[ofa-general] Coalescing interrupts in SRP

[Roland Dreier]

 > Does the "ib_req_notify_cq" call do the rearm?

Yes.

 > If so, does that mean that until it rearms, the interrupt is disabled?

Yes.

 > If so, shouldn't a delay before the rearm (after clearing the current
 > queue) in effect coalesce?

One could do that... probably in conjunction with
IB_CQ_REPORT_MISSED_EVENTS to avoid races on some hardware.  However, if
one is going to add a timer and delay before rearming, it seems you
might as well use that timer to poll the CQ again and see if there's
more work, and go to a NAPI/polling model instead.