Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Robert Hancock]

David Rees wrote:
> On 7/30/06, bert hubert <bert.hubert@netherlabs.nl> wrote:
>> On Sun, Jul 30, 2006 at 09:46:51PM +0200, Arjan van de Ven wrote:
>>
>> > as a side note ... you realize that clockmod doesn't actually save you
>> > any power right? ;)
>>
>> Indeed, and I've measured that too. But it saves an awful amount of 
>> noise!
> 
> If it doesn't save you power, how does it reduce noise? I guess it
> keeps you from overheating your processor which causes the fan to spin
> up?

The aim of power-saving techniques is to allow the same amount of 
computational work to be done while consuming less power. For example, 
if a task only uses 20% of the CPU at its maximum speed, the clock speed 
can be dropped down to half of the maximum. The main advantage of this 
is that it allows reducing the core voltage which I think accounts for 
most of the power savings. Essentially the lower performance settings 
have higher performance per watt than the higher settings.

Clock modulation doesn't reduce the CPU core voltage, nor does it reduce 
the rate at which power is consumed when the CPU is in the active state. 
It just causes the CPU to periodically stop its clock for a while, 
during which no work is done. This means that the power and heat 
produced is reduced but the work that can be done is also reduced by an 
proportional amount, so there is little or no improvement in performance 
per watt. As well, when the CPU has nothing to do it will be halted 
anyway which does pretty much the same as what clockmod is doing.

Essentially clockmod is there as a way to limit the thermal output of 
the CPU in thermal emergencies, it's not really very good as a 
power-saving feature.

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[bert hubert]

On Mon, Jul 31, 2006 at 01:14:05PM -0600, Robert Hancock wrote:

> per watt. As well, when the CPU has nothing to do it will be halted 
> anyway which does pretty much the same as what clockmod is doing.

You are empirically wrong in my case. I can't make you hear the difference
in noise, but perhaps this will convince you:

$ sensors | grep CPU
CPU_Fan:   2359 RPM  (min = 4000 RPM)                     
CPU:         +61 C  (low  =   +10 C, high =   +50 C)     

$ sudo /etc/init.d/powernowd start
 * Starting powernowd...  			[ ok ]

$ sensors | grep CPU
CPU_Fan:   1411 RPM  (min = 4000 RPM)                     
CPU:         +57 C  (low  =   +10 C, high =   +50 C)     

$ sudo /etc/init.d/powernowd stop
 * Stopping powernowd:				[ ok ]

$ sensors | grep CPU
CPU_Fan:   2452 RPM  (min = 4000 RPM)                     
CPU:         +61 C  (low  =   +10 C, high =   +50 C)     

And this with only seconds in between starting and stopping, on an idle
system.

> Essentially clockmod is there as a way to limit the thermal output of 
> the CPU in thermal emergencies, it's not really very good as a 
> power-saving feature.

That I've measured, and indeed, the same amount of power is drawn from the
mains. But the reduction in fan noise is worth it.

On this CPU, HLT may simply not be doing as good a job. Who knows.

processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 4
model name      : Intel(R) Pentium(R) 4 CPU 3.00GHz
stepping        : 1
cpu MHz         : 3000.000
cache size      : 1024 KB

It does reduce effective CPU to 300MHz though, which is a lot.

Kind regards,

bert

-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services

2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[bert hubert]

Hi everybody,

Since 2.6.18-rc1, up to and including -rc3, cpufreq has died on me. It
worked fine in 2.6.16.9.

