this patch adds two new config flags to the ZMQ driver that allows to:
* configure the default ZMQ trx timeout in ms
* turn on error logging if the timeout occurs
Use with, e.g.:
device_args = log_trx_timeout=true,trx_timeout_ms=3333
Removing unecessary \n for logs and changed log of PDCP info queue
capacity to debug to avoid log spam.
Changed log level for unhandled S1AP messages from error to warning
in EPC to avoid failing tests because of error message.
Changed usage of allocate_unique_buffer to make_unique_buffer()
This includes:
- Adding a queue (implemented with std::map) for undelivered PDUs.
This queue uses the SN used for TX as the key.
- Added discard timer that is started upon reception of the SDU. Upon
expiry of the timeout a discard callback removes undelivered PDUs
from the queue.
- Added the mechanisms to the notify_delivery to remove PDUs from the
undelivered queue when the PDU is ACK'ed.
- Added test case for both timer expiry and acknowledgment.
- Fix up the getter for buffered SDUs to return the undelivered SDUs
- Changed default PDCP discard timer, so AM has a discard timer by
default.
test. The race condition was being cause by write_sdu being called
simultanously to read_pdu, which could cause the read_pdu to try to get
the SDU info before it had been written by the write_sdu.
Changes in this commit include:
- Make sure PDCP sn in included in RLC AM stress test.
- Stop handling control PDU when TX is not enabled in RLC AM.
- Fixed issue with length of the PDCP SN in rlc_stress_test.
- Moved the place were sdu info erase was called to avoid double calls to erase
- Tentative fix for race condition in rlc_am_stress test.
- Added function to print information about undelivered_sdu_info_queue
for debugging.
PDCP PDUs. This includes:
- Modifying the byte_buffer_t to include PDCP SN meta-data. This way,
the RLC can keep track of the ack'ed bytes for a specific PDCP PDU.
- Added in the RLC an `undelivered_sdu_info queue`, to keep track of the
amount of ack'ed bytes and the total size of the PDCP PDU,
so the RLC can know when delivery is finished.
- Added an interface between the PDCP and the RLC so that the RLC can
notify the PDCP when it receives an ack from the status PDUs. The RLC
passes to the PDCP a vector of all the ack'ed pdus in a rx'ed status PDU.
- Added some tests to the notify functionality. This includes some tests
where the PDUs are acked imediatly, and one test where the PDU is
retx'ed.
after adding the thread-safe PCAP writer functionality to the EUTRA
MAC object it became clear that we don't wont to replicate the
same for the NR object.
This patch therefore refactors the class that now supports both EUTRA and
NR rats. The old mac_nr_pcap.[h/cc] has been deleted. All test-cases
and usages now use the new object that needs to pass the RAT type in
the ctor.
this patch addresses the last open point of #2160
* Match hex dump indentation with srslog.
* Port asn1 utils to use srslog. Converted logging functions from taking varargs to variadic template functions.
* Remove trailing new lines in log lines from asn1 utils.
* First round of upgrades to srslog to asn1 tests.
* Second round of porting asn1 tests to srslog.
- multiple tunnels per E-RAB
- data forwarding between connected GTPU tunnels
- forwarding GTPU End Marker between connected tunnels
- TeNB GTPU handles in-sequence delivery when multiple tunnels for the same ERAB exist.
* - Started porting the enb PHY layer to use srslog loggers.
- Updated srslog to manage the none level.
* Finished porting enb phy layer including the NR part.
* Ported MAC, GTPU, PDCP, RLC and S1AP enb classes to use srslog.
* Use new stack logger.
* Ported the enb RRC clases to use srslog.
* Remove unused log macros.
* Replace loggers in sched, sched_carrier, sched_helpers.
* Replaced loggers in sched grid.
* Replaced loggers in sched harq.
* Replaced loggers in sched ue.
* Replaced loggers in sched ue ctrl.
* Replace loggers in sched ue ctrl TPC.
* Replaced loggers in sched subclasses.
* Replaced loggers in rrc_meascfg_test
* Configure loggers in rrc_mobility_test.
* Fix compilation errors left out after the rebase.
* - Implement a custom log sink that will serve as a test spy to intercept and count the number of error and warning log entries.
