Re: Re: Centrino question

[Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>]

Stefan
The LAPIC TSC is not reliable when C3 runs.

Indeed, we need to either disable the TSC if C3 is enabled,
or disabled C3 if the TSC is used.

-Len

http://bugme.osdl.org/show_bug.cgi?id=2560

On Tue, 2004-05-25 at 03:27, Stefan Seyfried wrote:
> On Mon, May 24, 2004 at 09:11:18AM -0700, Ow Mun Heng wrote:
> > On Sun, 2004-05-23 at 06:21, Cyril Wattebled wrote:
> > > On 22 May 2004 23:24:10 -0400
> > > Len Brown <len.brown-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org> wrote:
> > > > Linux slows the processor to 600MHz when Windows gets down to 200MHz?
> > 
> > Len,
> > 
> > 	On my system, either 2.4 or 2.6 kernel, running on battery, cpufreq is at 600Mhz but
> > if i were running gkrellm, the x86info plugin will show that the cpufreq is actually fluactuating 
> > between 99 to 200 mHz. I'm not sure where this plugin gets it's data but..
> 
> I am quite sure that this is due to the tool you are using and due to C3 usage.
> I don't know about the gkrellm x86info plugin, but the suse 9.1 powersaved
> has a function to get the "real" CPU speed. It goes something like this:
> 
> float getRealProcessorSpeed() {
> 
>         /* get this function in cached memory */
>         gettimeofday(&tvstart, NULL);
>         __asm__ volatile ("rdtsc" : "=A" (x));
>         cycles[0] = x;
>         gettimeofday(&tvstart, NULL);
> 
>         /* we don't trust that this is any specific length of time */
>         sleep_left = sleep(1);
>         if (sleep_left || errno == EINTR)
>                 return -1.0;
>         __asm__ volatile ("rdtsc" : "=A" (x));
>         cycles[1] = x;
>         gettimeofday(&tvstop, NULL);
>         microseconds = ((tvstop.tv_sec-tvstart.tv_sec)*1000000) +
>                 (tvstop.tv_usec-tvstart.tv_usec);
> 
>         ret = (float)(cycles[1]-cycles[0])/microseconds;
>         return ret;
> }
> 
> There is of course some more checking around, it is only working on i386, 
> not on ia64.  This shows me:
> 
> seife-4N56LNXDvWs@public.gmane.org:~> powersave -r
> 82.559082 MHz
> 
> but from /sys/.../scaling_setspeed i see that my cpu is running on 600mhz.
> 
> seife-4N56LNXDvWs@public.gmane.org:~> cat /proc/acpi/processor/CPU0/power 
> active state:            C3
> default state:           C1
> bus master activity:     00000000
> states:
>     C1:                  promotion[C2] demotion[--] latency[000] usage[00020000]
>     C2:                  promotion[C3] demotion[C1] latency[002] usage[00737909]
>    *C3:                  promotion[--] demotion[C2] latency[205] usage[35772579]
> 
> if i force C2 state (by plugging in a usb mouse or playing sound), it shows
> the frequency correct:
> 
> seife-4N56LNXDvWs@public.gmane.org:~> powersave -r
> 598.058594 MHz
> 
> seife-4N56LNXDvWs@public.gmane.org:~> cat /proc/acpi/processor/CPU0/power 
> active state:            C2
> default state:           C1
> bus master activity:     ffffffff
> states:
>     C1:                  promotion[C2] demotion[--] latency[000] usage[00020850]
>    *C2:                  promotion[C3] demotion[C1] latency[002] usage[00770669]
>     C3:                  promotion[--] demotion[C2] latency[205] usage[35825285]
> 
> So the effects you are seeing are clearly related to a tool that is measuring
> wrong.



-------------------------------------------------------
This SF.Net email is sponsored by: Oracle 10g
Get certified on the hottest thing ever to hit the market... Oracle 10g. 
Take an Oracle 10g class now, and we'll give you the exam FREE.
http://ads.osdn.com/?ad_id=3149&alloc_id=8166&op=click