The whole cpufreq hierarchy in /sys/ is missing, nothing is reported during
bootup mentioning 'cpufreq', and when I build with all the cpufreq stuff
configured as modules, I see this:

$ lsmod
Module                  Size  Used by
capability              3464  0
commoncap               5120  1 capability
acpi_cpufreq            4748  0
speedstep_lib           3716  0
cpufreq_userspace       3092  0
cpufreq_stats           3588  0
freq_table              3588  2 acpi_cpufreq,cpufreq_stats
cpufreq_powersave       1664  0
cpufreq_ondemand        4716  0
cpufreq_conservative     5280  0
b2c2_flexcop_pci        6936  0
b2c2_flexcop           22412  1 b2c2_flexcop_pci
mt352                   5636  1 b2c2_flexcop
mt312                   6660  1 b2c2_flexcop
bcm3510                 8196  1 b2c2_flexcop
stv0299                 8840  1 b2c2_flexcop
stv0297                 6272  1 b2c2_flexcop
nxt200x                11396  1 b2c2_flexcop
lgdt330x                7196  1 b2c2_flexcop
psmouse                31624  0

# modprobe p4_clockmod
FATAL: Error inserting p4_clockmod
(/lib/modules/2.6.18-rc3/kernel/arch/i386/kernel/cpu/cpufreq/p4-clockmod.ko):
Device or resource busy

dmesg is silent on why this is. I've grepped the files in the cpufreq
directory, but none of them contain EBUSY.

Under 2.6.16.9, this was logged at startup:
[17179582.044000] p4-clockmod: P4/Xeon(TM) CPU On-Demand Clock Modulation
available

cpuinfo:
$ cat /proc/cpuinfo
processor       : 0
vendor_id       : GenuineIntel
cpu family      : 15
model           : 4
model name      : Intel(R) Pentium(R) 4 CPU 3.00GHz
stepping        : 1
cpu MHz         : 3000.298
cache size      : 1024 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 5
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 pbe nx
constant_tsc pni monitor ds_cpl cid xtpr
bogomips        : 6004.95

>From .config:

#
# CPU Frequency scaling
#
CONFIG_CPU_FREQ=y
CONFIG_CPU_FREQ_TABLE=m
# CONFIG_CPU_FREQ_DEBUG is not set
CONFIG_CPU_FREQ_STAT=m
# CONFIG_CPU_FREQ_STAT_DETAILS is not set
CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set
CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
CONFIG_CPU_FREQ_GOV_POWERSAVE=m
CONFIG_CPU_FREQ_GOV_USERSPACE=m
CONFIG_CPU_FREQ_GOV_ONDEMAND=m
CONFIG_CPU_FREQ_GOV_CONSERVATIVE=m

#
# CPUFreq processor drivers
#
CONFIG_X86_ACPI_CPUFREQ=m
CONFIG_X86_POWERNOW_K6=m
CONFIG_X86_POWERNOW_K7=m
CONFIG_X86_POWERNOW_K7_ACPI=y
CONFIG_X86_POWERNOW_K8=m
CONFIG_X86_POWERNOW_K8_ACPI=y
CONFIG_X86_GX_SUSPMOD=m
CONFIG_X86_SPEEDSTEP_CENTRINO=m
CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI=y
CONFIG_X86_SPEEDSTEP_CENTRINO_TABLE=y
CONFIG_X86_SPEEDSTEP_ICH=m
CONFIG_X86_SPEEDSTEP_SMI=m
CONFIG_X86_P4_CLOCKMOD=m
CONFIG_X86_CPUFREQ_NFORCE2=m
CONFIG_X86_LONGRUN=m

dmesg output is on http://ds9a.nl/tmp/dmesg , full .config on
http://ds9a.nl/config

Please let me know how I can help solve this problem. Zwane, you are welcome
to ssh to my system if that helps, let me know.

Kind regards,

bert



-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Tomasz Torcz]

[-- Attachment #1: Type: text/plain, Size: 620 bytes --]

On Sun, Jul 30, 2006 at 02:08:44PM +0200, bert hubert wrote:
> Hi everybody,
> 
> Since 2.6.18-rc1, up to and including -rc3, cpufreq has died on me. It
> worked fine in 2.6.16.9.
> 
> # modprobe p4_clockmod
> FATAL: Error inserting p4_clockmod
> (/lib/modules/2.6.18-rc3/kernel/arch/i386/kernel/cpu/cpufreq/p4-clockmod.ko):
> Device or resource busy
> 

  I have similar problem with cpufreq-nforce2 -- http://lkml.org/lkml/2006/7/7/234
  I haven't do a git-bisect yet.

