-Add gevent rtnetlink listener for interface and IP changes.
-Refresh cached transport info immediately after network changes.
-Extend x4xx cache TTL to 300 seconds with epoch-based invalidation.
get_xport_info() enumerates network interfaces via netlink/pyudev on every
GetDeviceInfo call (~350ms, non-yielding). Since GetDeviceInfo is
unclaimed and gets flooded during discovery, this could starve claimed
RPCs past their short deadlines like 2 seconds.
Cache the result for 10s (X400_XPORT_INFO_CACHE_TTL); it's stable per
session, and init() drops the cache since IPs may change. This turns most
GetDeviceInfo calls much faster.
Add a dedicated lock for the xport_info_cache so concurrent
GetDeviceInfo calls don't race on refresh/invalidations
pb_utils.py contains several helpers to translate between
Protobuf and Python datatypes/messages including other utility
functions. This change adds a bunch of unit tests to cover
various aspects of its usage by rpc_server.py.
Some blocks (e.g., the fosphor block) don't have parameters in their
respective YAMLs, but the GRC file importer makes the assumption that
there are always parameters, and will crash on a faulty dictionary
lookup when that's not the case. This conditionally checks parameters,
making the lack of parameters non-fatal.
The current firmware missed descriptors for B200-mini and
B205-mini. Added the missing descriptors.
Also this commit ammends the firmware howto. It specifies
the SDK version we are using as well as some additional
command applied to the SDK files to make the patch successful.
Co-authored-by: Lars Amsel <lars.amsel@emerson.com>
This is an upgrade to the latest FX3 SDK (1.3.5) and to the ARM gcc in
version 15.3.rel1 for the build of the B2xx firmware and bootloader. All
required steps are explained in the updated readme file.
Additionally, some bounds checks were added both in the firmware and the
bootloader, as well as some more validations and checks to prevent over-
and underflows. Newly introduced checks were documented in the code.
Update MPM GPIO handling for the updated X4xx GPIO SPI register
interface.
Require the FPGA compatibility number bump introduced with the new
SPI support so MPM does not operate against an incompatible FPGA image.
Bump MPM compat number.
Introduce 64-bit SPI transaction methods into the common spi_iface API,
default implementation throws an uhd::unsupported_operation exception.
Add explicit 32-bit aliases for the old/existing SPI transactions
methods.
Expose the width-specific operations through the RFNoC Python bindings.
Extend GPIO SPI transactions up to 64 bits and update the X4xx FPGA
register map and RTL to support the new access width.
Migrate the GPIO SPI register definitions to SystemVerilog.
Add testbench for X4xx GPIO SPI module covering both exisiting
32bit functionality and added 64bit functionality.
Separate TB Makefiles for X4xx and X410 implementation variant.
Bump the FPGA compatibility number so host and MPM software can reject
images that do not provide the required GPIO SPI register interface.
Extend the SPI slave model to propagate simulation failures to the test
execution framework.
Extract the timing monitor from the control-port SPI testbench so it
can be shared by the GPIO SPI simulations while keeping timing checks
centralized and consistent.
msg_task_impl's constructor waits on _start_cv until _running is set,
checking the predicate under _mutex. task_loop() set _running and
called notify_one() without taking _mutex, so a notification sent
after the constructor had checked the predicate but before it started
to wait was lost, and the constructor then blocked forever.
The condition-variable handshake replaced a boost::barrier, which
cannot lose a wakeup, in 3855516f6f ("lib: Remove boost::thread usage
from USB transports"), so UHD 4.10.0.0 and 4.11.0.0 are affected. On
a B210 the hang leaves b200_impl's constructor blocked in
msg_task::make() right after "Operating over USB 3.". It shows up
most when the device is opened at a real-time priority, as SDR
applications that raise their priority before opening the device do.
Set _running while holding _mutex, then notify.
Treat a RuntimeError raised while the periph manager is resetting as
transient. Log once at WARNING and abort with gRPC status UNAVAILABLE
so the client retries, instead of logging a scary ERROR.
Real runtime errors still log ERROR and abort when periph manager
object is live.
usrp_hwd runs all gRPC greenlets on a single gevent hub, so any
non-yielding C call blocks every other RPC. Under a uhd_find_devices
discovery flood, the GetDeviceInfo interface enumeration run long enough
And may starve most claimed RPC past its default 2s client deadline,
causing "Deadline Exceeded" failures.
Three places that this commit optimizes, all in net.py:
- get_local_ip_addrs() used pyroute2 IPDB, which spawns helper threads
and mirrors the full link/addr/route/neighbour tables on every call
(~100s of ms, non-yielding) --> Replace it with a single stateless
IPRoute get_addr() dump. Keep IFA_LOCAL (falling back to IFA_ADDRESS)
so the returned set matches IPDB's behaviour, including point-to-point
links.
- get_link_speed() used to create a new pyudev.Context() per interface,
which forks on creation, tripping gRPC's fork handlers. gRPC then
freezes all workder threads and restarts after fork is done. Fix is to
reuse one PID-guarded pyudev context per process.
- get_valid_interfaces() did an O(N) netlink round-trip per interface.
This stalls gRPC server. --> Dump all links and IPv4 addresses once
and filter in Python.
IPRoute stays per-call: it opens its own netlink socket and does not
fork, so there is no shared state to lock.
