On Thu, Jun 09, 2016 at 04:17:15PM +0300, Moni Shoua wrote:
2. Trying to force SoftRoCE on this model will end up with rxe driver
registers to rvt with all ib_device hooks implemented. This makes the
reason to use rvt irrelevant in this case.
Everything? What about AH, MR, PD? Aren't those pretty generic constructs.
In fact you folks submitted the change for AH.
3. Modifying rvt model to make it more suitable for a generic lower
driver model can have a big effect on QIB/HF1 performance. This
probably will raise objections from QIB/HF1 owners. In that context I
want to say that attempts to influence rvt model months ago, when rvt
was still an idea, were rejected with this exact reason (being bad for
QIB/HF1 performance).
Yes, performance trumps, that's the nature of the game.
I'm not saying that you have to use rdmavt, however I believe it could be
possible. Also I want to make it clear that we are not against changes in
rdmavt to further its use in other drivers. In fact quite the contrary.
Changes are welcomed, of course with the caveat that it does not harm
performance.
-Denny
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html