cortex_m_systick programmed cycles are 1.7 tick less than expected, for tickless idle

[zephyrbot]

Reported by Sharron LIU:

The issue is detected when I tried to verify tickless idle @arm with unified kernel.

After time idle enter, the test case check "__scs.systick.strvr" programmed, and expect the value should be around "ticks * cycles_per_tick", with an error range that not greater than a tick (in ARM case, 840000 cycles).

However, for a 20-tick idle, now the "__scs.systick.strvr" programmed is "15352506", while "[15960000, 16800000]" is expected.

Could you review if this error is as expected? Since when read into the code I noticed two steps added to buffer time for reaction.

• timer_idle_skew is introduced
• (ticks - 1) is used

I attaching the test code.

# source zephyr-env.sh
# tar -zxvf test_tickless_concept.tgz
# make -C test_tickless_concept BOARD=arduino_due
# bossa ....
# verify from console output "programmed cycles 15352506 err 1447494", failed

Actual programmed cycles: err 1447494 (which is around 1.7 tick)
Expected programmed cycles: err within 840000 (cycles per tick)

NOTE: the test case works on X86 @quark_se and @quark_d2000

(Imported from Jira ZEP-1072)

[zephyrbot]

by Sharron LIU:

• test_tickless_concept.tgz (size: 3161)

[zephyrbot]

by Dmitriy Korovkin:

The attached test demonstrates on Arduino Due, FRDM-k64f and qemu-x86 platforms the fact that sleep takes 200 ms (as the test requires) plus one timer tick.
The behavior the test demonstrates is exactly what is expected: kernel waits for the system clock tick and then waits for 200 ms, which is exactly 20 system clock ticks.

[zephyrbot]

by Anas Nashif:

correct behaviour per developer

[zephyrbot]

by Mark Linkmeyer:

Fixing incorrect priority