sys/pm: Fix behavior of fallback pm_off() implementation

[maribu]

Contribution description

• pm_off() should prevent other threads from getting executed after it is called, as it should turn off the MCU completely according to its doc
  • Previously, the fallback implementation would still allow other threads to continue working
  • Simply disabling IRQs and never enabling them again should make sure the MCU behaves like it would be completely off
• pm_off() should reduce the power consumption as much as possible
  • Previously, when IRQs came after the call to pm_set_lowest() in the fallback implementation of pm_off(), while(1) {} got executed at full power consumption
  • Just calling pm_set(0); in a loop while make sure that lowest power mode is restored if the MCU wakes up again.
  • The check if the lowest power mode is available is skipped, as no code gets executed afterwards anyway

Testing procedure

The fallback implementation of pm_off() should now prevent any code from being executed after it is called.

I'll add a test application to this PR later on. Update: tests/periph_pm is fully sufficient for testing this.

Issues/PRs references

Split out of #13973 (but improved afterwards)

[maribu]

I'll add a test application to this PR later on.

After a second look into tests/periph_pm I changed my mind: tests/periph_pm already contains everything needed. The callback for BTN0 should no longer be reachable if pm_off() is called (using the pm shell command).

[waehlisch]

@maribu, can you improve the title of this PR?

[maribu]

Btw: The implementation of pm_off() in drivers/periph_common/pm.c is now almost identical with the one here.

[maribu]

Maybe @benpicco could test this using his mcb2388 board? However, in order to reproduce the issue (which in its full beauty can only occur if no off mode is implemented) the following patch needs to be (temporary) applied:

diff --git a/cpu/lpc2387/periph/pm.c b/cpu/lpc2387/periph/pm.c
index 2afbb92509..5fe853fdb6 100644
--- a/cpu/lpc2387/periph/pm.c
+++ b/cpu/lpc2387/periph/pm.c
@@ -33,7 +33,7 @@ void pm_set(unsigned mode)
         case 0:
             /* Everything except battery backup is powered down */
             DEBUG_PUTS("pm_set(): setting Deep Power Down mode.");
-            PCON |= PM_DEEP_POWERDOWN;
+            PCON |= PM_IDLE;
             break;
         case 1:
             /* PLL & Flash are powered down */

Note: The race condition fixed in this PR affects every user of pm_layered, including those implementing deep power down mode.

[benpicco]

Maybe @benpicco could test this using his mcb2388 board?

Yup that works:

2020-04-29 17:43:23,081 # mode 0 blockers: 1 
2020-04-29 17:43:23,082 # mode 1 blockers: 1 
2020-04-29 17:43:23,082 # mode 2 blockers: 1 
2020-04-29 17:43:23,082 # Lowest allowed mode: 3
2020-04-29 17:43:23,083 # using BTN0 as wake-up source
# button press is registered
> 2020-04-29 17:43:37,862 #  BTN0 pressed.
2020-04-29 17:43:47,657 # pm off
# button presses are ignored

Whereas on master

2020-04-29 17:45:14,518 # mode 0 blockers: 1 
2020-04-29 17:45:14,519 # mode 1 blockers: 1 
2020-04-29 17:45:14,520 # mode 2 blockers: 1 
2020-04-29 17:45:14,520 # Lowest allowed mode: 3
2020-04-29 17:45:14,521 # using BTN0 as wake-up source
> 2020-04-29 17:45:30,662 #  BTN0 pressed.
2020-04-29 17:45:33,447 # pm off
2020-04-29 17:45:35,415 # BTN0 pressed.
2020-04-29 17:45:35,767 # BTN0 pressed.
2020-04-29 17:45:36,072 # BTN0 pressed.