Messages in this thread | | | Date | Wed, 18 Dec 2013 10:59:10 -0800 | From | "Paul E. McKenney" <> | Subject | Re: [PATCH v7 1/4] qrwlock: A queue read/write lock implementation |
| |
On Wed, Dec 18, 2013 at 01:45:01PM -0500, Waiman Long wrote: > On 12/17/2013 02:21 PM, Paul E. McKenney wrote:
[ . . . ]
> >>+ signal_next(lock,&node); > >Good, this allows multiple readers to acquire the lock concurrently, > >give or take memory latency compared to critical-section duration. > >When the first writer shows up, it presumably spins on the lock word. > > > > Yes, that was the intention. The first writer that shows up will > block succeeding readers from getting the lock. > > BTW, what was the status of the TSO memory barrier patch? This patch > has some partial dependency it.
I am hoping that it makes it into -tip soon, as I also have a patch that depends on it.
Thanx, Paul
| |