1 Commits

Author	SHA1	Message	Date
Carlos Valdes	4a1ff18f57	feat: add Nordic nRF54L15-DK variant (Zephyr + BLE + LoRa) (#10193) ... * feat: add Nordic nRF54L15-DK variant (Zephyr + BLE + LoRa) Adds a community hardware variant for the Nordic nRF54L15-DK (PCA10156) with an external EBYTE E22-900M30S (SX1262) LoRa module. First Meshtastic port running on the Zephyr RTOS; all other Nordic targets use the nRF5 SoftDevice stack. Scope ----- - New Zephyr-based platform layer under src/platform/nrf54l15/ providing Arduino-compatible shims (Arduino.h, SPI, Wire, Print, Stream) over the Zephyr APIs plus a LittleFS-backed InternalFileSystem on SPIM20. - Bluetooth LE peripheral (NRF54L15Bluetooth.*) built on the Zephyr BT host stack, exposing the Meshtastic GATT service with legacy connectable advertising, just-works pairing, dynamic MTU exchange (up to 247 bytes), and iOS connection-parameter tweaks. - Variant directory variants/nrf54l15/nrf54l15dk/ with pin map for the E22 module on connector J1, PlatformIO env (nrf54l15dk), Zephyr DT overlay and a wiring README. - Zephyr project config (zephyr/prj.conf + board overlay) tuned for BT + LoRa: 16 KB main stack, 4 KB BT RX thread, RTT logging in immediate mode, newlib-nano heap sized to leave room for the GATT pools while still allowing ATT MTU=247. - extra_scripts/nrf54l15_linker.py works around a PlatformIO + old Ninja issue where Zephyr's two-pass linker script generation does not run automatically; the post-script parses build.ninja and invokes the gcc -E step directly before the final link. - boards/nrf54l15dk.json board definition (PlatformIO needs it for the DK; the Seeed platform only ships the XIAO variants). - variants/rp2350/rp2350.ini excludes platform/nrf54l15/ from RP2350 build_src_filter so the shared platform tree does not leak between targets. - .gitignore: add nRF J-Link / RTT debug artifacts (flash.jlink, rtt_*.txt). Shared source changes --------------------- - src/main.{cpp,h}, src/RedirectablePrint.cpp, src/FSCommon.{cpp,h}, src/mesh/{Channels,NodeDB,RadioLibInterface,MeshService,PhoneAPI}.cpp, src/mesh/RadioLibInterface.h, src/modules/AdminModule.cpp: add small guards / helpers so the Zephyr build compiles alongside the Arduino targets. Behavior on existing boards is unchanged. Hardware model -------------- HW_VENDOR maps to meshtastic_HardwareModel_PRIVATE_HW until a dedicated protobuf enum value is assigned upstream. The variant declares custom_meshtastic_hw_model = 132 so the maintainers can wire the new enum value through the protobufs repo after merge. Hardware note ------------- The E22-900M30S does not connect its DIO2 pin to TXEN internally — a wire/solder bridge between DIO2 and TXEN on the module is required for TX to work. Details and full pin map are in the variant README. Validation ---------- Built clean against develop. On real hardware (April 2026) the port passes end-to-end: iOS companion app pairs and connects, configuration round-trip works, LoRa TX/RX reaches a canonical tbeam on the same mesh channel, NodeDB updates propagate both ways, and traceroute completes. * fix(nrf54l15): use atomic fs_rename instead of copy fallback Zephyr LittleFS on nrf54l15 supports fs_rename natively, so route it through the same atomic path as ESP32. The previous copyFile+remove fallback truncated the destination before copying, leaving 0-byte files if interrupted mid-write. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): expand storage_partition from 36KB to 700KB LittleFS on the default 9-block (36KB) storage_partition ran out of space during copy-on-write of config.proto, causing fs_write to return ENOSPC and pb_encode to surface "io error" when saving configuration via the mobile app. Reclaim slot1_partition (the MCUboot secondary slot — unused since we flash directly via J-Link) and grow storage_partition to span 0xb6000..0x165000 (~175 blocks). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): drop USERPREFS_LORACONFIG_* so LoRa config stays mutable NodeDB rewrites LoRa config from USERPREFS_LORACONFIG_* on every boot, which prevented reconfiguration via the BLE/serial app. Drop the variant-level defaults; users configure region and modem preset through the app like every other variant. