The main RPC channel can sit idle for a long time (e.g. an interactive
Python session) while only the claim-loop channel sees traffic. Without
keepalive, gRPC self-closes the idle transport (~30 min) and the next
RPC fails re-establishing within its per-call deadline.
With this change the client now pings every 30 s to keep the channel
warm, and the server is configured to accept that cadence instead of
replying GOAWAY "too_many_pings".
The MPM gRPC server runs on a single-threaded gevent hub. gRPC's BDP
(bandwidth-estimation) probe fires repeatedly while the reclaim loop
keeps the link busy, and its ACK must return within grpcio's short,
RTT-derived internal BDP deadline. Any blocking hardware register read
(e.g. a ClkMgr PLL-lock poll) stalls all socket I/O on the hub until it
returns; , but these polls also run periodically in steady state.
When a stall outlasts the BDP deadline the server sends GOAWAY
"ping_timeout" and drops the transport. The fix is to disable BDP on
the server. The default flow-control window is fine on the LAN link,
so this costs nothing here. BDP is left enabled on the client, whose
native-threaded stack never freezes.
host:
Replace the msgpack-RPC (rpclib) client with a gRPC-based
implementation. Add auto-generated C++ client and Python bindings
from protobuf definitions. Remove rpclib dependency and integrate
FindgRPC into CMake.
mpm:
Replace the msgpack-RPC server with a gRPC server generated from
protobuf definitions. Add pb_utils.py for type conversion between
Python and protobuf. Update rpc_server.py to dispatch calls via
gRPC stubs. Update mpm_shell to use gRPC. Replace pyudev with sysfs
scan in uio device discovery.
Bump MPM compatibility number to 7.0 and RPC version to 2.
This change requires new depedencies. E.g. for Ubuntu 24.04 or 25.10:
libgrpc++-dev, libprotobuf-dev, protobuf-compiler-grpc, python3-grpcio
This adds support for the new USRP X420. It is the latest addition to
our X4xx series and comes with two HBX daughterboards. The HBX is
a one-channel direct conversion transceiver that covers 5G NR FR3 and
X-/KU-Band needs for Research and Prototyping in a frequency range of 10
MHz up to 20 GHz. It supports analog bandwidths of up to 1 GHz.
This commit is the MPM portion of the X420 support.
Co-authored-by: Martin Anderseck <martin.anderseck@emerson.com>
Co-authored-by: Ahmed Salman <ahmed.salman@emerson.com>
Co-authored-by: Max Köhler <max.koehler@emerson.com>
Co-authored-by: Markus Schill <markus.schill@emerson.com>
With this change, it is possible to have different sub-channels within
one channel to serve different purposes. For instance it is possible to
have one converter run in Real mode while two other converters serve I
and Q respectively. They all count as the same channel for UHD and the
driver software needs to take care to switch between them if required.
For the IQ architecture we also disable the NCO in the RFDC.
Upon reset, MPM Server tries to destroy the current instance of
Periperal Manager and re-create a new one.
Due to external references to bound methods, Peripheral Manager
and its member DioControl could not get destroyed which resulted
in reference leaks and file desciptor leaks.
This PR clears all external references to bound methods such that
PeripheralManager and its members are cleanly destroyed before
re-creation.
This fixes the "Too Many Open Files" when running 'uhd_image_loader'
repeatedly.
setproctitle is not available by default on our embedded systems. We therefore use it only if the module is installed and leave the title as is otherwise.
To get the process title one has to install setproctitle via pip and then restart MPM.
During debug, it is often helpful to distinguish the processes by
name. By default, Python processes inherit the name of the parent,
which is the name of the script executed. This change sets explicit
names for all our MPM processes.
When registering a signal, the current registered signal is removed.
This is often an unwanted behavior. Instead, one typically wants to
add code to the existing handler. This helper remembers the current
handler and calls it after the newly registered handler is executed.
Co-authored-by: Martin Anderseck <martin.anderseck@ni.com>
Importing from usrp_mpm.rpc_server imports a lot of dependencies which are
unnecessary if the rpc_server functionality is not actually needed.
Move the decorator functions to a new file rpc_utils.py and import only
from this file inside periph_manager.
usrp_hwd.py needs additional dependencies which are specifically imported:
- usrp_mpm.rpc_server.spawn_rpc_process
- usrp_mpm.discovery.spawn_discovery_process
This change ensures that ADC selfcal is triggered during session start
if the clocking was reconfigured. So it saves time if several
consecutive sessions are opened with always the same settings.
It also makes the ADC self cal run if in an open session either
set_clock_source(), set_time_source() or set_sync_source() are called
and a clock reconfiguration was triggered by that.
