Bug 13995 - Incorrect CPU frequency reported
Summary: Incorrect CPU frequency reported
Status: CLOSED INVALID
Alias: None
Product: Power Management
Classification: Unclassified
Component: cpufreq (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: cpufreq
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2009-08-16 09:54 UTC by Ben Morris
Modified: 2011-03-03 01:13 UTC (History)
2 users (show)

See Also:
Kernel Version: 2.6.29.5
Subsystem:
Regression: Yes
Bisected commit-id:


Attachments
My /proc/cpuinfo, with performance governer (1.34 KB, text/plain)
2009-08-16 09:55 UTC, Ben Morris
Details
Output of cpufreq-info (1.06 KB, text/plain)
2009-08-16 09:56 UTC, Ben Morris
Details
/proc/cpuinfo not showing the bug (1.34 KB, text/plain)
2009-08-17 08:28 UTC, Ben Morris
Details
Output of cpufreq-info not showing bug (1.08 KB, text/plain)
2009-08-17 08:30 UTC, Ben Morris
Details

Description Ben Morris 2009-08-16 09:54:49 UTC
Both cpufreq-info and /proc/cpuinfo are incorrectly reporting CPU frequencies on my system.

They both say my 1.86GHz Core 2 Duo is capable of running at either 600 or 700MHz.

Reasons I believe this to be incorrect:

The BIOS configuration screen confirms that the processor is not underclocked (or overclocked).
I have tested with Live CDs and got much higher reported frequencies.
The bogomips score is consistent with what I think the frequency should be, and nearly identical to that recorded when running a Live CD which reports sensible frequencies, suggesting that the live CD is not actually making the CPU faster :-).
The system does not feel like a 600MHz system; I like to think I would notice that sort of performance.

I am running Gentoo Linux (with a "vanilla" kernel). The LiveCD which reported high frequencies was Slax using kernel 2.6.27.8 (I don't know how they patch it), which claimed 1903.992MHz (which seems too high to me). The slax kernel did not seem to support frequency scaling, so I could not determine which other frequencies should be supported. I also tested the Sabayon LiveCD using kernel "2.6.29-sabayon", which had an identical (save for very small bogomips differences) /proc/cpuinfo to my real system.

I would like to test various old kernels properly on my actual file system with similar configs, but I am using an ext4 root partition.
Comment 1 Ben Morris 2009-08-16 09:55:38 UTC
Created attachment 22743 [details]
My /proc/cpuinfo, with performance governer
Comment 2 Ben Morris 2009-08-16 09:56:35 UTC
Created attachment 22744 [details]
Output of cpufreq-info
Comment 3 Ben Morris 2009-08-17 08:24:55 UTC
I'm sorry to say that I gave incorrect information above: the CPU *was* slightly overclocked, because the BIOS config screen seems to be buggy (choosing something like "default setting - 1866 (266)" for the FSB frequency actually caused it to be set at 272MHz for a CPU clock speed of 1904MHz - yes I know this is not a kernel issue; just explaining how I got the incorrect information). This meant that the figure from Slax was correct. At no point was the CPU overheated, nor were there any signs of instability, even when compiling many packages (with -j3). The motherboard is an Abit AB9.

Setting the memory frequency to 266 manually in the BIOS makes the kernel report correctly again.

I presume this is still a bug, since when slightly overclocked the CPU frequency is misread, when the machine is running very stably and older kernels are able to display it correctly.
Comment 4 Ben Morris 2009-08-17 08:28:49 UTC
Created attachment 22751 [details]
/proc/cpuinfo not showing the bug

Attached cpuinfo when not overclocked.
Comment 5 Ben Morris 2009-08-17 08:30:49 UTC
Created attachment 22752 [details]
Output of cpufreq-info not showing bug

Attached output of cpufreq-info when not overclocked.
Comment 6 Dave Jones 2009-08-17 22:20:10 UTC
when overclocked, there is no way of getting an accurate reading of the processor speed other than running a benchmark. What you're seeing in /proc/cpuinfo is the best guess that the cpufreq code could come up with based on the wrong information that it had to work with from the BIOS.

(The BIOS typically only has information about the stock cpu speeds, not the overclocked ones)

There's no bug here afaics. You ran your hardware outside of spec, and stuff broke as a consequence. Expected behaviour.
Comment 7 Ben Morris 2009-08-17 23:32:12 UTC
Fair enough. I probably wouldn't have filed the bug if I'd realised the BIOS configuration was buggy.

Note You need to log in before you can comment on or make changes to this bug.