[indiana-discuss] b111a: Power Management and nwam-issue ?

[Mark Haywood]

Brock Pytlik wrote:
> Mark Haywood wrote:
>> Mark Haywood wrote:
>>> Brock Pytlik wrote:
>>>> Bill Nesheim wrote:
>>>>> On 04/24/09 11:55, Mark Haywood wrote:
>>>>>> Bill Nesheim wrote:
>>>>>>> Detlef Drewanz at Sun.COM wrote:
>>>>>>>> Hi,
>>>>>>>> with b111a now I can boot again my Toshiba M9.
>>>>>>>>
>>>>>>>> 1.I discoverred that Power Management seems no longer to work.
>>>>>>>> (The fan is blowing like hell and "kstat -m cpu_info" shows in
>>>>>>>> currentClock_Hz 2001000000
>>>>>>>> but
>>>>>>>> supported_frequencies_Hz
>>>>>>>> 800000000:1200000000:1600000000:2000000000:2001000000
>>>>>>>>
>>>>>>>> New Bug with b111a ? 
>>>>>
>>>>> <snip>
>>>>>>
>>>>>> The current_clock_Hz is a little deceiving. The Power Aware 
>>>>>> Dispatcher results in much more frequent P-State transitions. Try 
>>>>>> running /usr/bin/powertop for better information.
>>>>>>
>>>>>> Mark
>>>>>
>>>>> Indeed.  powertop shows the system switching nicely between 800Mhz 
>>>>> and 1401 Mhz), with most time spent at 800.  Thanks.
>>>>>    -- Bill
>>>>>
>>>> Unfortunately, I don't see the same thing on my Tecra M10. Perhaps 
>>>> I haven't twiddled the right switch to turn power management on, 
>>>> but I would've expected this to have been on by default. In any 
>>>> case, powertop reports that my machine's always at 2531 Mhz (more 
>>>> detailed output below in case that helps).
>>>>
>>>> Brock
>>>>
>>>>                                        OpenSolaris PowerTOP version 
>>>> 1.1
>>>>
>>>> Cn            Avg    residency    P-states (frequencies)
>>>> C0 (cpu    running)        (20.3%)         800 Mhz    0.0%
>>>> C1            0.5ms    (79.7%)        1600 Mhz    0.0%
>>>>                                                2530 Mhz    0.0%
>>>>                                                2531 Mhz    100.0%
>>>>
>>>> Wakeups-from-idle per second: 1715.5    interval: 5.0s
>>>> Power usage (ACPI estimate): 0.000W (running on    AC power, 
>>>> fully    charged)
>>>>
>>>> Top causes for wakeups:
>>>> 20.7% (354.7)               <kernel> :    genunix`cv_wakeup
>>>> 9.7% (166.8)         <interrupt> :    audiohd#0
>>>> 7.5% (129.4)                  sched :    <cross calls>
>>>> 5.8% (100.2)               <kernel> :    genunix`clock
>>>> 2.4% (    41.8)         <interrupt> :    iwh#0
>>>> 1.4% (    23.2)               <kernel> :    
>>>> uhci`uhci_handle_root_hub_status_change
>>>> 0.8% (    14.4)         <interrupt> :    i8042#0
>>>> 0.8% (    13.6)         firefox-bin :    <cross calls>
>>>> 0.6% (    10.0)               <kernel> :    genunix`delay_wakeup
>>>> 0.5% (     8.2)        gnome-netstatus- :    <cross calls>
>>>> 0.5% (     7.8)               <kernel> :    
>>>> ehci`ehci_handle_root_hub_status_change
>>>> 0.3% (     6.0)               <kernel> :    uhci`uhci_cmd_timeout_hdlr
>>>> 0.3% (     4.4)               <kernel> :    genunix`lwp_timer_timeout
>>>> 0.2% (     4.0)               <kernel> :    genunix`schedpaging
>>>> 0.1% (     2.0)               <kernel> :    cpudrv`cpudrv_monitor_disp
>>>> 0.1% (     1.8)         <interrupt> :    e1000g#0
>>>> 0.1% (     1.2)               <kernel> :    
>>>> sd`sd_pm_idletimeout_handler
>>>> 0.1% (     1.0)               <kernel> :    nvidia`nvidia_rc_timer
>>>
>>> Your idle system looks a lot more busy than mine. I'm spending 
>>> almost 90% of my time in C1 while you are almost 80%.  Eric knows 
>>> the intimate details of how the dispatcher determines to change 
>>> P-States, so maybe he can explain.
>>>
>>> In the meantime, you *might* get better power savings in poll mode. 
>>> Just change
>>>
>>> cpudrv enable
>>>
>>> to
>>>
>>> cpudrv enable poll-mode
>>
>> Sorry, make that "cpupm" instead of '"cpudrv" above.
> Thanks, that made a huge difference. I'm now spending about 65-70% of 
> time in C1 (down from 80%), but now I'm seeing my P-states change 
> instead of being pegged at 2531Mhz. When I'm essentially idling, I'm @ 
> 800Mhz, and popping up to higher P states periodically.

That seems like a lot of time to be spending in C0. Is your system 
supposed to be essentially idle? I see a lot of wakeups due to audiohd?

>
>
> One other thing I'm noticing that seems suspicious is that my 
> percentage is always at 100% for one P state. It seems somewhat 
> unlikely (to me at least) that it's changing P states exactly on every 
> 5 second interval that powertop's using to sample. Is it possible that 
> the percentage isn't doing quite what's expected?

This doesn't surprise me all that much. In poll-mode the P-State 
transitions are much less frequent. You probably won't see them change 
often at all unless you either go from being idle to pretty busy or from 
busy to idle. In event-mode you should see transitions many times a 
second even on a somewhat idle system.

Mark

>
> Brock
>>
>>>
>>> in /etc/power.conf and run /usr/sbin/pmconfig.
>>>
>>> Mark
>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> indiana-discuss mailing list
>>>>> indiana-discuss at opensolaris.org
>>>>> http://mail.opensolaris.org/mailman/listinfo/indiana-discuss
>>>>
>>>
>>> _______________________________________________
>>> indiana-discuss mailing list
>>> indiana-discuss at opensolaris.org
>>> http://mail.opensolaris.org/mailman/listinfo/indiana-discuss
>>
>