Move uart writes from ISR on STM32

[alteman]

EDIT: related issue #1873

OK, here's a draft PR - I'm still testing, so at this point unsure if the issue is actually fixed.

The alternative workaround is increasing Arduino tx buffer to 128 bytes using -DSERIAL_TX_BUFFER_SIZE=128

[alteman]

Testing in progress; test script should get to 1000 RX power cycles in about a week. It powers the receiver for 5-10 minutes (random) then powers off, waits 30-60 seconds and repeats. Serial activity is monitored and logged; serial data timeout is 10 seconds; no data for 10 seconds marks the attempt as a failure.

[alteman]

Preliminary test results:
With old firmware (3.0.0 tag) I got 6 failsafes out of 32 test runs (didn't seem to make sense to run it for much longer).
test_log_3.0.0.txt

On this PR branch I got 0 errors out of 100 test runs; so far so good.
test_log_PR.txt

Fun fact: I had to switch to R9M TX module because Happymodel ES900TX doesn't have external power and TX16S battery was discharging even when connected to a USB charger putting out 11W (5V 2.1A).

[robustini]

Finally maybe someone has found where the problem was, I'm testing now, I'll update you, thank you very much for now.
If you read my comments in the issue you will find that I had reported that it could have been a buffer saturation problem or something like that, I shot it at random but maybe I had caught it.

[robustini]

After 5 hours of testing no problems found, normally no one has ever exceeded 3 hours without a firmware lock, sometimes even after few minutes.
So this fix looks very promising (my ten R9Mini are happy)!

[robustini]

Made about ten FPV flights, never a problem, when before at least once out of three the rx stopped working.
In my opinion this draft can be switched to PR. excellent hunting, thanks again!

[alteman]

No errors after 400 test cycles, I think it's ready for merging :)

[pkendall64]

Nice catch @alteman

[JyeSmith]

Should this be targeting the v3 maintenance branch and we can get it in the 3.0.1 release?

[pkendall64]

Should this be targeting the v3 maintenance branch and we can get it in the 3.0.1 release?

I think that would be a very sensible idea!

[robustini]

Should this be targeting the v3 maintenance branch and we can get it in the 3.0.1 release?

If there are no other things immediately to enter 3.0.1 should be released because this problem is not irrelevant imho.
Those who own R9MM or R9Mini may have serious problems with 3.0, the update should be done before unexpected crashes.
Or simply a warning not to use 3.0 with the rx ones, even if many have already understood this at their own expense.