On Thu, 2005-10-06 at 13:01, Sean Hefty wrote: > Hal Rosenstock wrote: > >>I didn't see any other way to retrieve the pkey associated with an IP > >>address > >>without this. > > > > Yes, and I looked at getting the ib_device but there is no easy way so I > > added them into the structure returned. Is CMA keeping a list of > > ib_devices that it walks for this ? > > The CMA maintains a list of devices. The address translation code takes an > IP > address and returns the corresponding GID. The CMA looks up the GID against > its > list of devices. All synchronization for device removal is handled by the > CMA. > > Currently, the address translation code isn't aware of ib_devices. It's > almost > a device independent IP to HW address translation mechanism. > > A question that I have is how does the user know if the ib_device pointer is > valid?
The only way I see is that a user needs to register as a client and track device removals. Is there another way ? > >>For SDP, if we layered it over the CMA, would it still need to access this > >>information? > > > > I'm not 100% sure. It partially depends on the CMA APIs. How is the > > PathRecord request done ? That's what it's needed for. > > Right now, the CMA issues a path record request based on the SGID/DGID only. > It > would be fairly easy to add the PKey to the request once the address > translation > code returns it. How would the address translation code get it ? -- Hal _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
