Messages in this thread | | | Date | Tue, 14 Jul 2015 21:39:03 -0400 | From | Waiman Long <> | Subject | Re: [PATCH 3/7] locking/pvqspinlock: Implement wait-early for overcommitted guest |
| |
On 07/13/2015 03:50 PM, Davidlohr Bueso wrote: > On Sat, 2015-07-11 at 16:36 -0400, Waiman Long wrote: >> /* >> + * Queued Spinlock Spin Thresholds >> + * ------------------------------- >> + * Because of the cacheline contention effect of the ticket spinlock, the >> + * same spin threshold for queued spinlock will run a bit faster. So we set >> + * a slight larger threshold for the queue head (1.25X) while the other queue >> + * nodes will keep the same threshold. > How did you come to 1.25x? These sort of magic numbers scare me. >
Yes, this is kind of arbitrary. I am going to get rid of it in the v2 patch.
Cheers, Longman
| |