Re: Thermal scheduling
Re: Thermal scheduling
- Subject: Re: Thermal scheduling
- From: Paul Ripke <email@hidden>
- Date: Fri, 1 Oct 2004 00:04:30 +1000
On Thursday, Sep 30, 2004, at 01:00 Australia/Sydney, John Siracusa
wrote:
On Wed, 29 Sep 2004 16:27:38 +0200, Markus Hitter <email@hidden>
wrote:
Am 29.09.2004 um 15:54 schrieb John Siracusa:
One theory presented to explain the results was that the kernel is
taking CPU temperature into account when doling out CPU cycles.
Well, I'm not really a kernel expert, but the scheduler's task is to
decide which process/thread gets the next CPU quantum, where a CPU
quantum is about 10 ms. If there's no process in need of processing
time, it sends the CPU into doze mode.
The observed result of the test program is that it runs at X
iterations/sec
for some time, then drops to X - N iterations/sec for the remainder.
The
drop seems to coincide with a CPU temp that passes some threshold.
Well, it appears the thermal throttling that could be done in the
T_THERMAL
interrupt handler in:
xnu/osfmk/ppc/interrupt.c: interrupt()
is #ifdef'ed out, at least in the older publicly available versions I've
checked. I note that this code has been removed somewhere between 10.3.2
and 10.3.5. Looks like something else is doing this throttling now,
judging by the "deprecated" comments in
xnu/osfmk/ppc/machine_routines_asm.s,
and the fact that the only reference I can find in the 10.3.5 xnu source
to CPU thermal interrupts is in CHUD, and that's just a counter.
Just curious: Are any "THERMAL" type messages dumped to the syslog?
Another
option may be to look at CHUD - maybe the above counter is accessible; I
don't have a 10.3.x capable machine ATM to check.
Cheers,
--
stix
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden