Re: [PATCH] sched: account system time properly

[Eric Dumazet <dada1@cosmosbay.com>]

Ingo Molnar a écrit :
> * Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:
> 
>> On Wed, 29 Apr 2009 11:20:03 +0200
>> Eric Dumazet <dada1@cosmosbay.com> wrote:
>>
>>> Martin Schwidefsky a écrit :
>>>> On Wed, 29 Apr 2009 09:46:17 +0200
>>>> Eric Dumazet <dada1@cosmosbay.com> wrote:
>>>>
>>>>> Eric Dumazet a écrit :
>>>>>> Andrew Morton a écrit :
>>>>>>
>>>>>> So, if IRQs are interrupting idle task, I guess if (p != rq->idle) will be false.
>>>>>>
>>>> If an IRQ interrupts the idle task the tick is supposed to be accounted
>>>> as an idle tick. Only if the IRQ interrupted the system while it has
>>>> been in hardirq or softirq processing then it should be accounted as
>>>> system tick.
>>>>
>>>>> Maybe following patch is needed ?
>>>>>
>>>>> [PATCH] sched: account system time properly
>>>>>
>>>>> When idle task is interrupted by an IRQ, time accounting considers CPU is idle,
>>>>> even while it should account for hard or softirq.
>>>>>
>>>>> Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
>>>>>
>>>>> diff --git a/kernel/sched.c b/kernel/sched.c
>>>>> index b902e58..26efa47 100644
>>>>> --- a/kernel/sched.c
>>>>> +++ b/kernel/sched.c
>>>>> @@ -4732,7 +4732,7 @@ void account_process_tick(struct task_struct *p, int user_tick)
>>>>>  
>>>>>  	if (user_tick)
>>>>>  		account_user_time(p, one_jiffy, one_jiffy_scaled);
>>>>> -	else if (p != rq->idle)
>>>>> +	else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
>>>>>  		account_system_time(p, HARDIRQ_OFFSET, one_jiffy,
>>>>>  				    one_jiffy_scaled);
>>>>>  	else
>>>> That patch makes a lot of sense to me. Does it fix the problem?
>>>>
>>> Yes it does, on my machine at least :
>>>
>>> 11:18:48 AM  CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest   %idle
>>> 11:18:58 AM  all    0.00    0.00    0.00    0.00    0.21    0.69    0.00    0.00   99.10
>>> 11:18:58 AM    0    0.00    0.00    0.00    0.00    1.70    5.50    0.00    0.00   92.80  << HERE >>
>>> 11:18:58 AM    1    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    2    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    3    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    4    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    5    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    6    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>>> 11:18:58 AM    7    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
>> Very good. Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com> 
> 
> Thanks.
> 
> Eric, mind (re-)sending the patch with Martin's ack included, and 
> with either a suitable impact-line footer or an extra paragraph that 
> describes the bug you found (and how it shows up in practice) and 
> how the patch fixed that problem.
> 

No problem, here it is :

[PATCH] sched: account system time properly

Andrew Gallatin reported that IRQ and SOFTIRQ times were sometime not reported
correctly on recent kernels, and even bisected to commit 
457533a7d3402d1d91fbc125c8bd1bd16dcd3cd4 ([PATCH] fix scaled & unscaled cputime
accounting) first bad commit.

Further analysis pointed that commit 79741dd35713ff4f6fd0eafd59fa94e8a4ba922d
([PATCH] idle cputime accounting) was the real cause of the problem.

account_process_tick() was not taking into account timer IRQ interrupting
the idle task servicing a hard or soft irq.

On mostly idle cpu, irqs were thus not accounted and top or mpstat could tell
user/admin that cpu was 100 % idle, 0.00 % irq, 0.00 % softirq, while it was not.

Reported-by: Andrew Gallatin <gallatin@myri.com>
Re-reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Tested-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

diff --git a/kernel/sched.c b/kernel/sched.c
index b902e58..26efa47 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -4732,7 +4732,7 @@ void account_process_tick(struct task_struct *p, int user_tick)
 
 	if (user_tick)
 		account_user_time(p, one_jiffy, one_jiffy_scaled);
-	else if (p != rq->idle)
+	else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
 		account_system_time(p, HARDIRQ_OFFSET, one_jiffy,
 				    one_jiffy_scaled);
 	else