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): enforce MITM passkey pairing on GATT service - Add MESH_PERM_READ/MESH_PERM_WRITE macros (READ_AUTHEN/WRITE_AUTHEN) on all mesh service characteristics so clients must complete passkey exchange before accessing fromNum/fromRadio/toRadio/logRadio. - Wire FIXED_PIN mode to bt_passkey_set() so the device advertises a known PIN (config.bluetooth.fixed_pin); RANDOM_PIN keeps default per-pairing random passkey. - Reduce BleDeferredThread HARD_WATCHDOG_MS from 3min to 1min. - prj.conf: CONFIG_BT_SMP_ENFORCE_MITM=y, CONFIG_BT_FIXED_PASSKEY=y, CONFIG_BT_SMP_SC_PAIR_ONLY=n (legacy fallback for clients that abort SC pairing with reason 0x01 within 150ms). * fix(nrf54l15): resolve develop-merge conflict + cppcheck warnings The `Merge branch 'develop'` left two ~RadioLibInterface() declarations in src/mesh/RadioLibInterface.h: the inline version added upstream by PR #10254 (which independently applied the same UAF guard this PR was carrying) and the out-of-line version this PR introduced. GCC rejects the duplicate, breaking every platform build. Drop the out-of-line declaration + definition; keep upstream's inline form. Also silence the 13 cppcheck low warnings introduced by the new nrf54l15 Arduino shim — Arduino's `String`/`SPISettings` API contract relies on implicit single-arg constructors used pervasively by existing Meshtastic code, so suppress `noExplicitConstructor` inline with a comment instead of breaking the API. The few mechanical wins (`const tmp[2]`, `const uint32_t *sp`) are applied directly. * fmt: fix Trunk Check lint issues on nrf54l15-port - extra_scripts/nrf54l15_linker.py: move regular imports above Import("env") to silence E402, add trunk-ignore-all(F821) for the PIO/SCons SConstruct injection (matches esp32_pre.py / nrf52_extra.py convention) - src/platform/nrf54l15/NRF54L15Bluetooth.cpp: clang-format 16.0.3 - boards/nrf54l15dk.json + variants/nrf54l15/nrf54l15dk/README.md: prettier 3.8.3 (also resolves markdownlint MD060 on README tables) No behavior change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): address Copilot review comments + correct clang-format style Six review threads from the 2026-04-30 Copilot review: - src/platform/nrf54l15/nrf54l15_main.cpp: validate PSP against the nRF54L15 SRAM range (0x20000000..0x20040000) and 4-byte alignment before walking the faulting thread's stack, and clamp the walk so it never reads past the end of RAM. Prevents a second fault inside the fatal handler when PSP is corrupted (common in real faults). - src/platform/nrf54l15/nrf54l15_arduino.cpp: gate the bring-up printk traces in digitalWrite/digitalRead (CS/NRESET toggle log, BUSY-before-NRESET snapshot, BUSY periodic timeline) behind a new -DNRF54L15_GPIO_DEBUG flag that is off by default. The "dev NOT READY" message stays unconditional — it indicates a genuine hardware/DTS misconfig. - src/modules/AdminModule.cpp: don't mutate config.device.output_gpio_enabled from handleGetConfig(). Reflect the live pin state in the response payload only — a getter must not write back to disk-persisted state. - src/platform/nrf54l15/InternalFileSystem.h: derive totalBytes() from FIXED_PARTITION_SIZE(storage_partition) at compile time so it tracks the DK overlay's ~700 KB partition instead of the stale 36 KB hard-coded value. Updated the file header comment accordingly. - extra_scripts/nrf54l15_linker.py: make _extract_gcc_command() handle the POSIX Ninja COMMAND format (no `cmd.exe /C "..."` wrapper) in addition to the Windows form, so the script doesn't hard-fail on Linux/macOS hosts. - src/platform/nrf54l15/NRF54L15Bluetooth.cpp: clamp NO_PIN to RANDOM_PIN with a one-shot LOG_WARN. The mesh GATT permissions are declared with BT_GATT_PERM_*_AUTHEN and prj.conf sets CONFIG_BT_SMP_ENFORCE_MITM=y, so NO_PIN with no auth callbacks would leave every characteristic returning BT_ATT_ERR_AUTHENTICATION. Falling back to RANDOM_PIN keeps the link usable instead of silently broken. Also re-formatted this file with the project's .trunk/configs/.clang-format (Linux braces, 4-space indent, 130-col) — the previous lint-fix commit a2aca3234 accidentally used the default LLVM style here, which CI's clang-format would have rejected. Build verified: pio run -e nrf54l15dk passes, RAM 47.4%, Flash 28.6%. