schedule: ll: handle completed task state in library scheduler

[singalsu]

Update the library LL scheduler in schedule_ll_run_tasks() to check the task_run() return state and remove tasks returning SOF_TASK_STATE_COMPLETED from the scheduler list.

Following commit 002e499 ("ipc4: pipeline: split trigger and copy processing"), IPC4 uses a one-shot trigger task that returns SOF_TASK_STATE_COMPLETED upon trigger completion, while the copy task waits for task_is_active(p->trigger_task) to clear before copying. Previously, the library LL scheduler ignored the return value of task_run() and reset the task state back to SOF_TASK_STATE_QUEUED, leaving trigger_task permanently active and causing testbench pipeline copy tasks to stall indefinitely.

[Copilot]

Pull request overview

This PR updates the testbench/library low-latency (LL) scheduler to respect a task’s task_run() return state, specifically removing tasks that report SOF_TASK_STATE_COMPLETED from the scheduler list. This aligns the library scheduler behavior with IPC4’s one-shot trigger tasks so completed triggers don’t remain “active” and block dependent copy tasks in the testbench.

Changes:

• Capture task_run() return state in schedule_ll_run_tasks() and delete tasks that return SOF_TASK_STATE_COMPLETED from sched_list.
• Prevent re-scheduling tasks that are already queued/running in schedule_ll_task().
• Adjust cancel/free behavior to avoid deleting a task list node unless it is considered scheduled.

Suppressed comments (1)

src/platform/library/schedule/ll_schedule.c:101

• schedule_ll_task_free() has the same active-state gap as schedule_ll_task_cancel(): tasks freed from states other than QUEUED/RUNNING would stay linked on sched_list. Using task_is_active() ensures all active states are unlinked safely.

	if (task->state == SOF_TASK_STATE_QUEUED ||
	    task->state == SOF_TASK_STATE_RUNNING)
		list_item_del(&task->list);

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[kv2019i]

Merged quickly as this was blocking testbench testing in CI and causing longrunning CI jobs running for hours...