Pentium M and 2.4. kernel: does not work

Pentium M and 2.4. kernel: does not work

[Georg Sauthoff]

Hi,

I am just testing the 2.4.22-pre6 patch from Dominik Brodowski. I applied it 
to 2.4.22-rc2 (without failures) and selected following options:
[*] CPU Frequency scaling
[*]  CPU frequency table helpers
[*]  "userspace" for userspace frequency scaling
<*>  Intel Pentium-M Enhanced SpeedStep

So I have /proc/sys/cpu/0/, but "cat speed speed-max speed-min" prints only 0 
to the console.

A dmesg | grep -i cpu prints this:
Initializing CPU#0
Intel machine check reporting enabled on CPU#0.
CPU:     After generic, caps: a7e9f9bf 00000000 00000000 00000000
CPU:             Common caps: a7e9f9bf 00000000 00000000 00000000
CPU: Intel(R) Pentium(R) M processor 1300MHz stepping 05

Ok, perhaps I oversee something ... 

Regards
Georg Sauthoff

Re: Pentium M and 2.4. kernel: does not work

[Dominik Brodowski]

Hi Georg,

On Sat, Aug 16, 2003 at 12:43:30AM +0200, Georg Sauthoff wrote:
> So I have /proc/sys/cpu/0/, but "cat speed speed-max speed-min" prints only 0 
> to the console.
That's not a bug, it's a feature. If you want to set the CPU frequency
manually, you need to enable this "governor" first. [Other "governors" set
to lowest and highest speed permanently (powersave/performance), and there
could be many more such "governors" which adapt the CPU frequency to current
system load etc. These only need to be written....]

So, 
# echo -n "0%0%100%userspace" > /proc/cpufreq
first, then you can use the
/proc/sys/cpu/0/speed
file to set the CPU speed manually.

	Dominik

Re: Pentium M and 2.4. kernel: does not work

[Jeremy Fitzhardinge]

On Fri, 2003-08-15 at 14:30, Dominik Brodowski wrote:
> Hi Georg,
> 
> On Sat, Aug 16, 2003 at 12:43:30AM +0200, Georg Sauthoff wrote:
> > So I have /proc/sys/cpu/0/, but "cat speed speed-max speed-min" prints only 0 
> > to the console.
> That's not a bug, it's a feature. 

Isn't -pre6 the one which has the problem reading the EST cap bit?  Or
have you fixed it?

	J

Re: Pentium M and 2.4. kernel: does not work

[Dominik Brodowski]

On Sat, Aug 16, 2003 at 12:34:51AM -0700, Jeremy Fitzhardinge wrote:
> On Fri, 2003-08-15 at 14:30, Dominik Brodowski wrote:
> > Hi Georg,
> > 
> > On Sat, Aug 16, 2003 at 12:43:30AM +0200, Georg Sauthoff wrote:
> > > So I have /proc/sys/cpu/0/, but "cat speed speed-max speed-min" prints only 0 
> > > to the console.
> > That's not a bug, it's a feature. 
> 
> Isn't -pre6 the one which has the problem reading the EST cap bit?  Or
> have you fixed it?

Oh, indeed, it is. Will update the backport RSN.

	Dominik

Re: Pentium M and 2.4. kernel: does not work

[Georg Sauthoff]

On Friday 15 August 2003 23:30, Dominik Brodowski wrote:
> That's not a bug, it's a feature. If you want to set the CPU frequency
> manually, you need to enable this "governor" first. [Other "governors" set
> to lowest and highest speed permanently (powersave/performance), and there
> could be many more such "governors" which adapt the CPU frequency to
> current system load etc. These only need to be written....]

Ok, I did not do RTMF enough I think ...

> So,
> # echo -n "0%0%100%userspace" > /proc/cpufreq
             ^^^^^^^^ What does this mean?

> first, then you can use the
> /proc/sys/cpu/0/speed
> file to set the CPU speed manually.

Hm, unfortunetly after I set the userspace governor (like above) the 
/proc/sys/cpu/0/s* gives me 0 again ...

In the kernel logs I don't see any messages from cpufreq. Ok, after a very 
quick look at the patch sources I don't see printk's there, too.

Regards
Georg Sauthoff

Re: Pentium M and 2.4. kernel: does not work

[Dominik Brodowski]

On Sat, Aug 16, 2003 at 10:28:50AM +0200, Georg Sauthoff wrote:
> > So,
> > # echo -n "0%0%100%userspace" > /proc/cpufreq
>              ^^^^^^^^ What does this mean?

set the CPU 0  0%
to something between
                 0%
and                100%
and let the             userspace
governor decide which speed to use in between. It's much easier with the
2.6. sysfs interface -- one of the reasons I'm not really too excited about
the cpufreq backport for 2.4. [besides, I _only_ use 2.[56] kernels...]

