all: Add the Morse Micro Wi-Fi HaLow driver. by kwagyeman · Pull Request #1 · micropython/mm-halow-driver

kwagyeman · 2026-09-14T18:23:28Z

Summary

The drivers/halow/ content from micropython/micropython#19617, reorganised
along the lines of cyw43-driver as discussed there:

src/ -- the driver, with the C API renamed mm_halow_* and all
configuration macros MM_HALOW_*. Everything the driver previously took
from MicroPython (machine-SPI, the GC-heap root for its private heap, the
pin-interrupt callback, atomics/ticks/errnos/printf) now arrives through
the mm_halow_config.h / mm_halow_configport.h contract: the embedder
supplies atomic sections, a millisecond tick, pin accessors,
interrupt-context detection, and the mm_halow_port_* functions (SPI
transport, heap backing, hardware RNG, fallback MAC, optional edge IRQ).
mmport.h keeps its name -- morselib's own headers include it.
lib/mm-iot-sdk -- the Morse Micro MM-IoT-SDK as a submodule, pinned to
the exact version the hardware numbers in the MicroPython PR were measured
with (2.12.3).
mm_halow.mk -- build fragment for make-based embedders: prebuilt-morselib
linkage (grouped with the target multilib's libc/libm for -nostdlib
hosts) and firmware/BCF blob embedding.
tests/host -- the allocator suite compiled natively from the real
sources (233 checks), plus an ASan/UBSan variant.
tests/qemu -- the scheduler suite on a Cortex-M55 under QEMU (51
checks); the context switch is naked asm, so it can't run on the host.
tests/hardware -- the on-device throughput/soak harness the driver was
validated with (20-minute TCP/UDP soaks against a HaLow AP), run by hand.
.github/workflows/test.yml -- runs the host and QEMU suites in CI.

Testing

Both suites pass locally: host with gcc -m32 and under ASan/UBSan, QEMU
with arm-none-eabi-gcc 14.3.1 on mps3-an547 (Cortex-M55 needs GCC >= 14).
The MicroPython-side integration (submodule + configport + port glue) is the
follow-up, after which the N6/AE3/RT1062 hardware gauntlet re-runs.

kwagyeman · 2026-09-14T18:39:24Z

@dpgeorge

dpgeorge · 2026-09-14T23:23:38Z

Looks great!

I'd like to add more CI: code formatting (C and Python), code spell, building under clang. And putting CI related functions in tools/ci.sh like how cyw43-driver does it. I'm happy to do that myself after this PR is merged, but if you want to add it here, please do so.

kwagyeman · 2026-09-14T23:33:58Z

I can add. Give me a second.

kwagyeman · 2026-09-14T23:58:22Z

Almost done, adding a QEMU build for the device and not the host.

dpgeorge · 2026-09-15T00:03:49Z

No such thing as too much CI 😂

dpgeorge · 2026-09-15T00:07:19Z

The CI added here doesn't seem to be running and I'm not sure why. It should work...

kwagyeman · 2026-09-15T00:08:48Z

@dpgeorge - Okay, finished. I think the PR needs to be merged first, as the CI for the PR doesn't run on itself.

Break this PR into CI edits and code?

dpgeorge · 2026-09-15T00:12:15Z

I think the PR needs to be merged first, as the CI for the PR doesn't run on itself.

I'm quite certain it should run here...

dpgeorge · 2026-09-15T00:15:17Z

I checked github settings and the actions are enabled. Not sure what's wrong.

Break this PR into CI edits and code?

Yes, please open a separate PR in this repo to add just the code formatting and spelling workflows. Then I'll merge that first.

kwagyeman · 2026-09-15T01:29:22Z

Rebased

kwagyeman · 2026-09-15T01:34:36Z

Wait, sending CI stuff, AI just decided to drop it for no reason. Readding...

kwagyeman · 2026-09-15T01:56:40Z

@dpgeorge - Ready

dpgeorge · 2026-09-15T01:58:23Z

CI is working, nice!

dpgeorge · 2026-09-15T02:43:52Z

+#
+#   make          build + run
+#   make asan     build + run under ASan/UBSan (catches what the checks miss)
+CC      ?= gcc


This line does nothing because CC is set by make to the default value of cc. So maybe just remove this line.

dpgeorge · 2026-09-15T02:47:41Z

+# Cortex-M55 needs arm-none-eabi-gcc 14 or newer (Ubuntu 24.04 ships 13.2,
+# which rejects -mcpu=cortex-m55 with hard float).  Override CC to point at a
+# suitable toolchain if the default is too old.
+CC       ?= arm-none-eabi-gcc


This line does nothing: building with just make doesn't work because CC defaults to cc and is not overridden by this line.

