Most recent kernel where this bug did *NOT* occur: unknown Hardware Environment: UP PC with two FireWire cards, one card has an external FireWire node attached Steps to reproduce: for ((x=0; x<20; x++));do modprobe ohci1394 && sleep 2 && modprobe -r ohci1394 || break; done modprobe -r will soon get stuck in D state, hanging on kthread_stop with hl_drivers_sem and kthread_stop_lock held. But different to bug 6706, knodemgrd sleeps *interruptibly* in nodemgr_host_thread. As soon as the external node is unplugged, modprobe gets going again. I.e. the bug can be recovered from with physical access to the machine. I cannot reproduce this _without_ an external node attached.
PS: This was an uniprocessor kernel, configured without kernel preemption.
Als happens on UP PC with SMP PREEMPT kernel.
Created attachment 10376 [details] ieee1394: fix another deadlock in nodemgr
fix is in 2.6.21-rc1