> > first, then you can use the
> > /proc/sys/cpu/0/speed
> > file to set the CPU speed manually.
> 
> Hm, unfortunetly after I set the userspace governor (like above) the 
> /proc/sys/cpu/0/s* gives me 0 again ...
> 
> In the kernel logs I don't see any messages from cpufreq. Ok, after a very 
> quick look at the patch sources I don't see printk's there, too.

Probably a BUG I introduced in the latest backport. Hopefully I'll have
the time tomorrow to update it. In the meantime: what does /proc/cpufreq
tell?

	Dominik

Re: Pentium M and 2.4. kernel: does not work

[Georg Sauthoff]

On Saturday 16 August 2003 20:53, Dominik Brodowski wrote:
> On Sat, Aug 16, 2003 at 10:28:50AM +0200, Georg Sauthoff wrote:
> > > So,
> > > # echo -n "0%0%100%userspace" > /proc/cpufreq
> >
> >              ^^^^^^^^ What does this mean?
>
> set the CPU 0  0%
> to something between
>                  0%
> and                100%
> and let the             userspace
> governor decide which speed to use in between. It's much easier with the
> 2.6. sysfs interface -- one of the reasons I'm not really too excited about
> the cpufreq backport for 2.4. [besides, I _only_ use 2.[56] kernels...]

Well, I tested a last 2.5er and it worked very stable with my hardware, and 
the interface for cpufreq was easy to use (ok - the Pentium M support was not 
so far at this time I think). So everything just fine, but the §"!*$^ Cisco 
VPN at the university needs a §$=?)"# closed source cisco kernel module ...
Ok, the 2.5er has IPsec support included, but I didn't find so much 
documentation about it (specially how to configure the userspace programs). 
And btw it seems that an encrypted group passward of a vpn-gateway doesn't 
make it easier to use other client software instead of the buggy cisco 
client.

> > > first, then you can use the
> > > /proc/sys/cpu/0/speed
> > > file to set the CPU speed manually.
> >
> > Hm, unfortunetly after I set the userspace governor (like above) the
> > /proc/sys/cpu/0/s* gives me 0 again ...
> >
> > In the kernel logs I don't see any messages from cpufreq. Ok, after a
> > very quick look at the patch sources I don't see printk's there, too.
>
> Probably a BUG I introduced in the latest backport. Hopefully I'll have
> the time tomorrow to update it. In the meantime: what does /proc/cpufreq
> tell?

It does tell (AFAIRC):
min-policy max-policy ... (sorry don't remember the end of the line)

I will mail the exact line tommorow.

Regards
Georg Sauthoff

Re: Pentium M and 2.4. kernel: does not work

[Georg Sauthoff]

On Saturday 16 August 2003 21:57, Georg Sauthoff wrote:
> > Probably a BUG I introduced in the latest backport. Hopefully I'll have
> > the time tomorrow to update it. In the meantime: what does /proc/cpufreq
> > tell?
>
> It does tell (AFAIRC):
> min-policy max-policy ... (sorry don't remember the end of the line)
>
> I will mail the exact line tommorow.

Ok, here it comes:
# cat /proc/cpufreq 
          minimum CPU frequency  -  maximum CPU frequency  -  policy
# echo -n "0%0%100%userspace" > /proc/cpufreq
# cat /proc/cpufreq 
          minimum CPU frequency  -  maximum CPU frequency  -  policy

Hope that helps.

Regards
Georg Sauthoff

Re: Pentium M and 2.4. kernel: does not work

[Georg Sauthoff]

On Monday 18 August 2003 19:50, you wrote:

> Yes it did. Indeed no cpufreq driver got properly loaded -- a bug in
> speedstep-centrino prevented this. Can you try the attached version,
> please?

Sure, no problem:

# /proc/cpufreq
          minimum CPU frequency  -  maximum CPU frequency  -  policy
CPU  0       600000 kHz ( 46 %)  -    1300000 kHz (100 %)  -  performance
# echo -n "0%0%100%userspace" > /proc/cpufreq
# cat /proc/cpufreq
          minimum CPU frequency  -  maximum CPU frequency  -  policy
CPU  0       600000 kHz ( 46 %)  -    1300000 kHz (100 %)  -  userspace
# cat /proc/sys/cpu/0/speed
1300000
# cat /proc/sys/cpu/0/speed-min
600000
# cat /proc/sys/cpu/0/speed-min > /proc/sys/cpu/0/speed
# cat /proc/sys/cpu/0/speed
600000

So it seems to work now.

A 'speed rsa' in openssl is nearly perfectly about a factor of 2 slower in 
min-speed in comparison to max-speed. Like it should be! At max-speed the fan 
turns on and and min-speed not (while openssl ist testing).

Great.

Regards
Georg Sauthoff