can reproduce it here. the problem is 9fronts implementaiton of ioprocs. instead of posting notes, we added a "interrupt" and "nointerrupt" ctl messages to /proc/n/ctl that interrupts without posting a note.
the problem was that notes could be scheduled before we even did the syscall making them unreliable. this mechanism simplified the ioproc implementation quite a lot, but relied on the fact that one can open /proc/n/ctl of the ioproc(). this breaks because nfactotum makes its ctl file readonly so it can't be killed. fortunately, it doesnt use iointerrupt() at all so we could just ignore the failing open of /proc/n/ctl in the case of factotum. -- cinap
