Getting AMD64 powernow up and working under Gentoo

[danyer]

CnQ lowers or rise the frequency of the processor, also the voltage.
fancontrol, part of lm_sensors, adjust the speed of the fan, based on the motherboard sensors

when your computer is idle, is no point for CPU to run full speed. so it will lower its frequency. Now, the temperature will go down (motherboard sensor, cpu sensor, or other sensors). This is the Cool part of CnQ.
The quiet part is usually implemented in BIOS (set the Fan speed to Smart, in my case). But I don't like the way BIOS manages the speed of the fan, I use fancontrol. This script, running all the time and checking the temperature of a sensor you specify, at the interval you specify, will detect the drop in temperature and will decrease the speed of the fan, thus producing less noise. This is the Quiet part, implemented in software (as I said, you can let BIOS do it if you like).

Let's take it step by step:
1. First you have to enable power management (see this thread), and Cool functionality in the kernel, using the right governor.
2. Then you have to install and configure lm_sensors, to be able to monitor the board (temperature, fan speed)
3. The last step, configure fancontrol.

Which step are you at? 1 and 2 are documented everywhere. for 3, search Gentoo forum for fancontrol.

Tell me if you need further help.

Best,
Dan.

[SAngeli]

Hi danyer,

thank you for your explanations.

So far, I have lm_sensors installed and cofugured. I need clarifications on:
1) CPU scripts (which governor to use) and how to run the scripts
2) CPU fan scripts, how to write them and implement them
3) how to monitor and verify both CPU and CPU fan work together.

So far I solved my issue by manually setting the CPU fan speed based on what lm_sensors reports for CPU heat, and I always live it running at more and less 5V. It would be nice to do this through scripts.
Do you know how much I can lower in rpm the CPU fan? I ask this as currently I can manually set it from 0 to 12V

Thank you,
Spiro

[stdPikachu]

1) CPU scripts (which governor to use) and how to run the scripts

IMHO, the best CPU governor is the "ondemand" governor that you can build into the kernel. You can then activate it at boot by adding this to /etc/conf.d/local.start:

Code: Select all

echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

If you want the ondemand governor to ramp up the clock speed with niced processes (instead of running them at reduced speed) add this line just below it:

Code: Select all

echo 1 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/ignore_nice

If you have an SMP config, you just need to put multiple entries (one for each CPU), replacing the cpu0 bit with cpu1, cpu2, cpu3, etc etc

2) CPU fan scripts, how to write them and implement them
3) how to monitor and verify both CPU and CPU fan work together.

Not sure about this meself, since the BIOS in both of my AMD64 boards does it for me. But as I understand it, you first need to set up lm_sensors properly, then you can run pwmconfig to set up a config file for the fan control utility in /etc/fancontrol. After that's done, just add something like "/usr/sbin/fancontrol &> /var/log/fancontrol.log &" to /etc/conf.d/local.start.

[Defiance]

I can't seem to get the ondemand governor to work. When I echo ondemand into the scaling_governor file, my CPU stays at the same speed as indicated by /proc/cpuinfo. CNQ is enabled in my BIOS, my dmesg shows that powernow is activated without any errors. The governor just doesn't work. I had a similar problem on my P4 laptop that I never resolved. Any ideas?

[jcmorris]

Hi everyone! I thought this little bit might be helpful. I noticed lots of people are using "ondemand", but, here is what the kernel help for "conservative" says:

If you have a desktop machine then you should really be considering the 'ondemand' governor instead, however if you are using a laptop, PDA or even an AMD64 based computer (due to the unacceptable step-by-step latency issues between the minimum and maximum frequency transitions in the CPU) you will probably want to use this governor

The way I interpret this, there will be a high latency when switching from the min to the max frequency as the ondemand governor does, but the Athlon 64 has lower latency when gradually stepping up frequency, as the conservative governor does. Any thoughts?

jcm

[ankhcraft]

I can't seem to get the ondemand governor to work. When I echo ondemand into the scaling_governor file, my CPU stays at the same speed as indicated by /proc/cpuinfo. CNQ is enabled in my BIOS, my dmesg shows that powernow is activated without any errors. The governor just doesn't work. I had a similar problem on my P4 laptop that I never resolved. Any ideas?

Same issue here, but using the "conservative" governor, as instructed by the kernel config menus (since this a laptop ... and AMD64).

EDIT: let me also add that I *do* have cpufreq statistics support built-in to the kernel and I can see clearly that 0 state transitions have taken place.

