cpu/atmega_common: RTT support

[benpicco]

Contribution description

This is just the RTT commit split off from the RTT/RTC PR by @ZetaR60.
I made small adjustments & cleanups to make it compile with the latest master branch and made the RTT frequency configurable.

Without the RTC emulation I hope this should be less controversial.

The RTT requires a 32kHz oscillator to be present on the board.
The only boards I know where this is the case are mega-xplained, atmega256rfr2-xpro and avr-rss2. It's possible that I missed some.
#12799 will add two more.

Testing procedure

Run tests/periph_rtt.
The RTT on ATmega will only work if the board is equipped with a 32kHz crystal.

Issues/PRs references

split off from #8842 which is based on #7604

[maribu]

Code contains a data race, see inline explanation. But trivial to fix :-)

[maribu]

With the current implementation the returned time could be up to 255 ticks lower than correct in a race condition. See inline comment

[maribu]

Two more suggestions to get _safe_cnt_get() race-free (at least unless I didn't overlook anything).

My comments have been addressed

[maribu]

I'll test tonight.

[maribu]

(If noone does so sooner ;-))

[maribu]

Sorry, I'm too tired to think straight now. I'll test tomorrow.

[benpicco]

Don't worry, there is no ticking clock 😆

[maribu]

OK. I hooked up a 32.768kHz and two 10 pF capacitors to TOSC1 and TOSC2 as pictured in the schematic in the data sheet to an ATmega1284P. Seems to work as advertised :-)

Output of tests/periph_rtt on the ATmega1284P on a breadboard

2019-11-30 23:39:57,760 # main(): This is RIOT! (Version: 2020.01-devel-1017-gdabc8-ben)
2019-11-30 23:39:57,761 # 
2019-11-30 23:39:57,794 # RIOT RTT low-level driver test
2019-11-30 23:39:57,842 # This test will display 'Hello' every 5 seconds
2019-11-30 23:39:57,843 # 
2019-11-30 23:39:57,872 # Initializing the RTT driver
2019-11-30 23:39:57,884 # RTT now: 0
2019-11-30 23:39:57,927 # Setting initial alarm to now + 5 s (5120)
2019-11-30 23:39:57,975 # Done setting up the RTT, wait for many Hellos
2019-11-30 23:40:02,877 # Hello
2019-11-30 23:40:07,877 # Hello
2019-11-30 23:40:12,877 # Hello
2019-11-30 23:40:17,878 # Hello
2019-11-30 23:40:22,878 # Hello
2019-11-30 23:40:27,878 # Hello
2019-11-30 23:40:32,878 # Hello
2019-11-30 23:40:37,878 # Hello
2019-11-30 23:40:42,878 # Hello
2019-11-30 23:40:47,878 # Hello
2019-11-30 23:40:52,878 # Hello
2019-11-30 23:40:57,878 # Hello
2019-11-30 23:41:02,879 # Hello
2019-11-30 23:41:07,879 # Hello
2019-11-30 23:41:12,879 # Hello
2019-11-30 23:41:17,879 # Hello
2019-11-30 23:41:22,879 # Hello
2019-11-30 23:41:27,879 # Hello
2019-11-30 23:41:32,879 # Hello
2019-11-30 23:41:37,879 # Hello
2019-11-30 23:41:42,879 # Hello
2019-11-30 23:41:47,880 # Hello
2019-11-30 23:41:52,880 # Hello
2019-11-30 23:41:57,880 # Hello
2019-11-30 23:42:02,880 # Hello
2019-11-30 23:42:07,880 # Hello
2019-11-30 23:42:12,880 # Hello
2019-11-30 23:42:17,880 # Hello
2019-11-30 23:42:22,880 # Hello
2019-11-30 23:42:27,880 # Hello
2019-11-30 23:42:32,881 # Hello
2019-11-30 23:42:37,881 # Hello
2019-11-30 23:42:42,881 # Hello
2019-11-30 23:42:47,881 # Hello
2019-11-30 23:42:52,881 # Hello
2019-11-30 23:42:57,881 # Hello
2019-11-30 23:43:02,881 # Hello
2019-11-30 23:43:07,881 # Hello
2019-11-30 23:43:12,881 # Hello
2019-11-30 23:43:17,882 # Hello
2019-11-30 23:43:22,882 # Hello
2019-11-30 23:43:27,882 # Hello
2019-11-30 23:43:32,882 # Hello
2019-11-30 23:43:37,882 # Hello
2019-11-30 23:43:42,882 # Hello
2019-11-30 23:43:47,882 # Hello
2019-11-30 23:43:52,882 # Hello
2019-11-30 23:43:57,882 # Hello

(I might add a follow up PR to extend the doc of the ATmega1284P on a breadboard and of the ATmega328P on a breadboard to cover this feature. And also a way how to export an environment variable (e.g. ATMEGA1284P_HAS_TIMER_OSC=1) to let the board provide periph_rtt as feature.)

[maribu]

Please squash. (Also, please squash the fix for the nitpick right away.)

[maribu]

Nitpick: The comment should be moved inside the body of the if. (Or reword the comment, e.g."Check if overflow occurred just after reading TCNT2, in which case it will now have a value of zero and update cnt to correspondingly.)

[benpicco]

Reworded.

[maribu]

ACK. The code is well written and documented, it worked as expected during testing, and I gave my best to check that the code is race free.

[benpicco]

Thank you for the thorough review & suggestions w.r.t. race conditions!