Correction for broken MCFG tables on K8 breaks acpi for MSI S260

[Gerhard Schrenk <deb.gschrenk@gmx.de>]

Hi,

commit d6ece5491ae71ded1237f59def88bcd1b19b6f60 breaks acpi for my
Medion MD 95600 (aka MSI S260) notebook.

|gps@medusa:~/scratch/kernel-bisect-tree$ git bisect bad
|d6ece5491ae71ded1237f59def88bcd1b19b6f60 is first bad commit
|diff-tree d6ece5491ae71ded1237f59def88bcd1b19b6f60 (from 928cf8c62763349efc550a12f6518e52c3390906)
|Author: Andi Kleen <ak@suse.de>
|Date:   Mon Dec 12 22:17:11 2005 -0800
|
|    [PATCH] i386/x86-64 Correct for broken MCFG tables on K8 systems
|    
|    They report all busses as MMCONFIG capable, but it never works for the
|    internal devices in the CPU's builtin northbridge.
|    
|    It just probes all func 0 devices on bus 0 (the internal northbridge is
|    currently always on bus 0) and if they are not accessible using MCFG they are
|    put into a special fallback bitmap.
|    
|    On systems where it isn't we assume the BIOS vendor supplied correct MCFG.
|    
|    Requires the earlier patch for mmconfig type1 fallback

After this patch I can only boot with "acpi=off". Before this commit
I could boot with "pci=noacpi acpi_sleep=s3_bios resume=/dev/hda2"
without an oops. Unfortunately 

  git revert d6ece5491ae71ded1237f59def88bcd1b19b6f60

does not work without merge conflict on top-of-kernel.

You can find the call trace picture for first bad kernel
d6ece5491ae71ded1237f59def88bcd1b19b6f60 on 

  https://mc.1und1.de/mc/eLBNuN1jZx61ZrQL0dXS0RxhWfCjEv 
  (&& click on "MediaCenter starten"; valid until 2006-01-16)
  
There you will also find dmesg, config , lspci and lspci -vv with the
latest good commit 928cf8c62763349efc550a12f6518e52c3390906

My notebook is a "Medion MD 95600" (sometimes called "SIM 2100" or "SIM
2010"). It's actually a "MSI S260".  You may find more information for
this notebook on the tuxmobil page.

 http://tuxmobil.org/msi.html

-- Gerhard