Suggest making this (and the GCC variable below) and unconditional =:

CC = arm-none-eabi-gcc

That gets make working. And still allows overriding it via make CC=foo.

dpgeorge · 2026-09-15T02:48:27Z

+CC      ?= gcc
+HALOW   := ../../src
+SDK     := ../../lib/mm-iot-sdk/framework/morselib/include
+CFLAGS  := -m32 -std=c11 -Wall -Wextra -Werror -g -O1 \


I think these call all just use = (no need for :=).

dpgeorge · 2026-09-15T02:48:46Z

+CFLAGS := $(ARCH) -std=c11 -Wall -Wextra -Werror -Wno-unused-parameter -Og -g \
+          -ffreestanding -fno-common \
+          -Istub -I$(HALOW) -I../../lib/mm-iot-sdk/framework/morselib/include
+LDFLAGS := $(ARCH) -T$(LDSCRIPT) -nostartfiles -Wl,--gc-sections \


I think these (and others in this file) can all use = instead of :=.

dpgeorge · 2026-09-15T02:59:35Z

@@ -0,0 +1,14 @@
+// Host-test stub: halow.h embeds these by value, so they need to be complete


mm_halow.h

dpgeorge · 2026-09-15T03:16:27Z

+// (the driver task and the UMAC event loop), plus one more for the SDIO/SPI IRQ
+// task when the SDIO transport is in use.
+#ifndef MM_HALOW_MAX_TASKS
+#define MM_HALOW_MAX_TASKS  (4)


Suggest renaming this option to MM_HALOW_SCHED_MAX_TASKS`.

dpgeorge · 2026-09-15T03:16:43Z

+// Task turns per mm_halow_sched_run() call.
+// Ceiling on any single wait, however long the caller asked for.
+#ifndef MM_HALOW_WAIT_MAX_MS
+#define MM_HALOW_WAIT_MAX_MS (10000)


Suggest renaming this option to MM_HALOW_SCHED_WAIT_MAX_MS.

dpgeorge · 2026-09-15T03:17:49Z

+// an unevenly-dividing request yields a *faster* bus (40MHz becomes 50MHz on a
+// 200MHz AHB).  Check the rate you got.
+#ifndef MM_HALOW_SPI_BAUDRATE
+#define MM_HALOW_SPI_BAUDRATE   (12 * 1000 * 1000)


Move to mm_halow_config.h.

dpgeorge · 2026-09-15T03:18:40Z

+// Set by boards that invert the transceiver's BUSY output before it reaches the
+// MCU.  RESET_N is active low at the transceiver and is not configurable.
+#ifndef MM_HALOW_BUSY_INVERTED
+#define MM_HALOW_BUSY_INVERTED (0)


Move to mm_halow_config.h.

dpgeorge · 2026-09-15T03:23:11Z

+ * its own heap, so nothing else needs it.  morselib never calls these at run
+ * time, so the stubs only exist to satisfy the link, and _sbrk deliberately
+ * fails rather than hand out any of MicroPython's memory.  They are weak so that
+ * a port supplying real implementations takes precedence.


IMO this file does not belong in this driver. It is unfortunate that the mm-iot-sdk links against these, but it will need to be up to the host system (eg MicroPython) to provide them (and at least alif already does).

Instead I suggest removing this file, and moving this comment into the README, under the "Integrating" section.

Otherwise, it might be very confusing when someone integrates this driver, starts calling libc functions and finds they all fail (due to these weak stubs being pulled in behind the scenes).

Claude: osal.c still carries two weak libc stubs (_kill/_getpid) of the same kind as the removed file. I left them since only mm_halow_libc.c was flagged, but they're the same principle — I can pull those out too if the driver needs to carry none.

