[IBAL]path records pkey issue. was: [ofw] what's up with the OFAWindows project?

[Hal Rosenstock]

On Thu, 2008-02-14 at 10:22 -0600, Todd Rimmer wrote:
> 
> > -----Original Message-----
> > From: ofw-bounces at lists.openfabrics.org [mailto:ofw-
> > bounces at lists.openfabrics.org] On Behalf Of Alex Estrin
> > Sent: Thursday, February 14, 2008 11:18 AM
> > To: Leonid Keller; chas williams - CONTRACTOR; Gilad Shainer
> > Cc: ofw at lists.openfabrics.org
> > Subject: [IBAL]path records pkey issue. was: [ofw] what's up with the
> > OFAWindows project?
> 
> > 
> > I think we should talk a little bit more on this issue.
> > 
> > > here's one.  we have two partitions, 0xffff and 0xfff2.  this causes
> > > some confusion with the pnp layer.  when the pnp layer iterates the
> > > nodes/paths to the hardware, the first path returned isnt always on
> > > the default partition.  if this path is given to srp, then srp
> > > attempts to login to the storage on the non-default pkey.
> > 
> > I believe if SM returns path records with different pkeys it should
> mean
> > host and a storage are both members of both partitions, and obtained
> > paths are valid for this host.
> > Then the consumer of path records (SRP initiator in our case) could
> > validate what path has higher priority to use.
> > 
> > > > this patch makes the pnp layer only return the default paths.
> > 
> > With proposed patch all IOCs on the fabric will be visible only by
> hosts
> > having default pkey in it's pkey table.
> > For example, if IOC configured to have pkey1 only, then host on the
> same
> > partition with pkey1 won't ever find this IOC, since it will be
> looking
> > for it in default partition only.
> > 
> > Thoughts?
> 
> 
> I agree with Alex's comments above.  It should be up to the SM to return
> only valid paths in the preferred order of use, the initiator should be
> able to simply use the 1st path provided (additional paths if
> redundancy, etc is used).

Just to clarify:
It's up to SM to return only valid paths. Preferred order of use in the
ordering of those paths in a response is a SM value add rather than a
requirement. So I don't think that the client can assume an ordering.

-- Hal

> It's a bad idea to assume there is a default partition and that all
> hosts and IOCs are part of it.  Such a configuration would totally
> defeat the purpose of Partitions and prevent isolation access to
> specific IOCs for selected hosts.
> 
> Todd Rimmer
> Chief Architect 
> QLogic System Interconnect Group
> Voice: 610-233-4852     Fax: 610-233-4777
> Todd.Rimmer at QLogic.com  www.QLogic.com
> _______________________________________________
> ofw mailing list
> ofw at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ofw