-- 
Tomasz Torcz                 "God, root, what's the difference?"
zdzichu@irc.-nie.spam-.pl         "God is more forgiving."


[-- Attachment #2: Type: application/pgp-signature, Size: 229 bytes --]

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Zwane Mwaikambo]

On Sun, 30 Jul 2006, Tomasz Torcz wrote:

> On Sun, Jul 30, 2006 at 02:08:44PM +0200, bert hubert wrote:
> > Hi everybody,
> > 
> > Since 2.6.18-rc1, up to and including -rc3, cpufreq has died on me. It
> > worked fine in 2.6.16.9.
> > 
> > # modprobe p4_clockmod
> > FATAL: Error inserting p4_clockmod
> > (/lib/modules/2.6.18-rc3/kernel/arch/i386/kernel/cpu/cpufreq/p4-clockmod.ko):
> > Device or resource busy
> > 
> 
>   I have similar problem with cpufreq-nforce2 -- http://lkml.org/lkml/2006/7/7/234
>   I haven't do a git-bisect yet.

Could you fellows try it without;

CONFIG_X86_SPEEDSTEP_CENTRINO
CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
CONFIG_X86_SPEEDSTEP_CENTRINO_TABLE
CONFIG_X86_SPEEDSTEP_ICH
CONFIG_X86_SPEEDSTEP_SMI
CONFIG_X86_ACPI_CPUFREQ

It may likely be a driver registration thing.

Thanks,
	Zwane

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Tomasz Torcz]

[-- Attachment #1: Type: text/plain, Size: 1424 bytes --]

On Sun, Jul 30, 2006 at 10:45:13AM -0700, Zwane Mwaikambo wrote:
> On Sun, 30 Jul 2006, Tomasz Torcz wrote:
> 
> > On Sun, Jul 30, 2006 at 02:08:44PM +0200, bert hubert wrote:
> > > Hi everybody,
> > > 
> > > Since 2.6.18-rc1, up to and including -rc3, cpufreq has died on me. It
> > > worked fine in 2.6.16.9.
> > > 
> > > # modprobe p4_clockmod
> > > FATAL: Error inserting p4_clockmod
> > > (/lib/modules/2.6.18-rc3/kernel/arch/i386/kernel/cpu/cpufreq/p4-clockmod.ko):
> > > Device or resource busy
> > > 
> > 
> >   I have similar problem with cpufreq-nforce2 -- http://lkml.org/lkml/2006/7/7/234
> >   I haven't do a git-bisect yet.
> 
> Could you fellows try it without;
> 
> CONFIG_X86_SPEEDSTEP_CENTRINO
> CONFIG_X86_SPEEDSTEP_CENTRINO_ACPI
> CONFIG_X86_SPEEDSTEP_CENTRINO_TABLE
> CONFIG_X86_SPEEDSTEP_ICH
> CONFIG_X86_SPEEDSTEP_SMI

  I had those =n

> CONFIG_X86_ACPI_CPUFREQ

  I had this one =y. After setting =n, cpufreq-nforce2 (=m) works again.

powernowd: PowerNow Daemon v0.96, (c) 2003-2005 John Clemens
powernowd: Found 1 cpu:  -- 1 thread (or core) per physical cpu
/sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies: No
such file or directory
powernowd:   cpu0: 1228Mhz - 1753Mhz (7 steps)

-- 
Tomasz Torcz               "Never underestimate the bandwidth of a station
zdzichu@irc.-nie.spam-.pl    wagon filled with backup tapes." -- Jim Gray


[-- Attachment #2: Type: application/pgp-signature, Size: 229 bytes --]

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Zwane Mwaikambo]

On Mon, 31 Jul 2006, Tomasz Torcz wrote:

> On Sun, Jul 30, 2006 at 10:45:13AM -0700, Zwane Mwaikambo wrote:
> 
> > CONFIG_X86_ACPI_CPUFREQ
> 
>   I had this one =y. After setting =n, cpufreq-nforce2 (=m) works again.
> 
> powernowd: PowerNow Daemon v0.96, (c) 2003-2005 John Clemens
> powernowd: Found 1 cpu:  -- 1 thread (or core) per physical cpu
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies: No
> such file or directory
> powernowd:   cpu0: 1228Mhz - 1753Mhz (7 steps)

Hi Tomasz,

	Could you also please test Bert's patch which propogates error 
return values with your previous configuration?

