p4-clockmod doesn't seem to allow the highest possible speed

p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

Hi,

I can't seem to get my processor to run at full speed when using the
p4-clockmod driver:

root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_governor 
userspace
root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_max_freq 
2193205
root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# echo 2193205 > scaling_setspeed 
root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_setspeed 
1925000

What am I missing?

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

On Fri, Jan 09, 2004 at 07:18:51PM +0100, Wouter Verhelst wrote:
> Hi,
> 
> I can't seem to get my processor to run at full speed when using the
> p4-clockmod driver:
> 
> root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_governor 
> userspace
> root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_max_freq 
> 2193205
> root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# echo 2193205 > scaling_setspeed 
> root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_setspeed 
> 1925000
> 
> What am I missing?

Which kernel version are you using? What does "dmesg" tell us?

	Dominik

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

[-- Attachment #1.1: Type: text/plain, Size: 1399 bytes --]

On Sat, Jan 10, 2004 at 11:15:34AM +0100, Dominik Brodowski wrote:
> On Fri, Jan 09, 2004 at 07:18:51PM +0100, Wouter Verhelst wrote:
> > Hi,
> > 
> > I can't seem to get my processor to run at full speed when using the
> > p4-clockmod driver:
> > 
> > root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_governor 
> > userspace
> > root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_max_freq 
> > 2193205
> > root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# echo 2193205 > scaling_setspeed 
> > root@worldmusic:/sys/devices/system/cpu/cpu0/cpufreq# cat scaling_setspeed 
> > 1925000
> > 
> > What am I missing?
> 
> Which kernel version are you using?

Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
p4-clockmod on any kernel before that).

> What does "dmesg" tell us?

234 lines of output. Seems a bit excessive to put that in this mail, so
I've put it on <http://users.pandora.be/wouter.verhelst/cpufreq.tgz>,
together with the contents of /proc/cpuinfo and my .config. If you need
any further information, please ask.

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1.1: Type: text/plain, Size: 1050 bytes --]

On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> p4-clockmod on any kernel before that).

You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
choice for your system... Nonetheless, let us find the bug:

> > What does "dmesg" tell us?
> 
> 234 lines of output. Seems a bit excessive to put that in this mail, so
> I've put it on <http://users.pandora.be/wouter.verhelst/cpufreq.tgz>,
> together with the contents of /proc/cpuinfo and my .config. If you need
> any further information, please ask.

Thank you very much. Two questions: Do the init scripts modprobe either 
the acpi module (acpi.ko) or the speedstep-ich module (speedstep-ich.ko)
anyway? What does /proc/cpuinfo say before modprobing p4-clockmod, and
after modprobing p4-clockmod?

Also, could you please uncomment the printk in line 190 of
linux/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c and send me a diff of the
dmesgs?

Thank you very much,
	Dominik

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

[-- Attachment #1.1: Type: text/plain, Size: 2435 bytes --]

Op za 10-01-2004, om 14:18 schreef Dominik Brodowski:
> On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> > Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> > p4-clockmod on any kernel before that).
> 
> You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
> choice for your system...

Ah? Didn't know. OK, thanks :-)

(when checking the description now, it clearly says "P4-M"; must've
missed that before)

> Nonetheless, let us find the bug:
> 
> > > What does "dmesg" tell us?
> > 
> > 234 lines of output. Seems a bit excessive to put that in this mail, so
> > I've put it on <http://users.pandora.be/wouter.verhelst/cpufreq.tgz>,
> > together with the contents of /proc/cpuinfo and my .config. If you need
> > any further information, please ask.
> 
> Thank you very much. Two questions: Do the init scripts modprobe either 
> the acpi module (acpi.ko) or the speedstep-ich module (speedstep-ich.ko)
> anyway?

No. I compiled the ACPI module because I didn't want to convert my
scripts right away and the speedstep-ich one because I wasn't sure which
one to choose, but my /etc/modules (which lists modules to be loaded at
boot time) contained p4-clockmod alone (contained, because it's
speedstep-ich now ;-)

> What does /proc/cpuinfo say before modprobing p4-clockmod, and
> after modprobing p4-clockmod?

The only difference with the one you already have is the Mhz number
(which seems logical). Before it's loaded, it's at 2194.424; Immediately
after loading p4-clockmod, that number was still the same, but in the
mean time it went back to 1919.054

Note that I had to reboot my box to unload the module. This is probably
related to the above, but just in case it isn't: I cannot rmmod the
p4-clockmod module, trying anyway results in rmmod segfaulting -- at
least once the frequency has been altered. I didn't reboot again just to
test that, but if you want, it's no problem.

