> > This just emphasizes why we need to converge to a single method. > > In my opinion, we already have it. > > For local registrations, ib_reg_phys_mr()/ib_get_dma_mr(). These are not > quite equivalent, btw. Personally, I would work to eliminate local registration as part of the API. > For remote registrations, ib_post_send(FRMR). I slightly agree here. IMO, the rkey should be obtained by associating the memory buffer with the QP, but not explicitly as a 'send' operation. If queuing is a concern, we could expose max_active_registrations/max_total_registrations QP attributes. I would hide the protection domain concept entirely. If needed, a provider can allocate one PD per QP, with memory registration going to the PD. - Sean -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html