Re: [PATCH RFC 00/11] lock monitor: Separate features related to lock

[Hitoshi Mitake <mitake@dcl.info.waseda.ac.jp>]

On 03/19/10 05:30, Frederic Weisbecker wrote:
> On Thu, Mar 18, 2010 at 02:49:38PM +0900, Hitoshi Mitake wrote:
>> Unfortunately, we cannot use this detection method.
>> Because trylock series (e.g. spin_trylock()) only issues
>> lock_acquire() like this,
>>
>> static inline int __raw_spin_trylock(raw_spinlock_t *lock)
>> {
>> 	preempt_disable();
>> 	if (do_raw_spin_trylock(lock)) {
>> 		spin_acquire(&lock->monitor, 0, 1, _RET_IP_);<- spin_acquire() only
>> issues lock_acquire()
>> 		return 1;
>> 	}
>> 	preempt_enable();
>> 	return 0;
>> }
>>
>> So distinguishing trylocks and lock_acquire()/lock_release() pairs from
>> might_lock_read(), might_fault() and etc is hard.
>>
>> It seems that turning off PROVE_LOCKING must be required
>> for state machine of perf lock.
>
>
> No that's really not a problem. trylocks are pointless in latency
> profiling because by definition they don't content. OTOH, they
> grab the lock and other locks might wait and raise latencies. So
> they are part of the profile. But we don't care about having the
> usual acquire/aquired/release sequence as we have the flags that
> tell us if this is a trylock.
>
> So we just need to consider that acquire:try_lock - release is
> a right lock scenario, but that acquire - release is only a lockdep
> check.

Ah, I see. The argument flags of lock_acquire event can be used for
distinguishing try or read lock and pure verifying.

Thanks,
	Hitoshi