login
Search
Hi, I have an Intel Core 2 Quad and running kernel 2.6.25.11-97.fc9.i686. cpufreq doesn't seem to work. The cpufreq scaling monitor in Gnome says CPU Freq scaling is not supported in my CPU. The CPU can run at 2.0 and 2.5 GHz but mine always runs at 2.5 GHz in Linux. So I checked /sys and there is no cpufreq dir in /sys/... # ls /sys/devices/system/cpu/ cpu0 cpu1 cpu2 cpu3 cpuidle sched_mc_power_savings # ls /sys/devices/system/cpu/cpuidle/ current_driver current_governor_ro I have no clue what cpuidle directory is for. I added cpufreq.debug=7 in kernel boot params and saw the following in the dmesg: speedstep-smi: No supported Intel CPU detected. cpufreq-core: CPU 0: _PPC is 0 - frequency not limited cpufreq-core: CPU 3: _PPC is 0 - frequency not limited cpufreq-core: CPU 1: _PPC is 0 - frequency not limited cpufreq-core: CPU 2: _PPC is 0 - frequency not limited cpuidle: using governor ladder cpuidle: using governor menu I looked at arch/x86/kernel/cpu/cpufreq/speedstep-smi.c and it seems to detect only Pentium IIIs. Anyone know what files have the cpufreq code for Intel Core 2? Does cpufreq support Intel Core 2 Quads? Especially the Q9300? If not, anything I can do to help? Thanks, SK # cat /proc/cpuinfo processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 23 model name : Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz stepping : 7 cpu MHz : 2502.516 cache size : 3072 KB physical id : 0 siblings : 4 core id : 0 cpu cores : 4 fdiv_bug : no hlt_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 10 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx lm constant_tsc arch_perfmon pebs bts pni monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr sse4_1 lahf_lm bogomips : 5015.23 clflush size : 64 <snipped the rest> - shows 3 more similar sections. --
I have an Intel Core 2 Duo and it uses the ACPI cpufreq driver; your Quad should do the same. So this is likely an ACPI/BIOS issue. If you ask ACPI people they will ask you to post the output of acpidump. Also you should probably check if you have a BIOS option that needs to be enabled for this to work. BTW, cpuidle is something quite different, it is about how to save power when CPU is doing nothing (i.e. idle :-). Cpu frequency scaling is how to save power when CPU is working (but doesn't need to run flat out). Alan --
Hi, I'm attaching the acpidump output. Can someone (ACPI guys??) please me help figure this out? Is there any other info that's needed to debug this? I can be a tester for this and even do some development within the limits of my understanding. Thanks, SK
Will you please confirm whether the acpi_cpufreq driver is used on your machine? Please enable CONFIG_CPUFREQ_DEBUG in kernel configuration and boot the system with "cpufreq.debug=7". After the system is booted, please attach the output of dmesg. please attach the output of acpidump using the latest pmtools at http://www.lesswatts.org/patches/linux_acpi/ Of course you can open a new bug in bugzilla and attach the output of dmesg, acpidump, lspci -vxxx. http://bugzilla.kernel.org/enter_bug.cgi?product=ACPI --
Hi, I already did the cpufreq.debug thing and it was using the smi_speedstep driver instread of the acpi_speedstep (or whatever is the right name). Alan already told me this is likely an ACPI issue. So, I already sent the acpidump in a prev email (Aug 3rd 08). I'm don't have access to my home machine right now. Will send it again later if you can't get my prev email from the archives. Is lspci output needed for this? Thanks, SK --
