>Of course, there is a bug. Note, though, that the bug may be the lack of a timeout for some operations, rather than something the driver explicitly does to lose an interrupt; there may well be hardware botches that permanently lose completion interrupts.