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): address remaining Copilot review threads Round 2/3 review fixes — bugs first, then docs/portability: BLE concurrency (NRF54L15Bluetooth.cpp): - onNowHasData / sendLog / BleDeferredThread / shutdown: acquire active_conn under ble_mutex via new acquire_active_conn() helper so disconnected_cb can't free the conn between the null check and bt_conn_ref/bt_gatt_notify (use-after-unref). - write_toradio: reject writes that exceed MAX_TO_FROM_RADIO_SIZE with ATT_ERR_INVALID_ATTRIBUTE_LEN instead of returning success and silently dropping the payload (would hide failed config writes from the phone). - start_advertising: truncate the device name to fit the 31-byte legacy scan-response limit and switch to BT_DATA_NAME_SHORTENED so bt_le_adv_start() doesn't reject the payload when the name approaches CONFIG_BT_DEVICE_NAME_MAX=32. Linker / portability: - main.h: drop the rp2040Loop() forward declaration that had no definition and no callers — would surface as a link error if any RP2040 build added a call to the symbol. - nrf54l15_arduino.cpp: transfer16() now uses static __aligned(4) DMA buffers (matching transfer()), removing the EasyDMA-reachability hazard of caller-stack buffers on this part. Filesystem (InternalFileSystem.h): - usedBytes(): return real usage from fs_statvfs() instead of 0 so OTA / range-test free-space guards work. - rewindDirectory(): close the dir before reopening — Zephyr fs_dir_t has no rewind, and re-fs_opendir on an open handle leaks LittleFS state. Crash handler (nrf54l15_main.cpp): - After the stack walk, busy-wait 50 ms to flush RTT/printk and call sys_reboot(SYS_REBOOT_COLD) directly so the saved_crash record is actually reported on the next boot. Default Zephyr config has RESET_ON_FATAL_ERROR=n, so the previous k_fatal_halt() spun forever. Generalization / config: - PhoneAPI.cpp: replace the NRF54L15_DK ifdef with a MESHTASTIC_EXCLUDE_FILES_MANIFEST capability flag (defined in the nrf54l15dk env) so future variants can opt in/out without touching shared code. - variants/nrf54l15/nrf54l15.ini: parameterize libdeps include paths via ${PIOENV} so additional nRF54L15 envs sharing nrf54l15_base don't break. - prj.conf: drop the stale "36 KB storage_partition" comments — the DK overlay reclaims slot1 to ~700 KB and runtime size comes from FIXED_PARTITION_SIZE. - nrf54l15dk overlay: remove the zephyr,console / zephyr,shell-uart chosen entries that conflicted with CONFIG_UART_CONSOLE=n + RTT console; keep uart30 enabled so swapping the console is one Kconfig flip away. Build: nrf54l15dk SUCCESS (flash 28.6%, RAM 47.4%); wiznet_5500_evb_pico2 SUCCESS (verifies the shared main.h change). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15dk): use PRIVATE_HW (255) for custom_meshtastic_hw_model Per @thebentern's review: the nRF54L15-DK is a development kit, not a canonical Meshtastic SKU, so it falls under HardwareModel::PRIVATE_HW (255) — the same enum value already used at runtime via HW_VENDOR. The placeholder 132 is removed; no dedicated enum number will be assigned for DK boards. Slug stays NRF54L15_DK as a human-readable identifier. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(nrf54l15): unstarve bt_long_wq so SC pairing completes bt_pub_key_gen() runs the ECC P256 key generation on bt_long_wq. At default prio=10 (preemptible) and stack=1400 it gets starved by Meshtastic app threads at boot — sc_public_key stays NULL for minutes, smp_public_key() defers with SMP_FLAG_PKEY_SEND, and every SC pairing attempt stalls right after the public-key exchange. iOS shows "Connecting…" forever with no PIN prompt; bleak/CLI fails the first CCC notify write with "Protocol Error 0x05: Insufficient Authentication". Set CONFIG_BT_LONG_WQ_PRIO=0 (highest preemptible, ties with main) and CONFIG_BT_LONG_WQ_STACK_SIZE=4096 (margin for the P256M driver frames). Validated E2E with iOS Meshtastic app: bt_smp_pkey_ready fires within ~40 s of boot, 20 SC Passkey Entry rounds complete with matching pcnf/cfm, encrypt 0x01 / sec_level 0x04 (Authenticated MITM), bonded=1. * feat(nrf54l15): hardware I2C bus via TWIM30 + sensor telemetry Adds the Arduino TwoWire layer for the nRF54L15-DK so Meshtastic's sensor drivers can talk to external I2C devices over the hardware TWIM30 peripheral. Bus binding: - &uart30 disabled in the board overlay (peripheral instance 30 is shared between UARTE30 / TWIM30 / SPIM30 — pick one). Console stays on RTT via CONFIG_RTT_CONSOLE. - New i2c30_default / i2c30_sleep pinctrl with SDA=P0.03 / SCL=P0.04. External 4.7 kOhm pull-ups required on both lines. - &i2c30 enabled at I2C_BITRATE_FAST (400 kHz). - button_3 (SW3, P0.04) deleted from DTS so the pad can be claimed by i2c30 pinctrl; SW3 is still wired to the pad on the DK, do not press it during I2C use or it will short SCL