SMP+Tulip lockup, 2.0.36pre15, help requested

[Mark Hahn]

> any performance impact, but we have not benchmarked it, either; would
> you expect this to make a noticeable difference?

no.  the cost is trivial: a driver checks its hardware to see whether
there really is an interrupt to handle.  should take a microsecond or so.

> Either way, I would not think the system should lock up completely.

of course, if the driver is broken, it won't share an irq correctly.