CPU off power consumption

CPU off power consumption

[Giuliano Pochini]

Out of curiosity, what's the difference between a cpu that has never been
enabled and one that has been disabled with echo 0>/sys/.../online ? It
happens that when I boot with maxcpus=0 the temperature always stays low
enoung that the fan never spins up. If I enable and then I immediately
disable the 2nd cpu, the temperature goes a few degrees up. I have a dual
G4-MDD.

--
Giuliano.

Re: CPU off power consumption

[Paul Mackerras]

Giuliano Pochini writes:

> Out of curiosity, what's the difference between a cpu that has never been
> enabled and one that has been disabled with echo 0>/sys/.../online ? It
> happens that when I boot with maxcpus=0 the temperature always stays low
> enoung that the fan never spins up. If I enable and then I immediately
> disable the 2nd cpu, the temperature goes a few degrees up. I have a dual
> G4-MDD.

Interesting.  A cpu that has been disabled will be in sleep mode with
interrupts disabled and its caches flushed.  One that has never been
started may possibly be held in the reset state.  The way to check
would be to check the state of the GPIO register that controls the
soft reset line of the second CPU.

Paul.

Re: CPU off power consumption

[Benjamin Herrenschmidt]

On Mon, 2005-11-28 at 13:15 +1100, Paul Mackerras wrote:
> Giuliano Pochini writes:
> 
> > Out of curiosity, what's the difference between a cpu that has never been
> > enabled and one that has been disabled with echo 0>/sys/.../online ? It
> > happens that when I boot with maxcpus=0 the temperature always stays low
> > enoung that the fan never spins up. If I enable and then I immediately
> > disable the 2nd cpu, the temperature goes a few degrees up. I have a dual
> > G4-MDD.
> 
> Interesting.  A cpu that has been disabled will be in sleep mode with
> interrupts disabled and its caches flushed.  One that has never been
> started may possibly be held in the reset state.  The way to check
> would be to check the state of the GPIO register that controls the
> soft reset line of the second CPU.

No, I think CPUs that have not been started are held in a similar sleep
loop in ROM. I don't see right away why there would be any power
consumption difference unless some bug causing us to never actually call
the sleep loop ....

Ben.

Re: CPU off power consumption

[Giuliano Pochini]

On Mon, 28 Nov 2005, Benjamin Herrenschmidt wrote:

> No, I think CPUs that have not been started are held in a similar sleep
> loop in ROM. I don't see right away why there would be any power
> consumption difference unless some bug causing us to never actually call
> the sleep loop ....

Any hint on how to debug it ?  Where is the code that enables/disables the
cpus ?  Where is the sleep loop ?


--
Giuliano.

Re: CPU off power consumption

[Paul Mackerras]

Giuliano Pochini writes:

> Any hint on how to debug it ?  Where is the code that enables/disables the
> cpus ?  Where is the sleep loop ?

The initial sleep loop is in ROM.  A CPU that is being offlined will
enter low_cpu_die() in arch/ppc/platforms/pmac_sleep.c and will go to
sleep at the mtmsr about 40 lines into that function.

Paul.