Hello,
>
> Can we limit the number of span ports per bridge to a small number so that
> the instead of a heap object for the SLIST a simple stack array of
> MAX_SPAN_PORTS pointers could be used?
>
> Who needs more than a handfull of spanports per veb?
I just to make sure I follow your idea...
>
> > --------8<---------------8<---------------8<------------------8<--------
> > diff --git a/sys/net/if_veb.c b/sys/net/if_veb.c
> > index 2976cc200f1..a02dbac887f 100644
</snip>
> > + struct veb_span_port *sp;
> > + SLIST_HEAD(, veb_span_port) span_list;
so the span_list, will be turned to array of references, right?
> >
> > + SLIST_INIT(&span_list)
> > smr_read_enter();
> > SMR_TAILQ_FOREACH(p, &sc->sc_spans.l_list, p_entry) {
> > ifp0 = p->p_ifp0;
> > if (!ISSET(ifp0->if_flags, IFF_RUNNING))
> > continue;
> >
> > - m = m_dup_pkt(m0, max_linkhdr + ETHER_ALIGN, M_NOWAIT);
> > - if (m == NULL) {
> > - /* XXX count error */
> > - continue;
> > - }
> > + sp = pool_get(&span_port_pool, PR_NOWAIT);
> > + if (sp == NULL)
> > + continue; /* XXX count error */
> >
> > - if_enqueue(ifp0, m); /* XXX count error */
> > + veb_eb_brport_take(p);
> > + sp->sp_port = p;
> > + SLIST_INSERT_HEAD(&span_list, sp, sp_entry);
here we do instead of SLIST_INSERT_HEAD() something like:
span_list[i++] = p;
> > }
> > smr_read_leave();
> > +
> > + while (!SLIST_EMPTY(&span_list)) {
and here instead of walking list we just walk through array.
is that somewhat correct? If so than I need some 'qualified' suggestion
for upper limit of span ports. I'm afraid my real life experience
with network switches is zero, so I don't feel like a right person.
another idea I'm toying with is to add a 'service task' to every switch
so the underlying `vport_if_enqueue()` will put packet to queue and
let task to dispatch to ifp associated with vport. Once packet
will be dispatched the task will drop the reference to port.
I had no time to figure out how to wire things up for task.
I might give it a try tonight.
thanks and
regards
sashan
