Quoting Samuel Thibault (2016-02-28 22:35:14) > Justus Winter, on Sun 28 Feb 2016 00:33:20 +0100, wrote: > > 3/ ... instead the notification message carries a send right as the > > reply port, and the userspace server acknowledges the interrupt by > > deallocating this right. > > That makes much more sense to me than the current disable/enable. I > guess that this will allow on the long run to have shared interrupts?
I believe so, but I don't know how interrupt sharing works. I guess all the drivers have to acknowledge the irq before it can be reenabled? If so, then this interface will make it possible to implement this. It is also safer in the sense that only the interrupt can be reenabled that was disabled because of the notification. I also believe that we can make it faster this way, though I'm not sure it currently is because a no-senders notification has to be generated each time, and that involves notification message allocation and dispatch. If the need arises we can short-circuit that for in-kernel notifications. Justus