[ankhcraft]

I can't seem to get the ondemand governor to work. When I echo ondemand into the scaling_governor file, my CPU stays at the same speed as indicated by /proc/cpuinfo. CNQ is enabled in my BIOS, my dmesg shows that powernow is activated without any errors. The governor just doesn't work. I had a similar problem on my P4 laptop that I never resolved. Any ideas?

Same issue here, but using the "conservative" governor, as instructed by the kernel config menus (since this a laptop ... and AMD64).

EDIT: let me also add that I *do* have cpufreq statistics support built-in to the kernel and I can see clearly that 0 state transitions have taken place.

Okay, I've had to resort to sys-power/powernowd.

But now, finally I have a useful scaling of CPU frequencies, albeit by a userspace program.

[trueshanti]

Hi there ..

after following this Thread and also using all kernel-settings mentioned in http://gentoo-wiki.com/HOWTO_PowerNow! i run the powernowd for debug ( with -dv options [of course as root] )

Code: Select all

powernowd -dv 
PowerNow Daemon v0.90hun6, (c) 2003-2004 John Clemens
Settings:
  verbosity:        1
  mode:             1     (AGGRESSIVE)
  step:           100 MHz (100000 kHz)
  lowwater:        20 %
  highwater:       80 %
  poll interval: 1000 ms
Found 1 physical cpu and 2 virtual cpus:
cpu0: 1000000 kHz - 2000000 kHz

looks good so far

after starting some glxgears to produce some load the Frequenzy doesnot change

on very verbose debug i get this

Code: Select all

Before decide_speed
decide_speed()
decide_speed(cpu = 0x505880)
decide_speed(phys_cpuid = 0)
get_stat(cpu = 0x505880)
get_stat(phys_cpuid = 0)
before loop
Reading stats for physical cpu0)
Reading stats for virtual cpu0 (physical cpu0)
before strtol
Reading stats for virtual cpu1 (physical cpu0)
before strtol
after loop
phys cpu0 usage 86, total 99
phys cpu0 percentage = 0.868687

once , while playing around with the values I got this:

Code: Select all

Can't open scaling_setspeed: No such file or directory

the file is indeed nowhere .. but the error never occoured again

powernowd doesnt crash , nor does it change some frequenzy .. i trieded with all available governors

Code: Select all

conservative ondemand powersave userspace performance

nothing changed

i had to set my /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq value to "2000000" manually ( why that ?) and it boosted away .. but after stopping all glxgears .. the current scaling stay o 2GHz


it should go down to 1000000 again

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2000000 1800000 1000000

any idea how i can get this up and running ? ..

tnx4reply

-c-

my default /proc/cpuinfo is

processor : 0
vendor_id : AuthenticAMD
cpu family : 15
model : 35
model name : AMD Athlon(tm) 64 X2 Dual Core Processor 3800+
stepping : 2
cpu MHz : 1000.000
cache size : 512 KB
physical id : 0
siblings : 2
core id : 0
cpu cores : 2
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse
36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow pni lahf_l
m cmp_legacy
bogomips : 2012.83
TLB size : 1024 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp

processor : 1
vendor_id : AuthenticAMD
cpu family : 15
model : 35
model name : AMD Athlon(tm) 64 X2 Dual Core Processor 3800+
stepping : 2
cpu MHz : 1000.000
cache size : 512 KB
physical id : 0
siblings : 2
core id : 1
cpu cores : 2
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse
36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow pni lahf_l
m cmp_legacy
bogomips : 2012.83
TLB size : 1024 4K pages
clflush size : 64
cache_alignment : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp

[trueshanti]

ok . forget it .. anyway it is working suddenly ( after putting the optins -m0 -u75 into /etc/conf.d/powernow )

funnyO it didnt work in foreground mode

so be happy

[palu15]

WHO HAD THE SCALING PROBLEM PLEASE READ THIS!!!

please go to http://bugzilla.kernel.org/show_bug.cgi?id=4392#c1 and download the patch

follow the patch instruction ftom http://www.linuxhq.com/patch-howto.html and try!

(after applying the patch you need to do "make" and copy the bzImage to /boot , change the bootloader accordingly and reboot the new kernel)


This solved the problem for me

My system is a Sempron 2800 (1.6Ghz 256K cache) on a AsRock K8NF6G-VSTA

I applied the patch to powernow-k8.c and maked the kernel (I'm sure powernow-k8.o was recompiled) and updated the bzImage for grub, but for me the problem is the same , frequency scaling was not detected!!

