This is the first step towards the goal of moving the scheduler
into a separate library.
Change-Id: Ifa6137c239c215a3d323213ee74d34b419622be4
Related: OS#5599, OS#3761
Remove the paragraph about writing to the Free Software Foundation's
mailing address. The FSF has changed addresses in the past, and may do
so again. In 2021 this is not useful, let's rather have a bit less
boilerplate at the start of source files.
Change-Id: I73be012c01c0108fb6951dbff91d50eb19b40c51
Similar to what we do in osmo-bts-trx, group everything related to
an Uplink burst into a structure. Pass a pointer to this structure
to the logical channel handlers. This makes the code easier to read,
and facilitates sending NOPE indications to the transceiver
(will be introduced in the upcoming patch).
Get rid of sched_trx_handle_tx_burst(), and instead just call
sched_trx_a5_burst_enc() directly from sched_frame_clck_cb().
Change-Id: Id45b27180c233fdc42ae1ef0b195554dd299a056
Related: SYS#5313, OS#1569
When running together with fake_trx.py (mostly used back-end), it
is currently possible that Downlink bursts are received in a wrong
order if more than one transceiver is configured (multi-trx mode).
This is how it looks like:
DTRXD DEBUG trx_if.c:612 RX burst tn=3 fn=629 rssi=-86 toa=0
DSCHD DEBUG sched_lchan_tchf.c:60 Traffic received on TCH/F: fn=629 ts=3 bid=1
DTRXD DEBUG trx_if.c:612 RX burst tn=3 fn=630 rssi=-86 toa=0
DSCHD DEBUG sched_lchan_tchf.c:60 Traffic received on TCH/F: fn=630 ts=3 bid=2
DTRXD DEBUG trx_if.c:612 RX burst tn=3 fn=631 rssi=-86 toa=0
DSCHD DEBUG sched_lchan_tchf.c:60 Traffic received on TCH/F: fn=631 ts=3 bid=3
DTRXD DEBUG trx_if.c:612 RX burst tn=3 fn=633 (!) rssi=-86 toa=0
DSCHD NOTICE sched_trx.c:663 Substituting (!) lost TDMA frame 632 on TCH/F
DSCHD DEBUG sched_lchan_tchf.c:60 Traffic received on TCH/F: fn=632 ts=3 bid=0
DSCHD DEBUG sched_lchan_tchf.c:60 Traffic received on TCH/F: fn=633 ts=3 bid=1
DTRXD DEBUG trx_if.c:612 RX burst tn=3 fn=632 (!) rssi=-86 toa=0
DTRXD NOTICE sched_trx.c:640 Too many (>104) contiguous TDMA frames elapsed (2715647)
since the last processed fn=633 (current fn=632)
so here a burst with TDMA fn=633 was received earlier than a burst
with TDMA fn=632. The burst loss detection logic considered the
latter one as lost, and substituted it with a dummy burst. When
finally the out-of-order burst with TDMA fn=632 was received, we
got the large number of allegedly elapsed frames:
((632 + 2715648) - 633) % 2715648 == 2715647
Given that late bursts get substituted, the best thing we can do
is to reject them and log an error. Passing them to the logical
channel handler (again) might lead to undefined behaviour.
Change-Id: I873c8555ea2ca190b1689227bb0fdcba87188772
Related: OS#4658, OS#4546
It's not something that we should be trying to fix, if the whole
TDMA multi-frame is lost. For some yet unknown reason, sometimes
the difference between the last processed TDMA frame number and
the current one is so huge, so trxcon eats a lot of CPU trying
to compensate nearly the whole TDMA hyper-frame:
sched_trx.c:640 Too many (>104) contiguous TDMA frames elapsed (2715647)
since the last processed fn=633 (current fn=632)
Let's just print a warning and do not compensate more than one
TDMA multi-frame period corresponding to the current layout.
Change-Id: I56251d0d2f6fa19195ff105d3bdfbc22df6db8cd
As was noted by Pau Espin Pedrol, there is a theoretical chance
that lchan->tdma.num_proc would overflow, so as a consequence,
subst_frame_loss() will be unable to compensate one
(potentionally lost) Downlink burst.
On practice, given the size of unsigned long and duration of a
single TDMA frame, it would only happen once in roughly ~6 years.
