AW: Re: Xen BUG in mm / Xen 4.0.1 with 2.6.32.18/21 pvops Kernel?

["Jan Beulich" <JBeulich@novell.com>]

>>> On 14.09.10 at 11:10, "Carsten Schiers" <carsten@schiers.de> wrote:
> I think I now know. I found the BIOS option that causes the problem: 
> SpeedStep. The buggy table
> is the description of the C-States. But there is a different behavior 
> between native and Xen.
> See native here with acpi debug enabled:
> 
> [    0.358790] Table [SSDT](id 000C) - 6 Objects with 0 Devices 4 
> Methods 0 Regions
> [    0.358793]   nsload-0136 [10] ns_load_table         : **** Completed 
> Table Method Parsing and Object Initialization
> [    0.358944] processor_throttling-1143 [00] processor_get_throttli: 
> pblk_address[0x00000810] duty_offset[1] duty_width[0]
> [    0.358952] processor_throttling-0893 [00] processor_get_fadt_inf: No 
> throttling states
> [    0.358959] processor_idle-0361 [00] processor_get_power_in: No _CST, 
> giving up
> [    0.358963] processor_idle-0317 [00] processor_get_power_in: C2 
> latency too large [101]
> [    0.358967] processor_idle-0325 [00] processor_get_power_in: 
> lvl2[0x00000000] lvl3[0x00000815]
> [    0.358972] processor_idle-0548 [00] processor_power_verify: latency 
> too large [1001]
> 
> No _CST, giving up, C2 too large (0x80000000, I guess), giving up. Xen 
> doesn't give up.

Hmm, I wouldn't think it's the 0x80000000 value, but more likely
some random value they happen to read. The question really is
what ioremap()s the native kernel executes, and how that
compares to the Xen case.

> I did not manage to create the same with Xen, as it will not appear in 
> the serial log. It will show in dmesg, if it comes
> up, but unfortunately, it will crash with SpeedStep enabled, but not 
> interpret SSDT1 if disabled, thus no information. 

Sure. I have to admit that I don't know much about pv-ops' C-state
handling. Did you try a non-pv-ops kernel (e.g. XCP's or SLE11SP1's
or OpenSuSE 11.3's)?

But in any case - the root cause is broken firmware.

Jan