Fix STM RTC read so it's atomic.

[tannewt]

I turn off the shadow registers because 1) reading the SSR doesn't actually lock the other two (see the errata) and 2) the fallback is to read twice anyway. So, I might as well just read them twice because it'll be faster than waiting for the sync flag.

Fixes #3376

[jepler]

Specifically, see 2.2.3 in https://www.st.com/resource/en/errata_sheet/dm00037591-stm32f405-407xx-and-stm32f415-417xx-device-limitations-stmicroelectronics.pdf

This fix makes sense to me especially in light of the erratum. Assuming the test case originally supplied passes now (the one that sees time.monotonic going backwards), fine to merge. Thanks!

[tannewt]

Assuming the test case originally supplied passes now (the one that sees time.monotonic going backwards), fine to merge.

Yup! I ran it for a while with no issues.

[FoamyGuy]

I have not gotten the input/output error again running this build. So this does seem to solve that issue.

I've got the 33dee4a build from S3 installed on my Feather STM32F405 and I am seeing some odd behavior with time.sleep() taking longer than it should I think. I am running this code:

import time
for i in range(0,10):
    print(i)
    time.sleep(0.01)

This code is taking roughly 10 seconds to complete printing the numbers. At first I was thinking perhaps it's off by a factor of 10, but oddly enough if I tweak it to time.sleep(1) instead I am seeing pretty much the same behavior, no faster or slower than when I was sleeping for 0.01

If I use time.sleep(0.001) then everything will print very quickly, less than 1 second.

[tannewt]

Thanks @FoamyGuy! I used time.monotonic to measure that it's taking 7 seconds to run...... looking into it now.

[tannewt]

I think the sleep issue is fixed in #3549