> Also, could you please uncomment the printk in line 190 of
> linux/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c and send me a diff of the
> dmesgs?

The extra line reads:

cpufreq: P4 - MSR_EBC_FREQUENCY_ID: 0x16300a16 0x0

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
Most people have two reasons for doing anything -- a good reason, and
the real reason

[-- Attachment #1.2: Dit berichtdeel is digitaal ondertekend --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1: Type: text/plain, Size: 1488 bytes --]

> Note that I had to reboot my box to unload the module. This is probably
> related to the above, but just in case it isn't: I cannot rmmod the
> p4-clockmod module, trying anyway results in rmmod segfaulting -- at
> least once the frequency has been altered. I didn't reboot again just to
> test that, but if you want, it's no problem.

That's unrelated, and I was able to reproduce it on my notebook as well. The
attached patch [which should apply with some offset] should fix that. (Davej:
it applies on top of the p4-clockmod-fixup. Please queue for merging with 
Linus, thanks.)

On Sat, Jan 10, 2004 at 04:02:52PM +0100, Wouter Verhelst wrote:
> > What does /proc/cpuinfo say before modprobing p4-clockmod, and
> > after modprobing p4-clockmod?
> 
> The only difference with the one you already have is the Mhz number
> (which seems logical). Before it's loaded, it's at 2194.424; Immediately
> after loading p4-clockmod, that number was still the same, but in the
> mean time it went back to 1919.054
I think ACPI is messing with us. It detects a high temperature situation,
tries to scale down the CPU, and forces down the CPU speed. Could you send
me the output of /proc/acpi/processor/*/* and /proc/acpi/thermal_zone/*/*
when you notice this situation again?

BTW, could it be that this is still the 2.6.0 module, instead of the 2.6.1
variant? 2.6.1 should report more "clean" values in
/sys/devices/system/cpu/cpufreq/... it should report 2200000 instead of
2193205.

	Dominik

[-- Attachment #2: p4-clockmod-fix-oops.diff --]
[-- Type: text/plain, Size: 797 bytes --]

Fix an OOPS in unloading. Reverting frequency changes on unloading is
uncommon for cpufreq drivers, so let's remove this speciality.

 arch/i386/kernel/cpu/cpufreq/p4-clockmod.c |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)

diff -ruN linux-original/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c linux/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c
--- linux-original/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c	2004-01-10 16:26:06.914893464 +0100
+++ linux/arch/i386/kernel/cpu/cpufreq/p4-clockmod.c	2004-01-10 16:25:25.983116040 +0100
@@ -246,7 +246,7 @@
 static int cpufreq_p4_cpu_exit(struct cpufreq_policy *policy)
 {
 	cpufreq_frequency_table_put_attr(policy->cpu);    
-	return cpufreq_p4_setdc(policy->cpu, DC_DISABLE);
+	return 0;
 }
 
 static struct freq_attr* p4clockmod_attr[] = {

[-- Attachment #3: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

[-- Attachment #1.1: Type: text/plain, Size: 2442 bytes --]

On Sat, Jan 10, 2004 at 04:38:34PM +0100, Dominik Brodowski wrote:
> > Note that I had to reboot my box to unload the module. This is probably
> > related to the above, but just in case it isn't: I cannot rmmod the
> > p4-clockmod module, trying anyway results in rmmod segfaulting -- at
> > least once the frequency has been altered. I didn't reboot again just to
> > test that, but if you want, it's no problem.
> 
> That's unrelated, and I was able to reproduce it on my notebook as well. The
> attached patch [which should apply with some offset] should fix that.

OK, thanks.

[...]
> On Sat, Jan 10, 2004 at 04:02:52PM +0100, Wouter Verhelst wrote:
> > > What does /proc/cpuinfo say before modprobing p4-clockmod, and
> > > after modprobing p4-clockmod?
> > 
> > The only difference with the one you already have is the Mhz number
> > (which seems logical). Before it's loaded, it's at 2194.424;
> > Immediately after loading p4-clockmod, that number was still the
> > same, but in the mean time it went back to 1919.054
>
> I think ACPI is messing with us. It detects a high temperature
> situation, tries to scale down the CPU, and forces down the CPU speed.

