Re: Problem with ondemand on Athlon [and P4 too]

Re: Problem with ondemand on Athlon [and P4 too]

[Mattia Dongili]

On Fri, Oct 31, 2003 at 02:14:32PM +0100, Thierry Lathuille wrote:
> Hi,
[...]
> But echoing "ondemand" to 
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor has unexpected effects:
> - the keyboard stops working - but the mouse still works, I can use apps in 

happens the same here on a P4 1.8GHz with 2.6.0-test9

> - I can't kill any xterm anymore. But the other applications work and can be 
> killed normally.
> - trying to reboot hangs the system.

I tried from console (no X running). 
echoing ondemand > /sys/.../scaling_governor never returns, all the
consoles are freezed, keyboard too. I can ssh in and can cat
/sys/.../scaling_governor obtaining *ondemand* correctly. The *ondemand*
directory is also created. I forgot to check dmesg, I'll check it this
evening or tomorrow

I'll add more info when I'll go back home :)
-- 
mattia
:wq!

Problem with ondemand on Athlon

[Thierry Lathuille]

Hi,

I just tried Venkatesh's ondemand patches for kernel 2.6.0-test8 on my Acer 
laptop with Athlon XP1600+.

Everything works fine with performance, powersave and userspace governors, I 
can switch between frequencies without problem. 

But echoing "ondemand" to 
/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor has unexpected effects:
- the keyboard stops working - but the mouse still works, I can use apps in 
KDE without problem - as long as I don't need to type anything...
- I can't kill any xterm anymore. But the other applications work and can be 
killed normally.
- trying to reboot hangs the system.

I tried to figure out how far ondemand was executed, and added some debug 
messages to yours ; So I got :

ONDEMAND START: cpu0, max 1400000, min500000, cur <the last one>
(  then dbs_check_cpu is executed once)
ONDEMAND LIMIT: <idem>
And then, I don't get anything logged after the call to 
down(&dbs_sem);


I can reproduce the problem every time, whatever the frequency was before.
Please tell me what infos you need to help solve the problem !

(BTW : I installed cpufreqd-1.1-rc1 since then : this works perfectly !)



Thierry Lathuille

Re: Problem with ondemand on Athlon

[Dominik Brodowski]

Unfortunately, I can't reproduce the problem locally on my p4.
Can both of you try out the following, please:

pass "clock=pit" as a command line argument

The p4-clockmod driver [currently] needs to be built as a module then, 
and you need to pass stock_freq as an argument then. AFAIK, no such
requirement exists for powernow-k7.

Thanks,
	Dominik


On Fri, Oct 31, 2003 at 02:14:32PM +0100, Thierry Lathuille wrote:
> Hi,
> 
> I just tried Venkatesh's ondemand patches for kernel 2.6.0-test8 on my Acer 
> laptop with Athlon XP1600+.
> 
> Everything works fine with performance, powersave and userspace governors, I 
> can switch between frequencies without problem. 
> 
> But echoing "ondemand" to 
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor has unexpected effects:
> - the keyboard stops working - but the mouse still works, I can use apps in 
> KDE without problem - as long as I don't need to type anything...
> - I can't kill any xterm anymore. But the other applications work and can be 
> killed normally.
> - trying to reboot hangs the system.
> 
> I tried to figure out how far ondemand was executed, and added some debug 
> messages to yours ; So I got :
> 
> ONDEMAND START: cpu0, max 1400000, min500000, cur <the last one>
> (  then dbs_check_cpu is executed once)
> ONDEMAND LIMIT: <idem>
> And then, I don't get anything logged after the call to 
> down(&dbs_sem);
> 
> 
> I can reproduce the problem every time, whatever the frequency was before.
> Please tell me what infos you need to help solve the problem !
> 
> (BTW : I installed cpufreqd-1.1-rc1 since then : this works perfectly !)
> 
> 
> 
> Thierry Lathuille
> 
> 
> _______________________________________________
> Cpufreq mailing list
> Cpufreq@www.linux.org.uk
> http://www.linux.org.uk/mailman/listinfo/cpufreq
> 

On Fri, Oct 31, 2003 at 01:27:59PM +0100, Mattia Dongili wrote:
> On Fri, Oct 31, 2003 at 02:14:32PM +0100, Thierry Lathuille wrote:
> > Hi,
> [...]
> > But echoing "ondemand" to 
> > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor has unexpected effects:
> > - the keyboard stops working - but the mouse still works, I can use apps in 
> 
> happens the same here on a P4 1.8GHz with 2.6.0-test9
> 
> > - I can't kill any xterm anymore. But the other applications work and can be 
> > killed normally.
> > - trying to reboot hangs the system.
> 
> I tried from console (no X running). 
> echoing ondemand > /sys/.../scaling_governor never returns, all the
> consoles are freezed, keyboard too. I can ssh in and can cat
> /sys/.../scaling_governor obtaining *ondemand* correctly. The *ondemand*
> directory is also created. I forgot to check dmesg, I'll check it this
> evening or tomorrow
> 
> I'll add more info when I'll go back home :)
> -- 
> mattia
> :wq!
> 
> _______________________________________________
> Cpufreq mailing list
> Cpufreq@www.linux.org.uk
> http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: Problem with ondemand on Athlon

[Mattia Dongili]

On Tue, Nov 04, 2003 at 07:12:27PM +0100, Dominik Brodowski wrote:
> Unfortunately, I can't reproduce the problem locally on my p4.
> Can both of you try out the following, please:
> 
> pass "clock=pit" as a command line argument
> 
> The p4-clockmod driver [currently] needs to be built as a module then, 
> and you need to pass stock_freq as an argument then. AFAIK, no such
> requirement exists for powernow-k7.

ok, I hope I can do it this evening

bye
-- 
mattia
:wq!

