Using the new libosmovty features allow for:
* Setting different cpu-affinity masks for each thread in the process,
both at startup through .cfg file as well as changing it at runtime.
* Unified VTY interface to change the scheduling policy of the process
inherited by all osmocom processes enabling the feature.
Depends: libosmocore.git Change-Id If76a4bd2cc7b3c7adf5d84790a944d78be70e10a
Depends: osmo-gsm-masnuals.git Change-Id Icd75769ef630c3fa985fc5e2154d5521689cdd3c
Related: SYS#4986
Change-Id: I3798603779b88ea37da03033cf7737a6e4751d6e
Since there's now a rate counter, we can drop log level for those events
which can be bursty and hence print lots of output in short periods of
time, which may affect performance. This way setting them to INFO it's
enough to avoid getting them in stderr unless explicitly configured by
the user (for instance to debug stuff), while still allowing a good
enough level to be enabled for other targets such as gsmtap.
Related: OS#4679
Change-Id: I000f7112e35ac68d3d922444f78468b1ea74cbba
This way i most usual rate_ctr related internal logging is disabled by default
(notice), and it can b eeasily enabled by switching the category to info
or debug.
Change-Id: Id6c36432da7e7ce673c585bcae6772a695028ec5
This ones together with rate counters already available in lower layers
allows to understand better the source of the problem with stalled tx
bursts.
Change-Id: Ia34f7e7d780ad1e12f24638a07f05fe91f2afea5
This allows checking if there's timing issues on the downlink side
between osmo-bts-trx and osmo-trx. This counter is useful to find
information about osmo-bts-trx 'fn-advance' setting, since this counter
basically counts if burstrs from it arrived too late to osmo-trx.
Change-Id: Id6df00da81f6d6884f4dddc5a2c4b354dca3af97
RadioInterface ones will be added in next commit, so let's differentiate
the structs required for each one.
Change-Id: Ib0e142a1dd4bedefdb4c5f15c34132da872c0975
For instance, use in VTY:
"ctr-error-threshold tx_underruns 5 per-second"
If the condition becomes true (eg 5 underruns happened in one sec), the
statement inside OSMO_MAX would become -1, but it was being handled as
an unsigned when doing the OSMO_MAX internal comparison. As a result,
OSMO_MAX((unsigned)-1, 1) was returning -1 (unsigned) stored in
threshold_timer_sched_secs which then became and int -1, which was
handled by osmo_timer_schedule as a 0, hence having an immediate
trigerring all the time.
While at it, make threshold_timer_sched_secs unsigned since it doesn't
make sense to have it as signed anyway.
Change-Id: I6ea3d64dff189a5bc924e72d846e02d50536a8ea
The log category DDEV ueses LOGL_INFO as debug level. This is too
verbose, lets use LOGL_NOTICE instead
Change-Id: I56d45ce5c3f55574491ffa6e4d902d6ba7499d46
Related: OS#2577
Under armv7l arch, size_t is actually an unsigned int and not a long
unsigned int, and compiler errors:
CommonLibs/debug.h:28:24: error: format ‘%lu’ expects argument of type
‘long unsigned int’, but argument 8 has type ‘size_t {aka unsigned int}’ [-Werror=format=]
Change-Id: I7f6ded5a984570b5267916d6c84eb7d019db73a8
Using %lu for pthread_t was wrong on armv7l arch. Even worse, according
to pthread_self() man page, pthread_t cannot be assumed to be of a simple
type and hence printable:
"""
POSIX.1 allows an implementation wide freedom in choosing the type used
to represent a thread ID; for example, representation using either an
arithmetic type or a structure is permitted.
"""
Let's use gettid() instead. According to glibc documentation:
"""
The pid_t data type is a signed integer type which is capable of
representing a process ID. In the GNU C Library, this is an int.
"""
It may not be the same on other libc's though, so let's better cast to a
long int just in case.
Accordign to gettid() man, the libc function was only added recently
during glibc 2.30, however the system call has been around for quite
some time (linux 2.4.11). Let's accomodate use udner non-glibc or older
versions of it by having a direct syscall fallback.
Change-Id: I40265fd4c62e550014ba3ff3335ca053c5bc01f2
Add an enum containing each supported device type (LimeSDR-USB,
LimeSDR-Mini and LimeNet-Micro) plus "unknown", to leave some room for
yet-to-come devices to run with some generic parameters without
rebuilding osmo-trx.
Each device type is assigned a dev_desc structure, and all of them are
put in HashMap, similar to what's already done in UHDDevice.cpp.
Device type is infered from string provided by LMS_GetDeviceInfo(), as
it was already done before in several places. From now on, we only need
to parse the string once since we store the device type after first
during open time.
