slb_flush_and_rebolt panic trace in 2.6.12-rc2-mm3

[Anton Blanchard]

> Hrm.. if interrupts are now enabled through context_switch(), are we
> safe if we were interrupted between switch_mm() and _switch(), where
> we update the bolted kernel stack slb?

So long as we change both the bolted SLB stack entry and the kernel
stack at the same time in _switch, and we disable interrupts around
there, I think we are OK.

It probably wants more thought but Im having too much fun trying to
build a toolchain to think :)

Anton