Hm. This one got me thinking; I think I just found it. I forgot that I
installed cpufreqd a while ago, which seems to be the culprit: when I
start cpufreqd, the system falls back to 1925Mhz. I've removed it now,
however, as killing cpufreqd does not release the maximum speed back to
it's normal value of 2.2Ghz, I'm not entirely sure; so...

> Could you send me the output of /proc/acpi/processor/*/* and
> /proc/acpi/thermal_zone/*/* when you notice this situation again?

... I've still put those at
http://users.pandora.be/wouter.verhelst/proc-acpi.tgz

> BTW, could it be that this is still the 2.6.0 module, instead of the
> 2.6.1 variant? 2.6.1 should report more "clean" values in
> /sys/devices/system/cpu/cpufreq/... it should report 2200000 instead
> of 2193205.

Yes, the first mail I sent (with the command line stuff in /sys) was
before I installed 2.6.1, still running 2.6.0 at that point. Sorry for
the confusion :-)

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

[-- Attachment #1.1: Type: text/plain, Size: 796 bytes --]

On Sat, Jan 10, 2004 at 02:18:24PM +0100, Dominik Brodowski wrote:
> On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> > Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> > p4-clockmod on any kernel before that).
> 
> You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
> choice for your system...

This doesn't seem to work, it says 'No such device'. I have a VIA 82C686
south bridge, not an Intel one, so I guess that's the reason.

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

[-- Attachment #1.2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1.1: Type: text/plain, Size: 747 bytes --]

On Sat, Jan 10, 2004 at 04:21:45PM +0100, Wouter Verhelst wrote:
> On Sat, Jan 10, 2004 at 02:18:24PM +0100, Dominik Brodowski wrote:
> > On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> > > Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> > > p4-clockmod on any kernel before that).
> > 
> > You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
> > choice for your system...
> 
> This doesn't seem to work, it says 'No such device'. I have a VIA 82C686
> south bridge, not an Intel one, so I guess that's the reason.

Unfortunately, you're right. VIA doesn't publish their spec sheets as Intel
does, so there's no speedstep-via driver (yet). Maybe acpi.ko works?

	Dominik

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

On Sat, Jan 10, 2004 at 04:41:10PM +0100, Dominik Brodowski wrote:
> On Sat, Jan 10, 2004 at 04:21:45PM +0100, Wouter Verhelst wrote:
> > On Sat, Jan 10, 2004 at 02:18:24PM +0100, Dominik Brodowski wrote:
> > > On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> > > > Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> > > > p4-clockmod on any kernel before that).
> > > 
> > > You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
> > > choice for your system...
> > 
> > This doesn't seem to work, it says 'No such device'. I have a VIA 82C686
> > south bridge, not an Intel one, so I guess that's the reason.
> 
> Unfortunately, you're right. VIA doesn't publish their spec sheets as Intel
> does, so there's no speedstep-via driver (yet). Maybe acpi.ko works?

It does, but when my CPU is throttled, it makes an uncomfortable sound
at high frequency, so I try to avoid that.

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Ducrot Bruno]

On Sat, Jan 10, 2004 at 05:32:12PM +0100, Wouter Verhelst wrote:
> On Sat, Jan 10, 2004 at 04:41:10PM +0100, Dominik Brodowski wrote:
> > On Sat, Jan 10, 2004 at 04:21:45PM +0100, Wouter Verhelst wrote:
> > > On Sat, Jan 10, 2004 at 02:18:24PM +0100, Dominik Brodowski wrote:
> > > > On Sat, Jan 10, 2004 at 12:08:00PM +0100, Wouter Verhelst wrote:
> > > > > Vanilla 2.6.1, but I had the same problem with 2.6.0 (didn't try
> > > > > p4-clockmod on any kernel before that).
> > > > 
> > > > You shouldn't need to use p4-clockmod at all... speedstep-ich is a better
> > > > choice for your system...
> > > 
> > > This doesn't seem to work, it says 'No such device'. I have a VIA 82C686
> > > south bridge, not an Intel one, so I guess that's the reason.
> > 
> > Unfortunately, you're right. VIA doesn't publish their spec sheets as Intel
> > does, so there's no speedstep-via driver (yet). Maybe acpi.ko works?
> 
> It does, but when my CPU is throttled, it makes an uncomfortable sound
> at high frequency, so I try to avoid that.
> 

Do you mean the processor is at high speed (performance), and throttling
is enabled?

So it should not be even surprising.

The P4 is able to throttle itself, so I guess the throttling
things done externally by the south bridge would have not even
been tested by the platform designer...

