x86: meaning of nolapic command line option

x86: meaning of nolapic command line option

[Jan Beulich]

Shouldn't that option imply that all APIC related activity, including that relating to IO-APICs or PCI MSI, should be circumvented? I'm finding that MSI must be disabled separately, and while most of the IO-APIC stuff is indeed not happening, acpi_get_override_irq() only checks skip_ioapic_setup, but that doesn't normally set without the noapic command line option.

Is there any reason pci_no_msi() and disable_ioapic_setup() shouldn't be called when !cpu_has_apic at the end of identify_cpu()?

Thanks, Jan

Re: x86: meaning of nolapic command line option

[Ingo Molnar]

* Jan Beulich <jbeulich@novell.com> wrote:

> Shouldn't that option imply that all APIC related activity, including 
> that relating to IO-APICs or PCI MSI, should be circumvented? I'm 
> finding that MSI must be disabled separately, and while most of the 
> IO-APIC stuff is indeed not happening, acpi_get_override_irq() only 
> checks skip_ioapic_setup, but that doesn't normally set without the 
> noapic command line option.
> 
> Is there any reason pci_no_msi() and disable_ioapic_setup() shouldn't be 
> called when !cpu_has_apic at the end of identify_cpu()?

Yes, both depend on a lapic and they might limp on with whatever the BIOS 
gave us, you are right that it should be disabled explicitly. Mind sending 
a patch?

	Ingo

Re: x86: meaning of nolapic command line option

[Jan Beulich]

>>> Ingo Molnar <mingo@elte.hu> 11.01.09 03:46 >>>
>
>* Jan Beulich <jbeulich@novell.com> wrote:
>
>> Shouldn't that option imply that all APIC related activity, including 
>> that relating to IO-APICs or PCI MSI, should be circumvented? I'm 
>> finding that MSI must be disabled separately, and while most of the 
>> IO-APIC stuff is indeed not happening, acpi_get_override_irq() only 
>> checks skip_ioapic_setup, but that doesn't normally set without the 
>> noapic command line option.
>> 
>> Is there any reason pci_no_msi() and disable_ioapic_setup() shouldn't be 
>> called when !cpu_has_apic at the end of identify_cpu()?
>
>Yes, both depend on a lapic and they might limp on with whatever the BIOS 
>gave us, you are right that it should be disabled explicitly. Mind sending 
>a patch?

I will - just wanted to see whether there's some hidden reason behind the
current way this is coded. Actually, I meanwhile realized that doing this
somply based on !cpu_has_apic wouldn't be right, it should (at least for
32-bits) also depend on APIC_INTEGRATED() - just like e.g. done in
APIC_init_uniprocessor().

Jan