fw: add Memfault SDK

[noahp]

Summary

Add the Memfault SDK as a git submodule, and integrate it into the
project.

This enables tracking the following in Memfault:

• reboots (translated pebble reboot reason codes)
• heartbeat metrics

Some console commands are included to help test the integration:

• mflt export : dump serialized Memfault chunks
• mflt heartbeat : trigger a heartbeat
• mflt metrics_dump : print current heartbeat values
• mflt device_info : print device serial, software type, hardware
  version

TODO:

• log capture
• Memfault coredumps

Testing

Flashed to an asterix board, and ran these test commands:

mflt heartbeat
mflt export

Uploaded the symbol file (build/src/fw/tintin_fw.elf) and chunks to
app.memfault.com and confirmed both processed correctly.

Note: reboot reason tracking is not implemented on the nRF5 platform
(see reboot_reason.c). We'll need a solution there, eg just a small
slice of .noinit ram ideally.

---

Signed-off-by: Noah Pendleton noah@memfault.com

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[ericmigi]

Don't worry about the CLA for PebbleOS, I need to disable that

[gmarull]

Initial comments: make sure to make this feature optional; otherwise we won't be able to merge back to the OSS repository.

[noahp]

Initial comments: make sure to make this feature optional; otherwise we won't be able to merge back to the OSS repository.

Thanks @gmarull , I added a ./waf configure --no-memfault option to support this! LMK if you had a different approach in mind 🙏

[gmarull]

Initial comments: make sure to make this feature optional; otherwise we won't be able to merge back to the OSS repository.

Thanks @gmarull , I added a ./waf configure --no-memfault option to support this! LMK if you had a different approach in mind 🙏

Sorry to come late, I've been busy with other urgent stuff. I think memfault should probably be disabled by default, maybe enabled just on platforms we want it to be default (asterix/obelix). Another question: how is communication authenticated? Is fw just reporting logs to the mobile and mobile holds credentials?

[ericmigi]

@noahp how's things looking on this PR? Can we help in any way? We're aiming to start a beta test in 2-3 weeks. Any chance we might have coredumps and heart beats by then? We could do the wiring up for data for the heart beat

[noahp]

Sorry to come late, I've been busy with other urgent stuff. I think memfault should probably be disabled by default, maybe enabled just on platforms we want it to be default (asterix/obelix). Another question: how is communication authenticated? Is fw just reporting logs to the mobile and mobile holds credentials?

Ok, I changed it to what you asked: enabled by default for asterix + obelix, otherwise opt-in.

how is communication authenticated? Is fw just reporting logs to the mobile and mobile holds credentials?

Yes, Memfault data is sent with a Project Key, which is typically stored on the data relay (mobile app in this system): https://api-docs.memfault.com/#a8d3e36f-62f0-4120-9fc6-544ee04f3bb5

This PR is ready for review. There are some remaining items:

• enabling coredumps. my plan is to wholesale swap the Memfault fault handlers here, when Memfault is enabled. we will need to hook the flash API as a Memfault coredump storage backend. shouldn't be much work.
• log instrumentation
• update reboot reason tracking to work on the nRF52

Data can be exported for upload by calling memfault_packetizer_get_chunk(), cc @sjp4 who knows the drill here https://docs.memfault.com/docs/mcu/data-from-firmware-to-the-cloud#normal-mode

[gmarull]

open GH issue to track

[noahp]

cleaned up in this patch- #51

[gmarull]

open github issue to track

[noahp]

👍 added here, thank you!
#52