Re: [PATCH] sched/deadline/rtmutex: Fix a PI crash for deadline tasks

[Xunlei Pang <xpang@redhat.com>]

On 2016/04/05 at 17:29, Peter Zijlstra wrote:
> On Tue, Apr 05, 2016 at 11:19:54AM +0200, Peter Zijlstra wrote:
>> Or did I miss something (again) ? :-)
>>
>> ---
>>  kernel/locking/rtmutex.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
>> index 3e746607abe5..36eb232bd29f 100644
>> --- a/kernel/locking/rtmutex.c
>> +++ b/kernel/locking/rtmutex.c
>> @@ -1390,11 +1390,11 @@ rt_mutex_fastunlock(struct rt_mutex *lock,
>>  	} else {
>>  		bool deboost = slowfn(lock, &wake_q);
>>  
>> -		wake_up_q(&wake_q);
>> -
>>  		/* Undo pi boosting if necessary: */
>>  		if (deboost)
>>  			rt_mutex_adjust_prio(current);
>> +
>> +		wake_up_q(&wake_q);
>>  	}
>>  }
> So one potential issue with this -- and this might be reason this code
> is the way it is -- is that the moment we de-boost we can get preempted,
> before having had a chance to wake the higher prio task, getting
> ourselves into a prio-inversion.
>
> But again, that should be fairly simply to fix.

This is cool, I think we should also init "pi_task" properly for INIT_MUTEX and fork,
otherwise looks good to me :-)

Besides, do you think we can kill "pi_waiters_leftmost" from task_struct, as we
can easily get it from "pi_waiters"?

I will test it further with these new changes soon.

Regards,
Xunlei

>
> --
>  kernel/locking/rtmutex.c | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
>
> diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
> index 3e746607abe5..1896baf28e9c 100644
> --- a/kernel/locking/rtmutex.c
> +++ b/kernel/locking/rtmutex.c
> @@ -1390,11 +1390,21 @@ rt_mutex_fastunlock(struct rt_mutex *lock,
>  	} else {
>  		bool deboost = slowfn(lock, &wake_q);
>  
> -		wake_up_q(&wake_q);
> -
> -		/* Undo pi boosting if necessary: */
> +		/*
> +		 * Undo pi boosting (if necessary) and wake top waiter.
> +		 *
> +		 * We should deboost before waking the high-prio task such that
> +		 * we don't run two tasks with the 'same' state. This however
> +		 * can lead to prio-inversion if we would get preempted after
> +		 * the deboost but before waking our high-prio task, hence the
> +		 * preempt_disable.
> +		 */
> +		preempt_disable();
>  		if (deboost)
>  			rt_mutex_adjust_prio(current);
> +
> +		wake_up_q(&wake_q);
> +		preempt_enable();
>  	}
>  }
>