- Adapt the erab_test_setup and rrc_mobility tests to use this new class and make them pass again.
* - Remove trailing new lines introduced in the rebase.
- Ported the sched_ue_cell class to srslog.
* Remove unused log member.
* Ported mac tests to srslog.
* - Removed remaining trailing newlines from log entries.
* Fix compiler errors detected in CI.
* Fix another static variable without definition passed to log lines.
* Fixed a bug in srslog::flush that would never end when the backend queue is full.
* Fetch the RRC logger instead of injecting it in the constructor.
* Refactor NR resource allocation classes. Use DCI instead of grant for dummy PDSCH UE/eNB test
* Minor refactors in NR workers and ue_dl
* Fix include issues
* fix compilation issues
we've checked the same when unpacking the subheaders but missed the
case where the payload was read beyond the PDU length, as has been
seen with a malformed RAR PDU.
the previous patch only introduced a thread-safe queue between
PHY workers (writers) and PCAP writer thread (consumer).
However, it is also required to protect the ctor (and close()) to
prevent corrupted PCAP files.
The patch also correctly handles the case where the PCAP couldn't
be openend for writing and doesn't start the thread.
* offload PCAP writing to background thread
* use blocking_queue between writer and clients to make it thread-safe
* add basic test case
this fixes point 1-3 of #2161
detected with ASAN trying to write negative number of padding bytes.
The patch checks the calculated length and returns with an error
if the length is negative.
=================================================================
==5759==AddressSanitizer: while reporting a bug found another one. Ignoring.
m==5759==ERROR: AddressSanitizer: negative-size-param: (size=-6)
- Handle received E-RAB S1AP at s1ap.cc.
- Added methods to rrc.cc, rrc_ue.cc and rrc_bearer_cfg.cc to handle erab modify request.
- Made RLC add_bearer() function capable of re-creating the RLC entity.
- Send RRC reconfiguration to the UE and reply to the EPC with S1AP
modify bearer response.
This commit also adds support to srsEPC to send S1AP modify bearer request for
testing purposes.
this patch adds support for measuring the:
* SDU rx latency (Average time in ms from first RLC segment to full SDU)
* amount of buffered bytes (sum of payload of PDUs buffered in rx_window)
the implementation is using std::chrono
* Reorder DCI FORMAT enum
* Fix endianness issue
* Fix return codes in phy_ue_db
* Log members should be destructed after the layers.
* Add JSON metrics and Events. Add Alarm log channel. Simplify MAC metrics struct.
* Restore metrics_stdout change
RLC TM can't be segmented so it may happen that a PDU
cannot be sent because the grant is too small. That is not
an error and should only be logged in info.
when retrieving the RLC metrics, the number of TTIs since the last
call are passed. This allows to calculate the actual rate
based on the LTE timing rather than only the system timing.
we've only incremented the retx when retransmitting a full PDU,
but not when only retransmitting segments.
This might lead to many more retx than allowed.
Before this PR, the eNB was not activating UL-64QAM if the UE release was below 12,
due to an erroneous convertion of the ASN1 EUTRA Capabilities struct into the flattened
srslte::ue_capabilities_t.
With this PR, the eNB activates UL-64QAM for the following cases:
- UE category 5
- UE category 8 and release >= 10
- UE category UL 8 and release >= 12
the RLC bearer name was empty after reestablishing a UM bearer.
we need to pass the RLF config and RB name to the internal configure()
call of the Rx entity, like we already do for the Tx side
when EVM measurements are enabled, instead of only logging them for
each decoded codeword, we also calculate the average.
this is useful for calibration.
hard killing of AGC thread causes issue on some RF devices,
such as the N310. The thread still seemed to access the device while
the radio was already killed (or the streamer object deleted).
It's unclear why this isn't causing similar issues on B210 or X310,
at least not visible, but it is obviously not the correct way to stop a thread.
The patch now correctly sets the stop flag for the AGC thread, wakes it
it up and waits until it has terminated.
the Lime has better performance when using the default LTE sample
rates due to filterering being done in the LMS RF chip.
Print a warning when using the Lime with the non-default LTE rates.
the bug causes the BSR to return a bsr_idx of 0, i.e. no data
pending, when the buff_size was reported to be 1 B only.
