kstopmachine thread panic in do_exit (2.6.9-rc1-bk1)

Hi-

Something crept in between 2.6.9-rc1 and 2.6.9-rc1-bk1 which reliably trips the "Aiee, killing interrupt handler!" panic in do_exit when offlining a cpu. I inserted a BUG() before the panic call to see what's causing it, looks like one of the kstopmachine threads spawned by stop_machine. I can't do a comparison with current -mm because offlining a cpu doesn't even get that far there.

Backing out the most recent sched.c change ("sched: smt fixes" as of this writing) fixes it. Turning off preempt seems to work around it. The bug is still present in -bk2. I'm seeing this on a 4-way ppc64 machine.

Nathan

Yes, this is fixed in the -mm tree by the two patches I posted yesterday. Those versions depend on nicksched, so Andrew dropped them. I'll make new ones and send shortly.

Rusty.

Anyone who quotes me in their signature is an idiot -- Rusty Russell