> Why do we need to fiddle with dev->irq? I think it should
> just be undefined after acpi_pci_irq_disable().
I had been considering what the "undefined dev->irq" was.
In fact, I had other ideas that was clearing it by zero or
-1 (0xffffffff). But I didn't know if we can use these values
as a undefined IRQ number. So I'm clearing it by the value
which was assigned by PCI core code (pci_read_irq()) before
acpi_pci_irq_enable() was called.
I think it has little sense in restoring value from the configuration
space to dev->irq or clearing it.
If we do preventive programming, it may be worth
trying to define some magic constant (e.g. PCI_UNDEFINED_IRQ) and
panic/BUG when the irq is being enabled.
Otherwise, leaving dev->irq as it is would be ok.