Re: sem_lock() vs qspinlocks

[Peter Zijlstra]

On Sat, May 21, 2016 at 12:01:00AM -0400, Waiman Long wrote:
> On 05/20/2016 08:59 PM, Davidlohr Bueso wrote:
> >On Fri, 20 May 2016, Peter Zijlstra wrote:
> >
> >>On Fri, May 20, 2016 at 04:47:43PM -0400, Waiman Long wrote:
> >>
> >>>>Similarly, and I know you hate it, but afaict, then semantically
> >>>>queued_spin_is_contended() ought to be:
> >>>>
> >>>>-       return atomic_read(&lock->val) & ~_Q_LOCKED_MASK;
> >>>>+       return atomic_read(&lock->val);
> >>>>
> >>
> >>>Looking for contended lock, you need to consider the lock waiters
> >>>also. So
> >>>looking at the whole word is right.
> >>
> >>No, you _only_ need to look at the lock waiters.
> >
> >Is there anyway to do this in a single atomic_read? My thought is that
> >otherwise
> >we could further expand the race window

Its inherently racy, arrival of a contender is subject to timing. No
point in trying to fix what can't be fixed.

> The existing code is doing that, but I would argue that including the
> locked, but uncontended case isn't a bad idea.

It _IS_ a bad idea, you get unconditional lock-breaks.

Its the same as:

#define spin_is_contended(l)	(true)

Because the only reason you're using spin_is_conteded() is if you're
holding it.