Roland Dreier wrote:
I think this is exactly the right plan. Maybe the names could be a little better -- no need to use odd terminology like "psp." But other than that sort of minor quibble, this is the same as my vague ideas.
It is my hope to help with the implementation of some sort of CM abstraction.
Also on the IB side the AT code probably needs to be reviewed and improved. The API should be simpler, and I don't like the way AT sticks its tentacles into the IPoIB driver and network stack.
I agree with the comments on the API, but haven't looked at the implementation in enough detail to say anything about its operation.
Somewhat on a side note, AT appears to be the only way to connect from userspace without hard-coding values or interfacing directly with the MAD layer itself. Do we want to expose SA query to userspace or rely on AT? I'm leaning towards the latter at the moment.
- Sean _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