Currently the E320's integrated GPSDO can be disabled using the
"enable_gps" argment at MPM initialization time (e.g. when the radio
powers on and launches the MPM daemon). However, there is no way for
a UHD application to disable the GPSDO at connection time.
This patch allows the existing "enable_gps" arg to be accepted when
initiating an MPM session from UHD, such that an application can pass
the arg "enable_gps=0" at connection time to disable the GPSDO. The
default power state of the GPSDO is restored when the session ends.
Edit: On top of the original commit by draeman-synoptic, the MPM minor
compat number was also increased to denote new non-breaking feature
support.
Co-authored-by: mattprost <matt.prost@ni.com>
Signed-off-by: mattprost <matt.prost@ni.com>
This prevents any concurrency issues that might arise from killing and
spawning a new timer in the rpc server.
Signed-off-by: mattprost <matt.prost@ni.com>
This removes the race condition that can occur by reading the value of
the rpc server's shared state claim token value without a lock. After
the lock is released, it is possible that another process could preempt
and modify the value with a different claim operation. Also, added
timeouts to acquire the shared state lock for claim, reclaim, and
unclaim operations.
Signed-off-by: mattprost <matt.prost@ni.com>
The rpc_server_process was spawning with its parameters reversed. The
calling function also had its parameters reversed, so this is
functionally the same. This change makes sure that the parameters are
aligned with the arguments throughout the callstack.
Signed-off-by: mattprost <matt.prost@ni.com>
This reverts the Raw UDP host changes which is breaking
multiple interface streaming.
Commits addb81aa5d2d6adcd3a0c7a8d59fcd96af0c1ec4^..b1ca51f97aaa2226ed6ef339fb26fbea54ab7593
are reverted.
Reverts:
doc: Update manual on streaming
examples: Add remote RX streaming example
rfnoc: Enable remote UDP streaming
mpm: e320: Enable raw UDP streaming
mpm: n3xx: Enable raw UDP streaming
mpm: x4xx: Enable transport manager adapter
mpm: Enable opt-in transport adapter control
rfnoc: Transition stream managers and mgmt_portal to topo_graph_t
rfnoc: Add topology graph object
uhd/mpm: Add API to set remote routes
mpm: xports: Add XportAdapterMgr class
This is an addition to both PeriphManagerBase (MPM) and mb_iface (UHD).
Main changes:
- Addition of mb_iface::add_remote_chdr_route() and
mb_iface::get_chdr_xport_adapters()
- In X3x0, these APIs are stubbed out.
- In mpmd, these APIs are implemented and call the new MPM APIs (see
below)
- Addition of PeriphManagerBase.add_remote_chdr_route() and
PeriphManagerBase.get_chdr_xport_adapters()
- The PeriphManagerBase implements these APIs fully when the
'remote_udp_streaming' FPGA feature is detected.
- The MPM compat number is bumped to 4.3. UHD will continue to work with
lower compat numbers. It will query the compat number and act
accordingly.
Add the Filter API to n3xx specifically for the AD937x device. The TX
filter is limited to 32 taps, and the RX filter is limited to 48 taps.
This feature requires MPM version 4.2 or later on the device.
Co-authored-by: bpadalino <bpadalino@gmail.com>
Signed-off-by: mattprost <matt.prost@ni.com>
Allow users to control the Mykonos frontend bandwidth settings for
Rx and Tx. Note that this operation requires the daughterboard to
re-initialize, so it may take some time. Values for frontend filter
settings were derived using ADI's AD9371 Filter Wizard.
This feature requires MPM version 4.1 or later on the device.
Co-authored-by: bpadalino <bpadalino@gmail.com>
Signed-off-by: mattprost <matt.prost@ni.com>
MPM devices currently do not report their device name in the discovery
process. This adds the device name. After the change the device name
is reported, e.g. when using uhd_find_devices, like so:
$> uhd_find_devices
--------------------------------------------------
-- UHD Device 0
--------------------------------------------------
Device Address:
serial: DEADBEEF
claimed: False
fpga: X4_200
mgmt_addr: <mpm device ip>
name: <mpm device host name>
product: x410
type: x4xx
MPM server needs to be reclaimed in regular intervals. This is
monitored by the server using a timer. If the timer hits, the server
unclaims itself assuming the client process died for whatever
reason. In previous versions of `gevent.greenlet` the timer was
killed in a non blocking manner. This changed in version 0.13.0
(see
http://www.gevent.org/api/gevent.greenlet.html#gevent.Greenlet.kill)
which now leads to a dead lock in `timer.kill`. The kill command
is therefore now called explicitly with `block=False`.
