> On May 2, 2023, at 9:50 AM, Zhouyi Zhou <zhouzhouyi@xxxxxxxxx> wrote:
>
> On Tue, May 2, 2023 at 9:40 PM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
>>
>>> On Tue, May 02, 2023 at 08:01:41AM +0800, zhouzhouyi@xxxxxxxxx wrote:
>>> From: Zhouyi Zhou <zhouzhouyi@xxxxxxxxx>
>>>
>>> In kfree_rcu_test, kfree_scale_shutdown will be detected as hung task
>>> if kfree_loops is too big. Replace wait_event with wait_event_idle
>>> to avoid false positive.
>>>
>>> Tested in the PPC VM of Open Source Lab of Oregon State University.
>>>
>>> Suggested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
>>> Signed-off-by: Zhouyi Zhou <zhouzhouyi@xxxxxxxxx>
>>
>> Good catch, thank you!
>>
>> However, this commit beat you to it:
>>
>> ef1ef3d47677 ("rcuscale: Move shutdown from wait_event() to wait_event_idle()")
> You are very welcome ;-) Still, this is a very fruitful learning
> process for me ;-)
Speaking of learning, can you explain why it fixes the issue? ;-)
Your change log lacked the real reason but that is Ok since change logs can only tell you so much. I admit I myself did not know the reason till I read some code.
Quiz: why exactly does this all work out in the end even though the hung task detector saw it hung?
Thanks,
- Joel
>
> Cheers
> Zhouyi
>>
>> Thanx, Paul
>>
>>> ---
>>> kernel/rcu/rcuscale.c | 2 +-
>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/kernel/rcu/rcuscale.c b/kernel/rcu/rcuscale.c
>>> index 91fb5905a008..d99c586939d1 100644
>>> --- a/kernel/rcu/rcuscale.c
>>> +++ b/kernel/rcu/rcuscale.c
>>> @@ -771,7 +771,7 @@ kfree_scale_cleanup(void)
>>> static int
>>> kfree_scale_shutdown(void *arg)
>>> {
>>> - wait_event(shutdown_wq,
>>> + wait_event_idle(shutdown_wq,
>>> atomic_read(&n_kfree_scale_thread_ended) >= kfree_nrealthreads);
>>>
>>> smp_mb(); /* Wake before output. */
>>> --
>>> 2.34.1
>>>
