Network interrupt after shutdown method has been called [kern/62889]

[Maxim Sobolev]

Folks,

We experience a 100% reproducible panic on one of our machines during 
shutdown+power off. We have found that it's caused by the interrupt 
which happens in re(4) after re_shutdown() method has been called. Quick 
googling reveals that we are not alone who experience this problem and 
such condition sometimes happens as a result of interaction with the 
particular ACPI implementation on shutdown+power off.

Some FreeBSD network drivers have been patched to workaround for the 
problem (i.e. vr(4), see kern/62889), but quick browsing through sources 
suggests that majority still can be affected by the exactly the same 
problem.

Hence the question:

Who is "guilty"? Can the network driver make an assumption that no 
interrupt will happen after its foo_shutdown() has been called? Or such 
assumption cannot be made? In the former case most of the network 
drivers have to be fixed (usually by turning foo_shutdown() into a 
wrapper to foo_detach() as with vr(4)), while in the latter the reason 
of this stray irq should be investigated further and fixed where 
appropriate.

Any comments/ideas?

-Maxim