Re: [PATCH] cpuset semaphore depth check optimize

[Nikita Danilov <nikita@clusterfs.com>]

Paul Jackson writes:

[...]

 >  
 >  static inline void cpuset_down(struct semaphore *psem)
 >  {
 > -	if (current->cpuset_sem_nest_depth == 0)
 > +	if (cpuset_sem_owner != current) {
 >  		down(psem);
 > -	current->cpuset_sem_nest_depth++;
 > +		cpuset_sem_owner = current;
 > +	}
 > +	cpuset_sem_depth++;
 >  }

Err... note that now cpuset_{down,up}() take semaphore as a parameter,
but use global cpuset_sem_{owner,depth} to track recursion. This, I
believe, is an inconsistent API---it only works for a single semaphore
as passing different @psem's would lead to deadlocks and meaningless
owner and depth.

What about making these functions 

        static void cpuset_{down,up}(void);

operating on cpuset_sem internally?

"I won't rest till it's the best ..." :-)

Nikita.