Later on, more fields will be moved to device-type specific structure,
such as Tx timing offset, clock rate, etc.
Change-Id: I7658615787c5bc41c365bab9c11733b701ac2ae5
Make sure old configs using "logging level lms <level>" are still accepted.
Initialization order of VTY componenets need to be resorted since newly
introduced command requires logging VTY node to be already setup
beforehand.
Change-Id: Ia195a74a62a8a3dd6267fb1359acaa5628208d8e
Take the chance to clean up logging lines in this file:
* Use LOGCHAN in more places where chan is useful
* Replace inherited (old osmo-trx) categories such as WARNING with
osmocom ones.
Change-Id: Ic8c218f050f35d48046ccf1561fb0bfc505d4f63
In the command line options time, filler table/filer burts settings
were a bit difficult to undertand because the number of one-letter
settings was limited. Now, with VTY configuration, there is no reason
to keep it so difficult.
Also, after the previous commit it was no longer posible to enable
random 8-PSK filler bursts. With this patch you can configure all
supported filler bursts in a simple and logical way.
Change-Id: I752eb2c1162d084e8769181f2fcd6c0877663448
The EGPRS switch in the VTY config enables 8-PSK burst detection on
uplink. Enabling it shouldn't turn on filler bursts.
Change-Id: I2786c768e038b769a80c8b78fe58cfa09eb322a9
Since libosmocore Id7711893b34263baacac6caf4d489467053131bb, a new API
log_enable_multithread() is available which takes care of protecting
logging infrastructure from us (and actually does it correctly since we
cannot protect internal libosmocore structures from osmo-trx).
Let's drop all mutex related code from osmo-trx logging infra and simply
rely on libosmocore's.
Related: OS#4088
Change-Id: I519d0f30bce871005ca26b90177ea4aa4839360a
Otherwise, it could happen that underrun events are lost:
TxLower (isUnderrun): RxLower (pullBuffer):
read(underrun)
read(underrun)
write(underrun, |val) [maybe underrun becomes TRUE]
write(underrun, false)
Similary, it could happen the other direction if atomic was only applied
to isUnderrun:
TxLower (isUnderrun): RxLower (pullBuffer):
read(underrun) -> true
read(underrun)-> true
write(underrun, false)
write(underrun, true|val) where val=false
So in here isUnderrun would return true twice while it should only
return one.
Change-Id: I684e0a5d2a9583a161d5a6593559b3a9e7cd57e3
After discussion in [1] and further look at the code, it became obvios
rx_underrun events are not happening in general for any SDR (don't
exist), so let's drop that counter. Instead, add Tx Dropped Packet counters,
which were not accounted prior to this commit.
[1] bde55afd29
Change-Id: Iff1535c219a4695a511d383d7c4b06ef6eff959d
We have a good socket API in libosmocore, let's drop osmo-trx socket API
and use libosmocore's one instead of maintaining the two of them.
Change-Id: Ib19856a3e0a7607f63436c4a80b1381a3f318764
Take the chance to improve text with author, SPDX tag and fix incorrect
copyright dates.
Related: OS#3515
Change-Id: Ic745312ed07db205b1cdc0f2fa130000319354c5
osmo-trx will validate over time that those thresholds are not reached.
If they are reached, osmo-trx will die. As a result, osmo-bts-trx will
notice and will end up notifying the BSC about it (for instance because
it will also restart its process).
For instance:
"""
ctr-error-threshold rx_drop_events 2 minute
ctr-error-threshold rx_underruns 10 second
"""
In those cases above, osmo-trx will die if rate_ctr rx_drop_events went
to a value higher than 2 per minute, or it will die to if rx_underruns
went higher than 10 per second.
Change-Id: I4bcf44dbf064e2e86dfc3b8a2ad18fea76fbd51a
The callback actually belongs there, since it's the code/thread in main the one
actually in charge of stopping everything. It simplifies current code,
and more important, allows for new clients of this signal to use it.
This callback will also be used in forthcoming commits by code
controlling rate_ctr thresholds to stop the process if the VTY
configured threshold is used.
Change-Id: Id4159e64225c6606fef34a74b24f37c3a071aceb
Introduce a unified implementation-agnostic interface for radioDevice to
signal SDR error counters to upper layers and manage them.
This patch only implements counters for osmo-trx-lms (other devices will
show all counters unchanged during time).
