Skip to content

Fix/missing libzstd on nix - #3890

Open
hmaier-dev wants to merge 221 commits into
meshcore-dev:mainfrom
hmaier-dev:fix/missing-libzstd-on-nix
Open

hmaier-dev wants to merge 221 commits into
meshcore-dev:mainfrom
hmaier-dev:fix/missing-libzstd-on-nix

Conversation

@hmaier-dev

Copy link
Copy Markdown

Enables nRF53 development in nix environment

Quency-D and others added 30 commits July 22, 2026 16:33
M5Stack_Unit_C6L_kiss_modem env was missing ARDUINO_USB_CDC_ON_BOOT,
ARDUINO_USB_MODE, and ENABLE_USB_INTERFACE build flags, present in
the companion_radio_usb env for the same board. Without these,
ESP32-C6 native USB serial output goes to UART0 (unconnected on this
board), making the KISS modem unresponsive over USB.

Hardware-verified: M5Stack Unit C6L, KISS modem confirmed operational
via /dev/ttyACM0 with openhop_repeater.
A RAK4631 core paired with a RAK13302 leaves the core's own SX1262
powered but never initialized, idling in STDBY_RC for the life of the
board. Firmware cannot cut its supply -- P1.05 is the RF switch rail,
not the radio's -- so bring it up on its own pins just long enough to
issue a cold sleep, before std_init() repins SPI onto the RAK13302.

Gated behind RAK4631_ONBOARD_SX1262_SLEEP and exposed through seven
RAK_4631_13302_* envs, so RAK_3401_* artifacts are unchanged.

The sleep is issued regardless of what begin() returns: Module::init()
runs first, so SPI stays usable even if TCXO calibration fails. NSS is
then held high, since the SX1262 wakes on an NSS falling edge.

Verified on RAK19007 + RAK4631 + RAK13302: begin=0 sleep=0, RAK13302
init clean (device_errors=0x0000), TX and RX both confirmed. The
current saving itself is derived from the datasheet (600 uA STDBY_RC
-> 160 nA cold sleep) and is not measured.
BUSY is driven low in STDBY_RC and high while asleep, so reading it either
side of the sleep gives a non-invasive readback of the state the chip was
found in and the state it was left in. Confirmed on hardware.

This is what makes the sleep observable at all. The only other symptom of
a failed sleep is ~600 uA, which is below what an inline USB meter can
resolve, so without this a regression would be silent.

It immediately earned itself: busy_before reads 0 on every boot, showing
the cold sleep does not survive an MCU reset. GPIOs revert to input during
reset, NSS stops being driven high, and the falling edge wakes the chip.
The helper re-sleeps it, so this is self-healing, but it means the chip is
never found already asleep.
Add SetHardware sub-commands so a host can probe and configure
per-board RF front-end settings:

- GetCapabilities (0x1B): uint32 feature bits (AGC, FEM RX, FEM TX)
- Set/GetAgcResetInterval (0x1C/0x1D): 0-1020 s, 4 s resolution,
  0 disables; default stays 30 s
- Set/GetFemState (0x1E/0x1F): apply/value masks for FEM RX (LNA)
  and TX (PA) gain; all-or-nothing, Unsupported (0x08) if any
  requested bit is not controllable on this board

The AGC scheduler moves from main.cpp into KissModem so it is
configurable and unit-tested. FEM changes that arrive mid-TX are
deferred until the packet ends, since board setters drive FEM pins
directly.

MainBoard gains virtual FEM hooks (canControl/set/is for LNA and PA
gain). Boards that already implemented these methods privately now
expose them as overrides: Heltec V4, V4 R8, T096, Tracker V2,
Station G3 and Wio Tracker L1 1W.

KISS_FIRMWARE_VERSION bumped to 2.
USBCDC (ARDUINO_USB_MODE=0, TinyUSB) has no setTxBufferSize, which broke
the Station G3 and Heltec Tracker V2 KISS builds.
…ntrol

- Wio Tracker L1 1W no longer advertises TX gain control: its setter
  drives LORA_PWR_EN, the shared LDO for the SX1262 and the PA, so
  "gain off" would power down the radio.
- A SetFemState that arrives mid-TX is now applied after TxDone is
  queued, so its reply follows TxDone.
- SetFemState replies (immediate or deferred) are retained until they
  fit in the host output queue instead of being dropped, so a change
  that took effect is always acknowledged.
- GetFemState replies queue behind an outstanding SetFemState reply;
  a further SetFemState returns TxBusy until it clears.
Add SetRxBoostedGain (0x20) / GetRxBoostedGain (0x21, reply 0xA1) and
capability bit 3, the KISS equivalent of the radio.rxgain CLI setting.
It is separate from FEM RX gain; a board may support either or both.

- Support is declared by the driver via a new
  RadioLibWrapper::supportsRxBoostedGain() (SX1262, SX1268, LLCC68,
  LR1110, LR1121, LR2021), so nothing is written to the radio at boot.
- The set is refused with TxBusy during TX or while host output is
  backed up. Otherwise any completed RX packet is delivered first,
  because some drivers (LR2021) drop to standby to apply the gain.
- The reply always carries the state read back from the radio.

