Re: [PATCH] sched/fair: Prevent dead task groups from regaining cfs_rq's

[Mathias Krause]

Am 05.11.21 um 18:27 schrieb Peter Zijlstra:
> On Fri, Nov 05, 2021 at 06:14:33PM +0100, Mathias Krause wrote:
>> Am 05.11.21 um 17:58 schrieb Peter Zijlstra:
>>> On Fri, Nov 05, 2021 at 05:29:14PM +0100, Mathias Krause wrote:
>>>>> Looks like it needs to be the kfree_rcu() one in this case. I'll prepare
>>>>> a patch.
>>>>
>>>> Testing the below patch right now. Looking good so far. Will prepare a
>>>> proper patch later, if we all can agree that this covers all cases.
>>>>
>>>> But the basic idea is to defer the kfree()'s to after the next RCU GP,
>>>> which also means we need to free the tg object itself later. Slightly
>>>> ugly. :/
>>>
>>> Can't we add an rcu_head to struct task_group and simply call_rcu() the
>>> thing with a free function?
>>
>> There is already one and this patch makes use of it for the second RCU
>> GP requirement. Basically, the patch is doing what you request, no? See
>> the new free_fair_sched_group().
> 
> For some reason I thought you still did kfree_rcu(), I suppose reading
> is hard. I'll give it another go after dinner.

I wanted to go for kfree_rcu() initially, true. But after realizing,
that adding a rcu_head to struct cfs_rg, sched_entity and task_group
(which already has one) might be too much for what's needed, I went the
call_rcu() route instead and re-used the rcu_head of task_group.

Actually re-using the rcu_head in task_group is safe, as we'll use it
only sequentially: first in sched_destroy_group() to schedule
sched_free_group_rcu() and then, when it's executing, in
free_fair_sched_group() to schedule free_tg().

rcu_head's get unlinked prior to getting their callback function
invoked, which makes the above a valid use case.

Thanks,
Mathias