Provide a way to safely reset the peripheral manager from uhd and as
a result, a mechanism to reload the fpga/dts components.
Signed-off-by: Virendra Kakade <virendra.kakade@ni.com>
Add a new image_loader argument delay_reload to provide a way to update
components but optionally delay the actual load.
Similarly add a new argument, just_reload, to enable uhd to reload
the fpga/dts components.
Signed-off-by: Virendra Kakade <virendra.kakade@ni.com>
Sequence is now:
1. Get _state lock
2. Kill reclaim timeout
3. Run deinit sequence
4. Clear claim token and session ID
5. Release _state lock
Before, we were not locking the mutex, and the timer was killed after
the deinit sequence. If the deinit sequence stalls for whatever reason,
that doesn't have to cause a claimer loss to be reported. UHD will
already have stopped the reclaim loop before unclaim() is called.
In the stall case, it would also have been possible the to acquire a new
claim while the deinit() is still running. This is prevented with the
combination of actually acquiring the mutex (like claim() and reclaim()
do) and moving the token/session ID clearing to the end.
MPM is Python3-only, but contains some remaining compatibility code for
Python2. Because this code requires extra dependencies (like six) and
could become obsolete in the future, we remove it to preempt that.
No functional changes.
This commit fixes a spelling mistake, moves logic for checking if a
connection is local to its own method, and calls gevent.signal.signal
instead of its deprecated alias gevent.signal.
Signed-off-by: Samuel O'Brien <sam.obrien@ni.com>
Sometimes when running usrp_hwd.py in a terminal and then canceling it
with Ctrl+C, it prints a really large stacktrace into the terminal
resulting from an uncaught gevent BlockingSwitchOutError. It seems like
there was an attempt to catch this in usrp_hwd.py:kill_time(). This
try-except was surrounding a call to Process.join() which, to the best
of my knowledge, can't ever throw this exception.
Based on my troubleshooting, this error comes from the SIGTERM signal
handler of the RPC process. The handler (defined in
rpc_server.py:_rpc_server_process), is just a direct call to
RPCServer.stop(). When the server's backed is a thread pool, this call
may block when joining the thread pool, causing gevent to complain about
execution attempting to block in a signal handler.
This commit resolves this issue by simply triggering an event in the
signal handler which prompts a different thread to clean up the server
and end the process.
Signed-off-by: Samuel O'Brien <sam.obrien@ni.com>
msgpack 0.6.1 suggests new default parameters which ensures compatibility
with the upcoming msgpack 1.0 release which will have breaking changes.
The parameter changes are described in
https://github.com/msgpack/msgpack-python/blob/v0.6.1/README.rst
The default parameters for msgpack 1.0 will be:
- packer: use_bin_type=True
- unpacker: raw=False
The packer use_bin_type=True option is already set in the client
(mpm_shell.py) but the unpacker option raw=False needs to be set
in the server (rpc_server.py)
This change allows the usage of a patched version of python3-mprpc
0.1.17 which removes passing the encoding option to the Packer and
Unpacker
Signed-off-by: Joerg Hofrichter <joerg.hofrichter@ni.com>
Msgpack version 0.6 reduced the default max buffer size to 1MB which is
smaller than the bitfiles. This change sets the max buffer size to 50MB
which is larger than the bitfiles.
Many small cleanups:
- Fix copyright headers
- Fix superfluous imports
- Pull some constants out of classes where appropriate
- Fix formatting
- Improve/fix some docstrings
- Disable specific Pylint warnings where appropriate
- Global catches use BaseException instead of Exception
- Don't use len() for empty checks
- Make sure to declare all self attributes in __init__ (note: this is
particularly of interest for E310, becuase its regular init happens
outside of __init__)
- Compacted some E310 code that had multi-DB checks
When updating a component like the FPGA, the timeouts for reclaiming get
disabled, because the update can potentially take a long time, during
which the RPC server might not be available.
There was a bug that didn't re-enable the timeouts. The most common case
where this causes issues was when the Ethernet connection was severed
during FPGA reloading, which could lead to UHD losing connection with
MPM altogether (for example because SFPs would come up with a different
IP address). In that case, MPM would remain unreachable until the next
reboot.
- Turns the E310 into an MPM device (like N3xx, E320)
- Factor out common code between E320 and E310, maximize sharing between
the two devices
- Remove all pre-MPM E310 code that is no longer needed
- Modify MPM to remove all existing overlays before applying new ones
(this is necessary to enable idle image mode for E310)
Co-authored-by: Virendra Kakade <virendra.kakade@ni.com>
Signed-off-by: Virendra Kakade <virendra.kakade@ni.com>
