On Fri, Jun 16, 2023 at 03:56:25PM +0800, Hao Jia wrote:

> I'm afraid not, the rq clock also needs to be updated before
> select_fallback_rq() is called.

> If we just remove update_rq_clock() from __balance_push_cpu_stop(), we will
> get this warning.
> 
> 
> [ 1260.960166] rq->clock_update_flags < RQCF_ACT_SKIP
> [ 1260.960170] WARNING: CPU: 25 PID: 196 at kernel/sched/sched.h:1496
> update_curr+0xf6/0x1f0
> 
> [ 1260.960318] Call Trace:
> [ 1260.960320]  <TASK>
> [ 1260.960359]  dequeue_entity+0x3b/0x410
> [ 1260.960361]  dequeue_task_fair+0xc7/0x3c0
> [ 1260.960363]  dequeue_task+0x30/0xf0
> [ 1260.960365]  __do_set_cpus_allowed+0x94/0x130
> [ 1260.960366]  do_set_cpus_allowed+0x38/0x60
> [ 1260.960368]  cpuset_cpus_allowed_fallback+0x70/0x80
> [ 1260.960372]  select_fallback_rq+0x20f/0x250            <----

Urgh... :-(
diff mbox series

Patch

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index a8be5415daba..1eca36299d8b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2398,7 +2398,6 @@  static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
 	if (!is_cpu_allowed(p, dest_cpu))
 		return rq;
 
-	update_rq_clock(rq);
 	rq = move_queued_task(rq, rf, p, dest_cpu);
 
 	return rq;
@@ -2456,10 +2455,12 @@  static int migration_cpu_stop(void *data)
 				goto out;
 		}
 
-		if (task_on_rq_queued(p))
+		if (task_on_rq_queued(p)) {
+			update_rq_clock(rq);
 			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
-		else
+		} else {
 			p->wake_cpu = arg->dest_cpu;
+		}
 
 		/*
 		 * XXX __migrate_task() can fail, at which point we might end