IOSAPIC cleanup and bugfixes patch
Now that ia64 has been switched over to using the ACPI PCI irq code in the the 2.5 tree, I'm wondering whether we shouldn't just make the Linux irq numbers equal to the GSIs. Both are (relatively) arbitrary, small numbers and making them equal would simplify things a bit. The only downside would be that the Linux irq numbers would no longer be equal to the IA-64 vectors, so that the irq->vector translation would have to go through a table-lookup even on "small" systems. The performance effect of this should be completely trivial (i.e., I bet it's not measurable).
Comments? Takers? ;-)
--david