On Tue, 2008-05-06 at 17:09 +0300, Moni Shoua wrote: > The purpose of this patch is to make the events that are related to SM change > (namely CLIENT_REREGISTER event and SM_CHANGE event) less disruptive. > When SM related events are handled, it is not necessary to flush unicast > info from device but only multicast info. This patch divides the events that > are > handled by IPoIB to three categories; 0, 1 and 2 (when 2 does more than 1 and > 1 > does more than 0). > The main change is in __ipoib_ib_dev_flush(). Instead of flagging to the > function > about pkey_events we now use leveling. An event that requires "harder" > flushing > calls this function with higher number for level. Besides the concept, > the actual change is that SM related events are not flushing unicast info and > not bringing the device down but only refresh the multicast info in the > background. > As far as I know, when an SM change event occurs, it could mean the SM changed and the new one "decided" to reprogram all the LIDs for example. In that case you will issue only level 0 and the all your neighbours can become invalid.
_______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