I could be completely wrong though! I'd not heard of speedstep_smi before (now I look, it doesn't seem to be an option in the cpu frequency scaling Kconfig menu?). I assumed the error messages you had about SMI were just flack. I thought acpi-cpufreq was the only option on recent machines - and that a quad core probably counts as recent. The error messages could be due to Fedora scripts unconditionally loading several different modules, to find one that works. I'm not sure how you can answer Zhao's question if you don't already know. I expect Fedora C7 does try loading acpi_cpufreq (as well as speedstep_smi). Whether your machine actually supports acpi_cpufreq is the question we're trying to answer. (This is where looking at Windows, or BIOS settings, would help). Alan --
On Fri, Aug 8, 2008 at 1:41 AM, Alan Jenkins My machine has acpi-cpufreq module, but when I try to load it, I get: FATAL: Error inserting acpi_cpufreq (/lib/modules/2.6.25.11-97.fc9.i686/kernel/arch/x86/kernel/cpu/cpufreq/acpi-cpufreq.ko): No such device Btw, yes, cpu freq scaling works in Windows XP. What should I look for in Windows? I build my own system, so I have gone through my system BIOS several times and don't remember anything related to freq scaling. But I will check again. Btw, my machine is a Shuttle SG33G5. It has other issues too but right now heat is the biggest concern. Thanks, SK --
There is a "PPM mode" option which has possible values: Native - Native mode is for fully support ACPI OS (ex. Win XP, Vista) SMM - SMM mode is for legacy OS (ex. Win 2k) Setting it to SMM didn't help. Gave the same error for acpi_cpufreq. Btw, you were right about FC trying all the module, but it apparently tried only speedstep modules for some reason. It's clear from my dmesg.txt that all the speedstep modules are attempted. Forgot one obvious thing when I ran dmesg -- forgot to run it after trying modprobe acpi_cpufreq. This is what I get when I do that: acpi-cpufreq: acpi_cpufreq_init acpi-cpufreq: acpi_cpufreq_early_init cpufreq-core: trying to register driver acpi-cpufreq cpufreq-core: adding CPU 0 acpi-cpufreq: acpi_cpufreq_cpu_init cpufreq-core: initialization failed cpufreq-core: adding CPU 1 acpi-cpufreq: acpi_cpufreq_cpu_init cpufreq-core: initialization failed cpufreq-core: adding CPU 2 acpi-cpufreq: acpi_cpufreq_cpu_init cpufreq-core: initialization failed cpufreq-core: adding CPU 3 acpi-cpufreq: acpi_cpufreq_cpu_init cpufreq-core: initialization failed cpufreq-core: no CPU initialized for driver acpi-cpufreq cpufreq-core: unregistering CPU 0 cpufreq-core: unregistering CPU 1 cpufreq-core: unregistering CPU 2 cpufreq-core: unregistering CPU 3 Attaching the files I sent to Zaho so others can look at it too. Thanks, SK --
Try updating the BIOS.
I expect your CPU is too new and the BIOS cannot export the correct freq
values for the specific CPU yet.
Thomas
--
The BIOS is already at the latest version :( Is the above comment the only possible reason for the issue? -SK P.S: Thoman, I initially replied only to you. Sorry for the double email. --
It is very likely. your BIOS does not export basic CPU freq data. The only way to get cpufreq working on latest Intel HW is through ACPI and acpi-cpufreq driver. It may happen that above functions are somewhat hidden in SSDT tables but loading and finding the CPU freq functions generally works quite well. Mark Langsdorf reported some acpi breakage in .27-rc1, this could also be the reason. That a BIOS update is needed for latest CPUs is common. I expect your machine is rather new? There also is a cpufreq list which moved recently, the latest one you should post to is this one: cpufreq@vger.kernel.org Thomas --
Hi,
just that in case you try out kernels...
----------
So it's rc2 not rc1, but AFAIK your report should be older than rc2.
There also was a similar report from Laurence Darby <ldarby@tuffmail.com>.
He still posted on the old cpufreq list.
Subject: Core 2 Duo E8400 stepping 6 not recognised
A BIOS update helped.
You may be able to google this out and compare his CPU_FREQ_DEBUG logs.
