Re: [PATCH 2/4] sched: Account per task_group nr_iowait

[Kirill Tkhai <ktkhai@virtuozzo.com>]

On 06.11.2017 19:06, Peter Zijlstra wrote:
> On Mon, Nov 06, 2017 at 05:40:32PM +0300, Kirill Tkhai wrote:
>> The patch makes number of task_group's tasks in iowait state
>> be tracked separately. This may be useful for containers to
>> check nr_iowait state of a single one.
>>
>> Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
>> ---
>>  kernel/sched/core.c  |   45 +++++++++++++++++++++++++++++++++++++++++++++
>>  kernel/sched/sched.h |    5 +++++
>>  2 files changed, 50 insertions(+)
>>
>> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
>> index 712ee54edaa1..86d1ad5f49bd 100644
>> --- a/kernel/sched/core.c
>> +++ b/kernel/sched/core.c
>> @@ -796,12 +796,32 @@ void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
>>  
>>  static void task_iowait_start(struct rq *rq, struct task_struct *p)
>>  {
>> +#ifdef CONFIG_CGROUP_SCHED
>> +	struct task_group *tg = task_group(p);
>> +
>> +	/* Task's sched_task_group is changed under both of the below locks */
>> +	BUG_ON(!raw_spin_is_locked(&p->pi_lock) && !raw_spin_is_locked(&rq->lock));
> 
> We have lockdep_assert_held for that.
> 
>> +	while (task_group_is_autogroup(tg))
>> +		tg = tg->parent;
>> +	atomic_inc(&tg->stat[rq->cpu].nr_iowait);
> 
> You're joking right, more atomic ops on the fast paths..

There should be a synchronization... It's modified under rq->lock everywhere, except try_to_wakeup().
Would it be better to use one more rq->lock at try_to_wakeup() instead of atomic?
 
>> +#endif
>> +
>>  	atomic_inc(&rq->nr_iowait);
>>  	delayacct_blkio_start();