[openib-general] disable_irq WARN_ON

[Roland Dreier]

    Grant> I suspect this is a result of mthca_init_one bailing
    Grant> WITHOUT having succesfully called request_irq().
    Grant> request_irq() implies a call to enable_irq().  The
    Grant> WARN_ON() may be unneccessary since I would expect it to be
    Grant> safe to call disable_irq() multiple times for a given
    Grant> device. I'll ask if that's a reasonable expectation (or
    Grant> not).

If you look at the stack trace, it seems iosapic_unregister_intr() is
being called from pci_disable_device(), not disable_irq().  I don't
see any way for mthca_init_one() to call pci_disable_device() without
a balancing pci_enable_device(), which is why I'm confused.

(The IRQ stuff gets involved because of Bjorn Helgaas's recent changes
to wait to route IRQs until pci_enable_device() is called)

 - R.