Messages in this thread | | | Date | Tue, 24 May 2011 08:15:25 +0200 | From | Ingo Molnar <> | Subject | Re: [GIT PULL] more x86/apic changes for v2.6.40 |
| |
* Ingo Molnar <mingo@elte.hu> wrote:
> Possibly the resume fix in the APIC merge :-/ > > It's these commits: > > 31dce14a3269: x86, ioapic: Use ioapic_saved_data while enabling intr-remapping > 4c79185cdb14: x86, ioapic: Allocate ioapic_saved_data early > b64ce24daffb: x86, ioapic: Fix potential resume deadlock > > The first one fixes the resume bug in an easily backportable way (although > the GFP_ATOMIC is not nice), the later two do it cleaner. > > So if b64ce24daffb works for you and 31dce14a3269 breaks this would signal > that the fix from Suresh is the source of the Atom regression.
In particular the lapic_resume() bit looks suspect. Does save_ioapic_entries() get called? It's called in enable_IR_x2apic() but your Atom is probably not x2apic. Non-x2apic system do not get save_ioapic_entries() called, so there's nothing to restore at resume time AFAICS ... Suresh?
Thanks,
Ingo
| |