Re: [PATCH net-next 1/6] ethtool: Extend link modes settings uAPI with lanes

[Johannes Berg]

Hi,

Sorry, somehow didn't see this until now.

> > +/* Lanes, 1, 2, 4 or 8. */
> > +#define ETHTOOL_LANES_1                    1
> > +#define ETHTOOL_LANES_2                    2
> > +#define ETHTOOL_LANES_4                    4
> > +#define ETHTOOL_LANES_8                    8
> 
> Not an extremely useful set of defines, not sure Michal would agree.
> 
> > +#define ETHTOOL_LANES_UNKNOWN              0
> >  struct link_mode_info {
> >     int                             speed;
> > +   int                             lanes;
> 
> why signed?
> 
> >     u8                              duplex;
> >  };
> > @@ -274,16 +277,17 @@ const struct nla_policy ethnl_linkmodes_set_policy[] 
> > = {
> >     [ETHTOOL_A_LINKMODES_SPEED]             = { .type = NLA_U32 },
> >     [ETHTOOL_A_LINKMODES_DUPLEX]            = { .type = NLA_U8 },
> >     [ETHTOOL_A_LINKMODES_MASTER_SLAVE_CFG]  = { .type = NLA_U8 },
> > +   [ETHTOOL_A_LINKMODES_LANES]             = { .type = NLA_U32 },
> 
> NLA_POLICY_VALIDATE_FN(), not sure why the types for this
> validation_type are limited.. Johannes, just an abundance of caution?

So let me see if I got this right - you're saying you'd like to use
NLA_POLICY_VALIDATE_FN() for an NLA_U32, to validate against the _LANES
being 1, 2, 4 or 8?

First of all, you _can_, no? I mean, it's limited by

#define NLA_ENSURE_NO_VALIDATION_PTR(tp)                \
        (__NLA_ENSURE(tp != NLA_BITFIELD32 &&           \
                      tp != NLA_REJECT &&               \
                      tp != NLA_NESTED &&               \
                      tp != NLA_NESTED_ARRAY) + tp)

and the reason is sort of encoded in that - the types listed here
already use the pointer *regardless of the validation_type*, so you
can't have a pointer to the function in the same union.

But not sure I understood :-)

johannes