Hi:

I've been thinking about how rescheduling interrupts *might* work in 
Linux and how they relate to powertop's reporting.  First, a caveat:  I 
don't understand very much about either Linux kernel internals or Intel 
hardware.

It seems to me that on *some* hardware, a kernel IPI resulting from 
rescheduling interrupts shouldn't have any power implications.  Consider 
a Thinkpad T60, which has a twin-code processor.  I believe that with 
this hardware both processors always run at the same speed.  This means 
that running one processor is almost as energy-intensive as running two 
processors.  Since rescheduling interrupts just starts the "idle" 
processor, almost certainly while the other processor is working, 
rescheduling interrupts should be in this very-cheap situation.

So, given all that, maybe powertop shouldn't be reporting rescheduling 
interrupts (and thus causing lots of angst) for such hardware.  Perhaps 
powertop could do this in general - not reporting events that don't cost 
energy.  Of course this all depends on whether my analysis is correct.

Comments?

Peter F. Patel-Schneider

PS: Of course, it is still mystifying that rescheduling interrupts can 
cause more than 50% of wakeups, or even a significant fraction of 
wakeups with an extremely light load.  How can it make sense to wakeup 
the other core when all the current core is doing is finishing the work 
needed to service an interrupt.

_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power

Reply via email to