Thomas
--
But Windows XP is able to do the scaling properly. So how can it be a BIOS issue? While reading about the Foxconn BIOS fiasco that happened recently a kernel dev (forgot his name) stated that Linux tries to emulate Windows for ACPI operations. So I'm guessing this is a kernel bug. -SK --
_PCT,_PSS, _PPC. So acpi_cpufreq driver can't be used on your machine. This is a BIOS issue. But it is very strange that cpufreq scaling can work on Windows XP. Will you please try the P4 clock cpufreq driver and see whether the scaling can work ? In fact P4 clock cpufreq driver is already replaced by acpi_cpufreq driver for the latest cpu. Anyway please try p4 clock cpufreq driver. --
What is the name of the driver? I'm not able to find any such driver in Fedora 9 installed modules. There is only acpi-cpufreq and powernow-k8. Also, If someone doesn't mind, can they tell me what all files I need to replace in Fedora 9 once I get a kernel.org kernel compiled? I know how to edit grub.conf, and that I should replace vmlinuz, but do I need to replace initrd too? Thanks, SK --
Yes. There is no definition of _PSS,_PCT,_PPC in DSDT table, which are required by acpi_cpufreq driver.At the same time there is no extra SSDT The driver is P4-clockmod, which is located in the directory of Maybe you should enable "CONFIG_X86_P4_CLOCKMOD" in kernel configuration and compile a new kernel. --
So what's the significance of "no extra SSDT table"? Does that mean that XP Thanks. I just looked at the kernel.org kernel (what's the term used in lkml for this?) and found out the switch. I got the kernel.org kernel compiled with the .config used for the F9 kernel. Now I need figure out how to get it to boot. Btw, on your suggestion, I opened a bug for this issue. http://bugzilla.kernel.org/show_bug.cgi?id=11299 Does the LKML community consider the Fedora 9 kernels as too different? I would prefer using the F9 kernel for testing this issue since it's easier for me to compile and install. I got some important stuff to deal with for the next couple of days. So I will get back on this thread in 2-4 days. Thanks, SK --
The F9 kernels are very very close to the linux-kernel ones. See http://docs.fedoraproject.org/release-notes/f9/en_US/sn-Kernel.html http://fedoraproject.org/wiki/Docs/CustomKernel http://www.howtoforge.com/kernel_compilation_fedora --
It sounds like the BIOS just doesn't support frequency scaling. Perhaps it's disabled in the BIOS setup. Or the system doesn't support it at all. If he can't find an option to enable it there he'll have to update it and if there is no update he's out of luck. Using p4 clockmod doesn't make sense because it doesn't really safe any power and is often very slow. --
It works in Windows XP, so I'm not sure if it's only a BIOS issue. If it's not that simple, then care to impart some knowledge? Thanks, SK --
On Windows platform drivers sometimes provide a replacement DSDT that might enable cpufreq. If that's the case there's nothing we can do really on the Linux side. You might be able to extract that DSDT from somewhere and supply it to Linux as a replacement If you can't get your BIOS to provide the necessary methods then useful Linux cpufreq won't work. It's that simple. p4-clockmod aka throttling might work, but it doesn't actually save energy and often causes severe performance problems. -Andi --
Wow! Didn't mean to start a "tech war" here (point's at the long I don't remember installing any special driver for Windows XP. Since you claim that as a possibility, is there something that I can look for to check if this is the case? The reason I'm questioning this so much is because I plan to contact Shuttle tech support and try to get them to fix this if it really is a BIOS issue. But I need valid points to make sure they don't pass on the blame and slip away. They do seem to release BIOS updates now and then I definitely don't want to use anything that will have a drastic effect on performance. Thanks, SK --
On Sun, 10 Aug 2008 21:44:50 -0700 yeah don't bother with p4-clockmod, it's not something that saves you energy... it only gives you a fake feeling that you would. -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
Why it is still in tree? It's likely that you'll waste more time explaining everybody to not use p4-clockmod than send a patch. --
Because it's useful for thermal throttling on hardware that doesn't implement voltage scaling. -- Matthew Garrett | mjg59@srcf.ucam.org --
This statement is only true for HPC environments and when you measure
Watts per work done.
But if you have some desktop or server with a load that is determined by
when it gets used, then p4-clockmod can actually save energy when the
computer is not under high load or even idle.
Or do I miss anything?
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
--
On modern hardware, the C states will win you more than downclocking the processor would. -- Matthew Garrett | mjg59@srcf.ucam.org --
On Mon, 11 Aug 2008 15:07:34 +0100 actually.. p4_clockmod is *NOT* downclocking the processor.. that's the whole problem with it.. -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
On Mon, 11 Aug 2008 17:02:06 +0300 yes you're missing quite a few things... lets build this up 1) when the cpu is idle (as in "idle loop C states/hlt"; p4_clockmod doesn't mean anything.. the clock is stopped not just skipped. 2) when the cpu is executing code (eg non-idle), it takes more power for a unit of time than it takes when it's idle 3) when the cpu is in thermal throttle (p4_clockmod) and does a throttle cycle, it'll take less power for that cycle than a full execution (at least of the power intensive type of instruction), but more than if it would have been in idle. now lets take your "Light load" example (light could mean anything less than 100% cpu use for all I care) and lets assume a 50% throttle (just because it makes the explenation simpler, not fundamental) in 50% thermal throttle, after every cycle a throttle cycle is inserted, this means that code that gets executed will take 2x as long basically. take your lightly loaded case, and lets say you have 20% CPU utilization in the unthrottled case (again, random number, works with any number). Now with the thermal throttle, you suddenly take 40%. so before you had for one second "20% expensive, 80% low power" now you have for one second "20% expensive, 20% throttle power, 60% low power" since throttle power is higher than low/idle power.. you lose. -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
So what is the intended use case?
