[openib-general] Re: [PATCHv2] OpenSM: OpenIB vendor layer: Implement osm_vendor_d elete

Liran Sorani liran at mellanox.co.il
Tue Sep 6 09:15:26 PDT 2005


> Hal Rosenstock wrote:
> > On Tue, 2005-09-06 at 11:04, Eitan Zahavi wrote:
> > 
> >Hal Rosenstock wrote:
> >
> >>[same patch just generated with diff -up]
> >>
> >>OpenSM: OpenIB vendor layer: Implement osm_vendor_delete
> >>
> >>[I've done some testing of this; are there any regressions for this ?]
> >>
> >>OpenSM call osm_vendor_delete during osm_opensm_destroy.
> >>It is invoked during exit.
> > 
> > 
> > Actually, it should have said "Implement osm_vendor_unbind" rather than
> > delete.
> Well the semantics are very old. We stick to the old osm_vendor_api.h.
Maybe we should not have.
> But now we have too much depending on this API that I urge you not to
modify it if possible.
> > 
> > OpenSM calls it in a similar place (as it stops the SA and SM MAD
> > controllers). I went through this starting and stopping the OpenSM a
> > number of times although did not do this "infinitely".
> > 
> > I was asking about any other explicit regressions.
> No we do not have tests for the osm_vendor_api.h. It might have been a
good idea.
> We do have now some stuff for the Windows code. Maybe we should try and
make a test suite
> from it.  Liran - what do you think ?
I think that the vendor api should be tested with several bad flows (under
OpenSM / Osmtest we use only the good flow ) , althrough not tested , I'm
guiding a verficator (Yoav) to write directly over umad api test that send /
receive SMP / GSI packets .

Liran.
> > 
> > -- Hal
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20050906/77aa2793/attachment.html>


More information about the general mailing list