AMD Mobile Semprons (3500+, 3600+,...) break with nohz and highres enabled

[Thomas Renninger <trenn@suse.de>]

Hi,

we got a bit further, Evgeny was able to get C-states working with
"nohz=off highres=off" boot parameters.

A short summary what has been found out that far:
  - apic timer breaks on C2 or deeper
  - noapictimer workaround helps to come a bit further, but results in sever
    other errors (see comment #13)
  - "nohz=off highres=off" works (tested on 2.6.25.4-8-pae and
    2.6.22.17-0.1-default SUSE kernels)
  - processor.max_cstate=1 works
  - This affects AMD Mobile Semprons (at least 3500+ and 3600+)


The output of /proc/acp/processor/*/power without highres/nohz
looks promising:
states:
    C1:                  type[C1] promotion[C2] demotion[--] latency[000]
usage[00000030] duration[00000000000000000000]
    C2:                  type[C2] promotion[C3] demotion[C1] latency[005]
usage[00002135] duration[00000000000011279621]
   *C3:                  type[C3] promotion[--] demotion[C2] latency[020]
usage[00034222] duration[00000000000228307174]


Also see bug:
https://bugzilla.novell.com/show_bug.cgi?id=396220
and some posts from Richard on the linux-acpi list.


This seem to be a mainline kernel regression since the nohz/highres
patches have been added.
I do not own such a machine myself (Richard, Evgeny, Uli (and Anvin?)
have such a machine and already helped to come that far).

Also the highres/nohz parts are a dark area to me... Any pointer how to
debug (or debug patches or patches to test), where to look at, etc. is
very much appreciated.

Thanks,

   Thomas