High basal cpu load on 686 but not 386 kernel

[Harm van Bakel]

When running the latest 686 kernel (v20 or v21, haven't tried earlier versions) I noticed that the cpu load was very high, typically 20-30% without top showing any process that could be responsible for this. This high cpu load caused some noticable performance problems, especially in keyboard responsiveness. Keyboard repeat became very sluggish and kscreensaver showed lots of stuttering. After rebooting with the 386 kernel I don't see any problem anymore, so this seems to be specific for the 686 kernel.

A screenshot of top, showing high system load without any process showing much cpu usage, and additional information can be found in the following thread: http://www.ubuntuforums.org/showthread.php?t=165742

I'm running kubuntu on a Dell Latitude D810 (though the problem is not linked to either kde/gnome, since the performance problems are also evident in console mode when X is not started).

[santo (santo-prive)]

I can confirm this.
As hvbakel mentioned, please refer to http://www.ubuntuforums.org/showthread.php?t=165742 for more details

My system is Ubuntu Dapper (with kubuntu-desktop installed afterwards) on a Dell Inspiron 9300

[David Corrales (davidcorrales)]

I'm also running Ubuntu Dapper on an Inspiron 9300 and the 686 is giving me high cpu utilization. The temporary fix is to use the 386 one.

[Harm van Bakel (hvbakel)]

Just checked kernel version 2.6.15-22-686, unfortunately the problem still persists... If anyone has a clue how to get more info on what could be the problem it would be greatly appreciated.

[cellstije (marco-grimaldi)]

The same problem on a VAIO FS-115M using 2.6.15-22-686

Using the 2.6.15-22-386 kernel, the cpu usage is right

[towsonu2003 (towsonu2003)]

http://www.ubuntuforums.org/showthread.php?t=172319 suggests this is a widespread issue.

[Harm van Bakel (hvbakel)]

Did some more searching on the forums, this issue seems to be related to the fact that the dapper 686 kernel has SMP processing enabled by default: http://www.ubuntuforums.org/showthread.php?t=149348&page=2&highlight=cpu+686 (thanks to no1wantdthisname). Could this problem perhaps be specific for laptops?

[Pausanias (pausanias)]

Somebody please officially confirm this bug. I experience the same problem on a Precision M70. User CPU load while idle in GNOME is 7-20% using the 686 kernel, and 1% when using the 386 kernel. User CPU load when running idle in single user (text terminal) mode is 0%. What is going on?

[Awaka (egf-vd-geer)]

Same here.

Acer Aspire 3633lc Kernel 2.6.15-22-686.

[89c51 (barz621)]

the same here on a vaio s4hp

[Ben Collins (ben-collins)]

I haven't noticed this bug on similar setups. Can you all confirm what X driver you are using (or if you are using one of the binary drivers such as fglrx or nvidia-glx)?

[Harm van Bakel (hvbakel)]

I have an ATI X600 card and I'm using the latest fglrx driver (8.24.8)

[Ben Collins (ben-collins)]

Can you test with stock xorg ati driver?

[Harm van Bakel (hvbakel)]

Just tested this by using the 'ati' driver instead of 'fglrx' and rebooting with the 686 kernel, but this doesn't make any difference. Cpu usage with the 686 kernel remains high. Even if I shut down X completely, the high cpu usage still remains. A post by 'no1wantdthisname' (extract pasted below) suggests that this problem is somehow related to the fact that SMP processing is enabled in the 686 kernel by default. Turning of SMP in a custom compiled kernel solved the high cpu load problem in his case. Haven't tried yet if this also works in my case...

-----

"Just a note: Check your CPU load on the 686 kernel. If it's unusually high when you're not doing anything, then you should switch back to 386.

I have a sony vaio laptop with a pentium-m processor. Using the 686 kernel causes my cpu to stay steady at ~50%. Using the 386 kernel: my cpu stays steady at ~1%.

I went ahead and tried building my own kernel. I changed the processor to pentium-m, thinking that the 686 kernel was slow because it was compiling for pentium4. So I went ahead and changed it to pentium m. Even then, the cpu load was still around 50%.
After multiple tries of compiling my own kernel, I noticed that I forgot to turn off SMP. Just by turning off SMP, the load went down to 1%.

Back on breezy, I used the 686 kernel without any problems. However, the dapper 686 kernel includes SMP by default since it theoretically shouldn't be causing any problems even on single processor computers. It seems to be causing the really high cpu load on my laptop though. "

[Pausanias (pausanias)]

If you read the thread linked to above, it is claimed that the problem goes away after recompiling the 686 kernel without SMP support. See the post by no1wantdthisname.

In any case, on my system I can confirm that I experience the problem with *both* the stock xorg nv driver *and* the proprietary nvidia driver.

It seems so far that the problem is limited to Pentium M processors but I could be wrong.

[cellstije (marco-grimaldi)]

I compiled myself a kernel from the linux sources in the dapper
repository and it is working fine by disabling the SMP support. But,
it is optimized for Pentium-M, not generic 686 (Pentium-pro).

On 5/16/06, Pausanias <email address hidden> wrote:
> If you read the thread linked to above, it is claimed that the problem
> goes away after recompiling the 686 kernel without SMP support. See the
> post by no1wantdthisname.
>
> In any case, on my system I can confirm that I experience the problem
> with *both* the stock xorg nv driver *and* the proprietary nvidia
> driver.
>
> It seems so far that the problem is limited to Pentium M processors but
> I could be wrong.
>
> --
> High basal cpu load on 686 but not 386 kernel
> https://launchpad.net/bugs/41373
>

[Pausanias (pausanias)]

I can also verify that recompiling the kernel with SMP disabled fixed this issue.

However, this is not an acceptable solution as you know, because it would mean recompiling the kernel each time a security update is released. Plus there is also the headache of recompiling any closed source drivers as well.

Downgrading to the 386 kernel is not a possibility for me---there is a measurable performance loss in my critical apps with the 386 kernel.

Is there a chance that we could see an SMP-disabled 686 kernel hit the repositories?

Also note that this is a duplicate of bug 30570

[hardyn (arlenn)]

Ditto,

general performace while using the 686 kernel is also very poor, better performance with 386 kernel.

I am on a P-M system.

[cellstije (marco-grimaldi)]

this bug is still present on feisty herd2 on a sony vaio fs11m (centrimo 1.6GHz)
kernel:
2.6.20-5-generic

changing max_cstate as follows, resolve the issue:

echo 1 > /sys/module/processor/parameters/max_cstate

On 8/8/06, hardyn <email address hidden> wrote:
> *** This bug is a duplicate of bug 30557 ***
>
> Ditto,
>
> general performace while using the 686 kernel is also very poor, better
> performance with 386 kernel.
>
> I am on a P-M system.
>
> --
> High basal cpu load on 686 but not 386 kernel
> https://launchpad.net/bugs/41373
>

[Launchpad Janitor (janitor)]

[Expired for linux-source-2.6.15 (Ubuntu) because there has been no activity for 60 days.]

[sibidiba (sibidiba)]

Higy load avarage on idle Gutsy with kernel 2.6.22-14-generic too.