[DrDoverylittle]

I have an AMD Sempron 2800+ and ASUS K8V-VM mobo and having trouble with powernow-k8 too.
Both cpu and mobo are meant to suport PowerNow! and CnQ, but powernow-k8 fails to load.
BIOS is latest version, kernel version 2.6.17.
Is this a kernel bug ? Do you think it should work given the processor info below ?

Code: Select all

$ dmesg | grep powernow
powernow-k8: Power state transitions not supported

Code: Select all

$ cat /proc/cpuinfo 
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 44
model name      : AMD Sempron(tm) Processor 2800+
stepping        : 2
cpu MHz         : 1596.367
cache size      : 256 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt lm 3dnowext 3dnow up pni lahf_lm ts ttp tm stc
bogomips        : 3195.24

Code: Select all

$ cat /proc/acpi/processor/CPU1/info 
processor id:            0
acpi id:                 1
bus mastering control:   no
power management:        no
throttling control:      yes
limit interface:         yes
$ cat /proc/acpi/processor/CPU1/limit 
active limit:            P0:T0
user limit:              P0:T0
thermal limit:           P0:T0
$ cat /proc/acpi/processor/CPU1/power 
active state:            C1
max_cstate:              C8
bus master activity:     00000000
states:
   *C1:                  type[C1] promotion[--] demotion[--] latency[000] usage[00000000] duration[00000000000000000000]
$ cat /proc/acpi/processor/CPU1/throttling 
state count:             16
active state:            T1
states:
    T0:                  00%
   *T1:                  06%
    T2:                  12%
    T3:                  18%
    T4:                  24%
    T5:                  31%
    T6:                  37%
    T7:                  43%
    T8:                  49%
    T9:                  55%
    T10:                  62%
    T11:                  68%
    T12:                  74%
    T13:                  80%
    T14:                  86%
    T15:                  93%

The relavent directories don't seem to be made either

Code: Select all

$ ls /sys/devices/system/cpu/cpu0/
crash_notes  topology

[giantrabbit]

I have an AMD Sempron 2800+ and ASUS K8V-VM mobo ....

The CnQ thingy is disabled by default on the ASUS mobo, turn it on in ur BIOS.

[DrDoverylittle]

I have an AMD Sempron 2800+ and ASUS K8V-VM mobo ....

The CnQ thingy is disabled by default on the ASUS mobo, turn it on in ur BIOS.

This option is missing on the BIOS menu, but thanks for stating the obvious.

Actually according to this thread
http://forums.gentoo.org/viewtopic-t-34 ... ernow.html
only the 3000+ upwards supports CnQ on the Sempron series.
Actually I'm really pissed about this as nothing is said about this until you dig deep into the spec docs.
On the main Sempron page at http://www.amd.com/gb-uk/Processors/Pro ... 99,00.html
is gives the impression that all Sempron have this.
I should send it back to the seller as i have been duped but i can't afford the down time.

[hvengel]

I think that this is a bit of an over reaction. The semperon 2800 is only a 1.6Ghz CPU and with its small cash and simplified single channel memory controller at that speed already has a fairly low power consumption. It appears that AMD engineers decided that there is are CPU speeds slow enough that the added over head of speed stepping was not worth it. It appears that this is at the 1.8Gz level. I have an AMD64 3000 and it's ACPI speed states are 1Ghz and 1.8Ghz. And an amd64 3500 which has speed states of 1.0Ghz, 1.8Ghz. 2.0Ghz and 2.2Ghz. I also have a 4800+ and its speed states are 1Ghz, 1.8Ghz, 2.0Ghz, 2.2Ghz and 2.4Ghz. I also suspect that by disabling speed stepping on these slower CPUs that power consumption is reduced a little since it should be possible to disable some additional circuits on the chip.

[DrDoverylittle]

I think that this is a bit of an over reaction. The semperon 2800 is only a 1.6Ghz CPU and with its small cash and simplified single channel memory controller at that speed already has a fairly low power consumption. It appears that AMD engineers decided that there is are CPU speeds slow enough that the added over head of speed stepping was not worth it. It appears that this is at the 1.8Gz level. I have an AMD64 3000 and it's ACPI speed states are 1Ghz and 1.8Ghz. And an amd64 3500 which has speed states of 1.0Ghz, 1.8Ghz. 2.0Ghz and 2.2Ghz. I also have a 4800+ and its speed states are 1Ghz, 1.8Ghz, 2.0Ghz, 2.2Ghz and 2.4Ghz. I also suspect that by disabling speed stepping on these slower CPUs that power consumption is reduced a little since it should be possible to disable some additional circuits on the chip.