Is it for people who bought an Intel Heating 4 but forgot to buy an
appropriate cooling?
That sounds absurd.
There must be a reason why Intels CPUs support this throttling?
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
--
On Mon, 11 Aug 2008 19:03:46 +0300 it's thermal throttling. To forcefully reduce the number of cycles that have the full "execute" yes there is.. for cases where there is overtemperature. Think of it as the emergency break in the subway. You really don't want to use it but when you need it you're glad it's there. -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
But this only helps when the cooling is working but not good enough,
for the classic "fan broken" case it won't cool down enough.
Does the Pentium 4 predate CPUs that shutdown when becoming too hot
(I experienced such a shutdown recently on my Athlon...)?
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
--
On Mon, 11 Aug 2008 19:28:39 +0300 I would call this an unclassified statement for which you have no data thermal shutdown is the last measure for sure (and it's in cpus yes)... but sysadmins frown on that.. before you go down that path there's various things that happen first (thermal throttle being one of them) -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
Sorry, I was a bit too surprised after your description.
cu
Adrian
--
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon Seed
--
Hi, This statement might be true, but might also be wrong: a) on systems where only C1 is exported, p4-clockmod most often equals the state the CPU is in when in C1[*], so we're in a win-win, or lose-lose situation. b) IIRC 50% throttling is not "execute-one-statement skip-one-statement execute-one-statement, etc." but instead work for N us, skip for N us, work for N us, etc. Therefore, the situation is a bit more compilcated. Still, I agree that p4-clockmod is useful mostly in corner cases (and for developing the cpufreq infrastructure in the first case, but that's another story) [*] or C2 even? Best, Dominik --
On Mon, 11 Aug 2008 21:33:17 +0200 even if it were EXACTLY identical (which it isn't).. you would be at doesn't change the fundamental math though. -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
Hi Arjan,
well, the spec isn't really clear about this. It says (IA32 Intel
Architecture Software Developer's Manual, Volume 3, section 13.14.3) that
P6 family processors did this using STPCLK#. And STPCLK# was also used by the
chipset to force the CPU to enter C2, IIRC.
Do P4s only do an C1-equivalent (or even less than that) now, as they do the
If it's C2-equivalent vs. C1, it's a win. So throttling would be a win from
this perspective on a only C1-capable PIII, but not on a P4? Is that what
you're trying to hint at here?
To summarize:
(1) p6 family processors use STPCLK# initiated by the chipset for thermal
throttling.
(2) STPCLK# is also used by the chipset to make the CPU enter C2.
(3) p4-clockmod uses the STPCLK#-equivalent in p4 CPUs.
(4) Therefore, it is as effective as STPCLK#, and as effective as C2.
(5) STPCLK#/C2 has higher energy savings than hlt/C1.
(6) therefore, p4-clockmod might make sense on systems which only export C1
as an idle state.
Where's my mistake?
Thanks,
Dominik
--
On Tue, 12 Aug 2008 21:03:02 +0200 not quite.. if it does for a certain cpu, then it's only for cpus that it's basically always less (or really best case equal) than c1 just due not even on PIII is it a win.. it's just too short a duration This is not correct. C2 is only effective if you stay in it "long incorrect -- If you want to reach me at my work email, use arjan@linux.intel.com For development, discussion and tips for power savings, visit http://www.lesswatts.org --
Please don't top-post! According to the online ACPI spec, the BIOS should expose _PCT, _PSS and PPC objects if it supports multiple CPU performance states (aka P-states, cpufreq). I ran your acpidump.txt through acpixtract to generate the DSDT, and decompiled it using iasl. It did not contain any of _PCT, _PSS and _PPC. Again, it is possible this is an option in your BIOS which is currently disabled. You need to go into your BIOS setup screen and see if you can find any relevant options. If possible, find out whether cpu frequency scaling works in windows. If it doesn't work in windows, it's very unlikely to work on linux. Linux ACPI aims for bug-for-bug compatibility with Windows, which is what manufacturers test their product against. There may be other possibilities. Your acpidump output did not appear to include a SSDT, which is another place these objects might be found. My computer suffered from a bug where linux could not find the SSDT (<http://bugzilla.kernel.org/show_bug.cgi?id=8630>). This bug was relatively simple and was fixed. However there might be something similar going on. IIRC from last time, SSDTs or similar can also be loaded dynamically by the AML code in the DSDT, so maybe something goes wrong there. Regards Alan --
