Issues with power management introduction into Intel CAVS25

[nashif]

Describe the bug

Few Kernels tests started failing after 95168e6 was merged:

INFO     - 1) tests/boards/intel_adsp/ssp/boards.intel_adsp.ssp on intel_adsp_cavs25 failed (Timeout)
INFO    - 2) tests/kernel/tickless/tickless_concept/kernel.tickless.concept on intel_adsp_cavs25 failed (Timeout)
INFO    - 3) tests/kernel/timer/timer_api/kernel.timer.tickless on intel_adsp_cavs25 failed (Timeout)

To Reproduce
Steps to reproduce the behavior:

1. west twister -p intel_adsp_cavs25 -s tests/boards/intel_adsp/ssp/boards.intel_adsp.ssp

Expected behavior
Tests shall pass

Impact

Multiple kernel tests failing.

Logs and console output

In the case of SSP, no output. With tickless tests, the test just times out.

[nashif]

@kv2019i FYI

[ceolin]

There are a bunch of hw specific flow in this commit, looking Zephyr's pm, the way core 0 is handled in pm_state_set raises some suspicious.

+       if (state == PM_STATE_SOFT_OFF) {
+               core_desc[cpu].intenable = XTENSA_RSR("INTENABLE");
+               z_xt_ints_off(0xffffffff);
+               soc_cpus_active[cpu] = false;
+               z_xtensa_cache_flush_inv_all();
+               if (cpu == 0) {
+                       uint32_t ebb = EBB_BANKS_IN_SEGMENT;
+                       /* turn off all HPSRAM banks - get a full bitmap */
+                       uint32_t hpsram_mask = (1 << ebb) - 1;
+                       /* do power down - this function won't return */
+                       power_down_cavs(true, uncache_to_cache(&hpsram_mask));
+               } else {
+                       z_xt_ints_on(core_desc[cpu].intenable);
+                       k_cpu_idle();
+               }

how core 0 is supposed to wake up if all interruptions are masked ?

One more thing, if is not necessary to do anything in cores !=0 other than k_cpu_idle you should just remove cpu-power-states = <&d3>; in the device tree for these cores, then when these cores go idle (idle thread runs) the function pm_system_suspend will return false and the idle thread will invoke k_cpu_idle. This way you simplify a lot this code path.

[juimonen]

@ceolin you know is the test framework setting the PM_STATE_SOFT_OFF? In sof we are forcing it in driver power off message... I thought it was not set automatically.

[juimonen]

@ceolin hmm I guess the "new" power-off handling should be surrounded with #ifdef CONFIG_PM_POLICY_CUSTOM in power.c then the tests seem to work. it is kind of assumed here that this should be only called by custom handler in the app... this is my blind copy from ace side which I think we are not testing with zephyr tests...

So cavstools can't handle currently the ipc response sent from assembly, also for example the ssp test would need next_state handler for custom policy.

[juimonen]

fix here: #56843

[nashif]

this is fixed now