k_cpu_atomic_idle failed @ARM

[zephyrbot]

Reported by Sharron LIU:

I added test point to cover "k_cpu_atomic_idle", following the usage mentioned in https://www.zephyrproject.org/doc/kernel/other/cpu_idle.html.

Here's the test case:
https://gerrit.zephyrproject.org/r/#/c/11187

The test passed @x86 @arc, but failed @arm (qemu_cortex_m3). When look into "arch/arm/core/cpu_idle.S:159" this feature supposed to be supported.

(Imported from Jira ZEP-1720)

[zephyrbot]

by Maureen Helm:

I've started taking a look at this, and reproduced on FRDM-K64F and FRDM-KW41Z.

[zephyrbot]

by Benjamin Walsh:

Dunno why it fails on real hardware, but I'm 99% sure WFE is implemented as a NOP on QEMU. Actually, the ARM specs says that WFE can be implemented as a NOP. Same for WFI BTW.

[zephyrbot]

by Maureen Helm:

The test fails because it assumes the tick will increment after returning from k_cpu_atomic_idle(). I can get the test to pass on real hardware by changing the wfe to sev; wfe; wfe. This handles the case that the event register is already set before the wfe. But it doesn't help qemu.

[zephyrbot]

by Benjamin Walsh:

Test is bogus on QEMU. :-/

I think they expect the only event to occur to be the timer interrupt, and they align on a tick just before going idle to be sure that no interrupt is pending. Is it possible the hardware is receiving other interrupts ?

[zephyrbot]

by Anas Nashif:

test is bogus, so lower priority to avoid blocking a release because of a test...

[zephyrbot]

by Sharron LIU:

test case updated. test point "k_cpu_atomic_idle" disabled for CONFIG_ARM=y.

[nashif]

closing, nothing to be done here it seems.