Re: Problem with ondemand on Athlon

[Mattia Dongili]

On Tue, Nov 04, 2003 at 07:12:27PM +0100, Dominik Brodowski wrote:
> Unfortunately, I can't reproduce the problem locally on my p4.
> Can both of you try out the following, please:
> 
> pass "clock=pit" as a command line argument

this semms not to be a valid argument here
Linux inferi-2 2.6.0-test9-1 #3 Thu Oct 30 19:40:17 CET 2003 i686 GNU/Linux


> 
> The p4-clockmod driver [currently] needs to be built as a module then, 
> and you need to pass stock_freq as an argument then. AFAIK, no such
> requirement exists for powernow-k7.

having a P4 1.8GHz do I need stock_freq=1800000 ? (tried, it works)

let' start:

# modprobe p4_clockmod stock_freq=1800000

# echo *ondemand* > /sys/.../scaling_governor 
--> never returns

some infos from an ssh login:

Module                  Size  Used by
p4_clockmod             4488  3 
...

cpufreq: P4/Xeon(TM) CPU On-Demand Clock Modulation available
ONDEMAND START: cpu 0, max 1800000, min 225000, cur 1800000
ONDEMAND LIMIT: cpu 0, max 1800000, min 225000, cur 1800000

# cat /proc/cpuinfo 
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 2
model name      : Intel(R) Pentium(R) 4 CPU 1.80GHz
stepping        : 4
cpu MHz         : 1800.146
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 2
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm
bogomips        : 3547.13


after that echoing something else (tried with powersave) never
returns too...

what other info can I supply?

I tried playing with cpu-usage (compiling kernel) but
/sys/devices/system/cpu/cpu0/cpufreq/ondemand/current_freq always shows
1800000

bye
-- 
mattia
:wq!

Re: Problem with ondemand on Athlon

[Dominik Brodowski]

This is really strange... will provide a DEBUG()-printk patch soon, would be
great if you could test it then...

On Wed, Nov 05, 2003 at 07:21:10PM +0100, Mattia Dongili wrote:
> On Tue, Nov 04, 2003 at 07:12:27PM +0100, Dominik Brodowski wrote:
> > Unfortunately, I can't reproduce the problem locally on my p4.
> > Can both of you try out the following, please:
> > 
> > pass "clock=pit" as a command line argument
> 
> this semms not to be a valid argument here

? clock=pit specifies that the PIT should be used instead of the TSC as
timing source... so it should be a valid boot argument.

> Linux inferi-2 2.6.0-test9-1 #3 Thu Oct 30 19:40:17 CET 2003 i686 GNU/Linux
> > 
> > The p4-clockmod driver [currently] needs to be built as a module then, 
> > and you need to pass stock_freq as an argument then. AFAIK, no such
> > requirement exists for powernow-k7.
> 
> having a P4 1.8GHz do I need stock_freq=1800000 ? (tried, it works)

yes.

> # modprobe p4_clockmod stock_freq=1800000
> 
> # echo *ondemand* > /sys/.../scaling_governor 
> --> never returns

strange...

	Dominik

Re: Problem with ondemand on Athlon

[Mattia Dongili]

On Mon, Nov 10, 2003 at 09:59:12PM +0100, Dominik Brodowski wrote:
> This is really strange... will provide a DEBUG()-printk patch soon, would be
> great if you could test it then...

of course I can. In the meantime if you can tell a couple of interesting
functions and variables, I could add some printk-s by myself (if I get
some spare time before you do)

[...]
> > > pass "clock=pit" as a command line argument
> > 
> > this semms not to be a valid argument here
> 
> ? clock=pit specifies that the PIT should be used instead of the TSC as
> timing source... so it should be a valid boot argument.

sorry... my bad. I used it as a p4_clockmod argument and did not
bothered to check Documentation/kernel-parameters.txt

I'll re-test it as soon as possible.

bye
-- 
mattia
:wq!

Re: Problem with ondemand on Athlon

[Dominik Brodowski]

On Tue, Nov 11, 2003 at 10:51:43AM +0100, Mattia Dongili wrote:
> On Mon, Nov 10, 2003 at 09:59:12PM +0100, Dominik Brodowski wrote:
> > This is really strange... will provide a DEBUG()-printk patch soon, would be
> > great if you could test it then...
> 
> of course I can. In the meantime if you can tell a couple of interesting
> functions and variables, I could add some printk-s by myself (if I get
> some spare time before you do)

Basically the ->governor function in the ondemand() governor, the
__cpufreq_governor, cpufreq_governor, cpufreq_set_policy after the call to
__cpufreq_set_policy, and the sysfs fuctions (show / store /
store_scaling_governor).

> > > > pass "clock=pit" as a command line argument
> > > 
> > > this semms not to be a valid argument here
> > 
> > ? clock=pit specifies that the PIT should be used instead of the TSC as
> > timing source... so it should be a valid boot argument.
> 
> sorry... my bad. I used it as a p4_clockmod argument
Sorry for my unclear comment "command line argument"...

> I'll re-test it as soon as possible.
Many thanks.

	Dominik