The timer code used to trigger periodic statistics message sending used
ITTI. Together with a high frequency reporting (e.g. 1ms), this resulted
in high CPU usage. This commit reimplements this:
* using a thread that is synchronized to the RU (if there is no MAC, we
assume there is no RU generating a tick so we create the tick source
ourselves)
* no RB trees but plain arrays
* smaller interface for users
* slight user interface changes
* remove stopped timer state: such timer might be simply removed
* remove one-shot timers
* only set up timer for sending requests
* simplify interface for timer setup
* on timer expiry, do not pass through stats request function but
directly send the reply
* Remove whatever report structure and use stats request directly for
inferring the data to send
* Instead of partly initializing in read_config_and_init(), everything of
PDCP is initialized in main() of lte-softmodem
* Omit RC.rrc access in RRU through checking of RC.nb_inst > 0
* Includes FlexRAN, i.e. do not start FlexRAN in RRU
* Function pointers to PDCP functions have to be set explicitly (avoid
confusion of multiply setting function pointers)
This resolves the following warning:
"error: invalid suffix on literal; C++11 requires a space between literal and
identifier"
Since this file might be included from a C++11 file. It only adds spaces.
Although not used in the UE, we currently need to link the F1 library into it.
This is because we do not define ASN_DISABLE_OER_SUPPORT (otherwise, F1 has
compilation problems). In this case, some functions are apparently not
correctly defined, giving linker errors in the UE. Linking F1 in solves the
problem.
This is a hack. It would be better to be able to define ASN_DISABLE_OER_SUPPORT
and have no compilation errors or compile everything with
ASN_DISABLE_OER_SUPPORT except F1. I did not do this because I did not find
another way than add target_compile_definitions() to every target except F1
which adds many lines just so that one target (F1 ASN.1) compiles correctly (in
other words, F1 needs to be fixed).
link_send_packet() has been extended to use for multiple protocols (added UDP,
SCTP after TCP). The test was not update, though, which is fixed in this
commit.
This changes the destroy_link_manager() function to:
* call a new function message_get_unlock() to unlock the sending thread from
its blocking message_get()
* calls pthread_cancel() on the receiving thread, because this unlocks any
blocking read from a file descriptor
The same message_get_unlock() function has been added ringbuffer_queue to keep
the interfaces the same (it does nothing, because ringbuffer_queue is
non-blocking).
This commits isolates the S1 UE Ctxt Rel Complete and GTP Tunnel Delete
Requests in functions and calls these function in the original place
(rrc_rx_tx()). The same functions are then used in the F1 UE Ctxt Rel
Complete handler to perform the same functionality, i.e. forward the UE
Ctxt Rel Complete message to the MME and free outstanding resources (RRC
context, GTP tunnels, S1 context).
When a phone loses connection, it should not simply be removed but first
CU should be informed which will send a ContextRelCmd. Correspondingly,
when the RRC releases the UE, it does not send a UE Context Release Req
sctp_create_new_listener() takes a parameter server_type which is NOT
the IP version but whether it is an SCTP multihoming server. There was a
bug that when multihoming is not wanted, it created an IPv6 socket
instead of IPv4. This commit fixes this.
IPv6 is not supported yet.
* MAC differentiates between DU/eNB
* sends ITTI message to DU task for F1AP_UE_CONTEXT_RELEASE_REQ
* message is dispatched to correctly in DU_TASK
* sender correctly fills IDs and cause and sends
* RRC and MAC might sort users differently
* therefore, we have to "reverse lookup" the MAC ID from the RNTI when filling
user info
* For this purpose, add RAN API function flexran_get_mac_ue_id_rnti()
Robert Schmidt