Re: 2.6.19-rc1 genirq causes either boot hang or "do_IRQ: cannot handle IRQ -1"

[Muli Ben-Yehuda <muli@il.ibm.com>]

On Fri, Oct 06, 2006 at 09:14:53AM -0600, Eric W. Biederman wrote:

> Muli Ben-Yehuda <muli@il.ibm.com> writes:
> 
> > My x366 no longer boots with 2.6.19-rc1. The boot either hangs in
> > uhci_hcd_init or dies with 'do_IRQ: cannot handle IRQ -1". Bisection
> > says this one is bad:
> 
> Ok.  So at least the second case is because some irq is being delivered
> to a cpu that was not expecting it.
> 
> The hang case is weird because the kernel does not get told about
> the irqs on your second ioapic.
> 
> When it gets the 'do_IRQ: cannot handle IRQ -1' how long
> has the system been in user space?  (It doesn't look like
> init got started but that is hard to tell, shutting off irqbalanced
> for testing purposes would be interesting)

In some cases we haven't made it to userspace at all. In other, we're
in the initrd.

> Seeing the failure case is really weird because this early in boot
> everything should be routed to cpu 0.
> 
> What happens if you boot with max_cpus=1?

Trying it now... woohoo, it boots all the way and stays up!

> The change the patch introduced was that we are now always
> pointing irqs towards individual cpus, and not accepting an irq
> if it comes into the wrong cpu.  
> 
> The only hypothesis I have so far is that there may be an issue
> with the x366 chipset ioapics that this patch reveals.
> 
> I would suspect a wider issue but in several months of testing
> this is the first bug report I have seen.

I'm trying to find out if other x366's are also seeing it.

> If simple tests don't reveal what is going on then we will
> have to instrument up that BUG and print out the per
> cpu vector to irq tables, the cpu number, and the vector
> the unexpected irq came in on.

I'm certainly game for any debugging you have in mind - this is my
main Calgary development machine so getting it booting is a pretty
high priority :-)

Cheers,
Muli