Sample use through VTY:
"""
OsmoTRX> show rate-counters
osmo-trx statistics 0:
device:rx_underruns: 0 (0/s 0/m 0/h 0/d) Number of Rx underruns
device:rx_overruns: 0 (0/s 0/m 0/h 0/d) Number of Rx overruns
device:tx_underruns: 0 (0/s 0/m 0/h 0/d) Number of Tx underruns
device:rx_drop_events: 4 (0/s 2/m 3/h 0/d) Number of times Rx samples were dropped by HW
device:rx_drop_samples: 513 (0/s 196/m 425/h 0/d) Number of Rx samples dropped by HW
"""
Change-Id: I78b158141697e5714d04db8b9ccc96f31f34f439
Maximum number of carriers is fixed to 3 channels on a single
physical RF channel in multi-ARFCN mode. For some reason, it
was limited to 5.
Let's fix this, and also follow this limitation in the
following VTY command handlers:
- cfg_multi_arfcn_cmd,
- cfg_chan_cmd.
Change-Id: I66a1462f368458afd313ee6f0bc0abc496dde817
Since I838c21db29c54f1924dd478c2b34b46b70aab2cd we have both TS1
and TS2 synch. sequences, in addition to "default" TS0. Let's
finally introduce the VTY configuration parameter, that can
be used to toggle optional detection of both TS1 and TS2.
Note: we keep this optional because of potentially bad impact on
performance. There's no point in paying the performance penalty
unless upper levels (BTS, PCU) actually make use of it.
Change-Id: I1aee998d83b06692d76a83f79748f9129a2547e8
Related: OS#3054
According to gettimeofday manual:
"Applications should use the clock_gettime() function instead of the
obsolescent gettimeofday() function."
Furthermore, it may be desirable in the future to use other clocks such
as monotonic.
Change-Id: I2286998c5eefbf3c3dfb105c223daec7a1083803
Make the "opt" argument const. This function will also be used by
LMSDevice.cpp in a follow-up commit.
Related: OS#3654
Change-Id: If3f0f682ca453c2b0a06175ec9626567932cfce6
This log category is applied to messages related to TRX CTRL socket
interface, and it's printed in yellow, same color used in osmo-bts-trx
for TRX category (so same messages are printed with same color in both
sides).
Change-Id: I98ec5e416272783ad3fbadf70478a4e48ae64983
Original issue: In order to use SSE instructions, 16-byte aligned memory
chunks are needed, and C++ version < C++11 doesn't provide for a native
new/delete store. For that reason, memalign() must be used in the
implementation of convolve_h_alloc() for some buffers.
On the other side, The C++ code relies on C++ "new T[]" operator to
allocate a chunk of memory containing an array of class instances. As
classes are complex types, they cannot be allocated through C structures
(calling malloc). Experimentally can be seen too that it's unreliable
and the process will crash during startup if malloc() is used and then a
Complex<> deferred from it.
Previous implementation allowed for use of convolve_h_alloc or new[]
based on how the (signal)Vector is called, because then the buffer is
not going to be managed internally. But that's unreliable since resize()
calling resize() on it could use "delete" operator on a malloc'ed
buffer, and end up having a new new[] allocated buffer. It was also
found that some of the callers were actually leaking memory through ASan (because the
buffer is not managed by the Vector instance).
IMHO best option would be to rewrite all this code using C structures
and malloc/free exclusively, since it would make all this cod eeasier to
maintain.
But for now, let's extend the Vector class to allow specifying an
external alloc/free function and let the Vector instance take care of
the ownership of the buffer in all scenarios.
Change-Id: Ie484a4762a7f77fe1b105188ea03a6f025730b82
Found by ASan. when PointerFIFO::release() is called, alloicated node
being released is actually stored into an internal list for later-reuse
without having to access memory allocator. However, nodes from this
list are never freed.
Change-Id: I40e5e28603cde67005d9d92772967b05465ea2b8
This way the dependencies are passed after the .la object, which seems
to be the correct order. Some setups may fail to find some symbols from
libosmocore otherwise (OBS i586).
Change-Id: I22c80055bcffd4179a0a8ca76533ba7aaa38c859
osmo-trx can start a considerable amount of threads that can make
debugging it challenging at least. By using phtread_setname_np, the
system sets a meaningful name to the thread which can be seen while
debugging with gdb or by printing /proc/$pid/task/$tid/comm.
Now we also log system TID when setting the name so we can identify
different tasks in /proc even if pthread_setname_np fails.
Change-Id: I84711739c3e224cb383fd12b6db933785b28209e
Avoids this type of compilation warnings:
‘void* memcpy(void*, const void*, size_t)’ writing to an object of non-trivially copyable type ‘class Complex<float>’; use copy-assignment or copy-initialization instead [-Werror=class-memaccess]
Change-Id: I9724454dfb7b87f74f39074e4004580ac3b5fe5c
Pau Espin
Philipp Maier
Alexander Chemeris
Martin Hauke
Harald Welte
Eric Wild
Oliver Smith