> > > +static int airoha_alloc_gdm_port(struct airoha_eth *eth, struct device_node *np)
> > > +{
> >
> > > + port = netdev_priv(dev);
> > > + mutex_init(&port->stats.mutex);
> > > + port->dev = dev;
> > > + port->eth = eth;
> > > + port->id = id;
> > > +
> > > + err = register_netdev(dev);
> > > + if (err)
> > > + return err;
> > > +
> > > + eth->ports[index] = port;
> >
> > eth->ports[index] appears to be used in
> > airoha_qdma_rx_process(). There is a small race condition here, since
> > the interface could be in use before register_netdev() returns,
> > e.g. NFS root. It would be better to do the assignment before
> > registering the interface.
>
> actually I check eth->ports[] is not NULL before accessing it in
> airoha_qdma_rx_process():
>
> p = airoha_qdma_get_gdm_port(eth, desc);
> if (p < 0 || !eth->ports[p]) {
> ...
> }
Yes, you check it is not NULL, so you don't de-reference anything. But
i did not spend enough time to check if you leek something as a result
of it being NULL?
> Moreover, in airoha_alloc_gdm_port(), I set eth->ports[index] pointer just if
> register_netdev() is successful in order to avoid to call unregister_netdev()
> on an not-registered net_device in the airoha_probe() error path. I guess we can
> even check reg_state for this:
>
> for (i = 0; i < ARRAY_SIZE(eth->ports); i++) {
> ...
> if (dev->reg_state == NETREG_REGISTERED)
> unregister_netdev(dev);
> }
I prefer checking reg_state. Problems with this race condition are
hard to track down, so it is better to not have it in the first place.
Andrew
