[openib-general] SRP device management client (and a few opensmglitches)
Hal Rosenstock
halr at voltaire.com
Thu Nov 17 22:23:42 PST 2005
On Thu, 2005-11-17 at 23:50, Roland Dreier wrote:
> Hal> It's not LocalPortNum. I don't see any easy way to query all
> Hal> port 2s in the network as for HCAs and routers the PortNum
> Hal> component is reserved (and you need to supply port
> Hal> LID). Doing that would only get you all the switch port
> Hal> 2s. Perhaps the spec should be amended if this is useful. If
> Hal> the spec were amended like this, you would get all the port
> Hal> 2s in the network and would then need to determine whether
> Hal> they were on a switch on xCA. Do you think this is useful ?
> Hal> Shall I file a comment on this ? We need to move fast to get
> Hal> this into 1.2 errata.
>
> I think there are two comments that I would file against the spec:
>
> - LocalPortNum for SA PortInfoRecord queries is not well defined.
Rather than not well defined, it's not very useful as an SA search field
> It
> doesn't make sense to me that it should be whichever port the SM
> happened to discover a switch through, since that could change
> essentially at random if a switch is multiply connected to the fabric.
I think that's why you would want to use PortNum in the RID rather than
this. However, this is a reserved field for HCAs (and routers). Hence,
my original comment to make this useful as a search field.
> - It would be nice to be able to do an SA GetTable() query to get all
> PortInfo records with the IsDeviceManagement bit set in the
> capability mask (that's what I'm really after in this case). As it
> stands now, all that it is possible to do is to set the component
> mask so that all ports with an exact match for the full capability
> mask field are returned, and that's pretty useless.
Yes, that's a better way to get this and would be a perfect query for
this.
I will file both of these tomorrow. If there are others you know about
that would be useful, now's a good time to discuss them with Ted and/or
me.
-- Hal
More information about the general
mailing list