I suggest removing both those _kill and _getpid stubs. According to the comments they aren't even used.

dpgeorge · 2026-09-15T04:15:20Z

+# The transceiver firmware, and optionally a board configuration file holding
+# calibration data, are linked in as binary blobs.  A board picks its BCF by
+# name with MM_HALOW_BCF; the SDK keeps them per chip under morsefirmware.
+MM_HALOW_CHIP ?= mm8108


might also need to be changed to MM_HALOW_CHIPSET to match the C-level config

dpgeorge

Thanks for updating. Just a few more comments and it should be good to merge.

dpgeorge · 2026-09-15T06:17:00Z

-This is a driver for the Morse Micro MM6108/MM8108 802.11ah (Wi-Fi HaLow) transceivers.
+A portable driver for the Morse Micro MM6108/MM8108 802.11ah (Wi-Fi HaLow)
+transceivers, modelled on the structure of
+[cyw43-driver](https://github.com/georgerobotics/cyw43-driver): the driver is


suggest removing all references to cyw43-driver. the fact it's based on that is just to get a good head-start, it's not something that needs to stay in the docs/code

dpgeorge · 2026-09-15T06:19:49Z

+// mm_halow_ticks_ms() -> uint32_t
+//   A free-running millisecond tick counter.
+#ifndef mm_halow_ticks_ms
+#error "port must define mm_halow_ticks_ms"


This doesn't need to be a macro. It could instead be just declared here as:

uint32_t mm_halow_ticks_ms(void);

and then a port can define it as inline:

static inline uint32_t mm_halow_ticks_ms(void) { return ... }

The benefit of that way is it gives a clear signature for the function, and prevents use of macros which can lead to surprising outcomes.

(This should work fine, you can define a static inline function (done in mm_halow_configport.h prior to defining it as non-static (but not the other way around).)

dpgeorge · 2026-09-15T06:20:12Z

+// Pin accessors, applied to the MM_HALOW_CS/RESET/WAKE/BUSY/IRQ pin values
+// defined by the port.
+#ifndef mm_halow_hal_pin_write
+#error "port must define the mm_halow_hal_pin_read/write/input/output accessors"


As above, suggest defining these as proper C function declarations instead.

dpgeorge · 2026-09-15T06:20:26Z

+
+// mm_halow_in_irq() -> bool
+//   True when executing in interrupt context (on Cortex-M: IPSR != 0).
+#ifndef mm_halow_in_irq


As above, a proper C declaration.

dpgeorge · 2026-09-15T06:21:08Z

+#endif
+
+/*******************************************************************************/
+// Functions the port must implement (see README.md).


These could arguably be declared in a separate header eg mm_halow_port.h. But keeping them here (for now) is a reasonable pragmatic solution.

dpgeorge · 2026-09-15T06:21:25Z

+// Called when the driver hits an unrecoverable internal failure (a morselib
+// assertion).  The port routes this to its own assert/fatal-error handling; it
+// must not return.
+void mm_halow_assert_fail(void);


Suggest mm_halow_port_assert_fail() to match other naming here.

A portable driver for the Morse Micro MM6108/MM8108 802.11ah (Wi-Fi HaLow) transceivers, extracted from the MicroPython network.HALOW pull request and reorganised along the lines of cyw43-driver: - src/ holds the driver: the SD-over-SPI transport to the transceiver mapped onto a small mm_halow_port_* interface, an lwIP netif, a packet-memory pool, and an OSAL/scheduler shim that runs the vendor library (morselib) cooperatively off the host's scheduler with no vendor RTOS. - mm_halow_config.h defines the host integration contract; the embedder provides mm_halow_configport.h (or MM_HALOW_CONFIG_FILE) with atomic sections, a millisecond tick, pin accessors, interrupt-context detection, and the port functions (SPI, private heap backing, hardware RNG, fallback MAC, optional edge IRQ). - lib/mm-iot-sdk is the Morse Micro MM-IoT-SDK submodule, providing the prebuilt morselib and the transceiver firmware/BCF blobs; mm_halow.mk is a build fragment for make-based embedders covering morselib's newlib linkage, the blob embedding, and fetching the mm-iot-sdk submodule on demand. - tests/host compiles the real allocator natively (233 checks, plus an ASan/UBSan variant); tests/qemu runs the real scheduler, whose context switch is naked asm, on a Cortex-M55 under QEMU (51 checks); tests/hardware carries the on-device throughput soak harness the driver was validated with. Everything the driver needs from MicroPython in the original pull request now arrives through the configuration contract: MICROPY_* macros became MM_HALOW_*, the halow_* API became mm_halow_*, and the machine-SPI, GC-heap and pin-interrupt couplings became port hooks. Tested: both test suites pass locally (host: gcc -m32 + ASan; qemu: arm-none-eabi-gcc 14.3.1 on mps3-an547). Signed-off-by: Kwabena W. Agyeman <kwagyeman@live.com>

