This was causing oaisim to fail way too often at the startup
of a connection.

Things should go better with this patch.

It may not be the end of the story, the use of retrans_num_bytes_to_retransmit
is not clear to me. The problem I see is: what if the max number of
retransimission has been reached (if that has a meaning at all in RLC AM)?
will this value be decreased or not?

The code was changing the rnti in the red-black tree without
removing/reinserting the context. This is wrong because the
rnti is a key of the red-black tree and you cannot change the
key in a stored item at will.

There are also modifications of the function mac_eNB_rrc_ul_failure
which has to remove the UE in any case.

And rrc_mac_remove_ue has been changed for the case where
the UE is not found.

iqlog.c should change its name, it's too specific.
This new one is more generic and should take over the name.

It creates more problems than it solves.

We replace this approach by a straightforward
synchronous one.

Now in the downlink, at frame f and subframe s
we use HARQ process (f * 10 + s) % 8.
This is similar to uplink scheduling.