Messages in this thread | | | Date | Mon, 14 Sep 2015 15:37:32 -0400 | From | Waiman Long <> | Subject | Re: [PATCH v6 6/6] locking/pvqspinlock: Queue node adaptive spinning |
| |
On 09/14/2015 10:10 AM, Peter Zijlstra wrote: > On Fri, Sep 11, 2015 at 02:37:38PM -0400, Waiman Long wrote: >> In an overcommitted guest where some vCPUs have to be halted to make >> forward progress in other areas, it is highly likely that a vCPU later >> in the spinlock queue will be spinning while the ones earlier in the >> queue would have been halted. The spinning in the later vCPUs is then >> just a waste of precious CPU cycles because they are not going to >> get the lock soon as the earlier ones have to be woken up and take >> their turn to get the lock. >> >> This patch implements an adaptive spinning mechanism where the vCPU >> will call pv_wait() if the following conditions are true: >> >> 1) the vCPU has not been halted before; >> 2) the previous vCPU is not running. > Why 1? For the mutex adaptive stuff we only care about the lock holder > running, right?
The wait-early once logic was there because of the kick-ahead patch as I don't want a recently kicked vCPU near the head of the queue to go back to sleep too early. However, without kick-ahead, a woken up vCPU should now be at the queue head. Indeed, we can remove that check and simplify the logic.
BTW, the queue head vCPU at pv_wait_head_and_lock() doesn't wait early, it will spin the full threshold as there is no way for it to figure out if the lock holder is running or not.
Cheers, Longman
| |