Extends tools/ci.sh (added with the formatting and spelling CI) with the test entry points, and wires them up: - tests.yml runs the host allocator tests (plus ASan) and the QEMU Cortex-M55 scheduler test. - tests_clang.yml builds and runs both with clang too; the QEMU test cross-compiles the sources with clang and links with GCC, so clang also covers mm_halow_sched.c, which the host tests do not build. Signed-off-by: Kwabena W. Agyeman <kwagyeman@live.com>

dpgeorge

Looks good, let's get this in.

micropython/mm-halow-driver#1 has merged, so re-pin the submodule to the merged commit and follow the driver's reviewed integration contract: - The tunable options were renamed: MM_HALOW_AP -> MM_HALOW_ENABLE_AP, MM_HALOW_PIN_IRQ -> MM_HALOW_ENABLE_PIN_IRQ, MM_HALOW_CHIP -> MM_HALOW_CHIPSET, MM_HALOW_MAX_TASKS -> MM_HALOW_SCHED_MAX_TASKS. - The tick, interrupt-context and pin hooks are now C function declarations rather than macros, so the configport provides them as static inline functions and defines mm_halow_pin_t. - The driver's assert hook is mm_halow_port_assert_fail(), provided here. - The driver no longer ships newlib syscall stubs (mm_halow_libc.c was removed), so the port provides the weak back-end morselib links against, and extmod.mk drops mm_halow_libc.c from the source list. Builds green for OPENMV_N6 and OPENMV_AE3 (M55_DUAL). Signed-off-by: Kwabena W. Agyeman <kwagyeman@live.com>

kwagyeman force-pushed the add-driver branch from df960ad to cbba856 Compare September 15, 2026 00:07

kwagyeman force-pushed the add-driver branch 2 times, most recently from cf788cc to 8422aef Compare September 15, 2026 01:29

kwagyeman force-pushed the add-driver branch 2 times, most recently from af156dd to 33c722a Compare September 15, 2026 01:54

dpgeorge reviewed Sep 15, 2026

View reviewed changes

kwagyeman force-pushed the add-driver branch 2 times, most recently from 84c71cd to 3d9ddd7 Compare September 15, 2026 05:32

dpgeorge reviewed Sep 15, 2026

View reviewed changes

kwagyeman added 2 commits September 15, 2026 13:05

kwagyeman force-pushed the add-driver branch from 3d9ddd7 to e2c5154 Compare September 15, 2026 20:13

dpgeorge approved these changes Sep 15, 2026

View reviewed changes

dpgeorge merged commit 00d26f6 into micropython:main Sep 15, 2026
6 checks passed

kwagyeman deleted the add-driver branch September 15, 2026 23:44

		@@ -0,0 +1,14 @@
		// Host-test stub: halow.h embeds these by value, so they need to be complete

Conversation

kwagyeman commented Sep 14, 2026

Summary

Testing

Uh oh!

kwagyeman commented Sep 14, 2026

Uh oh!

dpgeorge commented Sep 14, 2026

Uh oh!

kwagyeman commented Sep 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kwagyeman commented Sep 14, 2026

Uh oh!

dpgeorge commented Sep 15, 2026

Uh oh!

dpgeorge commented Sep 15, 2026

Uh oh!

kwagyeman commented Sep 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dpgeorge commented Sep 15, 2026

Uh oh!

dpgeorge commented Sep 15, 2026

Uh oh!

kwagyeman commented Sep 15, 2026

Uh oh!

kwagyeman commented Sep 15, 2026

Uh oh!

kwagyeman commented Sep 15, 2026

Uh oh!

dpgeorge commented Sep 15, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kwagyeman Sep 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kwagyeman Sep 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

kwagyeman commented Sep 14, 2026 •

edited

Loading

kwagyeman commented Sep 15, 2026 •

edited

Loading

kwagyeman Sep 15, 2026 •

edited

Loading

kwagyeman Sep 15, 2026 •

edited

Loading