microcontroller.cpu.voltage hangs on latest 5.4.0 RC

[bd34n]

microcontroller.cpu.voltage hangs on latest 5.4.0 RC (5.4.0-beta.0-76-g0f55f5865-dirty), and only a hard reset appears to bring it back. Both .temperature and .frequency work, though.

This appears new - it didn't happen on 5.3.

Here is how to reproduce:

Press any key to enter the REPL. Use CTRL-D to reload.
Adafruit CircuitPython 5.4.0-beta.0-76-g0f55f5865-dirty on 2020-05-14; BDMICRO Vina M4 v004 with samd51n20

import os
import microcontroller

os.uname()
(sysname='samd51', nodename='samd51', release='5.4.0', version='5.4.0-beta.0-76-g0f55f5865-dirty on 2020-05-14', machine='BDMICRO Vina M4 v004 with samd51n20')

microcontroller.cpu.temperature
39.0517

microcontroller.cpu.frequency
120000000

microcontroller.cpu.voltage
[locked up at this point]

I apologize that I don't have time to attach my j-link and trace it. But thought it worthwhile to at least file the issue.

[siddacious]

Can't reproduce on an m0 so it may be an M4 specific thing?

[dhalbert]

The SAMD51 ADC is odd and flaky in a number of ways, sadly, so not so surprising.

[DavePutz]

Disclaimer: All I know about SAMD51 ADC's is what I picked up while looking at this issue.

1. This is not just 5.4 related. I was able to reproduce on a PyBadge back to CP 5.0
2. The hang is in ports/atmel-samd/asf4/samd51/hal/src/hal_adc_sync.c, line 121:

while (!_adc_sync_is_channel_conversion_done(&descr->device, channel))
			;

This can loop forever. I added code to put in a loop counter and exit after some number of attempts. While this eliminated the hangs it did not resolve the real issue. It does, perhaps,
look like timing related; since when I added a print statement in the loop, the first read being done (from common_hal_mcu_processor_get_voltage() line 312 ) exited after hitting the loop limit, but the second read then succeeded on the first attempt. My current code, which works but is an enormous kludge, is as follows:

uint8_t  loop_ctr = 0;
                while (!_adc_sync_is_channel_conversion_done(&descr->device, channel)) {
                        loop_ctr++;
                        mp_printf(&mp_plat_print, "");
                        if (loop_ctr > 50) {
                           return 0;
                        }
                 }

So, it would appear that the print statement does enough to let the adc_sync_is_channel_conversion_done() test succeed. There must be a better way - any suggestions?

[DavePutz]

So, testing shows that if the loop limit is increased to 255 and the mp_printf is removed then Python calls to microprocessor.get_voltage() will succeed. @dhalbert , that would require making changes to the asf4 library. Is that something we could/would do?

[dhalbert]

We have already forked asf4 and fixed a small number of things, so there's no problem. I haven't had a chance to look at this in detail yet.

[dhalbert]

We may be forgetting some sync-wait we need during the setup of the ADC. We may not need to change the ASF4 code.

[dhalbert]

I have a number of SAMD51 boards, including two Metro M4 boards that are identical. There has been a chip revision, but the SAMD51 chips on them have identical revision numbers in the DSU DID device field. They vary only in the date code on the package.

On one, microcontroller.cpu.voltage hangs, and on the other, it does not. So this varies per chip sample (ugh), and I'll investigate further.

I also tested several other SAMD51 boards and .voltage doesn't hang. So it's less-common problem.