Instead, the UE should return BSR idx 1, i.e. between 0 < BSR <= 10.
LBSR packing is fine but this makes the eNB do wrong things
because it thinkgs LCG2 has no or too much data to send.
LCG2 is our default LCG for DRB1. One occasion we saw the issue
is doing full rate TCP DL which requires quite a bit of ACK traffic
back to the sender. With the BSR for LCG2 being wrong, providing
UL grants was delayed and so the TCP DL throuhgput was very low.
In the log below we see that the buff_size[2]=1 so there is data to send.
Interestingly, the packing was correct so Wireshark displays it correctly.
But the unpacking not, as can be seen in the MAC log below which reports
0 buffer state for all LCGs
22:08:54.804792 [MAC ] [I] [ 5084] pdu_get: sdu_len=1212, channel.sched_len=1213, bsr.buff_size[2]=1
22:08:54.804840 [MAC ] [I] [ 5084] UL LCID=3 len=1212 LBSR: b=0 0 0 0
* rlc_am: add TC for pollRetx timer handling
* rlc_am: fix stopping of pollRetx timer
stopping the pollRetx timer when receiving a status PDU
without checking the acknowledged PDUs is wrong.
if an ACK for another PDU, for which the polling bit has been set,
is still pending, it won't be rescheduled until another
PDU is transmitted that again starts the pollRetx timer.
this fixes the issue with missing RLC AM segment retx in #1992,
and #2003
* mac_test: add extended TBSR unit test
unit test to MAC UL packing after sending a TBSR
this fixes the MAC issues described in issue #2002
* mux: fix updating of LCG buffer state after packing PDU
we've previously lowered the buffer state of the LCG according
to the bytes that have been scheduled, but not according to
those that have been actually included in the PDU.
* proc_bsr: fix LCG buffer state updating for TBSR
when sending a TBSR do not update the internal buffer
state of the BSR proc.
This caused issues because the buffer state for all LCG that
are not included in the TBSR are set to zero, although at least
one LCG does have data to transmit.
* rlc_am: include LCID when logging retx of SN
I noticed that the app didn't configure the process with
RT thread prio. But it then turned out that in order to do
that we need to actually spawn a new thread. Also a few other things
didn't work out of the box.
In summary the changes are:
* fix duration to support more than 120s
* move main functionality in thread
* set high priority for radio thread
* only return success if no radio errors occured
this makes sure that the reconstructed RLC AM header
has the poll bit set when any of its segments had it set.
this triggered a bug in RLC AM because the Status PDU
isn't transmitted for a RLC PDU that was segmented and
only the first segment had the P flag set.
reported/provided by user softdev86 in https://github.com/srsLTE/srsLTE/issues/566
author tested with local 4 port cell. I am not able to verify locally but
it looks ok, we'll revise later if needed.
When using srsLTE with Lime devices, calibration was performed before any configuration steps have happened, thus making calibration values invalid. Removing Lime specific calibration step from rf_soapy_imp makes so that devices will be automatically calibrated by SoapyLMS on rf_soapy_start_stream call.
Tested and working with srsENB using LimeSDR-USB v1.4 and LimeSDR-Mini v1.2 boards.
when the default RF args are used, NULL is passed as argument
to the RF driver. In this case, we should use the previous
SoapySDRDevice_enumerate() and let Soapy pick the device
This patch adds version string to the 'libsrslte_rf' library.
This results in the library being built as
libsrslte_rf.so.20.04.2
with the following symbolic links
libsrslte_rf.so.0 -> libsrslte_rf.so.20.04.2
libsrslte_rf.so -> libsrslte_rf.so.0
The SRSLTE_SOVERSION variable should be increased with each backwards
incompatible change of the library.
Signed-off-by: Jan Remes <jan.remes@invasys.com>
this patch adds a check to drop all PDCP control PDUs
in order to prevent handling them as data PDUs.
This could happen when the size exceeded the arbitrary length check.
This should fix#1787
* Fix deadlock caused by update_measurements calling in_sync
* Fix overlapping SRS condition
* Do not use shortened PUSCH in transmissions and retx from RAR
* Revert "Fix deadlock caused by update_measurements calling in_sync"
This reverts commit f58c8c8c766f8f95baa3a3bf8287d8e25b2057ba.
