[ia64 R&D] Re: [openib-general] disable_irq WARN_ON

[Roland Dreier]

    Matthew> The iosapic_unregister_intr() function is brand new, so I
    Matthew> would expect that to be buggy, not the driver.  I'm not
    Matthew> sure who is responsible for the
    Matthew> CONFIG_ACPI_DEALLOCATE_IRQ; someone at Intel, I think.

Was it Intel or was it part of Bjorn Helgaas's changes to interrupt routing?

In any case at first look it does seem like iosapic_unregister_intr()
is complaining about a perfectly valid pci_disable_device() balanced
by an earlier pci_enable_device().  Perhaps the fact that there's
_not_ a request_irq() in the middle is confusing things.  In any case
I have to confess that I don't understand ia64 interrupt routing and
the relationship between gsi, irq and vector well enough to follow
what is going on here.

All of this leaves open why mthca_init_one() is giving up and trying
to disable the device on Andras's system, which is probably the more
important question.  Answering this will require more information
about how far the device initialization is getting before it fails.

 - R.