It seems libatlas-base-dev provided an `enum CBLAS_TRANSPOSE`, but we
don't install it anymore. Now, in Ubuntu, there is only a typedef'd
version in cblas.h; for RHEL(-derivatives), the same typedef'd version
is in cblas/cblas.h.
Ubuntu 18.04 is now in "Extended Security Maintenance", but otherwise
gets no updates. Remove it from supported distributions (regarding
installing packages); users can of course still use it.
Users are encouraged to upgrade to Ubuntu 20+.
Apart from the obvious change of adding ubuntu24.04 to supported
distributions, we need to remove liblapacke-dev as it is incompatible
with libatlas3-base. Installing both aborts with:
The following packages have unmet dependencies:
liblapacke : Breaks: libatlas3-base (< 3.10.3-14) but 3.10.3-13ubuntu1 is to be installed
It seems that we don't need it though, as liblapacke-dev installs the
right atlas library as a dependency.
Avoid UBSan error:
openair1/PHY/NR_ESTIMATION/nr_measurements_gNB.c:183:22: runtime error: signed integer overflow: 2122747784 + 27403601 cannot be represented in type 'int'
avoid UBsan error:
openair1/PHY/NR_TRANSPORT/pucch_rx.c:266:21: runtime error: signed integer overflow: 2113190528 + 117251752 cannot be represented in type 'int'
avoid UBsan error:
openair1/PHY/NR_TRANSPORT/nr_ulsch_demodulation.c:459:77: runtime error: signed integer overflo w: 2009540296 + 668087372 cannot be represented in type 'int'
xran's xran_bm_init() uses rte_pktmbuf_pool_create() to allocate DPDK
buffers. For a 1x1 configuration, the number of buffers might be too
small, and rte_pktmbuf_pool_create() might fail. This commit ensures
that we increase the number of buffers to have enough even in a 1x1
configuration. This works by chosing the next power of two, which is
recommended as per DPDK documentation.
Closes-Bug: #750
With the capabilities now given as part of this MR, it happens that when
checking the "docker compose config", the container_name might not be
close to the service name. Something like
service:
cap_drop:
...
cap_add:
...
container_name:
This is inconvenient, as the previous grep for the service name + 3
lines might not cover the service name. To circumvent this, give the
service name directly in "docker compose config". (no clue why it has
not been done like this from the start)
- sa_b200_gnb
- sa_aw2s_gnb
- sa_e1_b200
- sa_f1_b200
- sa_sc_b200_gnb
- nsa_b200_gnb
By default, the container has no access to /dev. Mount all USB devices
into the container to ease usability: by default, USB devices are
mounted under /dev/bus/usb using the bus ID, which is different on every
system and might change on reboots or with (un-)plugging USB devices.
Hence, mount all USB; the process has not many capabilities anyway, and
thus might not do much harm.
A proper solution would be to write a udev rule to mount a device using
its vendor and product ID, and only mount the corresponding device into
the container.
- SA 2x2 tests on 60 and 100 MHz
- N310-based OAI gNB + UE
Give SYS_NICE and IPC_LOCK for performance. On the UE side, give
NET_ADMIN (for interface bringup) and NET_RAW (for ping an TUN).
It might be desirable to run the UE without superuser rights. More
concretely, it might run without NET_ADMIN. In this case, various
corresponding operations on the TUN socket fd might fail. In this
commit,
- improve error reports to make them more clear
- do not automatically exit(1), but return from the function to ensure
continuity of the softmodem
Use common threadCreate() function to create P7 thread with high(er)
priority. Remove old calls to set_priority(), as this is already done
with threadCreate().
Use common threadCreate() function to create P7 thread with high(er)
priority. Remove old calls to set_priority(), as this is already done
with threadCreate().
Use common threadCreate() function to create P7 thread with high(er)
priority. Remove old calls to set_priority(), as this is already done
with threadCreate().
The main thread of nr-uesoftmodem (executing main()) is mostly working;
it is thus not necessary to set high priority. Instead, the threads
created with threadCreate() will receive higher priority, if possible
(as determined by the availability of SYS_NICE capability).
To inform if SYS_NICE is not present, print a warning in main().
SYS_NICE is a capability that allows a process to set thread affinity
and priority, among other things (see capabilities(7) for more info).
In this commit, add a function that allows to determine if the process
has this capability, and try to change the thread affinity and priority,
if requested. If the capability does not exist, the function will simply
not attempt to change the corresponding thread attributes.
To determine if the process has SYS_NICE, libcap can be used. However,
it might not be installed by default. To avoid requiring another
dependency, if we detect that libcap is not present, use a workaround by
try to set a real-time scheduling policy; if it's present, or can be
changed, we assume that the process has SYS_NICE (and clean up, if
relevant).