* Take into account CRS from neigbhour cells when measuring interference
* fix std::isnormal compilation
* Fixed compilation of test
* Address comments
* Remove unused overrides
this patch refactors the MAC PDU class, the main changes are:
* add to_string() method to pretty print PDU and subheaders
This allows to have a single log entry per MAC PDU with all its
contents.
It removes the C-style fprint() method
* Simplity payload vs. w_payload_ce
Before we've used payload when reading the PDU and w_payload_ce
as a buffer when writing. In all getters we needed to differentiate
between both. Now payload points to w_payload_ce initially and is
only updated when parsing a new PDU.
* add various helpers, e.g. to get subheader index, update a BSR, ..
* fix PDU test to use new to_string() method
simplify logic to check whether a BSR fits and, if so, which type.
before the check has been done in two places.
we now also accomodate for the CE subheader size.
with the B210 and 2 RF ports, i.e. MIMO mode, we have stopped the
Rx stream after changing the rx_samp_rate but didn't start it again.
Either the issue doesn't exist in SISO mode or we never saw it but for MIMO
it can be reproduced easily with rate changes during streaming, i.e.:
$ ./lib/src/radio/test/benchmark_radio -p 2 -t 10 -x -y -s 23.04e6
Instantiating objects and allocating memory...
Initialising instances...
Opening 2 channels in RF device= with args=default
[INFO] [UHD] linux; GNU C++ version 9.2.1 20200304; Boost_107100; UHD_3.15.0.0-2build5
[INFO] [LOGGING] Fastpath logging disabled at runtime.
Opening USRP channels=2, args: type=b200,master_clock_rate=23.04e6
[INFO] [B200] Detected Device: B210
[INFO] [B200] Operating over USB 3.
[INFO] [B200] Initialize CODEC control...
[INFO] [B200] Initialize Radio control...
[INFO] [B200] Performing register loopback test...
[INFO] [B200] Register loopback test passed
[INFO] [B200] Performing register loopback test...
[INFO] [B200] Register loopback test passed
[INFO] [B200] Asking for clock rate 23.040000 MHz...
[INFO] [B200] Actually got clock rate 23.040000 MHz.
[INFO] [MULTI_USRP] 1) catch time transition at pps edge
[INFO] [MULTI_USRP] 2) set times next pps (synchronously)
Warning: TX gain was not set. Using open-loop power control (not working properly)
Setting manual TX/RX offset to 0 samples
Start capturing 10000 frames of 23040 samples...
Changing sampling rate to 23.04 Msamps/s
Setting manual TX/RX offset to 0 samples
Changing sampling rate to 1.92 Msamps/s
Setting manual TX/RX offset to 0 samples
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1211: Error timed out while receiving samples from UHD.
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1211: Error timed out while receiving samples from UHD.
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1211: Error timed out while receiving samples from UHD.
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1211: Error timed out while receiving samples from UHD.
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1211: Error timed out while receiving samples from UHD.
it fixes#1623.
this happens more often with MIMO since lates are more likely here.
after a late, the Rx stream must not be stopped on the B2xx either.
<log>
RF status: O=3, U=0, L=1
/home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1209: Error timed out while receiving samples from UHD.
stop rx stream
./home/anpu/src/srsLTE/lib/src/phy/rf/rf_uhd_imp.cc.1209: Error timed out while receiving samples from UHD.
stop rx stream
/home/anpu/src/srsLTE/lib/src/phy/ue/ue_sync.c.775: Error receiving samples
/home/anpu/src/srsLTE/lib/src/phy/ue/ue_cell_search.c.312: Error calling srslte_ue_sync_work()
/home/anpu/src/srsLTE/lib/src/phy/ue/ue_cell_search.c.272: Error searching cell
</log>
before the BSR was extracted but the actual index (between 0 and 63)
was not stored but directly converted into bytes.
for log parsing and debugging it is easier to follow the index
value. this patch therefore adds both values to the log message
and extends the API accordingly.
commit #51c6e8d1 introduced a regression in which the 2nd antenna
port was never considered. All MIMO ZMQ based tests therefore failed.
this fixes#1608
the enable_rx_timestamps() for the radio_ctrl object has only been
introduced in Nov 2019 for UHD 3.15 and therefore needs
to be excluded when compiling for 3.14 and below.
