Data ports and maximum number of bearers

[Bjørn Mork]

Carlo Lobrano <c.lobrano at gmail.com> writes:

> Hi Aleksander,
>
> that's interesting, even if I am not sure about what/where to look for more
> information.
> I was looking into network subinterfaces, starting from this discussion on
> the mailing list
>
>     [PATCH 1/2] broadband-modem-mbim: allow "max_sessions" bearers
> <https://lists.freedesktop.org/archives/modemmanager-devel/2014-June/001187.html>
>
> is this the correct direction?

Thanks. I had totally forgotten about that. Thought I hadn't even
started. Now my conscience is much better :)

FWIW, I still believe we should add proper support for up to
"max_sessions" bearers on MBIM devices.  But there is quite a bit of
remaining work to do, and those patches are nothing more than a
proof-of-concept.

I guess the first issue is to decide who is responsible for allocating
and deallocating data ports for devices with a dynamic set of data
ports, like MBIM.  I do not think it is necessarily wise to preallocate
all "max_sessions" data ports.  That would just create a large number of
unused network interfaces for most users.  So it would be nice if we
could create the data ports automatically when needed.  But it is not
clear if ModemManager is the right place for this, or libmbim, or
somewhere else?

And then there is the issue of IP session 0.  The driver implementation
is a bit ugly for compatibility reasons, allowing the shared transport
interface to be used as an IP interface for session 0.  But it would be
best to do the dynamic data port allocation for session 0 too, mapping
it to VLAN 4094.  The reason is that this will decouple the MTU of the
underlying transport IP session 0.



Bjørn