Re: [PATCH] Fix when booting Xen+Linux under QEMU.

[Jeremy Fitzhardinge <jeremy@goop.org>]

On 12/09/2010 01:01 PM, Konrad Rzeszutek Wilk wrote:
> Please take a look. The first patch just cleans up the find_unbound_irq
> so it is easier to understand what it does.
>
> The second patch hardness the Xen IRQ (event channels) allocation finder
> if you try to boot a minimalistic 32-bit Linux kernel along with a
> Xen hypervisor under QEMU.
>
> We end up with a weird scenario where the nr_irq_gsi was greater than nr_irq
> by 16. This code hardness it by pointing out the issue to the user and tries
> to continue - BUT this might impact the PCI device allocation.
>
> Tested also on normal machine with no regressions found.

I think I'd prefer to:

   1. move to using all dynamic irqs, and use the core kernel irq
      allocator (ie, resurrect IanC's patches)
   2. kill the ioapic dummy page hack (I'm assuming that would be
      trivial - at least to make it all zero - because the kernel won't
      care about the number of GSIs at that point)

    J