On Wed, Sep 07, 2005 at 09:33:06PM +0100, Ceri Storey wrote:
> I've got a box here running NetBSD 3.99.8 (built 2005-09-05), which is
> all fine and dandy, aside from devices hme[0-3]. I suspect that there's
> a problem with the interrupt setup; for several reasons. Initially I
> noticed that pinging a device physically connected to the same LAN was
> somewhat slow, somtimes taking ~1s to respond, other times ~10ms.
> However,  when running a non Xen kernel on the same hardware, the norm was 
> < 1ms.
> 
> I suppose alarm bells should have began to ring when I had to increase
> NR_PIRQS, as the card had been allocated irqs 40-43, wheras they were
> normally allocated lower down. 
> 
> I've attached both the regular and Xen (domain 0) dmesgs, and I'm rather
> hoping that some kind soul would be able to point me in the right
> direction. 
> 
> [...]
> hme0 at pci3 dev 0 function 1: Sun Happy Meal Ethernet, rev. 1
> hme0: interrupting at irq 40, event channel 13

Hum, can we have more than 32 IRQs on an i386 ? The i386 port seems
to only know about IRQ 0-31. Using irq 40 shouldn't be a problem in Xen,
but I suspect that irq 40 just doesn't exists at the hardware level.

With an i386 kernel, the IRQ seems to be properly remapped. Is your
GENERIC a plain GENERIC, or did you add options ?

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--