Messages in this thread | | | Date | Wed, 30 Jun 2010 12:25:26 +0100 | From | "Jan Beulich" <> | Subject | Re: [PATCH 4/4, v2] x86: enlightenment for ticket spin locks - improve yield behavior on Xen |
| |
>>> On 30.06.10 at 12:10, Jeremy Fitzhardinge <jeremy@goop.org> wrote: > On 06/30/2010 10:49 AM, Jan Beulich wrote: >>>>> On 30.06.10 at 10:11, Peter Zijlstra <peterz@infradead.org> wrote: >>>>> >>> On Tue, 2010-06-29 at 15:35 +0100, Jan Beulich wrote: >>> >>>> The (only) additional overhead this introduces for native execution is >>>> the writing of the owning CPU in the lock acquire paths. >>>> >>> Uhm, and growing the size of spinlock_t to 6 (or 8 bytes when aligned) >>> bytes when NR_CPUS>256. >>> >> Indeed, I should have mentioned that. Will do so in an eventual >> next version. >> > > Rather than increasing the lock size, why not just disable the > enlightenment if the number of (possible) cpus is > 256? I don't think > a VM will ever have that many cpus, so it will only apply in the case of > booting the kernel on large physical machine.
While that would be an option, I think the decision to go either way should really be left to the user (after all that's why there is a config option in the first place).
Jan
| |