On Wed, Sep 26, 2018 at 08:34:44PM +0300, Leon Romanovsky wrote:
> > You showed me IFLA_INFO_KIND, which was introduce to filter link dumps.
> > dc599f76c22b ("net: Add support for filtering link dump by master device and kind")
> >
> > It is not what we are doing here.
>
> More on that, in netdev, they are not creating any KABI dependencies on
> the input provided by IFLA_INFO_KIND, however, you are suggesting that
> rdmatool will be forever bounded to the kernel by specific names rxe, siw
> e.t.c.
What are you talking about? netdev uses strings.
When you do
ip link add link ib0 ipoib
It passes string 'ipoib' in a IFLA_INFO_LIND attribute of a
RTM_NEWLINK message. The kernel calls rtnl_link_ops_get("ipoib") and
matches it to
static struct rtnl_link_ops ipoib_link_ops __read_mostly = {
.kind = "ipoib",
And that kind name string is forever part of the kernel ABI and part
of the netlink command interface.
They don't use for this numbers, that is just pointless extra work to
marshall.
Jason
