Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1682889 > unrolled thread

[PATCH] sched/cputime: Fix using smp_processor_id() in preemptible

Back to article view | Back to linux.kernel

Contents

  [PATCH] sched/cputime: Fix using smp_processor_id() in preemptible Wanpeng Li <kernellwp@gmail.com> - 2017-07-07 05:20 +0200
    Re: [PATCH] sched/cputime: Fix using smp_processor_id() in  preemptible Peter Zijlstra <peterz@infradead.org> - 2017-07-07 10:20 +0200
      Re: [PATCH] sched/cputime: Fix using smp_processor_id() in preemptible Wanpeng Li <kernellwp@gmail.com> - 2017-07-07 10:50 +0200

#1682889 — [PATCH] sched/cputime: Fix using smp_processor_id() in preemptible

From: Wanpeng Li <wanpeng.li@hotmail.com>

 BUG: using smp_processor_id() in preemptible [00000000] code: 99-trinity/181
 caller is debug_smp_processor_id+0x17/0x19
 CPU: 0 PID: 181 Comm: 99-trinity Not tainted 4.12.0-01059-g2a42eb9 #1
 Call Trace:
  dump_stack+0x82/0xb8
  check_preemption_disabled+0xd1/0xe3
  debug_smp_processor_id+0x17/0x19
  vtime_delta+0xd/0x2c
  task_cputime+0x89/0xdb
  thread_group_cputime+0x11b/0x1ed
  thread_group_cputime_adjusted+0x1f/0x47
  wait_consider_task+0x2a9/0xaf9
  ? lock_acquire+0x97/0xa4
  do_wait+0xdf/0x1f4
  SYSC_wait4+0x8e/0xb5
  ? list_add+0x34/0x34
  SyS_wait4+0x9/0xb
  do_syscall_64+0x70/0x82
  entry_SYSCALL64_slow_path+0x25/0x25

As Frederic pointed out:

| Although those sched_clock_cpu() things seem to only matter when the
| sched_clock() is unstable. And that stability is a condition for nohz_full
| to work anyway. So probably sched_clock() alone would be enough.

This patch fixes it by replacing sched_clock_cpu() in vtime_delta() by 
sched_clock() to avoid to call smp_processor_id() in preemptible context.

Reported-by: Xiaolong Ye <xiaolong.ye@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
---
 kernel/sched/cputime.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 6e3ea4a..d86b94e 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -683,7 +683,7 @@ static u64 vtime_delta(struct vtime *vtime)
 {
 	unsigned long long clock;
 
-	clock = sched_clock_cpu(smp_processor_id());
+	clock = sched_clock();
 	if (clock < vtime->starttime)
 		return 0;
 
-- 
2.7.4

[toc] | [next] | [standalone]

#1683014 — Re: [PATCH] sched/cputime: Fix using smp_processor_id() in preemptible

On Thu, Jul 06, 2017 at 08:18:41PM -0700, Wanpeng Li wrote:
> --- a/kernel/sched/cputime.c
> +++ b/kernel/sched/cputime.c
> @@ -683,7 +683,7 @@ static u64 vtime_delta(struct vtime *vtime)
>  {
>  	unsigned long long clock;
>  
> -	clock = sched_clock_cpu(smp_processor_id());
> +	clock = sched_clock();
>  	if (clock < vtime->starttime)
>  		return 0;
>  

This is broken. sched_clock_cpu() and sched_clock() do not have the
exact same value :-/

So this ends up doing:

	vtime->starttime = sched_clock_cpu(cpu);

	/* ... */

	clock = sched_clock();

	delta = clock - vtime->starttime;


And you compute a delta using two different clocks.

[toc] | [prev] | [next] | [standalone]

#1683029

2017-07-07 16:11 GMT+08:00 Peter Zijlstra <peterz@infradead.org>:
> On Thu, Jul 06, 2017 at 08:18:41PM -0700, Wanpeng Li wrote:
>> --- a/kernel/sched/cputime.c
>> +++ b/kernel/sched/cputime.c
>> @@ -683,7 +683,7 @@ static u64 vtime_delta(struct vtime *vtime)
>>  {
>>       unsigned long long clock;
>>
>> -     clock = sched_clock_cpu(smp_processor_id());
>> +     clock = sched_clock();
>>       if (clock < vtime->starttime)
>>               return 0;
>>
>
> This is broken. sched_clock_cpu() and sched_clock() do not have the
> exact same value :-/
>
> So this ends up doing:
>
>         vtime->starttime = sched_clock_cpu(cpu);
>
>         /* ... */
>
>         clock = sched_clock();
>
>         delta = clock - vtime->starttime;
>
>
> And you compute a delta using two different clocks.

How about adding preempt_disable/enable() pair around vtime_delta() in
task_cputime()?

Regards,
Wanpeng Li

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web