:(a) add a USB IPL which can be used by the ?HCI to protect internally
: against races and has a lower priority than all the other masks
: [read: is masked by them]
:(b) add a facility to specify the interrupt mask with nexus_setup_intr
: directly and be allowed to change that later. This is the easiest
: solution and would allow us to implement something similiar to
: intr_mux for the *HCI interrupt handling. USB drivers could use
: the same protection mechanism as normal drivers and specify the own
: mask directly.
:(c) switch to soft interrupt style handling like NetBSD already does.
: This has the advantage of fitting into the DF style, but I'm not
: sure yet how much work it is. I have to look at the implementation
: of the taskqueue framework.
:
:Joerg
Well, we're trying to get away from the SPL model. We could use a token
to simulate the SPL (it should be ok for the USB interrupt code to obtain
and release tokens), but I think the cleanest solution is to follow
NetBSD.
You could temporarily use a critical section instead of spl*() in the USB
code.. or at least, that would be preferable to doing (a).
-Matt
Matthew Dillon
<dillon@backplane.com>