to GND. Arduino layer: - src/platform/nrf54l15/Wire.cpp resolves the DT node at compile time via DEVICE_DT_GET(DT_NODELABEL(i2c30)) and dispatches Arduino's beginTransmission / write / endTransmission / requestFrom to i2c_write / i2c_write_read / i2c_read. Buffer is sized to 256 bytes for forward compatibility with the SE050 secure element on the custom PCB. - Wire.h drops the prior compile-only stubs and exposes the real TwoWire surface. - Arduino.h: BitOrder becomes an enum (not #define) so Adafruit_BusIO's `typedef BitOrder BusIOBitOrder;` compiles. Variant + build flags: - nrf54l15.ini flips HAS_WIRE / HAS_SENSOR / HAS_TELEMETRY from 0 to 1 and cherry-picks the sensor libs Meshtastic needs (BusIO, Sensor, BMP280, BME280, INA219/226/260/3221, SHT4X). The full environmental_base group is avoided because it pulls Adafruit_SSD1306 / Adafruit_GFX which rely on Arduino pin macros the Zephyr shim does not implement. - nrf54l15dk variant.h defines PIN_WIRE_SDA / PIN_WIRE_SCL for parity with the Arduino convention used by other variants. The actual bus wiring is fixed by the overlay pinctrl above. Validated 2026-05-14/15 on the DK with BMP280 @ 0x76 (temperature + barometric pressure) and INA3221 @ 0x42 (rail voltage / current); EnvironmentTelemetry / PowerTelemetry packets transmit successfully over LoRa. Footprint cost on nrf54l15dk: +45 KB flash, +1.7 KB RAM. * feat(nodedb): honor USERPREFS for environment telemetry on first boot installDefaultConfig() now respects two new compile-time prefs: USERPREFS_CONFIG_ENV_TELEM_UPDATE_INTERVAL USERPREFS_CONFIG_ENVIRONMENT_MEASUREMENT_ENABLED The mobile apps enforce a 30 min floor on environment_update_interval in the settings UI, which makes short-interval bring-up testing of new sensor hardware painful — you have to wait half an hour for the first LoRa packet to confirm wiring + driver. With these prefs baked into the variant, the firmware can ship a freshly-flashed device that broadcasts on a shorter cadence (e.g. 900 s) the moment storage_partition is empty. Both prefs are gated on #ifdef so the behavior is unchanged for any variant that does not opt in. Documented in userPrefs.jsonc with the existing telemetry-interval pref block. * fix(nrf54l15): allow multiple bonded BLE peers CONFIG_BT_MAX_PAIRED defaults to 1, so once the first peer (e.g. an iOS phone) has paired and bonded, every subsequent pairing attempt from a different MAC fails inside bt_keys_get_addr() with no free key slot — the host returns BT_SECURITY_ERR_KEY_DOES_NOT_EXIST and the second peer never gets past SMP. Raise the slot count to 4 so the device can simultaneously hold an iOS phone, a Windows host, a Linux host, and one spare bond. Add BT_KEYS_OVERWRITE_OLDEST so that once the table fills, the LRU peer is evicted on the next pairing rather than rejecting the new peer. This matches the behavior other Meshtastic ports already provide (nRF52 uses CONFIG_BT_PERIPHERAL_PRIO_CONN with similar semantics). Discovered while bringing up the Python CLI on Windows alongside the existing iOS bond. * fix(nrf54l15): zero-initialize TwoWire buffers + clang-format Wire cppcheck on every CI target (esp32s3, rp2040, rp2350, nrf52840, ...) was failing the build with two `uninitMemberVar` warnings on TwoWire's constructor: `txBuf` and `rxBuf` (256-byte arrays) were not initialized. Even though the buffers are only read after txLen/rxLen is set, leaving them uninitialized is a footgun if any future caller bypasses the len-set step. Use C++11 value-initialization in the member initializer list — costs ~512 B of memset at boot, gains a clean cppcheck pass and defensive-against-future-changes semantics. Also reformat Wire.{cpp,h} with the project's `.trunk/configs/.clang-format` config so the Trunk Check Runner passes — clang-format moved the `<errno.h>` include before the Zephyr-namespaced ones in Wire.cpp and collapsed two inline overloads to single lines in Wire.h. * fix(AdminModule): remove dead OUTPUT_GPIO_PIN/GpioOutputModule references OUTPUT_GPIO_PIN is never defined and modules/GpioOutputModule.h doesn't exist in the codebase; all #ifdef OUTPUT_GPIO_PIN branches were dead code introduced by the nRF54L15-DK variant commit. Strips the include, the output_gpio_enabled OFF→ON/ON→OFF transition logic in handleSetConfig(), and the digitalRead() reflection in handleGetConfig(). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: Ben Meadors <benmmeadors@gmail.com>	2026-05-16 06:16:11 -05:00