lkml.org 
[lkml]   [2009]   [Mar]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [TIP][RFC 6/7] futex: add requeue_pi calls
Steven Rostedt wrote:
> On Thu, 5 Mar 2009, Darren Hart wrote:
>> As it turns out I missed setting RT_MUTEX_HAS_WAITERS on the rt_mutex in
>> rt_mutex_start_proxy_lock() - seems awfully silly in retrospect - but a
>> little non-obvious while writing it. I added mark_rt_mutex_waiters()
>> after the call to task_blocks_on_rt_mutex() and the test has completed
>> more than 400 iterations successfully (it would fail after no more than
>> 2 most of the time before).
>>
>> Steven, there are several ways to set RT_MUTEX_HAS_WAITERS - but this
>> seemed like a reasonable approach, would you agree? Since I'm holding
>> the wait_lock I don't technically need the atomic cmpxchg and could
>> probably just set it explicity - do you have a preference?
>>
>
>> +
>> +/**
>> + * rt_mutex_finish_proxy_lock - Complete the taking of the lock initialized
>> on
>> + * our behalf by another thread.
>> + * @lock: the rt_mutex we were woken on
>> + * @to: the timeout, null if none. hrtimer should already have been started.
>> + * @waiter: the pre-initialized rt_mutex_waiter
>> + * @detect_deadlock: for use by __rt_mutex_slowlock
>> + *
>> + * Special API call for PI-futex requeue support
>> + */
>> +int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
>> + struct hrtimer_sleeper *to,
>> + struct rt_mutex_waiter *waiter,
>> + int detect_deadlock)
>> +{
>> + int ret;
>> +
>> + if (waiter->task)
>> + schedule_rt_mutex(lock);
>> +
>> + spin_lock(&lock->wait_lock);
>> +
>> + set_current_state(TASK_INTERRUPTIBLE);
>> +
>> + ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter,
>> + detect_deadlock);
>> +
>> + set_current_state(TASK_RUNNING);
>> +
>> + if (unlikely(waiter->task))
>> + remove_waiter(lock, waiter);
>> +
>> + /*
>> + * try_to_take_rt_mutex() sets the waiter bit unconditionally. We
>> might
>> + * have to fix that up.
>> + */
>> + fixup_rt_mutex_waiters(lock);
>
> Darren,
>
> I take it you are talking about the above.

Actually no, I was talking about rt_mutex_START_proxy_lock():

/**
* rt_mutex_start_proxy_lock - prepare another task to take the lock
*
* @lock: the rt_mutex to take
* @waiter: the rt_mutex_waiter initialized by the waiter
* @task: the task to prepare
* @detext_deadlock: passed to task_blocks_on_rt_mutex
*
* The lock should have an owner, and it should not be task.
* Special API call for FUTEX_REQUEUE_PI support.
*/
int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
struct rt_mutex_waiter *waiter,
struct task_struct *task, int detect_deadlock)
{
int ret;

spin_lock(&lock->wait_lock);
ret = task_blocks_on_rt_mutex(lock, waiter, task, detect_deadlock);


I add the following line to fix the bug. Question is, should I use this atomic
optimization here (under the lock->wait_lock) or should I just do
"lock->owner |= RT_MUTEX_HAS_WAITERS" ?

=====> mark_rt_mutex_waiters(lock);

if (ret && !waiter->task) {
/*
* Reset the return value. We might have
* returned with -EDEADLK and the owner
* released the lock while we were walking the
* pi chain. Let the waiter sort it out.
*/
ret = 0;
}
spin_unlock(&lock->wait_lock);

debug_rt_mutex_print_deadlock(waiter);

return ret;
}



--
Darren Hart
IBM Linux Technology Center
Real-Time Linux Team


\
 
 \ /
  Last update: 2009-03-06 06:31    [W:0.639 / U:0.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site