Cpufreq hang

Cpufreq hang

[Johan Braennlund]

Hi! It seems like there's a bug in cpufreq in the 2.5 kernels.

When using recent 2.5-series kernels and running on battery, cpufreqd
locks up hard when it's started. If I run on AC power, everything works
fine.

If I use a 2.4 kernel, such as 2.4.20-wolk4.0s-rc3 everything works fine,
no matter whether I run off the battery or not.

When the machine hangs, nothing appears in the logs, it just locks up and
I have to power it off. I'm using an Acer Aspire with an Athlon XP 1800+
processor.

The version of cpufreqd I'm using is 1.0-1, installed as a debian package.
I've tried compiling from source, but the result is the same.

If there's anything I can do to pinpoint the bug, please let me know.

Regards,

Johan

Re: Cpufreq hang

[Mattia Dongili]

On Wed, Jul 16, 2003 at 04:31:34PM +0200, Johan Braennlund wrote:
> Hi! It seems like there's a bug in cpufreq in the 2.5 kernels.
> 
> When using recent 2.5-series kernels and running on battery, cpufreqd
> locks up hard when it's started. If I run on AC power, everything works
> fine.

so are you still compiling with the deprecated /proc interface?
(cpufreqd 1.0 is broken wrt 2.[56] sysfs - the CVS code is fixed though)

> If I use a 2.4 kernel, such as 2.4.20-wolk4.0s-rc3 everything works fine,
> no matter whether I run off the battery or not.
> 
> When the machine hangs, nothing appears in the logs, it just locks up and
> I have to power it off. I'm using an Acer Aspire with an Athlon XP 1800+
> processor.

can you write directly to /proc/cpufreq without using the daemon?
eg:
echo "0%0%100%performance" > /proc/cpufreq

-- 
mattia
:wq!

Re: Cpufreq hang

[Johan Braennlund]

On Wed, 16 Jul 2003, Mattia Dongili wrote:

> can you write directly to /proc/cpufreq without using the daemon?
> eg:
> echo "0%0%100%performance" > /proc/cpufreq

That works, but if I substitute "powersave" for "performance" the system
hangs. In fact, I tried various changes in the above and anything with
"performance" works and anything with "powersave" hangs. The kernel I'm
using is 2.5.74-mm1, but I've seen cpufreqd hang on other recent 2.5
kernels as well, so I imagine the situation would be the same there.

On a 2.4 kernel, all the commands I tried worked.

- Johan

Re: Cpufreq hang

[Mattia Dongili]

On Wed, Jul 16, 2003 at 07:53:44PM +0200, Johan Braennlund wrote:
> 
> On Wed, 16 Jul 2003, Mattia Dongili wrote:
> 
> > can you write directly to /proc/cpufreq without using the daemon?
> > eg:
> > echo "0%0%100%performance" > /proc/cpufreq
> 
> That works, but if I substitute "powersave" for "performance" the system
> hangs. In fact, I tried various changes in the above and anything with
> "performance" works and anything with "powersave" hangs. The kernel I'm
> using is 2.5.74-mm1, but I've seen cpufreqd hang on other recent 2.5
> kernels as well, so I imagine the situation would be the same there.

so it's most probably a driver issue, anybody from the gurus?

-- 
mattia
:wq!

Re: Cpufreq hang

[Dominik Brodowski]

On Wed, Jul 16, 2003 at 07:53:44PM +0200, Johan Braennlund wrote:
> 
> On Wed, 16 Jul 2003, Mattia Dongili wrote:
> 
> > can you write directly to /proc/cpufreq without using the daemon?
> > eg:
> > echo "0%0%100%performance" > /proc/cpufreq
> 
> That works, but if I substitute "powersave" for "performance" the system
> hangs. In fact, I tried various changes in the above and anything with
> "performance" works and anything with "powersave" hangs. The kernel I'm
> using is 2.5.74-mm1, but I've seen cpufreqd hang on other recent 2.5
> kernels as well, so I imagine the situation would be the same there.
> 
> On a 2.4 kernel, all the commands I tried worked.

Can you please re-test it with 2.6.0-test1, and especially check whether the
following works:

a) loading the "userspace" governor [please modprobe it first, of course]
cd /sys/devices/system/cpu/cpu0/cpufreq/
echo -n "userspace" > scaling_governor

b) slowly decrease the current frequency by echoing the corresponding values
[in kHz!] into scaling_setspeed. You can read the frequency states supported
by your notebook in the "PST table" which is printed in dmesg. 

c) if these slow transitions work, please change the frequency more
"randomly", and also try to switch from lowest to highest frequency and
backwards.

d) please tell us whether these tests worked, or at what stage [esp. at what
frequency] it failed. Also, please include the part of "dmesg" where the PST
table is printed.

Many thanks for testing cpufreq,
	Dominik

Re: Cpufreq hang

[Johan Braennlund]

Hi!

On Sat, 19 Jul 2003, Dominik Brodowski wrote:

> Can you please re-test it with 2.6.0-test1, and especially check whether the
> following works:
>
> a) loading the "userspace" governor [please modprobe it first, of course]
> cd /sys/devices/system/cpu/cpu0/cpufreq/
> echo -n "userspace" > scaling_governor

[...]

I tried the tests you suggested using 2.6.0-test1-mm1, since I had that
kernel handy and everything worked. Cpufreqd also works they way it's
supposed to now, so everything is fine. I noticed that the cpufreq
changes in -mm1 were pretty minor and probably wouldn't affect the hang I
saw, so plain -test1 is probably okay too.

> Many thanks for testing cpufreq,
> 	Dominik

I'm glad to be able to do my little part.

Regards,

Johan