-- 
Ducrot Bruno

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

On Sat, Jan 10, 2004 at 07:53:08PM +0100, Ducrot Bruno wrote:
> On Sat, Jan 10, 2004 at 05:32:12PM +0100, Wouter Verhelst wrote:
> > On Sat, Jan 10, 2004 at 04:41:10PM +0100, Dominik Brodowski wrote:
> > > Unfortunately, you're right. VIA doesn't publish their spec sheets
> > > as Intel does, so there's no speedstep-via driver (yet). Maybe
> > > acpi.ko works?
> > 
> > It does, but when my CPU is throttled, it makes an uncomfortable
> > sound at high frequency, so I try to avoid that.
> 
> Do you mean the processor is at high speed (performance), and throttling
> is enabled?
> 
> So it should not be even surprising.
> 
> The P4 is able to throttle itself, so I guess the throttling
> things done externally by the south bridge would have not even
> been tested by the platform designer...

Uh, I'm not entirely sure I understand what you mean here. The issue is:
when I put the processor in some other T-state than T0, it makes noise.
That happens both when it's in P0 or P1.

Not that it matters -- p4-clockmod seems to do it right :)

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Ducrot Bruno]

On Sat, Jan 10, 2004 at 08:35:26PM +0100, Wouter Verhelst wrote:
> On Sat, Jan 10, 2004 at 07:53:08PM +0100, Ducrot Bruno wrote:
> > On Sat, Jan 10, 2004 at 05:32:12PM +0100, Wouter Verhelst wrote:
> > > On Sat, Jan 10, 2004 at 04:41:10PM +0100, Dominik Brodowski wrote:
> > > > Unfortunately, you're right. VIA doesn't publish their spec sheets
> > > > as Intel does, so there's no speedstep-via driver (yet). Maybe
> > > > acpi.ko works?
> > > 
> > > It does, but when my CPU is throttled, it makes an uncomfortable
> > > sound at high frequency, so I try to avoid that.
> > 
> > Do you mean the processor is at high speed (performance), and throttling
> > is enabled?
> > 
> > So it should not be even surprising.
> > 
> > The P4 is able to throttle itself, so I guess the throttling
> > things done externally by the south bridge would have not even
> > been tested by the platform designer...
> 
> Uh, I'm not entirely sure I understand what you mean here. The issue is:
> when I put the processor in some other T-state than T0, it makes noise.
> That happens both when it's in P0 or P1.

A CPU have a pin which is used in order to stop execution.  By
generating a signal which toggle that pin, you throttle down a CPU.
If now the circuitry in order to generate that signal contains
some cheap capacitors, this may generate sounds.

All you do via ACPI T states is to configure a chipset (most likely
the south bridge) in order to generate the signal at a given
frequency, and T0 will disable this signal.

All P4 have the ability, though, to throttle without another hardware
circuitry.  p4-clockmod is the software interface for that feature.

In short, acpi T-state and p4-clockmod will give you the exact same
behaviour, but is done diferently, one requiring an external hardware
which seems, in your case, to contains cheap components (but that do
not matter, since you should use p4-clockmod instead of ACPI T-states
anyway).

-- 
Ducrot Bruno

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1.1: Type: text/plain, Size: 238 bytes --]

On Sun, Jan 11, 2004 at 04:48:25PM +0100, Ducrot Bruno wrote:
> (but that do
> not matter, since you should use p4-clockmod instead of ACPI T-states
> anyway).

What are the advantages of p4-clockmod in favour of ACPI T-states?

	Dominik

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Ducrot Bruno]

On Sun, Jan 11, 2004 at 06:30:03PM +0100, Dominik Brodowski wrote:
> On Sun, Jan 11, 2004 at 04:48:25PM +0100, Ducrot Bruno wrote:
> > (but that do
> > not matter, since you should use p4-clockmod instead of ACPI T-states
> > anyway).
> 
> What are the advantages of p4-clockmod in favour of ACPI T-states?
> 

1- Latency (MSR access instead of ISA IO space).
2- Safety (p4-clockmod do not rely on external circuitry, or hw bugs from
   the southbridge).
3- even though I don't think p4-clockmod do not support this, it is
   possible to throttle the CPU in case of overheating issue, without
   need to do that in software.
4- a little more power efficient, since generating a signal cost power
   (but I don't think that you gain so much).

-- 
Ducrot Bruno

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1.1: Type: text/plain, Size: 1893 bytes --]

On Sat, Jan 10, 2004 at 08:35:26PM +0100, Wouter Verhelst wrote:
> That happens both when it's in P0 or P1.

