[PATCH libosmocore 0/8] Cell Broadcast (SMSCB) MS support

[Sylvain Munaut]

Hi,


> Inertia would be one reason -- since my first patchset, LAPDm has moved
> from BB into libosmocore, so I just tagged along.
>
> On the practical side, I see two reasons now:
>
> i) if we want to hook it into LAPDm, it needs to be where lapdm.c is

After starting to read the spec (not done yet, so further comments my
come afterwards), I don't think "hooking" LAPDm is the right way.

>From the specs, LAPDm and the SMSCB link layer are not one above the
other but rather completely separate.
And again from the spec, should you expect one or the other on a
channel, you must ignore any packets with the wrong LPD. So AFAIK on a
CBCH channel if you ever get a LPD=00 it should be ignored and not fed
to a LAPDm processor. Same thing for a BTS side LAPDm instance if it
receives a LPD=01 it should be ignored.

>From that I think the best way is to split based on LPD before it's
fed to LAPDm at all. So when you get a L2 packet from L1, instead of
blindly feeding it to LAPDm, you should check a LPD handler table to
know who to feed it to, and for normal channels there would only be
handler for LAPDm and for CBCH channel there would be no LAPDm handler
and only a LPD=01 handler.


> In the future, we can also add the BTS-side implementation.  I don't
> think they can share much code,

If they can't share code, then, I would make sure to mark the method
to be 'ms' side so as to keep the namespace clear.
Also, I would stick to the convention we used for sms and name it
something like gsm412_ms_entity (and same for the methods)


> but it might make sense to have both in the same place.

That's not really the policy we used so far. Only shared code goes to
libosmo-xxx
If it ever comes a time where we need it in another project, it'll
still be time to move it there then.


> I can spin a patch series directly on top of osmocom-bb, but the
> testcases will probably not make it.

Why ?  layer23 is auto-tools based as well, copying the test suite
stuff from libosmocore shouldn't be too hard.


Cheers,

    Sylvain