FRAME_DURATION = 4615 * 10e-6
ULONG_MAX = 2 ** 32 - 1
FRAME_DURATION * ULONG_MAX -> ~198212740 seconds
-> ~55059 hours
-> ~2294 days
-> ~6 years.
Chances are that trxcon would crash much earlier, or even GSM
would be completely forgotten after such a long time run, but
let's work this around and simply start counting from 1
if that overflow eventually happens.
Change-Id: I3d40ef09b06039a85df52af06ab38de314e1a434
It may happen that one or more Downlink bursts are lost on their
way to the MS due to a variety of reasons. Modern transceivers
supporting TRXDv1 protocol would substitute lost bursts with
so-called NOPE indications. Hovewer, neither fake_trx.py nor
grgsm_trx do support this feature at the moment.
We can still detect and compensate TDMA frame loss per logical
channels in the same way as it's already done in osmo-bts-trx.
In short, we should keep TDMA frame number of the last received
burst in the logical channel state, and using the appropriate
multiframe layout, check if there were any gaps between TDMA
frame number of the current burst and the stored one.
Change-Id: I3551d79796a3730565c2c70577e9d134e636f275
Using TDMA frame number of a burst with bid=0 is fine for xCCH,
but not for TCH and FACCH, because they use the block-diagonel
interleaving. A single block on TCH may be interleaved over
8, 4 or even 6 consecutive bursts depending on its type.
Since we now have the measurement history, we can attach TDMA
frame number to each measurement set, and then look up N-th
one when averaging the measurements in sched_trx_meas_avg().
Change-Id: I9221957297a6154edc1767a0e3753f5ee383173f
So far we used to store the sums of ToA and RSSI measurements in the
logical channel state, and after decoding of a block, we did calculate
the average. This approach works fine for xCCH and PDTCH, but when it
comes to block-diagonal interleaving (which is used on TCH/F and TCH/H
channels), the results are incorrect. The problem is that a burst on
TCH may carry 57 bits of one encoded frame and 57 bits of another.
Instead of calculating the sum of measurements on the fly, let's push
them into a circular buffer (the measurement history), and keep them
there even after decoding of a block. This would allow us to calculate
the average of N last measurements depending on the interleaving type.
A single circular buffer can hold up to 8 unique measurements, so the
recent measurements would basically override the oldest ones.
Change-Id: I211ee3314f0a284112a4deddc0e93028f4a27cef
During the handover the MS needs to release the existing dedicated
channel(s), establish the new one(s) as indicated by the network,
and then, depending on the synchronisation state, send one or more
HANDOVER ACCESS messages carried by Access Bursts.
In order to implement this, trxcon needs to be able to transmit
Access Bursts on any TDMA timeslot regardless of the logical
channel type and the associated handler, i.e. != TRXC_RACH.
The controlling side on L1CTL (layer23 or TTCN-3) needs to send
one or more L1CTL_RACH_REQ message(s) with properly populated
UL info header. Otherwise a regular RACH on TS0 is assumed.
Change-Id: Ia967820a536c99966ba2c60b63d2ea9edb093f46
PDCH channel support was introduced quite a while ago, but there
was no way to activate it via L1CTL so far. Let's fix this.
Change-Id: I3b66cab26108ab999a7fe969365ab57dc661399c
Wherever possible, use #defines from libosmogsm as opposed to magic
numbers. Using magic numbers in several places has the danger of
different programs/repositories having different views on what those
values mean.
Change-Id: I7ab4958801b3422973b67ff0452b90afa8a3f501
Related: OS#4027
Depends: libosmocore Change-Id I93e557358cf1c1b622f77f906959df7ca6d5cb12
OsmoBTS, BSC and TTCN3 used cbits == 0x18 for dynamic PDCH, while
trxcon wanted to use 0x18 for CBCH on SDCCH/4. Let's fix this and
bring everyone in agreement.
Related: OS#4027
Change-Id: Ia9a415628c659cbc2dd5dc65b875b7f935d6e211
This change extends sched_trx_chan_nr2pchan_config() with Osmocom
specific cbits related to CBCH, so now one can to decode
CBCH channels in dedicated mode (see L1CTL_DM_EST_REQ).