So ACPI works to set the processor in P0 and P1? If so, you really really
should use the ACPI cpufreq driver instead of the p4-clockmod driver. The
former will offer you true "SpeedStep" support, the latter only some clock
modulation which won't save you much energy as ACPI supports the C2 idle
state anyway. 

> Not that it matters -- p4-clockmod seems to do it right :)

It's not as noisy, indeed. But it's not right -- as long as acpi.ko works.

> > I think ACPI is messing with us. It detects a high temperature
> > situation, tries to scale down the CPU, and forces down the CPU speed.
> 
> Hm. This one got me thinking; I think I just found it. I forgot that I
> installed cpufreqd a while ago, which seems to be the culprit: when I
> start cpufreqd, the system falls back to 1925Mhz. I've removed it now,
> however, as killing cpufreqd does not release the maximum speed back to
> it's normal value of 2.2Ghz, I'm not entirely sure; so...
> 
> > Could you send me the output of /proc/acpi/processor/*/* and
> > /proc/acpi/thermal_zone/*/* when you notice this situation again?
> 
> ... I've still put those at
> http://users.pandora.be/wouter.verhelst/proc-acpi.tgz

Hm, can't recognize anything strange there. When the situation appears
again, what does /sys/devices/sytem/cpu/cpu0/cpufreq/scaling_max_freq say?

> > BTW, could it be that this is still the 2.6.0 module, instead of the
> > 2.6.1 variant? 2.6.1 should report more "clean" values in
> > /sys/devices/system/cpu/cpufreq/... it should report 2200000 instead
> > of 2193205.
> 
> Yes, the first mail I sent (with the command line stuff in /sys) was
> before I installed 2.6.1, still running 2.6.0 at that point. Sorry for
> the confusion :-)

No problem :-)

	Dominik


[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

On Sun, Jan 11, 2004 at 06:35:47PM +0100, Dominik Brodowski wrote:
> On Sat, Jan 10, 2004 at 08:35:26PM +0100, Wouter Verhelst wrote:
> > That happens both when it's in P0 or P1.
> 
> So ACPI works to set the processor in P0 and P1?

Yes.

> If so, you really really should use the ACPI cpufreq driver instead of
> the p4-clockmod driver. The former will offer you true "SpeedStep"
> support, the latter only some clock modulation which won't save you
> much energy as ACPI supports the C2 idle state anyway. 

Ah, so what p4-clockmod does is not SpeedStep?

Hm. That sucks.

> > Not that it matters -- p4-clockmod seems to do it right :)
> 
> It's not as noisy, indeed. But it's not right -- as long as acpi.ko works.

It mostly does, although I've had lockups when using the /proc interface
to switch CPU frequencies -- yes, should've reported that, but I only
now remember :)

> > > I think ACPI is messing with us. It detects a high temperature
> > > situation, tries to scale down the CPU, and forces down the CPU speed.
> > 
> > Hm. This one got me thinking; I think I just found it. I forgot that I
> > installed cpufreqd a while ago, which seems to be the culprit: when I
> > start cpufreqd, the system falls back to 1925Mhz. I've removed it now,
> > however, as killing cpufreqd does not release the maximum speed back to
> > it's normal value of 2.2Ghz, I'm not entirely sure; so...
> > 
> > > Could you send me the output of /proc/acpi/processor/*/* and
> > > /proc/acpi/thermal_zone/*/* when you notice this situation again?
> > 
> > ... I've still put those at
> > http://users.pandora.be/wouter.verhelst/proc-acpi.tgz
> 
> Hm, can't recognize anything strange there. When the situation appears
> again, what does /sys/devices/sytem/cpu/cpu0/cpufreq/scaling_max_freq say?

That does not change.

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Dominik Brodowski]

[-- Attachment #1.1: Type: text/plain, Size: 1837 bytes --]

On Sun, Jan 11, 2004 at 08:56:19PM +0100, Wouter Verhelst wrote:
> On Sun, Jan 11, 2004 at 06:35:47PM +0100, Dominik Brodowski wrote:
> > On Sat, Jan 10, 2004 at 08:35:26PM +0100, Wouter Verhelst wrote:
> > > That happens both when it's in P0 or P1.
> > 
> > So ACPI works to set the processor in P0 and P1?
> 
> Yes.

