On Thu, 2007-02-15 at 10:53, yipeeyipeeyipeeyipee wrote: > Hi, > > It seems like I've stumbled into some sort of bug in the port info mad query. > I have several pc's connected to an IB switch. > On one of the machines I have an OpenIB installation, and on one pc I > continuously run a management utility that sweeps the fabric (using > ibnetdiscover from management/diags/ibnetdiscover/). At one point in time > after > another slow-booting pc boots, ibnetdiscover fails during its fabric sweep and > the IB_ATTR_PORT_INFO query to the sweeping node's ib port fails returning a > physstate == 6 (LinkErrorRecovery). > When I check the /sys/class/infiniband/mthca0/ports/1/state I get "4: ACTIVE".
That's because the initial smpquery (by ibnetdiscover) sees the LinkErrorRecovery PortPhysicalState, the port then comes up at the physical level, and then the SM moves it through the port states to active and when you look again locally (via sys/class/infiniband/mthca0/ports/1/state), it has been made active and I would expect an smpquery of portinfo of this or ibnetdiscover would now show this. > Is there some known issue with port info mad queries? Could this be somehow > related to mixed SDR/DDR switch and hcas? Maybe someone here knows how to > workaround this issue? Sounds like the way it is suppposed to work to me. -- Hal > Thanks > > > > _______________________________________________ > openib-general mailing list > openib-general@openib.org > http://openib.org/mailman/listinfo/openib-general > > To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general > _______________________________________________ openib-general mailing list openib-general@openib.org http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
