cvs commit: src/sys/dev/acpica acpi_cpu.c

[Nate Lawson]

On Wed, 17 Mar 2004, John Baldwin wrote:
> On Wednesday 17 March 2004 04:49 pm, Nate Lawson wrote:
> > njl         2004/03/17 13:49:47 PST
> >
> >   FreeBSD src repository
> >
> >   Modified files:
> >     sys/dev/acpica       acpi_cpu.c
> >   Log:
> >   Fix border error to allow systems that specify 100 for latency also use
> >   C2 and 1000 to use C3.
> >
> >   Submitted by:   Bruno Ducrot <ducrot at poupinou.org>
> >   Tested by:      Scott Lambert <lambert at lambertfam.org>
>
> Hmm, the BIOS on my laptop uses those values when the C states are disabled
> (according to the BIOS text).  That is, my BIOS lets me turn C2 and C3 on and
> off.  When turned on, it uses the values 1 and 85 for the latency, and when
> off it uses 100 and 1000.  Note that the kernel never uses C3 on this machine
> by the way with these latencies:
>
> hw.acpi.cpu.cx_supported: C1/1 C2/1 C3/85

That's incorrect behavior.  See Table 5-8 of the ACPI 2.0 spec (at or near
page 102).  Have you actually done an acpidump -t both with and without C2
disabled in your BIOS to see what it does?  Perhaps the manual is wrong
and it does the right thing.  In any case, your failure mode (not being
able to disable Cx states in BIOS) is not as bad as others (not being able
to use C2 when it's supported.)  You can always keep using the sysctl to
disable C2 and/or C3.

-Nate