Re: [PATCH v2 1/4] locking/mutex: Add waiter parameter to mutex_optimistic_spin()

[Waiman Long <waiman.long@hpe.com>]

On 02/16/2016 03:53 AM, Peter Zijlstra wrote:
> On Mon, Feb 15, 2016 at 06:22:14PM -0800, Jason Low wrote:
>> On Mon, 2016-02-15 at 18:15 -0800, Jason Low wrote:
>>> On Fri, 2016-02-12 at 14:14 -0800, Davidlohr Bueso wrote:
>>>> On Fri, 12 Feb 2016, Peter Zijlstra wrote:
>>>>
>>>>> On Fri, Feb 12, 2016 at 12:32:12PM -0500, Waiman Long wrote:
>>>>>>   static bool mutex_optimistic_spin(struct mutex *lock,
>>>>>> +				  struct ww_acquire_ctx *ww_ctx,
>>>>>> +				  const bool use_ww_ctx, int waiter)
>>>>>>   {
>>>>>>   	struct task_struct *task = current;
>>>>>> +	bool acquired = false;
>>>>>>
>>>>>> +	if (!waiter) {
>>>>>> +		if (!mutex_can_spin_on_owner(lock))
>>>>>> +			goto done;
>>>>> Why doesn't the waiter have to check mutex_can_spin_on_owner() ?
>>>> afaict because mutex_can_spin_on_owner() fails immediately when the counter
>>>> is -1, which is a nono for the waiters case.
>>> mutex_can_spin_on_owner() returns false if the task needs to reschedule
>>> or if the lock owner is not on_cpu. In either case, the task will end up
>>> not spinning when it enters the spin loop. So it makes sense if the
>>> waiter also checks mutex_can_spin_on_owner() so that the optimistic spin
>>> queue overhead can be avoided in those cases.
>> Actually, since waiters bypass the optimistic spin queue, that means the
>> the mutex_can_spin_on_owner() isn't really beneficial. So Waiman is
>> right in that it's fine to skip this in the waiter case.
> My concern was the 'pointless' divergence between the code-paths. The
> less they diverge the easier it is to understand and review.
>
> If it doesn't hurt, please keep it the same. If it does need to diverge,
> include a comment on why.

I will keep the preemption, but will still leave out the 
mutex_can_spin_on_owner() check for waiter. I will add a comment to 
document that.

Cheers,
Longman