K 10
svn:author
V 7
rwatson
K 8
svn:date
V 27
2009-05-31T13:52:17.537258Z
K 7
svn:log
V 583
Add a shutdown handler for device polling -- don't issue wakeups to the
netisr once file systems are done syncing, otherwise the scheduler may
generate IPIs to CPUs that have already been shutdown, leading to a
panic.

As similar panics (spin lock 0xc0d8e500 (sched lock 1) held by
0xc4d546c0 (tid 100005) too long) have been reported on both 7.x and 8.x
in other code, we might want to think about whether there's some missing
scheduler shutdown logic to handle this case for unpinned/unbound
threads by migrating them to the CPU managing the shutdown and allowing
them to preempt.

END