Excellent.
> 
> > If so, you really really should use the ACPI cpufreq driver instead of
> > the p4-clockmod driver. The former will offer you true "SpeedStep"
> > support, the latter only some clock modulation which won't save you
> > much energy as ACPI supports the C2 idle state anyway. 
> 
> Ah, so what p4-clockmod does is not SpeedStep?
> 
> Hm. That sucks.

No. p4-clockmod stops the CPU for certain short periods of time, it
"modulates" the CPU frequency. 

------------------> time  axis

x x x x x x x x x   normal clock ticks
x x x x - - - - x   50 % frequency modulation [ACPI T-States or P4-clockmod]
x  x  x  x  x  x    frequency scaling     [ACPI P-States]

Only if the frequency is scaled, the voltage can be scaled also. So only
that offers the best reduction of energy usage. For details, check the
cpufreq archives for a lengthy thread with many calculations...

> > > Not that it matters -- p4-clockmod seems to do it right :)
> > 
> > It's not as noisy, indeed. But it's not right -- as long as acpi.ko works.
> 
> It mostly does, although I've had lockups when using the /proc interface
> to switch CPU frequencies -- yes, should've reported that, but I only
> now remember :)

The /proc/ interface is deprecated, and the acpi module is full of errors
[I'm working on it, though...]. You should use the /sys-fs interface with
the acpi.ko module as well.

> That does not change.

Very strange... I'm out of ideas for the moment being.

	Dominik

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Wouter Verhelst]

On Sun, Jan 11, 2004 at 09:06:18PM +0100, Dominik Brodowski wrote:
> On Sun, Jan 11, 2004 at 08:56:19PM +0100, Wouter Verhelst wrote:
> > It mostly does, although I've had lockups when using the /proc interface
> > to switch CPU frequencies -- yes, should've reported that, but I only
> > now remember :)
> 
> The /proc/ interface is deprecated,

Yes, I know. In fact, it being deprecated is the reason I started
playing with p4-clockmod in the first place :)

> and the acpi module is full of errors [I'm working on it, though...].
> You should use the /sys-fs interface with the acpi.ko module as well.
> 
> > That does not change.
> 
> Very strange... I'm out of ideas for the moment being.

I'm more and more convinced it's got something to do with cpufreqd. I
shut it down yesterday for about an hour, and nothing happened. Then I
restarted it, and -surprise- less than a minute later, the CPU was
throttled below the maximum speed.

-- 
Wouter Verhelst
Debian GNU/Linux -- http://www.debian.org
Nederlandstalige Linux-documentatie -- http://nl.linux.org
"Stop breathing down my neck." "My breathing is merely a simulation."
"So is my neck, stop it anyway!"
  -- Voyager's EMH versus the Prometheus' EMH, stardate 51462.

Re: p4-clockmod doesn't seem to allow the highest possible speed

[Ducrot Bruno]

On Sun, Jan 11, 2004 at 08:56:19PM +0100, Wouter Verhelst wrote:
> On Sun, Jan 11, 2004 at 06:35:47PM +0100, Dominik Brodowski wrote:
> > On Sat, Jan 10, 2004 at 08:35:26PM +0100, Wouter Verhelst wrote:
> > > That happens both when it's in P0 or P1.
> > 
> > So ACPI works to set the processor in P0 and P1?
> 
> Yes.
> 
> > If so, you really really should use the ACPI cpufreq driver instead of
> > the p4-clockmod driver. The former will offer you true "SpeedStep"
> > support, the latter only some clock modulation which won't save you
> > much energy as ACPI supports the C2 idle state anyway. 
> 
> Ah, so what p4-clockmod does is not SpeedStep?
> 
> Hm. That sucks.

No.  p4-clockmod is not SpeedStep.  Normal.  As Dominik say, p4-clockmod
is *not* intended for power comsuption issues.

There is normally two reasons for slowing a CPU:

1- change voltage of processor (SpeedStep, Powernow, etc.)
   The usage for is to get more battery life, or to get less power comsuption
   for the system.

2- throttling the CPU via clock modulation (P4 only), or by generating
   a signal which toggle the STP# pin of a CPU.  The usage for it is
   to cool the CPU instead of turning on/off the CPU fan.  It is also
   used if the fan is not sufficiant to cool enough the CPU.

1 and 2 do not have the same goal.  The CPUFreq project currently only
handle case 1- (via policy, governor, etc.).  But I'm thinking about
designing case 2- for the cpufreq core (there is already something for,
but it is not exactly what I would expect), and that should rely on
lm-sensors or/and ACPI thermal.

Cheers,

-- 
Ducrot Bruno

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.