Change-Id: I9347c45638223cac34f4b48eb736e51a5055a36f
According to 3GPP TS 04.08, section 3.4.1, SACCH logical channel
accompanies either a traffic or a signaling channel. It has the
particularity that continuous transmission must occur in both
directions, so on the Uplink direction measurement result messages
are sent at each possible occasion when nothing else has to be sent.
The LAPDm fill frames (0x01, 0x03, 0x01, 0x2b, ...) are not
applicable on SACCH channels!
Unfortunately, 3GPP TS 04.08 doesn't clearly state which "else
messages" besides Measurement Reports can be send by the MS on
SACCH channels. However, in sub-clause 3.4.1 it's stated that
the interval between two successive measurement result messages
shall not exceed one L2 frame.
This change introduces a separate handler for SACCH primitives,
which dequeues a SACCH primitive from transmit queue, if present.
Otherwise it dequeues a cached Measurement Report (the last
received one). Finally, if the cache is empty, a "dummy"
measurement report is used. When it's possible,
a non-MR primitive is prioritized.
Change-Id: If1b8dc74ced746d6270676fdde75fcda32f91a3d
Related: OS#2988
Having access to a logical channel state is required by the
follow-up change, which will introduce a separate function
for dequeuing SACCH primitives.
Change-Id: Ibde0acf8e6be224b1007be707a636eaad68c8d36
The 'normal' math operations, such as addition and substraction,
are not applicable for TDMA frame numbers because they may result
in out-of-range values.
Having TDMA frame math helpers in a single place would allow
one to avoid possible out-of-range result mistakes.
Change-Id: Ibb66ba846cc3d6c2eaa88414569e5f3751128047
This field of the logical channel state structure was not used at
all as there is nothing related to A-bis / RSL in trxcon itself.
Change-Id: Iec1abf777a74cf57deadafa95e2337cba5d02842
Initially it was assumed that FACCH prioritization should be done
in the same way for both TCH/F and TCH/H. Moreover, it was not
possible to confirm this, because TCH/H was (and still) not
implemented yet. But according to the specs:
- unlike FACCH/F, FACCH/H transmissions shall be aligned
within a multiframe, i.e. can only be initiated on
particular frame numbers (see GSM 05.02, clause 7);
- unlike FACCH/F, a FACCH/H frame steals two TCH/F frames;
so the TCH/H (including FACCH/H) primitives should be handled
separately from the TCH/F (including FACCH/F) primitives.
Change-Id: I9b59f60e1cbac8fb8fd557b6c67b5e376c0a6bbb
the initial tch_mode is not always 0 (signalling) but can very well
be directly a codec mode, if the initial activation of the channel
is in speech mode as opposed to signalling
Change-Id: I96e4c89da1165e9c5287d863e0e65d811460c606
If at the moment of transmission there are no frames in TX buffer,
then either a dummy LAPDm frame (0x01, 0x03, 0x01, 0x2b ...) or a
silence frame (depending on a codec in use) shall be transmitted.
This is required for proper measurements on the BTS side.
Change-Id: Ie590990f2274ea476678f6b2079f90eeadab6501
The sched_frame_clck_cb() is responsible for UL burst transmission.
Iterating over each timeslot, it chooses a proper lchan handler
according to a current frame number and a multiframe layout in use,
takes a L2 UL frame from a TX buffer, and finally calls the chosen
handler in order to to encode and transmit a taken frame.
A handler should be called only for activated logical channels...
but for some long time, there was a bug, so each lchan was
processed, including inactive ones. It's time to fix this.
Change-Id: I33e3ecc14be3ae64dfd02789c7f0970c945582c9
The llist_for_each_entry_safe() should be used instead of the
llist_for_each_entry(), because it's safe against removal
of llist entry.
Found using Valgrind's memcheck tool.
Change-Id: I65234971ec152df038c5388da537a503060c215b
There's no need to express ToA value as a float. Let's turn it into
an int16_t with 1/256 symbol period accuracy throughout the code to
avoid both float arithmetic as well as loosing any precision.
Inspired by Idce4178e0b1f7e940ebc22b3e2f340fcd544d4ec.
Change-Id: I99c0f38db08a530d5846c474aba352aa0b68fe86
As there is no any order relation between logical channels, it's
better to use the linuxlist API instead of talloc array.
