IDC: fix a race

[lyakh]

fix an IDC race when using P4WQ

[Copilot]

Pull request overview

Fixes an IDC submission race in the Zephyr/P4WQ-based IDC path by changing how work items are allocated/submitted per target core, and aligns a notifier call site with the codebase’s “local core mask” convention.

Changes:

• Expand IDC per-target work storage to two slots and add a mutex intended to serialize submissions.
• Adjust IDC work selection logic to avoid re-submitting a work item that hasn’t been fully released by P4WQ yet.
• Replace a direct core-bitmask expression with NOTIFIER_TARGET_CORE_LOCAL in the DAI trigger notifier.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/idc/zephyr_idc.c	Attempts to prevent P4WQ re-submit race by adding 2 work items/core and a mutex around selection/submission.
src/audio/dai-zephyr.c	Uses NOTIFIER_TARGET_CORE_LOCAL instead of BIT(cpu_get_id()) when firing local-core notifier events.

Comments suppressed due to low confidence (1)

src/idc/zephyr_idc.c:156

• k_mutex_lock(&idc_mutex, ...) is not released on the error path when the target core is down. The early return -EACCES; will leave the mutex locked and can deadlock all subsequent IDC sends. Ensure the mutex is unlocked before returning (or check cpu_is_core_enabled() before taking the mutex).

	k_mutex_lock(&idc_mutex, K_FOREVER);

	if (work->thread)
		/* See comment above the idc_work[] array. */
		work++;

	/* Same priority as the IPC thread which is an EDF task and under Zephyr */
	work->priority = CONFIG_EDF_THREAD_PRIORITY;
	work->deadline = 0;
	work->handler = idc_handler;
	work->sync = mode == IDC_BLOCKING;

	if (!cpu_is_core_enabled(target_cpu)) {
		tr_err(&zephyr_idc_tr, "Core %u is down, cannot sent IDC message", target_cpu);
		return -EACCES;
	}

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[kv2019i]

This is a nice fix. I have a faint memory we've been looking at this problem as well, but can't find any PR now (I recall there was an attempt to have a pool of Idc_work items. But this seems simpler and does the job.

[jsarha]

LGTM, and also "fixes" the same assert when I try to use my "user" IPC thread to play audio on pipelines that are not on core 0 (it still does not work, as is expected, but does not hit an assert either).