Thanks,
	Zwane

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Tomasz Torcz]

[-- Attachment #1: Type: text/plain, Size: 1033 bytes --]

On Mon, Jul 31, 2006 at 07:04:07AM -0700, Zwane Mwaikambo wrote:
> On Mon, 31 Jul 2006, Tomasz Torcz wrote:
> 
> > On Sun, Jul 30, 2006 at 10:45:13AM -0700, Zwane Mwaikambo wrote:
> > 
> > > CONFIG_X86_ACPI_CPUFREQ
> > 
> >   I had this one =y. After setting =n, cpufreq-nforce2 (=m) works again.
> > 
> > powernowd: PowerNow Daemon v0.96, (c) 2003-2005 John Clemens
> > powernowd: Found 1 cpu:  -- 1 thread (or core) per physical cpu
> > /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies: No
> > such file or directory
> > powernowd:   cpu0: 1228Mhz - 1753Mhz (7 steps)
> 
> Hi Tomasz,
> 
> 	Could you also please test Bert's patch which propogates error 
> return values with your previous configuration?

  Bert's patch works. With CONFIG_X86_ACPI_CPUFREQ=y and
cpufre-nforce2=m I have working cpufreq (like in 2.6.17 and before).

-- 
Tomasz Torcz               "Never underestimate the bandwidth of a station
zdzichu@irc.-nie.spam-.pl    wagon filled with backup tapes." -- Jim Gray


[-- Attachment #2: Type: application/pgp-signature, Size: 229 bytes --]

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[Arjan van de Ven]

On Sun, 2006-07-30 at 18:07 +0200, Tomasz Torcz wrote:
> On Sun, Jul 30, 2006 at 02:08:44PM +0200, bert hubert wrote:
> > Hi everybody,
> > 
> > Since 2.6.18-rc1, up to and including -rc3, cpufreq has died on me. It
> > worked fine in 2.6.16.9.
> > 
> > # modprobe p4_clockmod
> > FATAL: Error inserting p4_clockmod
> > (/lib/modules/2.6.18-rc3/kernel/arch/i386/kernel/cpu/cpufreq/p4-clockmod.ko):
> > Device or resource busy
> > 

as a side note ... you realize that clockmod doesn't actually save you
any power right? ;)

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[bert hubert]

On Sun, Jul 30, 2006 at 09:46:51PM +0200, Arjan van de Ven wrote:

> as a side note ... you realize that clockmod doesn't actually save you
> any power right? ;)

Indeed, and I've measured that too. But it saves an awful amount of noise!

-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[David Rees]

On 7/30/06, bert hubert <bert.hubert@netherlabs.nl> wrote:
> On Sun, Jul 30, 2006 at 09:46:51PM +0200, Arjan van de Ven wrote:
>
> > as a side note ... you realize that clockmod doesn't actually save you
> > any power right? ;)
>
> Indeed, and I've measured that too. But it saves an awful amount of noise!

If it doesn't save you power, how does it reduce noise? I guess it
keeps you from overheating your processor which causes the fan to spin
up?

-Dave

Re: 2.6.18 regression: cpufreq broken since 2.6.18-rc1 on pentium4

[bert hubert]

On Mon, Jul 31, 2006 at 12:50:07AM -0700, David Rees wrote:
> >Indeed, and I've measured that too. But it saves an awful amount of noise!
> 
> If it doesn't save you power, how does it reduce noise? I guess it
> keeps you from overheating your processor which causes the fan to spin
> up?

I have no idea. Perhaps heat production moves somewhere else. But you are
free to come listen to the difference p4-clockmod makes for me. 

Thanks.

-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services