Teensy 4.1: implement USB remote wakeup

[stapelberg]

This allows waking up a computer from Suspend-to-RAM by pressing a key on the keyboard (with the QMK keyboard firmware, which uses ChibiOS for the Teensy).

related to kinx-project/kint#59

related to qmk/qmk_firmware#16934

[stapelberg]

I converted this PR to a draft for now, because I noticed that on a different PC, with this change, the keyboard seems to not work after suspend/resume (when the resume is not triggered by the keyboard). Will need to investigate.

[stapelberg]

I converted this PR to a draft for now, because I noticed that on a different PC, with this change, the keyboard seems to not work after suspend/resume (when the resume is not triggered by the keyboard). Will need to investigate.

Did some initial investigation: the symptom is that when the PC resumes from suspend-to-RAM, it resets all devices connected to my USB hub, and the keyboard does not enumerate any longer.

Unfortunately, as soon as I compile in printf_debug in lib/chibios-contrib/os/hal/ports/MIMXRT1062/MIMXRT1062/printf_debug.[ch], the problem no longer reproduces. I can replace the printf_debug() implementation with a 1ms sleep and the problem still doesn’t show up.

This indicates it’s either a missing delay somewhere, a race condition/timing issue, or adding extra sleeps papers over the issue. In any case — very annoying to figure out :(

I’ll see if I can find some time to set up my NXP devkit to see if anything obvious shows up in a debugger.

[stapelberg]

I’ll see if I can find some time to set up my NXP devkit to see if anything obvious shows up in a debugger.

When running it on the NXP devkit, the issue doesn’t reproduce :(

Worse, in my current testing, waking the PC up via USB didn’t work anymore!

I have a https://www.crowdsupply.com/great-scott-gadgets/luna ordered, so when/if it ships I might give this another look, but for now I’ll give up on this. I don’t have the right tools to effectively debug this.

[stapelberg]

Actually, one thing came to mind: I think my hack to use DWT->CYCCNT directly instead of hwTick is not quite correct: hwTick is measured in ms, not in cycles. So, probably the USB stack delays are too short, which might explain why the extra delays paper over the issue.

I’ll try to give this a shot over the weekend

[stapelberg]

So, probably the USB stack delays are too short, which might explain why the extra delays paper over the issue.

I have updated the code to convert milliseconds to cycles to fix the delay timing, but unfortunately the “hanging USB enumeration” symptom still happens :(