Note: According to the original UHD commit in
67dbaa41f2 (diff-60a9387c0fc8406fd5b39fa995dd8c14)
it looks like the Rx timestamps are disabled by default
for UHD versions before 3.15
This commit fixes#1602
- Import the srslog project into srslte.
- Ported srsue app to use the new logging framework.
- Implemented a wrapper that dispatches log entries to srslog.
- Renamed an existing log test to be more specific to avoid name clashes.
at the when applying the fading channel we copied the state back
to our buffer. Since we are offsetting the read buffer
by X samples, we need to make sure that we don't read after then
(len-X) samples of the temporary memory buffer
the PSS detection needs more temporary buffer than a full subframe.
we therefore need to allocate and initialize the sync object with
larger maximum size to support Scell search of large cells, e.g.
20 MHz
this fixes issue #1538
* add locked and unlocked version of has_data() since one is
called from stack and one from PHY threads
* add comments in each interface section as to why locking
is required or not
* remove RLC rwlock when not required
* move calls only used by RRC to RRC section
this patch refactors the SDU queuing and dropping policy of the RLC and PDCP layer.
the previous design had issues when packets have been generated at a higher
rate above the PDCP than they could be consumed below the RLC.
When the RLC SDU queues were full, we allowed two policies, one to block on the write
and the other to drop the SDU. Both options are not ideal because they either
lead to a blocking stack thread or to lost PDCP PDUs.
To avoid this, this patch makes the following changes:
* PDCP monitors RLC's SDU queue and drops packets on its north-bound SAP if queues are full
* a new method sdu_queue_is_full() has been added to the RLC interface for PDCP
* remove blocking write from pdcp and rlc write_sdu() interface
* all writes into queues need to be non-blocking
* if Tx queues are overflowing, SDUs are dropped above PDCP, not RLC
* log warning if RLC still needs to drop SDUs
* this case should be avoided with the monitoring mechanism
with ZMQ RF, we've seen the radio returning -1 and traced
the issue down to the ringbuffer.
in fact, pthread_cond_timedwait() returned 22 (EINVAL) indicating
a wrong timeout value. The issue was likely an overflow
in the timespec parameter. This patch checks for the return
value of pthread_cond_timedwait() and also calculates the
nanosecond wrap around for the timespec
amarisoft flags an RF error and exits after not receiving samples
for the radio. Running with ZMQ in the RFCI, the UE sometimes needs more
time to start up due to slow processing and no fftw wisdom file and
then Amarisoft eNB times out. This gives the whole process a little
bit more time. Should we still find it too short we may increase
it further.
This lock was removed in 1cbf7eac because it was considered unneeded.
However, as can be seen in issue #1503, we need to protect the access
to rx_window, for example.
Issue #1503 shows a stack trace where a PHY worker generates a status PDU.
While holding the mutex to access rx_window, the stack thread on the
other hand, happily accesses the rx_window member without acquiring
the lock. This commit protects all handle_*() functions in write_pdu().
This reverts commit 1cbf7eac04.
instead of logging PDU received from below, and SDU that is forwarded
to above, only log the PDU in rx in info mode.
The next layer will do the same and log the PDU received (which is the SDU coming from here)
so there is now loss of information in the logs.
during the PDU packing it could happen that a new SDU segment
was added but the resulting larger header was so big that not even
a single byte of the new PDU could be added. because this
corner case wasn't handled correctly an invalid PDU was transmitted.
the solution is to revert the addition of the new SDU.
* state funktion for eia1 (mitigate race condition)
* Extra file for snow3g
* just use mbedtls aes
* ref specification
* adding const keyword to security functions
Use the return value of parse_string function to determine whether a parameter was found.
I experienced issues when the rx_type was set, but the rx_format was not. This led to "Unsupported sample format pub" as the tmp variable is still set to the value of the rx_type.
clang complained about an uinit var that is returned and it turned
out that we don't even implement that function in NEON.