Simplify reading of capabilities
set_latency_target() attempts to set keep "low latency" by
- writing a specific latency to /dev/cpu_dma_latency
- setting manually the minimum CPU processor frequency to be the maximum
There is no functionality to undo this after stopping the softmodem(s),
and most users are probably not even aware that OAI does this. It is
generally preferable to set this beforehand using a governor or by
disabling sleep states (as outlined in the tutorial, and in a follow-up
commit in the performance tuning docs).
The previous mlockall() call, to lock memory to RAM, is retained in a
new function. There were additional mlockall() calls, which have been
replaced with lock_memory_to_ram(), where necessary.
Print a clear error if a config file does not exist (unclear previously,
when the user would get a confusing, unspecific, "file I/O error" on
line 0)
I tried to free the memory that had been allocated at that point, and
verified that using the address sanitizer. Nevertheless, the function
does too much, and is too complicated for refactoring.
Use correct baseTag for Unit Tests build in execution:
- if build_helper or othe files changed, which would trigger rebuild of
ran-base, use the corresponding ci-temp tag
- use that tag consistently afterwards
38.321 6.1.3.1 says: "The size of the RLC headers and MAC subheaders are
not considered in the buffer size computation" [for BSR]. However, prior
to this commit, we take the size signaled of the BSR and schedule the UE
with that amount of data, making it impossible for the UE to transmit
all its data at once (because it is the size signaled + headers).
In this commit, over-estimate the buffer status size. As we cannot know
the exact number and size of headers, we simply use "BSR size + n", as
done in function overestim_bsr_index().
It can happen that in the current DL slot, we cannot send an allocation
for Msg3 in a later UL slot. Up to this commit, MAC printed a warning.
However, this is not really a warning, as it will likely work in the
next or a later slot. So demote this to debug.
In some pipelines, it happens that iperf3 is hanging on 100% CPU load.
In these cases, a simple TERM signal does not seem to be enough to kill
it; KILL is necessary.
While e.g., RemoteCmd's driver paramiko has a timeout command, the
documentation [1] is not really clear how it is trying to stop commands.
Since it effectively does not stop, we have to assume it is only sending
TERM.
To remedy this, use the timeout command to start iperf3. We use option
-v for diagnose timeouts on stderr, and send KILL after an additional 3
seconds (-k3) if TERM does not succeed. The timeout is sent to the same
duration as the overall iperf timeout.
[1] https://docs.paramiko.org/en/3.4/api/channel.html#paramiko.channel.Channel.settimeout
A to-be-added UE's RA is added through nr_mac_add_test_ue() (the name is
misleading, it is not a "test UE") through its CellGroupConfig. Using
the process_CellGroupConfig() is unfortunate, because it is not really
clear what part of the CellGroupConfig gets evaluated (it is certainly
not all). Move the code for adding a new RA directly in the user-add
function, and refactor this one.
Introduce new LC configuration "priority", as specified in 38.321, and
use to sort LCs by priority. We set the priority of an SRB according to
38.331 Section 9.2.1, and for a DRB to be the minimum priority of the
QoS flows associated to this DRB.
Introduce new functions to add/release LCs without needing to go through
the intermediate CellGroupConfig. This harmonizes LC handling at the "DU
handler" to keep it consistent with RLC bearer handling.
These functions allow to group LC configuration handling, and we can
remove functions for setting NSSAIs and QoS by handling in one central
place (while adding LCs).
nr_mac_add_test_ue() is used by simulators and also in NSA mode to add a
new UE. Ideally, these users should also call the ue context setup
request handler to add a UE context; for the moment, respect the passed
list of LCs to add. Similarly, for do-ra, we require a dedicated
function to still parse the CellGroupConfig, so we keep that part of
functionality.
On reestablishment, since we do not process the CellGroupConfig from the
old UE, we need to add LCs for the new UE based on the onces from old
UE.
Use a common structure nr_lc_config_t to group LC configuration for one
logical channel. Use seq_arr for simple lookup, adding, removing of
elements without manual indexing logic.
This commit retains the previous logic of adding/releasing LCs through
the process_CellGroupConfig(), although we know while building the
CellGroupConfig if we add/release LCs. The next commit fixes this.
If we do not make this change, with the next commit, we get the
following error from gcc:
openair2/E2AP/RAN_FUNCTION/O-RAN/ran_func_kpm.c:299:46: error: assignment of read-only location '*(arr_ue_id.ue_info_list + (sizetype)(arr_ue_id.sz * 2444864))'
| arr_ue_id.ue_info_list[arr_ue_id.sz] = *ue;
In a later commit, we will use find_if in L2. In the case of building
with E2 agent, we link e2_agent which comes with its own implementation
of find_if_arr(), which results in a name clash. Rename the OAI
implementation, as nothing uses this, anyway.
This reverts commit 0d0e111135. It was
set to UE RX to TX time to 2 as it was deemed that the UE could sustain
this time. However, we noticed a degradation of performance, so reset
the RX to TX time to something higher to ensure proper performance.
Robert Schmidt