Re: 2.6.27 kernel disables frequency switching on x86_64 - stuck on lowest frequency

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, 26 Sep 2008, Jason Vas Dias wrote:
> 
> There appears to be a serious problem with every 2.6.27-X kernel I've 
> tried - and I've tried quite a few from the linux-2.6, linux-acpi-2.6, 
> and linux-2.6-tip GIT trees, and the latest Fedora 10 kernels over the 
> past few weeks trying to solve this problem:

So which trees work, and which don't? Also, since it seems entirely 
repeatable, this is a prime candidate for "git bisect" - it will be 
eventually faster to do than trying many different trees at random, even 
if you will likely need to reboot/compile about 12 times or so (assuming 
2.6.27-rc1 is broken, and 2.6.26 works, which it sounds like).

Bisecting really isn't that hard. Get the git tree, do

	git bisect start
	git bisect bad v2.6.27-rc1
	git bisect good v2.6.26

and off you go. You don't even need to know a lot about git, there's a few 
quick tutorials out there if you haven't used it. See for example

	http://www.kernel.org/doc/local/git-quick.html

which has an example git bisect run, in addition to just initial clone 
insutrctions.

> CPU frequency switching is completely disabled, the 2.2Ghz AMD TL-64 
> Dual Core CPU is stuck at 0.8Ghz, the machine cannot reboot or perform 
> any ACPI actions.
> 
> There are no obvious failures indicated in the logs, only this:
> 
> [    0.000000] ACPI Error (tbfadt-0453): 32/64X address mismatch in "Pm2ControlBlock": [00008800] [0000000000008100], using 64X [20080609]

Hmm. I don't know if it's ACPI-related, but the fact that it cannot even 
reboot etc sure seems to make it likely. Have you made the acpi lists 
aware of it (linux-acpi@vger.kernel.org and lenb are listed in 
MAINTAINERS)?

		Linus