Re: [PATCH] MSI interrupts: disallow when no LAPIC/IOAPIC support

[Grant Grundler <grundler@parisc-linux.org>]

On Mon, Sep 26, 2005 at 09:52:45PM -0700, Randy.Dunlap wrote:
> "nosmp" (currently) means 1 CPU and no LAPICs/no IOAPICs.

ok - ie we ignore HW that may be present.

In a way, this makes sense since LAPIC was used for
IRQ "load balancing" - ie IRQs could get redirected
to another "less busy" CPU.  IIRC, discussions about
"XPR" should explain how that works.

> It doesn't have to be that way, but yes, I suspect that it's
> mostly historical, plus a method of getting to a lowest common
> hardware boot sequence, which is sometimes nice for debugging
> or installation.
> 
> Nevertheless, there is a problem here.  What do you suggest
> to solve it?  Just making PCI_MSI depend on Local APIC support,
> or something else?

Yeah, my preference would be PCI_MSI only depend on Local APIC.
(Note that having a Local APIC implies having an IO APIC as well
- but MSI should completely ignore it.)

But I don't know enough x86 history to know if thats feasible
or not. And I'm not blessed with the time to unravel the x86
APIC support - if the dependencies are necessary, so be it.

> [Kernel is assigning MSI interrupts, but then they are "lost."
> Using "irqpoll" will find them, but that's a performance penalty.]

Interesting. I'm suprised an MSI can get "lost".
It implies MSI code is allocating a CPU vector but the vector is
not getting enabled/unmasked or the Local APIC is ignoring it.

Yeah, we don't want to be using irqpoll for this - entirely
defeats the purpose of MSI.

thanks,
grant