Driver fixes found along the way, which also affect the radio.rxgain
CLI in regular firmware:
- CustomLR2021 never updated its cached gain state, so the getter
  always reported off. It now tracks the applied level per RX path
  and rolls RadioLib's cached level back if a write fails.
- CustomLR1110/LR1121 cached the requested state even when the write
  failed.
- lr11x0ResetAGC only reapplied the gain when RX_BOOSTED_GAIN was
  defined, so a runtime setting was lost on the next AGC reset.
Boards own their FEM control through handleCommand() ("get/set
radio.fem.rxgain|txgain"). Use that instead of a parallel set of FEM
virtuals on MainBoard, so any board that implements the commands works
in KISS with no extra code.

- Remove the six FEM virtuals from MainBoard and the board overrides.
- KissModem talks to the board in one adapter. A gain counts as
  controllable only if "get" replies exactly "> on" or "> off". "set"
  is checked by reading the state back, never by its reply wording.
- SetFemState always replies with the read-back state once the
  requested bits pass the capability check.
- Guard the boards' prefs writes against a NULL store. KISS attaches
  no prefs, so "set radio.fem.*" would dereference NULL.
- Wio Tracker L1 1W TX gain is exposed again, matching its CLI.
A second SetFemState sent while the first was held for a TX got an
immediate TxBusy, which reached the host before the first request's
reply. With no request IDs, the host could not tell which Set failed.

FEM requests now go through a small ordered queue (Set, Get and their
error replies). Sets still wait for TxDone to be queued and apply
once. Each reply is kept until it can be queued. Capability checks
happen in order at processing time. Only an overflow (more than 4
waiting) replies TxBusy immediately.
The AGC interval is set in 4 s steps, matching MeshCore's
agc.reset.interval, so the old 30 s default could not be restored by a
host: setting 30 gives 28. Found on hardware. Use 32 s, the nearest
multiple of 4 above it.
sx126xResetAGC only restored the gain when the build defined
SX126X_RX_BOOSTED_GAIN, so a gain set at runtime was lost on reset.
Match lr11x0ResetAGC and always reapply. Also pass the gain state from
the LLCC68 wrapper, whose call did not match the helper's signature.
Housekeeping: removed unnecessary space
Use the existing RAK3401 firmware roles for RAK4631 cores paired with a
RAK13302. Probe the optional onboard radio and put it into cold sleep
before handing SPI to the primary radio, removing seven duplicate build
environments.

Preserve BUSY startup readiness, release probe pulls before sleep, and
disconnect unused GPIO input buffers during cleanup. Add focused tests
for sequencing, missing signatures, timeouts, and failed sleep readback.

Validated with four native tests, seven firmware-role builds, and repeater
boot checks on RAK4631 + RAK13302. Supply current was not measured.
Fix: tz_offset should allow for half hour timezone offsets
…ditional-display

Remove BLE screen when BLE not available
Move the onboard radio probe into RAK3401Board::begin() so a low-voltage
SYSTEMOFF at boot no longer leaves the RAK4631's onboard SX1262 in
standby. Add board startup tests and run them in CI.
netrunnereve and others added 29 commits November 14, 2026 15:17
RakLocationProvider wraps a MicroNMEALocationProvider but did not override
setFirstFixTimeout(), getFirstFixTimeout() or formatDiagnostics(), so on
RAK4631 boards with a serial GPS:
- `gps firstfix` always answered "0 (no limit)" (the LocationProvider
  default), while the limit in force was the build default (1 h);
- `gps firstfix <n>` saved the value but never applied it, at once or at boot;
- `gps diag` gave the short base form, without the UART/NMEA counters or fm:.

Forward all three to the NMEA provider. The RAK12500 (I2C) path has no
first-fix limit, so it keeps reporting 0 and the short diagnostics.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…e sync

In periodic mode a window closed after GPS_FIX_SETTLE_MS (10 s) from the
first fix even if the clock sync was still pending. The sync needs about
4 s of continuous valid fixes (time_valid > 2, reset on any pass without
a fix), so after a hot start, when the receiver reports a fix within
seconds but not yet steadily, the window could close unsynced. On a
board without an RTC the clock then stays at its default until the next
window, a whole period later.

Close the window only once the sync has been tried (done, not needed, or
refused by GPS_GUARD); the settle time and the satellite count apply
after that. The acquisition timeout still bounds the window.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…re-dev#3726)

* Initial attempt at fixing serial issue on Xiao ESP32-S3 Wio (meshcore-dev#1416)

* restrict Xiao ESP32-S3 WIO fix to just that board
…settle

gps: don't let the settle time close a periodic window before the time sync
…-diag

rak4631: forward the first-fix limit and gps diag to the serial GPS
Allow the user to set a custom bluetooth name using the cli
* added BaseUITask for common behaviours
* ui_task now instantiated in one of the variant's target-*.cpp modules
* only select variants .ini config refactored so far (Heltec V3, MeshTracker V1, LilyGo T-Echo Card)
…core-error-flags

docs: describe stats-core output and its error flags

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.