Dell D810/FreeBSD 6.1-PRERELEASE/interrupts??

[Dmitry Pryanishnikov]

Hello!

  (I've changed CC: from freebsd-current to freebsd-acpi since it's more
appropriate there).

  What this thread is about: many people (including me) have a hardware which 
gives unacceptable performance if the new defaults *_cx_lowest="LOW" aren't
overridden.

On Fri, 3 Mar 2006, Milan Obuch wrote:
> On Friday 03 March 2006 11:19, Dmitry Pryanishnikov wrote:
>> Hello!
>> On Fri, 3 Mar 2006, Anthony Maher wrote:
>>> In /etc/rc.conf, commenting out the following solved the problems.
>>>
>>> ###performance_cx_lowest="LOW"
>>> ###economy_cx_lowest="LOW"
>>>
>>> sorry for the noise.
>>
>>   It's not a noise, it's a real problem. My ASUS M5A notebook (Intel
>> Pentium M CPU model 750 1.86GHz), usually quite fast machine, becomes
>> dead slow if I leave now-default settings *_cx_lowest="LOW". Every
>> keystroke takes almost second to get echo on idle machine. I understand,
>> ACPI developers want to test non-C1 cx states, that's why default
>> settings have changed. But with this level of performance everybody will
>> just switch back to C1.
>>
>
> Could you eventually try another Cx state? I have similar trouble with VIA
> Samuel based system, available states are C1, C2 and C3. LOW means C3, but
> this is terribly slow and unusable, in effect. C2, set via sysctl works well.
> I think ACPI developers would like to know our experiences.

  My hardware claims only C1 and C2, if notebook was started with AC power:

hw.acpi.cpu.cx_supported: C1/1 C2/1

and C1-C3 if it was started on batteries:

hw.acpi.cpu.cx_supported: C1/1 C2/1 C3/1

(BTW, is it normal?)

If I switch to C2, it not only slows machine down, but also breaks timer-based
delays:

root at notebook# date;sleep 5;date
Sat Mar  4 03:21:44 EET 2006
Sat Mar  4 03:21:49 EET 2006
root at notebook# sysctl hw.acpi.cpu.cx_lowest=C2
hw.acpi.cpu.cx_lowest: C1 -> C2
root at notebook# date;sleep 5;date
Sat Mar  4 03:22:46 EET 2006
Sat Mar  4 03:23:18 EET 2006

(redraw delay in 'top -s 1' raises to 6-7 seconds). C3 (when it's available)
looks the same as C2.

  Of course I run my notebook forced to C1. I would be glad to provide any
additional info in order to help developers to fix the issue. I'm a novice
in ACPI-related stuff so I don't dig into it myself.


Sincerely, Dmitry
-- 
Atlantis ISP, System Administrator
e-mail:  dmitry at atlantis.dp.ua
nic-hdl: LYNX-RIPE