On Mon, Jul 19, 2010 at 09:32:58AM +0300, Jukka Ruohonen wrote: > This is a known issue. No clean solution exist in any implementation I am > aware of. The IDs may also vary between the processor object and MADT.
As I noted to Christos in private mail, the right solution is likely to include the ACPI processor object ID in (x86) cpu_info and operate with that value only. The specification imposes no restrictions on these IDs; some BIOSes may start counting the ACPI CPUs from zero and others from one. > Due to the reasons with the identification, I think we might want to > attach them all nevertheless. I also checked the ACPI tables of some servers; these may define even up to 0xFF processor objects by default, possibly in order to use the same tables with different models, etc. Another case is virtualized environments. So, yes, we should only attach those ACPI CPUs that have a real counterpart. - Jukka.