Hi,

Unfortunately the cpu_info's current_clock_Hz "snapshot"
is relatively useless in human time scale.  It only shows
the instantaneous p-state at the time of the kstat snapshot.
It is not useful to compute an average over some time period.
Additionally the kstat command itself may cause the CPU's
power domain to momentarily transition to a faster p-state.

The kstat would be usable if it provided a running count
of hrtime spent in each p-state....   ;-)

Regards,
Bill


On 07/26/10 12:56, Andrej Podzimek wrote:
Hello,

So it seems that Lord Kernel is right and I'm wrong:

            OpenSolaris PowerTOP version 1.2

C-states (idle power)    Avg    Residency    P-states (frequencies)
C0 (cpu    running)        (6.1%)        1199 Mhz    93.7%
C1            1.4ms   (86.2%)         1333 Mhz        0.0%
C2            1.5ms   (6.5%)      1466 Mhz    0.0%
C3            1.5ms   (1.2%)         1599 Mhz    0.0%
                                                1733 Mhz    0.0%
                                                1794 Mhz(turbo)    6.3%

I read a couple of blogposts saying that processor states would be reflected in the output from 'kstat cpu_info'. That's why I got confused by the fact that those numbers from 'kstat cpu_info' *never* change. Anyway, powertop explains this. Scaling obviously does work, although kstat doesn't say so.

Andrej

Hi,

what is powertop saying?

Also, pm-discuss@ is better place for this.

Best regards,

Milan

Andrej Podzimek píše v po 26. 07. 2010 v 06:38 +0200:
Hello,

There seems to be no freuqency scaling on my machine. The frequencies shown by kstat remain at the highest possible value.

The output from kstat (for the first one of the eight virtual CPUs):

    module: cpu_info                        instance: 0
    name:   cpu_info0                       class:    misc
brand Intel(r) Core(tm) i7 CPU Q 820 @ 1.73GHz
        cache_id                        0
        clock_MHz                       1733
        clog_id                         0
        core_id                         0
        cpu_type                        i386
        crtime                          41,803748237
        current_clock_Hz                1734000000
        current_cstate                  3
        family                          6
        fpu_type                        i387 compatible
        chip_id                         0
implementation x86 (chipid 0x0 GenuineIntel 106E5 family 6 model 30 step 5 clock 1733 MHz)
        model                           30
        ncore_per_chip                  4
        ncpu_per_chip                   8
        pg_id                           5
        pkg_core_id                     0
        snaptime                        61391,9925084
        socket_type                     Unknown
        state                           on-line
        state_begin                     1280057052
        stepping                        5
supported_frequencies_Hz 1199000000:1333000000:1466000000:1599000000:1733000000:1734000000
        supported_max_cstates           3
        vendor_id                       GenuineIntel

This is in my /etc/power.conf:

    autopm            default
    autoS3            default
    cpu-threshold        1s
    autoshutdown        30        9:00 9:00        noshutdown
    cpupm  enable

BTW, the frequency value seems to be the TurboBoost „virtual frequency“... This is quite surprising, especially when it's shown for all the eight virtual CPUs at once all the time.

Is this a known issue or have I just forgotten to configure something? This is a ThinkPad W510 and frequency scaling works just fine on this machine under Linux.

Andrej

_______________________________________________
on-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/on-discuss



------------------------------------------------------------------------

_______________________________________________
pm-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pm-discuss

_______________________________________________
pm-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pm-discuss

Reply via email to