[ofa-general] opensm - can/cannot set alternate default pkey?

Al Chu chu11 at llnl.gov
Wed Nov 12 14:59:53 PST 2008


Hey Hal,

Now its making more sense to me.  Thanks for clearing it up.

Al

On Wed, 2008-11-12 at 17:46 -0500, Hal Rosenstock wrote:
> Hi Al,
> 
> On Wed, Nov 12, 2008 at 2:26 PM, Al Chu <chu11 at llnl.gov> wrote:
> > Before I run off and write a patch I shouldn't, I thought I'd ask.
> 
> I don't think there's a need (see below).
> 
> > In 10.9.1.2 of the spec, it states, "The P_Key value of 0xFFFF shall
> > represent the default partition key."
> 
> Default in this sense is referring to the default partition (and it is
> not changeable in the same sense other defaults are).
> 
> All end ports _must_ be a member of the default partition either as a
> full or limited member. This is needed for SA communication. See p.882
> Table 185 P_KeyTable (initialization) for one citation on this. There
> are others in the spec.
>
> > (I couldn't find the glossary in the spec about what "shall" means, but
> > I assume it means "must" or "required" like RFCs.)
> 
> Yes.
> 
> > Does this mean that a P_Key of 0xFFFF must be in the P_Key_Table?
> 
> Either 0xffff or 0x7fff must be in the P_KeyTable of every end port.
> 
> > Currently, it seems that in opensm, no matter how you write your
> > partition.conf file, 0xFFFF will always be the P_Key_Table.  This is
> > because opensm inserts this in it's internal list by default, and
> > nothing (as far as I can find) can remove it/get rid of it out of that
> > internal list.
> 
> That's being a full member of the default partition. You should be
> able to change this to be a limited member of the default partition
> too.
> 
> -- Hal
> 
> > This seems wrong to me, but I'm getting confused on the wording.
> 
> > Thanks,
> > Al
> >
> > --
> > Albert Chu
> > chu11 at llnl.gov
> > Computer Scientist
> > High Performance Systems Division
> > Lawrence Livermore National Laboratory
> >
> > _______________________________________________
> > general mailing list
> > general at lists.openfabrics.org
> > http:// lists.openfabrics.org/cgi-bin/mailman/listinfo/general
> >
> > To unsubscribe, please visit http:// openib.org/mailman/listinfo/openib-general
> >
> 
-- 
Albert Chu
chu11 at llnl.gov
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory




More information about the general mailing list