Processor state C6 only used in battery mode, workaround possible?

Processor state C6 only used in battery mode, workaround possible?

[garkein]

Hello,
with recent powertop, I could confirm that my T61 uses the C6 processor
state only in battery mode:

(Kernel 2.6.25)

AC:
Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
Your BIOS reports the following C-states : C1 C2 C3

Battery:
Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
Your BIOS reports the following C-states : C1 C2 C6

Since the notebook runs quite a bit cooler, and thus less noisy (fan) in
battery mode, I would like to have it use that C6 state also in AC mode.
Is it possible, maybe with some slight tweaks, to convice the kernel (or
BIOS?) to also use that C6 state when in AC mode?

Thanks,
Daniel

Re: Processor state C6 only used in battery mode, workaround possible?

[Len Brown]

On Sun, 29 Jun 2008, garkein@mailueberfall.de wrote:

> Hello,
> with recent powertop, I could confirm that my T61 uses the C6 processor
> state only in battery mode:
> 
> (Kernel 2.6.25)
> 
> AC:
> Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
> Your BIOS reports the following C-states : C1 C2 C3
> 
> Battery:
> Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
> Your BIOS reports the following C-states : C1 C2 C6
> 
> Since the notebook runs quite a bit cooler, and thus less noisy (fan) in
> battery mode, I would like to have it use that C6 state also in AC mode.
> Is it possible, maybe with some slight tweaks, to convice the kernel (or
> BIOS?) to also use that C6 state when in AC mode?

Lenovo decided when they designed the platform to do extra work to
switch what ACPI C3 means on AC vs DC.  They chose to have lower latency
on AC, and save energy on DC.  (you will notice lower latency reported
for C3 when on DC if you look in /proc/acpi/processor/*/power or
in /sys/devices/system/cpu/cpu*/cpuidle/*/latency
eg. it may switch between 57usec and 17usec

if you boot with "processor.nocst=1"
then you'll use the legach FADT method of specifying C-states,
and will lose the AC/DC switching capability, and also lose
the MWAIT capability -- though you'll unlikely notice.
So you'll get stuck with a single C3 state -- dunno if it
is the one you prefer or the one you don't prefer...

If this doesn't help and you want to hack your driver,
then just go to acpi_processor_notify() in processor_core.c
and disable the call to acpi_processor_cst_has_changed()
and boot on DC...

The long term answer to your question is that we are hoping
to get permission to ship a native intel_idle driver which
will give Linux the option to ignore the ACPI BIOS in this 
area.

cheers,
-Len

Re: Processor state C6 only used in battery mode, workaround possible?

[garkein]

Am Mittwoch, den 18.03.2009, 14:08 -0400 schrieb Len Brown:
> On Sun, 29 Jun 2008, garkein@mailueberfall.de wrote:
> 
> > Hello,
> > with recent powertop, I could confirm that my T61 uses the C6 processor
> > state only in battery mode:
> > 
> > (Kernel 2.6.25)
> > 
> > AC:
> > Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
> > Your BIOS reports the following C-states : C1 C2 C3
> > 
> > Battery:
> > Your CPU supports the following C-states : C1 C2 C3 C4 C5 C6 
> > Your BIOS reports the following C-states : C1 C2 C6
> > 
> > Since the notebook runs quite a bit cooler, and thus less noisy (fan) in
> > battery mode, I would like to have it use that C6 state also in AC mode.
> > Is it possible, maybe with some slight tweaks, to convice the kernel (or
> > BIOS?) to also use that C6 state when in AC mode?
> 
> Lenovo decided when they designed the platform to do extra work to
> switch what ACPI C3 means on AC vs DC.  They chose to have lower latency
> on AC, and save energy on DC.  (you will notice lower latency reported
> for C3 when on DC if you look in /proc/acpi/processor/*/power or
> in /sys/devices/system/cpu/cpu*/cpuidle/*/latency
> eg. it may switch between 57usec and 17usec
> 
> if you boot with "processor.nocst=1"
> then you'll use the legach FADT method of specifying C-states,
> and will lose the AC/DC switching capability, and also lose
> the MWAIT capability -- though you'll unlikely notice.
> So you'll get stuck with a single C3 state -- dunno if it
> is the one you prefer or the one you don't prefer...
> 
> If this doesn't help and you want to hack your driver,
> then just go to acpi_processor_notify() in processor_core.c
> and disable the call to acpi_processor_cst_has_changed()
> and boot on DC...
> 
> The long term answer to your question is that we are hoping
> to get permission to ship a native intel_idle driver which
> will give Linux the option to ignore the ACPI BIOS in this 
> area.
> 
> cheers,
> -Len

Thank you Len for your explanation. The long term way would be my
preference as I don't want to be "stuck" with a single state. (I want
them all! ;-)).

Regards,
Daniel