Roland Dreier wrote:
As long as you can tell me that a thread blocked on get_cq_event will wakeup on a device close, signal, or CQ error then I don't need a wakeup call from userspace.> Here is my TODO list that I need some feedback on....
> - resize_cq
A fair bit of work, probably won't get done too soon.
> - query_device > - ib_query_gid
Both easy, I'll add them shortly.
> - ibv_get_cq_event(), need timed event call and wakeup
Can you explain what this means a little more? Is there something you need that you can't get by using select()/poll() with a timeout on the CQ event FD?
and yes, select() was exactly my thinking, but I was hoping we could get it added to the ibv_get_cq_event() code and just include a new timeout parameter (in usecs) with the call.
-arlin
> - current implementation supports one event per device, plans for more?
Yes, in the medium term I plan to add support for multiple MSI-X vectors so that multiple CQ events are possible.
> - memory window support
Right now I don't have any plans to implement this. All the feedback I've seen is that with current hardware, performance is not good enough to make MWs worth using.
- R.
_______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
