Closes#685 and #698
See merge request oai/openairinterface5g!2425
* !2404 NR UE MCS for retransmission v2
* !2420 Scan build: Argument with 'nonnull' attribute passed null
* !2422 Fix AssertFatal when using NR band n1
* !2418 NR UE resynchronization fix
* !2376 Save NSSAI params of PDU session coming from AMF
* add `build_oai` option for thread sanitizer
* update 60 MHz 2x2 phytest-timing threshold
* !2423 Free F1 UE data struct only on UE release
* !2428 Prevent CU-UP UE ID management if integrated CU
* !2390 NR PTRS improvements
* !2291 Pass config object instead of using implicit global
* !2421 Change UL_TTI.request PRACH PDU beamforming pack/unpack handling
* !2406 CI:AW2S - test with 10 AmariUEs
* !2401 ue-rrc-coarse-cleaning
* !2366 Remove F1 BWP switching logic
* !2426 F1: support multiple DUs per CU
* !2432 Remove cblas dependency for 5G targets
Multiple DUs host cells, whose cell ID cannot match (to identify them
uniquely). Hence, don't check for equality with CU, because otherwise,
we can never connect multiple DUs.
This is a workaround for a problem described in #706. In short, we
cannot reuse the cu_add/remove_f1_ue_data() functions if we are in an
integrated CU-UP/CP ("CU"), but then we can also not reliably use
different UE IDs in CU-UP and CU-CP (except if using even more hacks).
An assoc_id of 0 is used to mark UEs whose DU is "offline" (not
connected). In this patch, if a DU disconnects, we go through all UEs
and set their assoc_id to 0 if they belong to a UE that just went
offline.
This code printed BW masks (what is this) and number of MIMO layers. It
was dependent on the current DU, and only showed the information for the
current DU config (SCS, band). On DU disconnect, it leads to segfaults
to this dependency, and IMO is not useful in itself (it could rather be
printed when receiving the UE capabilities, for all bands etc.).
To allow to send F1 messages to multiple DUs from the CU, we need to
send these messages via different assoc_id (SCTP association ID,
basically akin to a file descriptor). Currently, the assoc_id is stored
in a context at the F1 module and retrieved in
f1ap_itti_send_sctp_data_req().
To facilitate a later change towards indicating (from the RRC) to which
assoc_id (hence, DU) to send a message, we refactor all
CU_send_*()/DU_send_*() functions to take the assoc_id as a parameter to
which to send messages. In this commit, we then pass the assoc_id from
the context. A later commit changes this (at the CU) to pass the
assoc_id from the message to be sent.
Previous to this commit, the CU creates an F1 instance and initializes
GTP after a successful connection establishment, and tears the F1
instance down after the DU disconnects. This does not work when trying
to connect multiple DUs, or have them reconnect, as e.g., GTP
re-initialization fails (socket already bound).
This commit makes the CU F1 instance creation and GTP init at CU task
init (only once), and also frees the F1 instance only after the CU task
exits, which is at termination of the application, typically.
Remove dead code, result of 2136ceec0d: we
cannot connect() in GTP on multiple remotes (as then only packets of one
host might be received), but might need to send to multiple remote
sockets.
The DU UE capability printing depends on data on SCS and band, which
currently is only stored with the DU. To avoid any assert, we check
that the DU is present. A follow-up commit will rework this.
Before this commit, as visible in the handler of an F1 UE release, it
can happen that we forward the RRC Release message to a UE and start the
release timer, while also removing the F1 UE data (containing e.g., the
CU UE ID). This can lead to asserts, since there might still be
subsequent traffic for such UE during a short time, while the DU does
not have the secondary UE ID stored anymore.
This commit changes the logic. It introduces function
nr_mac_release_ue() that removes the UE at MAC and RLC, frees the F1 UE
data (if applicable), and sends the complete, after timing expiry. If
the UE is out of sync, this function is used to free the UE immediately.
This fixes#685 and #698.
integration_2023_w43
See merge request oai/openairinterface5g!2415
* !2407 hotfix: UE DCI format 11: only 1 dl_config instead of 2
* !2372 chore(ci): testing up to 10 UEs in 5G RF simulation
* !2396 Add optimized function to count the number of bits set in uint64_t variable
* !2383 E1: implement release bearer, handle multiple CU-UPs
* !2413 Assigned value is garbage or undefined reported by scan-build
* !2387 Small fixes for UE stability and introduction of ITTI "no-thread" mode (separate bug fixes and ITTI changes?)
* !2410 Fix CellGroupConfig-related memory problems
* !2414 fix doxygen
- Set the correct QoS flow ID to associate a correct QoS flow to the
bearer: the UE would misassociate otherwise
- Set the DRB as the default: the handler applies bool with true==1, so
use bool value
Before this commit, the DMRS_UplinkConfig was stored in a separate
DMRS-related structure, although it is always referenced in the
PUSCHConfig of the current UL BWP.
Through certain code paths (e.g., retransmission), it could happen that
an old DMRS_UplinkConfig was accessed that was freed (through new
CellGroupConfig).
In this commit, we always look up the DMRS_UplinkConfig in the current
PUSCH config, which should always be up to date.
Until this commit, the SIB1's PDCCH_configCommon was set (via a pointer,
i.e., an indirection) to the one of the ServingCellConfigCommon (SCC).
Afterwards, the SIB1 code further populated this PDCCH configCommon,
indirectly populating the one of the the SCC, on which code later
depends.
Not only did this create possibilities for double-frees (since freeing
the memory of SIB1, then SCC would free the PDCCH configcommon twice),
but also it makes it harder to track where certain structures are
populated. Hence, this commit solves both issues:
- The SCC is correctly populated on initialization of the SCC
- The SIB1 makes a deep copy of the SCC' PDCCH configCommon
This resolves the two issues above.
This allows to communicate with multiple remote hosts. A further
improvement would be to refactor this piece of code into a function and
call it only when necessary, i.e., when we know the remote host, or want
to add one to a list of hosts.
This commit introduces the capability to handle multiple CU-UPs. It uses
a RB tree and puts the CU-UPs into the tree. Upon the connection, it
associates UEs to CU-UPs in a round-robin fashion.
A later commit will introduce an RB tree to manage multiple CU-UPs. The
RB tree implementation relies on some macros that generate RB tree
functions. Functions using the RB tree implementation will be grouped in
this file.
The E1AP Setup Request contained the network configuration (IP address,
ports) as well as the actual E1AP Setup Request application data
(Supported PLMNs, ...). This has the drawbacks that
- The E1AP Setup Request is stored to retrieve IP addresses in the E1AP
module, which is confusing as the Setup Request, per standard, has no IP
info
- The CU-CP received an E1 Setup Request for configuration during start
up, but it did not actually receive such Setup Request, but merely the
IP configuration to set up the socket
This commit splits the E1AP Setup Request into a "real" Setup Request
for application data, and creates a new type e1ap_net_config_t to group
all IP configuration data. Further, a new ITTI message type
E1AP_REGISTER_REQ is introduced to group both types. What happens is
- RCconfig_NR_CU_E1() reads both E1AP application-level data and IP
configuration, as previously
- The data is sent to the CU-CP. It discards the E1AP Setup Request
data, and only uses the network configuration to set up the socket
- The data is sent to the CU-UP. It uses the network configuration to
connect to the CU-CP, and then sends the E1AP Setup Request to the
CU-CP.
Currently, the CU-CP still stores the Setup Request locally, which will
be changed in the next commit to send it to the RRC.
Robert Schmidt