Re: [PATCH 2/2] mutex: use spin_[un]lock instead of arch_spin_[un]lock

[Ingo Molnar <mingo@kernel.org>]

* Yuanhan Liu <yuanhan.liu@linux.intel.com> wrote:

> On Thu, Jan 24, 2013 at 11:14:50AM +0100, Ingo Molnar wrote:
> > 
> > * Yuanhan Liu <yuanhan.liu@linux.intel.com> wrote:
> > 
> > > On Thu, Jan 24, 2013 at 10:58:07AM +0100, Ingo Molnar wrote:
> > > > 
> > > > * Yuanhan Liu <yuanhan.liu@linux.intel.com> wrote:
> > > > 
> > > > > Use spin_[un]lock instead of arch_spin_[un]lock in mutex-debug.h so
> > > > > that we can collect the lock statistics of spin_lock_mutex from
> > > > > /proc/lock_stat.
> > > > > 
> > > > > Cc: Ingo Molnar <mingo@kernel.org>
> > > > > Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
> > > > > ---
> > > > >  kernel/mutex-debug.h |    4 ++--
> > > > >  1 files changed, 2 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/kernel/mutex-debug.h b/kernel/mutex-debug.h
> > > > > index 0799fd3..556c0bc 100644
> > > > > --- a/kernel/mutex-debug.h
> > > > > +++ b/kernel/mutex-debug.h
> > > > > @@ -43,13 +43,13 @@ static inline void mutex_clear_owner(struct mutex *lock)
> > > > >  							\
> > > > >  		DEBUG_LOCKS_WARN_ON(in_interrupt());	\
> > > > >  		local_irq_save(flags);			\
> > > > > -		arch_spin_lock(&(lock)->rlock.raw_lock);\
> > > > > +		spin_lock(lock);			\
> > > > 
> > > > But in that case it could probably use the spin_lock_irqsave() 
> > > > primitive, right?
> > > 
> > > Right, in that case I should use spin_lock_irqsave.
> > > 
> > > But one question, why we use spin_lock at kernel/mutex.h, 
> > > while use 'local_irq_save(); arch_spin_lock' at 
> > > kernel/mutex-debug.h?
> > > 
> > > Shouldn't we keep it consistent? Say use spin_lock_irqsave?
> > 
> > I think we did it to increase performance with lockdep enabled - 
> > this particular lockdep annotation, given the short codepaths, 
> > is not that hard to verify - and if it breaks it will break a 
> > thousand mutex locking places in the kernel.
> 
> Thanks for the explanation.
> > 
> > So maybe it's better to leave it alone - maybe add a comment 
> > that explains the reason.
> 
> Sorry, I may not get your point clearly. Should I make another 
> patch to convert 'local_irq_save(..); arch_spin_lock(..);' at 
> kernel/mutex-debug.h to spin_lock_irqsave() then?

No, I'd suggest to add a comment that explains why there's no 
lockdep annotation in that place.

Thanks,

	Ingo