[ltp] Re: [gmane.linux.hardware.thinkpad] Re: T60p officially supported with SuSE SLED 10 by Levono

[Andrew Barr]

(Restoring the CC to the list, hope you don't mind)

On Wed, 2006-08-16 at 17:44 -0400, David Abrahams wrote:
> Andrew,
> 
> My posts don't seem to be showing up on the list, so here's the
> result.
> 
> How do I undo this stuff?  My sensors showed the GPU at 77 celsius
> while idling, which is higher than I've ever seen before, and the fan
> was running slower than it ever runs after being on for a few minutes
> in Linux.  The whole machine felt really hot.  It's off now, needless
> to say.

Well, we don't know what the batinfo module does, it's binary-only, but
to remove it, you need to 'modprobe -r batinfo', ideally delete the .ko
file so it doesn't get autoloaded somehow again on boot, and probably
reboot.

That's if you extracted the .ko file manually. If you're on SuSE, you
need to remove the RPM files you installed using the 'rpm' command line
tool or distro GUI tools.

I assume your graphics hardware is a Radeon chip. You're better off
asking other people about that GPU specifically. I don't know anything
about them.

> email message attachment

> > plain text document attachment (dmesgs.txt)
> > [17183159.440000] batinfo: no version for "struct_module" found: kernel tainted.
> > [17183159.440000] batinfo: no version magic, tainting kernel.

Well, maybe we need someone with SLED 10.1 or whatever it is, if that's
all that was output.

> > > Diff the two files--'diff -Naru pre.txt post.txt > ioports.diff'
> > 
> > No difference (really!)

Hm. I don't know what to make of this. Either this is a really strange
module (doesn't use kernel APIs much) or it's not activating for some
reason.

If it doesn't use kernel APIs, is it possible they're trying to avoid
being a "derived work" and be forced to release their source code? It
seems to me sysfs is the logical place for interfaces for something like
this, but apparently they use /proc (as far as we know). Sysfs is
GPL-only, right?

> > Since I'm unable to distinguish what's interesting, I've attached a
> > diff of "ls -lR1 /proc" before and after (see bottom).

> > Does NHC undervolt the CPU without any special configuration?  It
> > looked like that section of the NHC panel was not enabled (I'd have
> > had to check a box).  

I don't know. Anyone have NHC experience care to chime in?

> > I just turned NHC on and the machine cooled off
> > a little and the fan stopped, like magic.

Well, just remember it's not magic. I think people forget that sometimes
when they are having issues with Linux that don't show up in Windows. I
don't know exactly what that program does, it certainly doesn't do
anything that Linux (or other OSes, for that matter) cannot do. It's a
matter of figuring out what that program does and reproducing it in
Linux, or (in a worst-case scenario) finding out why that would not be
entirely possible.

I played around with the undervolting patch available at
http://linux-phc.sourceforge.net/ , and after finding some reasonable
values, my machine idles at about 43C if the fan cools it and can go as
low as 41. As long it it is idling or just lightly loaded, you can shut
off the fan and it will stay there. And I haven't even tested battery
life yet. There's also some patches out there that let you take over fan
policy from the firmware. Some people seem to think that the IBM
defaults for shutting off the fan are too low--the fan never goes off.
That has certainly been my experience--even at 41C the fan stayed on.
Look at ThinkWiki on the fan control pages, there are some fan control
scripts (userspace daemon) and also a patch that creates a kernel thread
to control the fan--this is for the paranoid who don't trust user space
to be reliable. Be forewarned, the patches, especially the kernelspace
daemon one, are somewhat out of date and require merging into recent
kernels by hand.

OTOH, the fan control scripts may not require any patching at all.

> > And what is it about Windows that allows undervolting to be done
> > without patching the kernel (or, what is it about Linux that requires
> > a kernel patch?)

You may be patching the kernel when you install NHC, in a manner of
speaking. If it installs any drivers (.sys files) then it is doing the
equivalent of patching Windows' kernel.

> > plain text document attachment (procdiffs.txt)
[snip]

Hm, nothing I haven't seen in a regular system.

(The list was huge, BTW, and might be why your posts aren't showing
up--the message was too big. So, if anyone wants to see it resend it to
them as a gzipped attachment)

Andrew