Re: [PATCH v3] memcg: fix soft lockup in the OOM process

[Chen Ridong]

On 2025/1/14 17:20, Vlastimil Babka wrote:
> On 1/14/25 09:40, Michal Hocko wrote:
>> On Mon 13-01-25 19:45:46, Andrew Morton wrote:
>>> On Mon, 13 Jan 2025 14:51:55 +0800 Chen Ridong <chenridong@xxxxxxxxxxxxxxx> wrote:
>>>
>>>>>> @@ -430,10 +431,15 @@ static void dump_tasks(struct oom_control *oc)
>>>>>>  		mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
>>>>>>  	else {
>>>>>>  		struct task_struct *p;
>>>>>> +		int i = 0;
>>>>>>  
>>>>>>  		rcu_read_lock();
>>>>>> -		for_each_process(p)
>>>>>> +		for_each_process(p) {
>>>>>> +			/* Avoid potential softlockup warning */
>>>>>> +			if ((++i & 1023) == 0)
>>>>>> +				touch_softlockup_watchdog();
>>>>>
>>>>> This might suppress the soft lockup, but won't a rcu stall still be detected?
>>>>
>>>> Yes, rcu stall was still detected.
> 
> "was" or "would be"? I thought only the memcg case was observed, or was that
> some deliberate stress test of the global case? (or the pr_info() console
> stress test mentioned earlier, but created outside of the oom code?)
> 

It's not easy to reproduce for global OOM. Because the pr_info() console
stress test can also lead to other softlockups or RCU warnings(not
causeed by OOM process) because the whole system is struggling.However,
if I add mdelay(1) in the dump_task() function (just to slow down
dump_task, assuming this is slowed by pr_info()) and trigger a global
OOM, RCU warnings can be observed.

I think this can verify that global OOM can trigger RCU warnings in the
specific scenarios.

>>>> For global OOM, system is likely to struggle, do we have to do some
>>>> works to suppress RCU detete?
>>>
>>> rcu_cpu_stall_reset()?
>>
>> Do we really care about those? The code to iterate over all processes
>> under RCU is there (basically) since ever and yet we do not seem to have
>> many reports of stalls? Chen's situation is specific to memcg OOM and
>> touching the global case was mostly for consistency reasons.
> 
> Then I'd rather not touch the global case then if it's theoretical? It's not
> even exactly consistent, given it's a cond_resched() in the memcg code (that
> can be eventually automatically removed once/if lazy preempt becomes the
> sole implementation), but the touch_softlockup_watchdog() would remain,
> while doing only half of the job?