Its not really the power consumption that i am bothered about, but it would be nice to have one step lower at around 800 MHz, as appears in some of the links from the page i showed, as i could slow down the cpu fan rate when not in use and reduce noise. Its not really an over reaction, if they don't think it is needed for certain processors then they should say that. All the numbers you quote have one step which is significantly smaller than all the rest, which makes a nice quite 'sleep' mode. I will just have to get used to it like sleeping with a snoring partner.

[dnadesign]

Just wanted to say, that this howto works nicely for Gentoo AMD64 with kernel 2.6.19 on AMD Turion64 X2 processor. Although these processors have only two frequency peaks (800 and 1600 MHz per core in my case), it works fine. When idle it's 800 MHz and when it feels the load it goes 1,6 GHz (with an additional fan turn on).

PS. I still need to test this, when I power up the notebook with no external power source. The (processor probably) issue is that: when I power it up without any external power source it goes down to 800 MHz and stays there until the notebook is restarted with an AC adapter pluged in (if anyone knows a solution to this, please let me know ).

Best regards,
DNA DesigN

[drescherjm]

Hi I am trying to archive what a few people wanted to avoid on page 3 of this thread...

Since I am running this on a desktop system (dual processor opteron at that), I want to minimize the performance loss of power management but still keep the processors cool. Basically I want the cpus to go up fast but down very slow. It frustrates me every time I see my gdesklet say I am running at 1GHz while I am transcoding a 2 GB mpeg.

On the third page there is a talk about /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_down_factor which I do not have on my 2.6.19 vserver sources kernel.

Code: Select all

ls /sys/devices/system/cpu/cpu0/cpufreq/ondemand/ -al
total 0
drwxr-xr-x 2 root root    0 Feb 23 10:00 .
drwxr-xr-x 4 root root    0 Feb 23 10:00 ..
-rw-r--r-- 1 root root 4096 Feb 23 10:00 ignore_nice_load
-rw-r--r-- 1 root root 4096 Feb 23 10:00 powersave_bias
-rw-r--r-- 1 root root 4096 Feb 23 10:00 sampling_rate
-r--r--r-- 1 root root 4096 Feb 23 10:00 sampling_rate_max
-r--r--r-- 1 root root 4096 Feb 23 10:00 sampling_rate_min
-rw-r--r-- 1 root root 4096 Feb 23 10:00 up_threshold

I have tried and have been running a few of the dozens of lightly documented userspace utilities but none seem to give me what I am looking for.

[Simius]

All amd64 other than the Opterons have speed scaling as a standard feature. Many desktop users of the amd64s setup their machines to speed scale because it reduces power consumption, heat and (if the fan is speed controlled) noise.

I take it this means Opterons don't have speed scaling.
This certainly explains this:

Code: Select all

 # modprobe powernow-k8
FATAL: Error inserting powernow_k8 (/lib/modules/2.6.18-gentoo-r6.Daniel.Jozsef.20070112.02/kernel/arch/x86_64/kernel/cpufreq/powernow-k8.ko): No such device

(I have an Opteron.)

[drescherjm]

I take it this means Opterons don't have speed scaling.

Some did not. I know mine does as I have been using this for over a year...

[Cyker]

All recent Opterons support CnQ.

According to the docs, you need to have the ACPI p-states driver compiled as well as the powernow-k8 driver for it to work 'tho.

[Beekster]

Hi I am trying to archive what a few people wanted to avoid on page 3 of this thread...

Since I am running this on a desktop system (dual processor opteron at that), I want to minimize the performance loss of power management but still keep the processors cool. Basically I want the cpus to go up fast but down very slow. It frustrates me every time I see my gdesklet say I am running at 1GHz while I am transcoding a 2 GB mpeg.
<snip>

Had similar thoughts with my X2 4200+. What works for me:

.config

Code: Select all

#
# CPU Frequency scaling
#
CONFIG_CPU_FREQ=y
CONFIG_CPU_FREQ_TABLE=y
# CONFIG_CPU_FREQ_DEBUG is not set
CONFIG_CPU_FREQ_STAT=y
# 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=y
CONFIG_CPU_FREQ_GOV_USERSPACE=y
CONFIG_CPU_FREQ_GOV_ONDEMAND=y
CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y

They're all compiled in, system starts booting with "Performance" the default.

In /etc/conf.d/local.start, include:

Code: Select all

echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
echo 25 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/up_threshold

This sets the governor back to "On Demand", but more importantly tells it to ramp up when CPU use (=core use average) goes over 25%. This value defaults to a much higher value (~75% ?) that a single thread can never get the cpu to hit! Very silly default value for a multi core system...