I've found a nice MIT-licensed header that was easy to integrate.
this fixes the subframe-accurate rx in tracking mode.
previously we've only alligned to the next subframe start in find and
when there was a single overflow we've lost the track.
this patch calculates the offset in samples to the actual start of the
subframe (full ms of the receive timestamp) vs. the actual
receive timestamp and uses the value as the "next_rf_sample_offset"
to realign on the next receive
we've used a macro that can return -1 as access index
for an array. this has now been converted in a member
that is initialized and checked during init
fixes Coverity issue 1480088 where srslte_ue_dl_find_dl_dci() expects
an array of struct srslte_dci_dl_t but is only given one element.
srslte_ue_dl_find_and_decode() itself only uses the first decoded
message but find_dl_dci() may still decode more than that
this patch adds the -y paramter that changes the sampling rate
during a radio benchmark to stress-test the API
it turned out that some devices/drivers are very prone to
sampling rate changes. since this happens during UE operation
when the UE is searching for cells, we need to make sure
the driver/device supports that operation.
this turns off some Lime specific tweaks that we did, for example
turning off streaming before changing the sampling rate.
those are now implemented inside the LimeSuite (as of 20.01)
and are no longer required here
patch also disable the explicit BW setting and async background thread
It allows to provide an index for tx_port, rx_port, tx_freq and rx_freq even
for the first (and possibly only) port. So for example a MIMO config would be this for example:
tx_port0=tcp://*:2001,tx_port1=tcp://*:2003,rx_port0=tcp://localhost:2000,rx_port1=tcp://localhost:2002
the missing mem zero'ing caused the unit test to fail on some
machines where the memory wasn't initialized by default.
for d and d16 allocate the maximum number of bits, not
only the max length of a DCI
this is an extension to the currently available security
configuration and activation methods provided by PDCP.
The new call allows to specify a Rx and Tx SN after which the
entire security config, i.e. integrity and ciphering, should
be enabled.
this is mainly required for the PDCP entity of the conformance
testing SS but could potentially also be used by the eNB
to enable security for PDCP after sending the SecModeCommand
for example.
the extra call was added to not break the existing API.
This commit implements support for the CLOSE_UE_TEST_LOOP message.
Currently the mode is not saved and looping the data is not implemented.
Only the *_COMPLETE message is sent back as a response.
during deinit of the UE/eNB it could happen that the log_ptr
was already set to null but the stack thread was still accessing
it through log_h->info()
i've moved the old macro into a new header and used as as much
as possible. there are still places in the code where similar
macros are used but we would need to change the variable name
to replace them
* Use task id to track old background tasks in RA procedure
* Improve robustness against RF overflow in PHY
* Increase SNR out-of-sync threshold
* Do not change frequency if it's the same
* Increase sync priority
* Increase time to start receiving to reduce input buffer occupation
* Use scoped lock in sf_worker
gcc8 added a -Wstringop-truncation warning that flags
an issue in phy_common because we are copying a str
to another str without explicitly saying it's null-terminated.
we could turn off the warning using a pragma in the code
but I guess it's safe to just null-terminate after
copying manually
[1] https://stackoverflow.com/questions/50198319/gcc-8-wstringop-truncation-what-is-the-good-practice
previously PDCP security (integrity and ciphering) could only
be enabled for both Rx and Tx at the same time.
this, however, caused an issue during the conformance testing in which,
in TC_8_2_1_1() for example, the eNB sends a SecModeCommand and a
RRC Reconfiguration in the same MAC TB. In this case, the eNB
needs to be able to enable DL security right after sending the SecModeCmd
in order to send the RRCReconfig encrypted. However, enabling UL security needs
to be postponed until after the SecModeComplete is received.
This patch allows to enable PDCP security for rx/tx independently if
that is needed. The default way is like before, enabling it for tx/rx at
the same time.
this patch adds a buffer len paramter to the receive_samples() call
that protects the (also) provided buffer from overflowing.
currently each call to srslte_ue_sync_zerocopy() which then calls receive_samples()
relies on a buffer that is "big enough". But that buffer is sometimes
2 subframes, sometimes 3 or 5, sometimes has space for the maximum PRB size, sometimes
only for 6 PRBs (i.e. during cell search).
By extending the interface to pass the buffer size we can make sure that
only samples are received that actually fit inside the provided buffer.