Proposal: Changing enable_irq/disable_irq to be two argument.
I've been putting together some patches to deal with stuck interrupts more intelligently. Instead of just deciding an interrupt is stuck the code tries to respond to stuck interrupts by running every shared IRQ handler registered. This is generally safe because SHIRQ handlers have to check if their hardware caused the interrupt anyway. It breaks for the corner case of disable_irq() on a shared interrupt.
Fixing this needs enable/disable_irq() to take the cookie we passed to request/free_irq. At that point we can both disable the irq and skip the handler when doing recovery.
Even without this change the hack test version of the change is working very well on test hardware that previously never ran Linux properly because of broken IRQ routing tables. Clearly it needs to be a user enabled option because it harms performance, we need to trap any bugs we cause in this area and to continue to provide appropriate pain to offending hardware vendors.
Alan