09/05/2018 20:07, Ferruh Yigit: > On 5/9/2018 10:43 AM, Thomas Monjalon wrote: > > @@ -3386,6 +3383,8 @@ rte_eth_dev_probing_finish(struct rte_eth_dev *dev) > > if (dev == NULL) > > return; > > > > + _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_NEW, NULL); > > + > > Technically we can have as many ethdev created as we want in probe() right?
Yes probing can create several ports. > Doesn't have to be a one to one mapping there, having user event in > rte_eth_dev_allocate() guaranties each ethdev created sends the event. Allocation is too early to notify a new port. We need to wait it is initialized before using it. > But when you moved this into probe() now one event sent for event, same > comment > for previous one, I don't think it is good idea to tie ethdev allocation with > probe() The PMD sends one event per port by calling the appropriate ethdev function. Event and allocation are not tied. I don't see the issue. Note the definition of this event has always been about probing, not allocation: RTE_ETH_EVENT_NEW, /**< port is probed */