Grant> Well, or other race. Ordering was just my first guess.
Grant> It's likely the race is not even here - but on the
Grant> completion side of things.
Hmm, I'll think about it too. The problem appears to be an overflow
of the FW command completion event queue (that's what the message
ib_mthca 0000:41:00.0: Unhandled event 0f(00) on eqn 3
is saying ... event 0f is EQ overflow)
It is true that we call mthca_cmd_event() on FW command completion
(which frees up a command slot and could let another command start)
before we update the consumer index and let the HCA know that there is
a free slot in the EQ. Still, it's a little hard to see how we could
overflow the EQ (which is created with 128 slots).
- R.
_______________________________________________
openib-general mailing list
[EMAIL PROTECTED]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
