Re: [PATCH]new ACPI processor driver to force CPUs idle

[Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>]

* Shaohua Li <shaohua.li@intel.com> [2009-06-29 10:54:55]:

[snip]

> > 
> > How do we handle interrupts and timers during this interval?  You seem
> > to disable interrupts and hold the cpu at idle for 0.95 sec.  It may
> > cause timeouts and overflows for network interrupts right?
> The x86 mwait/monitor instruction can detect interrupt and complete execution
> even interrupt is disabled, so this isn't an issue.

Cool, this will save a lot of trouble :)
 
> > Next issue is halting sibling threads belonging to a core at the same
> > time to have any power/thermal benefit.  Who does the coordination for
> > forced idle in this approach?
> Nobody does the coordination. Halt some threads even they belong to a core
> is the best we can provide now. For future, if the scheduler approach really
> works, we will happily use it.

Do you have some indicative data to show that arbitrary force-idling
of hardware threads reduce power and heat?  It will be very good if
this works without coordination among siblings.  Basically what you
are saying is that running one or two of the force-idle threads in
a 16-thread system provides significant reduction in power even
without explicit methods to ensure that they idle sibling threads
belonging to same core.

--Vaidy