Re: [PATCH net-next 1/3] net: lan966x: Add function lan966x_mac_cpu_copy()

[Vladimir Oltean <vladimir.oltean@nxp.com>]

On Mon, Jan 03, 2022 at 04:39:10PM +0100, Horatiu Vultur wrote:
> > > +int lan966x_mac_learn(struct lan966x *lan966x, int port,
> > > +                   const unsigned char mac[ETH_ALEN],
> > > +                   unsigned int vid,
> > > +                   enum macaccess_entry_type type)
> > > +{
> > > +     return lan966x_mac_learn_impl(lan966x, port, false, mac, vid, type);
> > 
> > If you call lan966x_mac_cpu_copy() on an address and then
> > lan966x_mac_learn() on the same address but on an external port, how
> > does that work (why doesn't the "false" here overwrite the cpu_copy in
> > the previous command, breaking the copy-to-CPU feature)?
> 
> Then you will overwrite the cpu_copy so the frames will not reach the
> CPU anymore.
> But you should not do that. The function lan966x_mac_cpu_copy() should be
> used for IPv4/IPv6 and lan966x_mac_learn() for the other types.
> 
> Maybe the function lan966x_mac_cpu_copy() is too generic. It should be
> something like lan966x_mac_ipv4(), lan966x_mac_ipv6() and these functions
> will call __lan966x_mac_learn with the correct parameters. Also I can
> add a WARN_ON(...) inside lan966x_mac_learn not to be used with the
> IPv4/IPv6 types.

The intended usage pattern isn't clear at all in the current series.