[drescherjm]

Beekster

Thanks. That sounds logical. When I get some time this weekend I will try that. I have since reverted back to using powernowd which works but not the way I want it...

[Nexces]

I've got a bit different problem.
My cpu is Athlon64 3500+ which normaly is running @ 2200
I've got 2.6.19 kernel
after aplying CnQ my spu is working @ 1800 and it is not changing it.

Code: Select all

$ dmesg | grep k8
powernow-k8: Found 1 AMD Athlon(tm) 64 Processor 3500+ processors (version 2.00.00)
powernow-k8: invalid freq entries 3900000 kHz vs. 65535000 kHz
powernow-k8: invalid freq entries 3900000 kHz vs. 65535000 kHz
powernow-k8:    0 : fid 0xe (2200 MHz), vid 0x2
powernow-k8:    1 : fid 0xc (2000 MHz), vid 0x6
powernow-k8:    2 : fid 0xa (1800 MHz), vid 0xa
powernow-k8:    3 : fid 0x2 (1000 MHz), vid 0x12

no 'cpu_init' or anything like that in dmesg which was present in other posts

Code: Select all

$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors
conservative ondemand powersave userspace performance
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2200000 2000000 1800000 1000000
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
ondemand

it doesn't matter if i'm using any other governor... speed remains the same

Code: Select all

$ cpufreq-info
cpufrequtils 002: cpufreq-info (C) Dominik Brodowski 2004-2006
Report errors and bugs to linux@brodo.de, please.
analyzing CPU 0:
  driver: powernow-k8
  CPUs which need to switch frequency at the same time: 0
  hardware limits: 1000 MHz - 2.20 GHz
  available frequency steps: 2.20 GHz, 2.00 GHz, 1.80 GHz, 1000 MHz
  available cpufreq governors: conservative, ondemand, powersave, userspace, performance
  current policy: frequency should be within 1.80 GHz and 1.80 GHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 1.80 GHz (asserted by call to hardware).

and this is the most puzzling thing to me... a "huge" range 1.80 - 1.80...
any ideas??

[gaga]

hi

I have a problem with C&Q on a A64 3000+ s754:

powernow-k8: Found 1 AMD Athlon(tm) 64 Processor 3000+ processors (version 2.00.00)
powernow-k8: invalid freq entries 3900000 kHz vs. 65535000 kHz
powernow-k8: invalid freq entries 3900000 kHz vs. 65535000 kHz
powernow-k8: invalid freq entries 3900000 kHz vs. 65535000 kHz
powernow-k8: 0 : fid 0xc (2000 MHz), vid 0x2
powernow-k8: 1 : fid 0xa (1800 MHz), vid 0x6
powernow-k8: 2 : fid 0x2 (1000 MHz), vid 0x12

I can change de frequency, but the Vcore is the same for each freqency : 1.5 V

I have all "auto" or "default" in my BIOS and I use powernowd ( I have already tried with a echo ondemand > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor and it's the same )

any idea ?
thks

EDIT : I have tried with the last vanilla kernel and now it's :

# dmesg |grep power
powernow-k8: Found 1 AMD Athlon(tm) 64 Processor 3000+ processors (version 2.00.00)
powernow-k8: 0 : fid 0xc (2000 MHz), vid 0x2
powernow-k8: 1 : fid 0xa (1800 MHz), vid 0x6
powernow-k8: 2 : fid 0x2 (1000 MHz), vid 0x12

but there aren't any voltage.....

[drescherjm]

Looks the same to me and the voltage changes. I believe the vid is Voltage ID.

Code: Select all

# dmesg |grep power
powernow-k8: Found 2 AMD Opteron(tm) Processor 850 processors (version 2.00.00)
powernow-k8:    0 : fid 0x10 (2400 MHz), vid 0x8
powernow-k8:    1 : fid 0xe (2200 MHz), vid 0x8
powernow-k8:    2 : fid 0xc (2000 MHz), vid 0xa
powernow-k8:    3 : fid 0xa (1800 MHz), vid 0xc
powernow-k8:    4 : fid 0x2 (1000 MHz), vid 0x12
powernow-k8:    0 : fid 0x10 (2400 MHz), vid 0x8
powernow-k8:    1 : fid 0xe (2200 MHz), vid 0x8
powernow-k8:    2 : fid 0xc (2000 MHz), vid 0xa
powernow-k8:    3 : fid 0xa (1800 MHz), vid 0xc
powernow-k8:    4 : fid 0x2 (1000 MHz), vid 0x12