"Michael S. Tsirkin" <[EMAIL PROTECTED]> wrote on 04/25/2006 08:23:12 AM:
> Quoting r. Shirley Ma <[EMAIL PROTECTED]>:
> > The more light weigh in hardware interrupts, the better.
>
> But we don't know whether the ULP will be doing lightweigh things incompletion
> handler or not.
>
> So it seems better to report completion events in IRQ context from low level
> driver and leave the decision on whether to use tasklet, work queue or polling
> like NAPI to the ULPs. No?
>
> --
> MST
Yes. That's I am working on right now. The first patch will use work queue in IPoIB
layer not driver layer to see the perfomance number with both send/recv polling
handler simultanously on mthca.
Thanks
Shirley Ma
IBM Linux Technology Center
15300 SW Koll Parkway
Beaverton, OR 97006-6063
Phone(Fax): (503) 578-7638
_______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
