Re: [Syllable-developer] Test version of an acpi processor driver

[Tim t. L. <ti...@sc...>]

On Thu, 26 Apr 2007, Arno Klenke wrote:

> Hi,
>
> here is an early test version of an acpi processor driver. It currently
> supports the following features (if the hardware supports it, of course):
>
> *C1 (hlt) and C2 idle states
> *The C3 idle state is currently supported on uniprocessor systems. The
> same codepath can also be used on intel smp systems and so it should
> work with them, too.
> *Frequency scaling using the generic acpi method and using the intel msr
> registers.

First off, the new USB driver doesn't seem to fully agree with my system; 
kernel.log is flooded with the following message:

0:kernel::usb_hub_thread : USB: get_port_status failed (err = -1)
0:kernel::usb_hub_thread : USB: get_port_status failed (err = -1)
0:kernel::usb_hub_thread : USB: get_hub_status failed

My USB mouse and keyboard still work though.

The ACPI Processor driver generates the following entries in kernel.log:

0:init::init(-1) : ACPI: Register driver ACPI Processor Driver
0:init::init(-1) : ACPI: Processor [CPU ID: 0x6a APIC ID:0 ACPI ID:1] found
0:init::init(-1) : ACPI: Processor PBLK @ 0x1010
0:init::init(-1) : ACPI: C1 state latency[001] address[0x0]
0:init::init(-1) : ACPI: C2 state latency[080] address[0x1014]
0:init::init(-1) : ACPI: Warning: pstates not supported on this hardware

and pmset's output (without arguments):
ACPI c-states:

[0] C1, 1 ticks latency, usage 49
[1] C2, 80 ticks latency, usage 590534

So the C2 state seems to be actually used!
As can be expected from the kernel.log above, pmset complains loudly if 
you want to set a frequency.

I'm not quite sure why the driver thinks pstates aren't supported; a dump 
from my ACPI tables shows my CPU0 object has _PCT, _PSS and _PPC objects, 
with _PSS actually showing 5 performance states. Perhaps it's because the 
_PCT contains only zeroes, thus not indicating a register the OS has to 
write to to change frequencies? Especially since afaik K7 (and K8?) 
processors only scale by writing to MSRs, and not by writing to I/O 
registers.

Tim.