> >Anything in the interrupt code, or the arch independent IDE code since >then? From the descriptions posted, there was a stray IDE interrupt after >the set multiword DMA failure in most cases. Should the SET_FEATURE >command that performs the set multiword DMA 2 operation generate an >interrupt? Can delayed handling of this interrupt result in the drive not >clearing the busy flag?
It should not generate an interrupt, but I may have forgot to tell the drive not to do so. But that's not the problem. The IDE interrupt problems are almost _never_ related to arch interrupt management, and almost always related to IDE configuration problems (or bogus drives). There have been no changes to interrupt management I know of affecting recent 2.2.x kernel on old ohare controllers anyway.