sys/evtimer: fix 0 offset events

[fabian18]

Contribution description

Something is weird in the evtimer. When a 0 offset event is added it seems like the timer IRQ does not fire.
I got the fix from AI actually 😅.
But I noticed in development that sometimes an event with 0 offset is stuck in the _nib_evtimer which prevents any further events.

This is not the case on native, but on same54-xpro.

Testing procedure

Added a test in tests/sys/evtimer Without the fix the 0 offset event does not trigger.
Try to run it with the fix and it triggers.

Issues/PRs references

[riot-ci]

Murdock results

✔️ PASSED

f892984 tests/sys/evtimer_msg: add test for 0 offset events

Success	Failures	Total	Runtime
11004	0	11004	09m:01s

Artifacts

• Documentation preview

[kaspar030]

I wouldn't want this to execute here (in calling context) without debate. It breaks the (maybe implicit) assumption that evtimer handlers are executed in the timer isr.

[fabian18]

More investigation shows that it works with increasing:

/**
 * @def     RTT_MIN_OFFSET
 *
 * @brief   The minimum offset to correctly set an rtt callback.
 *
 * If the callback is taking into account rtt_get_counter() then the rtt
 * might advance right between the call to rtt_get_counter() and
 * rtt_set_alarm(). If that happens with val==1, the alarm would be
 * set to the current time, which would then underflow. To avoid this,
 * the alarm should be set at least two ticks in the future.
 *
 * This value can vary depending on the platform.
 *
 */
#ifndef RTT_MIN_OFFSET
#define RTT_MIN_OFFSET (2U)
#endif

Confusing to me is that a timer offset of 1 works, but is also smaller than the default value of RTT_MIN_OFFSET.