Re: [PATCH 1/1 V2] workqueue: remove the del_timer_sync()s in maybe_create_worker()

[Lai Jiangshan <laijs@cn.fujitsu.com>]

On 07/15/2014 05:42 AM, Thomas Gleixner wrote:
> On Tue, 15 Jul 2014, Lai Jiangshan wrote:
>> On 07/14/2014 11:33 PM, Thomas Gleixner wrote:
>>> On Mon, 14 Jul 2014, Tejun Heo wrote:
>>>
>>>> Hello,
>>>>
>>>> On Mon, Jul 14, 2014 at 04:13:21PM +0800, Lai Jiangshan wrote:
>>>>> It is said in the document that the timer which is being
>>>>> deleted by del_timer_sync() should not be restarted:
>>>>>   Synchronization rules: Callers must prevent restarting of
>>>>>   the timer, otherwise this function is meaningless.
>>>>>
>>>>> Repeating timer may cause the del_timer_sync() spin longer,
>>>>> or even spin forever in very very very very extreme condition.
>>>>
>>>> I'm fairly sure del_timer_sync() can delete self-requeueing timers.
>>>> The implementation busy-waits if the queued timer is the currently
>>>> executing one and dequeues only while the timer isn't running which
>>>> should be able to handle self-requeueing ones just fine.  Thomas,
>>>> del_timer_sync() can reliably delete self-requeueing ones, right?
>>>
>>> Yes. 
>>
>> The comments of the del_timer_sync() needs to be updated
>> if I did not misunderstood?
>>
>>> If the timer callback is running on the other cpu, then it waits
>>> for the callback to finish before checking whether the timer is
>>> enqueued or not.
>>
>> The syncer may be interrupted here, after it comes back, the timer
>> may be running again (and maybe again and again).
> 
> No. The del_timer_sync() code holds the base lock with interrupts
> disabled. So it can't be interrupted.
> 

	WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
	for (;;) {
		int ret = try_to_del_timer_sync(timer);
		if (ret >= 0)
			return ret;
		cpu_relax();

How about when it is interrupted here?

	}

> Thanks,
> 
> 	tglx
> 
> 
> .
>