Change-Id: I5a78582c77ed1ab33817d240e065dc4cd4708199
Previously, when resetting or deleting a timeslot, we did not
deactivate the logical channels, relaying on talloc hierarchical
nature. This approach may cause some problems, e.g. on embedded
systems with emulated talloc API.
Change-Id: I8c34c793df87bd8c79b7bf1f05b949faf10520e8
Let's assume that a logical channel, which was already in use,
is activated again for a new connection. As we don't reset the
state variables, such as burst masks or ciphering data, it may
cause an unexpected behaviour.
In order to avoid this, let's always reset the logical channel
state after deactivation.
Change-Id: I91e736a97cb05b167614cb488a00d847a9a859e0
This change implements the A5/X ciphering support transparently
for the logical channel handlers. In other words, a DL burst is
deciphered before being passed to a handler, and an UL burst is
ciphered before being sent to transceiver.
The implementation mostly relays on the libosmocore's A5 API.
Change-Id: Ib53418d8c0f394fdece09cf5cc240887cb0bb5af
Having a possibility to preprocess UL burst before sending to
transceiver is required for the further ciphering support
integration and probably some other tasks.
Change-Id: Ia6eead5d4f51d7c0bf277b9d5ebb0a74676df567
Previously, each lchan handler used to obtain and delete primitives
from a timeslot's tranmit queue itself. This approach entails many
potential problems and bugs:
- The lchan handlers shall not do that by definition, they
should encode and decode frames according to GSM 05.03.
- In some cases (e.g. TCH), a single transmit queue may contain
primitives of different types (e.g. TCH, FACCH and SACCH). At
the same time, the lchan handlers don't care and don't even
know about each other. So, this could cause an unexpected
behaviour in some cases.
This change separates all primitive management routines,
providing a new API for obtaining and dropping them.
"Write programs that do one thing and do it well."
Change-Id: I29503ece51903784bc53541015285234471c8d15
It's good to write, keep and make the source code as much modular
as possible. So, Tte primitive management code was separated to
the 'sched_prim.c' and going to be extended in the near future.
Change-Id: Ifec8c9e4f2c95c72b00772688bcb5dc9c11d6de7
In order to get the transceiver more time to process bursts,
the L1 scheduler should process the frames and send the bursts
in advance (a few frames before), like OsmoBTS does. By default,
the advance value is 20 frames, but this can be adjusted using a
new command line option of trxcon '-f'.
Change-Id: Ic258a169f3554f931d6277e18ca060d029b77f32
For some reasons, the function, which is used to flush a queue of
transmit primitives, was intended to flush a list of msgb instances
instead of trx_ts_prim, so memory was being cleaned incorrectly.
Moreover, the items weren't actually removed from queue.
Change-Id: Ia84b57350a5c2eee0afebc65f62e30eaddb141d4
This change introduces shared primitive management functions,
exposed from the l1ctl_rx_data_req() implementation:
- sched_trx_init_prim() - allocates memory for a new primitive
and its payload. Initializes primitive's header, setting
the logical channel type and the payload length. After
initialization, the talloc context of a primitive is
a trx instance, which passed as the first argument.
- sched_trx_push_prim() - decodes the timeslot index from
chan_nr and pushes a primitive to its transimt queue.
The talloc context of primitive is changed to the
parent trx_ts instance after queuing.
Both functions will be used for handling both L1CTL_TRAFFIC_REQ
and L1CTL_RACH_REQ.
Change-Id: I8169a1ef4ef54d91b50f3e213e4842f54af8b499
It's better to pass a trx_lchan_state instance directly from
caller to lchan handler instead of passing trx_lchan_type. This
way a handler wouldn't need to find lchan itself.
Change-Id: I47a40542b03ab31da12b0abb1c263c83662ff018
Previously, the chan_nr, received from L2, was parsed in a wrong
way, so in some cases only one logical channel was activated or
some messages (such as Measurement Requests) were sent on
incorrect channel (e.g. on SDCCH instead of SACCH).
This change reimplements the sched_trx_chan_nr2lchan_type(), and
introduces a new function sched_trx_set_lchans(), whics will
parse chan_nr and (de)activate required channels.
Change-Id: I480311c65ef93bbc1644ec708dd2a68fd33091e0
Vadim Yanitskiy
Oliver Smith
Vadim Yanitskiy
Pau Espin
Harald Welte