- move following radios: AW2S, BladeRF, IRIS, LMSSDR, RFsimulator, USRP
- move following transports: Ethernet, benetel4g, benetel5g
- for each: create independent cmake variable to switch compilation
on/off, in order: OAI_AW2S, OAI_BLADERF, OAI_IRIS, OAI_LMSSDR,
OAI_SIMU, OAI_USRP, OAI_ETHERNET, OAI_BENETEL4G, OAI_BENETEL5G
- Remove RF_BOARD variable, allows to compile multiple boards at once
- where applicable: remove USERSPACE/LIB in path, i.e., move files up to
directories, as all are userspace libraries
Note that with the above, it is now mandatory to enable a radio
implementation before compiling it, e.g., for USRP:
./build_oai -w USRP
cmake .. -DOAI_USRP=ON && ninja oai_usrpdevif
Including sources in targets makes that file being compiled multiple
times. Instead, define an OBJECT library (i.e., it is only compiled, but
not linked), and link it into each target that uses it.
Instead of using the adhoc code for checking and including (globally!)
BLAS/LAPACKE, use PkgConfig to correctly identify libs (see deleted
comment: "Hope this gets fixed in future!" -- you are welcome). Next,
only link into those libraries that need it (PHY_UE, SIMU). Note that
only RHEL also needs cblas, so it is a non-REQUIREd dependency that is
only linked in if available. Further, take out linking into all other
modules, as they don't need it.
In the source code, take out preprocessor directives to correctly
include headers -- that is cmake's job, which it does just fine now.
First, we do not include openssl globally using `include_directories()`,
as most modules don't need it. Further, it is linked into SECURITY, and
removed from all other targets (they don't need it, only SECURITY).
Finally, remove a header include in NR PDCP that is not needed
- LINUX: option to optionally set a sighandler for certain simulators,
no practical relevance
- LINUX_LIST: not used anywhere
- DEBUG_DLSCH: referenced in the code, but those files have a separate
(commented) line `#define DEBUG_DLSCH`; probably never used
- LINK_ENB_PDCP_TO_IP_DRIVER, LINK_ENB_PDCP_TO_GTPV1U: commented in
cmake, although they are referenced. It is likely that code behaves
unexpected if re-enabled; we therefore remove these options to avoid
future problems
- Remove some more commented code
ENB_MODE is a cmake option to swap the order of include directories.
This has historical reasons; before OAI used cmake, it was very
sensitive to the order of includes. Nowadays, the order does not matter.
Remove this option, and fix the compilation issues that arise while
doing so.
CMake now takes the option ASN1C_EXEC to select an asn1c to use for
compilation. For instance, run with
cmake .. -GNinja -DASN1C_EXEC=/opt/asn1c/bin/asn1c
or
./build_oai --cmake-opt -DASN1C_EXEC=/opt/asn1c/bin/asn1c
to select the alternative asn1c. For instance, this is important for the
CI to allow to build multiple versions of OAI in parallel that might use
different commits of OAI.
Further, this patch introduces functionality to check for the support of
asn1c options that we require, and alerts the user if they are missing,
alongside a message of how to resolve the problem.
We don't work much on asn1c. The version we use has several bugs.
There is an active community out there that fixes bugs and improves
asn1c. Let's use their version.
Previously, in f1ap_decoder.c, we asserted if we did not know a message.
Apart form decoding, we did not do anything in that file.
Additionally, in f1ap_handlers.c, we were also checking for an unknown
message, and printing a warning if a handler is not known.
This commit moves the F1AP message decoding to f1ap_handlers.c, and
deletes f1ap_decoder.c. In summary, we don't assert on an unknown
message, and compile less files. The rest is unchanged.
This commit fixes a bug caused by global variables in the gNB_MAC_INST
structure used to transfer MAC scheduler's results to the PHY layer.
When using several threads in the PHY layer, it may happen that the
MAC scheduler is called a second time by the second PHY thread for the
next slot before the content of the global variables in the gNB_MAC_INST
structure used to store the result of the MAC scheduler for the current
slot have been consumed by the first thread, leading to unpredictable
results (all of them certainly wrong).
The solution is to introduce a small memory management module for the
interface between the PHY threads and the MAC scheduler. The MAC
scheduler now fills the data structures passed to it, and not a
global variable anymore. The PHY layer manages the structures and
ensures that no data is overwritten before being fully consumed by
the various threads of the PHY layer.
Some care has to be taken because several PHY threads will access the
same structures at the same time for a given slot (in read mode, so
it's okay). A mechanism of reference counting has thus been introduced.
Only one CC is supported. This restriction should not be too hard
to remove if/when needed. (The code before this commit does not seem
to work for more than one CC anyway.)
Most targets don't need the symbols from the compiled RRC ASN.1
libraries, but only the headers. In particular when compiling individual
libraries (e.g., for tests), linking the full library makes the RRC
ASN.1 libraries compile and link, when they are not needed (only headers
are needed).
For libraries that can do without these headers, don't link them.
The simulators have been changed: due to linker errors that would arise
otherwise, a lot of functions are defined in them to avoid such errors.
Some functions are also defined in header files; in this commit, we
define most functoins in nr_dummy_functions.c, which is compiled in, not
by including the C file, but by adding it to the target in
CMakeLists.txt
- nr_dlsim+nr_ulsim need NR_IF_Module_init(), so it cannot be in
nr_dummy_functions, but the others need it
- Link correcr NR_IF_Module_init() into these simulators, and reduce
overall link list
- Correctly link remaining simulators
Robert Schmidt