ATA driver races with interrupts

[Matthew Dillon]

    I should also note that there seems to be an assumption (at least in 4.x)
    that no ATA interrupt will occur once the ATA interrupt is masked.

    The problem with this assumption is that the ATA interrupt could have
    already occured and been queued/delayed due to a mutex, Giant, or other
    mechanism, just prior to the device interrupt being masked.

    It is exactly the same problem that one has with shared device 
    interrupts so hopefully solving the shared device interrupt case will
    also solve the interrupt masking race case.

						-Matt