[userspace LL] Scheduling part of #10945

[lyakh]

Most of the scheduling changes from the userspace LL PR #10945

[Copilot]

Pull request overview

This PR ports parts of the “userspace LL scheduling” work (from #10945) into the main scheduling code, primarily to allow user-space LL execution to trigger DP scheduling and to grant the required kernel object access across cores.

Changes:

• Expose DP scheduler hooks as Zephyr syscalls (notably scheduler_dp_ll_tick(core) and scheduler_dp_internal_free()), and switch DP scheduler data lookup to the user-scheduler registry.
• Wire LL → DP tick propagation with an explicit core parameter.
• Add a helper to retrieve the userspace IPC thread and extend access grants for DP task initialization when CONFIG_SOF_USERSPACE_LL is enabled.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
zephyr/CMakeLists.txt	Adds DP scheduling header to Zephyr syscall header generation.
src/schedule/zephyr_ll.c	Updates DP tick invocation to pass a core ID.
src/schedule/zephyr_dp_schedule.h	Removes declarations now expected to come from the public DP schedule header/syscall layer.
src/schedule/zephyr_dp_schedule.c	Converts DP tick to a syscall-style implementation and switches to user scheduler data lookup.
src/schedule/zephyr_dp_schedule_thread.c	Converts internal free to syscall-style implementation (thread variant) with a verify stub.
src/schedule/zephyr_dp_schedule_application.c	Adds IPC-thread access grants and adds syscall verification for internal free (app variant).
src/ipc/ipc-common.c	Adds ipc_thread_user(core) accessor for the userspace IPC thread.
src/include/sof/schedule/dp_schedule.h	Updates DP scheduler public API to syscall declarations under Zephyr full-app builds.
src/include/sof/ipc/common.h	Adds prototype for ipc_thread_user(core).

Suppressed comments (1)

src/schedule/zephyr_dp_schedule_application.c:641

• scheduler_dp_mod_vrfy() uses assert(alloc) in syscall verification code. If asserts are compiled out, a NULL alloc will be dereferenced, causing a kernel fault during syscall verification.

	struct mod_alloc_ctx *alloc = mod->priv.resources.alloc;

	assert(alloc);
	if (alloc->heap) {

---

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

[lyakh]

right, so this one depends on 671a2cf

[Copilot]

Pull request overview

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Suppressed comments (1)

src/schedule/zephyr_dp_schedule.c:392

• scheduler_get_task_info_dp() unconditionally uses scheduler_get_user_data(SOF_SCHEDULE_DP) and then dereferences dp_sch without a NULL check. With CONFIG_SOF_USERSPACE_LL=n, DP is registered in the kernel scheduler list, so this can become a NULL dereference.

	scheduler_props->processing_domain = COMP_PROCESSING_DOMAIN_DP;
	struct scheduler_dp_data *dp_sch = scheduler_get_user_data(SOF_SCHEDULE_DP);

	lock_key = scheduler_dp_lock(cpu_get_id());
	scheduler_get_task_info(scheduler_props, data_off_size,  &dp_sch->tasks);

[lgirdwood]

I think longer term we could just call the single function here and have it do the right thing based on the Kconfig.

[lyakh]

@lgirdwood yep, will do something about this

[softwarecki]

One more comment regarding the PR description.

It would be very helpful if mergeable PRs contained a description of the actual changes they introduce, instead of only linking to another draft PR. In this case, the description points to a draft #10945, which itself points to another draft #10558, and the only information available there is a very high-level overview of the planned functionality.

For reviewers, it is much easier when each PR clearly explains what was changed, why it was changed, and what part of the functionality is covered by that specific PR.