On 4/2/06, Tziporet Koren <[EMAIL PROTECTED]> wrote: > Leonid Arsh wrote: > > Of course, the FW doesn't handle the event, it handles the request by > > sending the port info to the SM. > > It also generates an appropriate event to SW. The actual > > re-registration is to be done by the SW in ULPs. > > An alternative way to generate the event is catching the request MAD > > in smp_snoop, as you suggested, but the port info still will be sent > > by the FW. > > > I actually prefer that the FW will generate the ClientReregister event > since its already generating this event and in this way we can save > logic in the driver.
I have to weigh in on the side of SW here - there's already logic to handle other events in SW, so supporting client reregister in SW is trivial. Further, the Windows stack handles this in SW, and I don't see any reason why that should change since it's already in place and working. It would be nice to have some similarities between the stacks, especially for cases where the FW would affect the behavior of the HCA. We don't want a FW for Linux and a different one for Windows. - Fab _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
