Messages in this thread | | | Date | Thu, 19 Dec 2013 19:23:14 +0100 | From | Ingo Molnar <> | Subject | Re: [PATCH] x86 idle: repair large-server 50-watt idle-power regression |
| |
* H. Peter Anvin <hpa@zytor.com> wrote:
> On 12/19/2013 10:09 AM, H. Peter Anvin wrote: > > On 12/19/2013 09:07 AM, Ingo Molnar wrote: > >> > >> Likewise, having a barrier before the MONITOR looks sensible as well. > >> Having it _after_ monitor looks weird and is probably wrong. [It might > >> have been the effects of someone seeing the spurious wakeup problems > >> with realizing the true source, or so.] > >> > > > > Does anyone know the history of this barrier after the monitor? I know > > Len is looking for a minimal patchset that can go into -stable, and it > > seems prudent to not preturb the code more than necessary, but going > > forward it would be nice to know... > > > > Hmm... it *looks* like it is intended to be part of the construct: > > smp_mb(); > if (!need_resched()) > ... > > I found a note in the HLT variant of the function saying: > > /* > * TS_POLLING-cleared state must be visible before we > * test NEED_RESCHED: > */
Yes, that makes sense: the need_resched test is a load, and MONITOR is a load as well. Can the two ever cross, or does the CPU guarantee that because it's the same address, the loads don't cross?
Thanks,
Ingo
| |