audio: pipeline: fix control init order and unrouted component prepare in static pipeline by lgirdwood · Pull Request #11315 · thesofproject/sof

lgirdwood · 2026-10-15T16:43:24Z

Description

This PR fixes two initialization and prepare issues in the declarative static pipeline loader subsystem (static_pipeline_loader.c), which operates strictly under IPC4 (CONFIG_IPC_MAJOR_4=y):

Premature kcontrol initialization order in sof_static_topology_init():
- Stage 7 (sof_static_topology_init_controls()) dispatches default control values via sof_static_kcontrol_set(), which verifies if (!s_active_topo) return -ENODEV;.
- s_active_topo was assigned only after Stage 7 succeeded, causing all initial kcontrol configurations at boot to fail with -ENODEV (-19).
- Fix: Publish s_active_topo = topo; prior to Stage 7, and reset it to NULL in err_unwind if topology construction fails.
Unrouted component prepare failure:
- In Stage 6 (sof_static_topology_prepare_comps()), comp_prepare() is called on all components in COMP_STATE_READY. For optional or unrouted components declared in a static topology that have no connected buffers (bsource_list and bsink_list both empty), comp_prepare() fails with -ENOTCONN (-107).
- Fix: Skip components without any buffer connections in the stage 6 prepare loop.

Changes

src/audio/pipeline/static_pipeline_loader.c:
- Set s_active_topo = topo; before calling sof_static_topology_init_controls(), and clear to NULL on error unwind.
- Skip prepare on components where list_is_empty(&dev->bsource_list) && list_is_empty(&dev->bsink_list).

All static pipeline logic remains strictly IPC4-based with zero IPC3 dependencies.

softwarecki · 2026-10-29T12:56:20Z

-		/* Only prepare components currently in READY state */
+		/* Only prepare components currently in READY state that are connected to buffers */
 		if (dev->state == COMP_STATE_READY) {
+			if (list_is_empty(&dev->bsource_list) && list_is_empty(&dev->bsink_list))


Could we check this for the sink/source buffer APIs instead of relying on comp_buffer?

will do, I want everything using the new API too

Updated and rebased on main:

Replaced the dev->bsource_list / dev->bsink_list check in sof_static_topology_prepare_comps() with sof_static_comp_has_sink_source(), which checks mod->num_of_sources > 0 || mod->num_of_sinks > 0 for module adapter components and source_get_comp_state() / sink_get_state() on sof_audio_buffer handles for endpoint components.

Added comp_bind() / comp_unbind() in sof_static_topology_init_routes() and sof_static_topology_unwind() so sof_sink and sof_source interfaces are bound to IPC4 processing_module instances before prepare().

Converted init_buffer_params(), sof_static_topology_unwind(), and sof_static_pipeline_set_sample_rate() to use the sof_sink, sof_source, and sof_audio_buffer APIs (sink_set_params(), source_set_params(), audio_buffer_reset(), audio_buffer_free(), source_get_frm_fmt(), source_get_channels()).

Copilot

🟡 Changes recommended

Critical loader defects can corrupt memory when connection limits are exceeded, misconfigure SRC/ASRC pipeline rates, and prevent static mixin-to-mixout routes from binding.

3 open issues

What changed in this PR

This PR updates SOF’s IPC4 static pipeline loader to initialize controls after publishing the topology and skip preparation of unconnected components. It also extends buffer binding, cleanup, and sample-rate handling.

Changes:

Publish the active topology before control initialization and clear it on failure.
Bind sink/source interfaces to routed components and unbind them during cleanup.
Use buffer APIs for initialization and update module configurations and buffer thresholds during sample-rate changes.

File	Description
`src/audio/pipeline/static_pipeline_loader.c`	Updates initialization order, connection-aware preparation, interface binding and cleanup, and sample-rate configuration.

_{Review effort: Balanced}

lgirdwood · 2026-11-04T19:39:35Z

+			if (topo->routes[j].sink_comp_id == r->sink_comp_id)
+				dst_queue++;
+		}
+		buf_sink->audio_stream_params->id = IPC4_COMP_ID(src_queue, dst_queue);


Fixed in 80c0fb7 — added per-component queue bounds validation against CONFIG_MODULE_MAX_CONNECTIONS, 3-bit IPC4 queue ID limits, and src_mod->max_sinks / sink_mod->max_sources before pipeline_connect() and comp_bind().

lgirdwood · 2026-11-04T19:39:37Z

+
+		bind_data.bind_type = COMP_BIND_TYPE_SOURCE;
+		bind_data.source = buf_source;
+		ret = comp_bind(sink, &bind_data);


Fixed in 80c0fb7 — static components are now registered in ipc_get()->comp_list (with fallback IPC initialization when needed) during sof_static_topology_init_comps() and removed via list_item_del() in sof_static_topology_unwind(), allowing ipc4_get_comp_dev() to resolve endpoints in mixout_bind()/mixout_unbind().

lgirdwood · 2026-11-04T19:39:40Z

+					bcfg->audio_fmt.sampling_frequency = rate;
+					if (frame_bytes) {
+						bcfg->ibs = frames * frame_bytes;
+						if (!cdesc->caps.sink_rate)
+							bcfg->obs = bcfg->ibs;


Fixed in 80c0fb7 — added sof_static_resolve_comp_in_rate() so components and route buffers downstream of a fixed-output rate converter (caps.sink_rate != 0) preserve sink_rate during validation, module base_cfg updates, and route buffer reconfiguration.

intel-sofci · 2026-11-04T17:51:28Z

PR 11315: test results

Run date: 2026-11-04 20:03 UTC

Tested commit: 80c0fb7aaa48e0d426c01bbd724c6591ec9726db

lgirdwood · 2026-11-04T19:01:05Z

+	if (dev_comp_type(dev) == SOF_COMP_MODULE_ADAPTER) {
+		struct processing_module *mod = comp_mod(dev);
+
+		return mod && (mod->num_of_sources > 0 || mod->num_of_sinks > 0);
+	}
+
+	if (!topo)
+		return false;
+
+	uint32_t comp_id = dev_comp_id(dev);
+
+	for (size_t i = 0; i < topo->num_routes; i++) {
+		const struct sof_static_route *r = &topo->routes[i];
+
+		if (r->src_comp_id != comp_id && r->sink_comp_id != comp_id)
+			continue;
+
+		struct comp_buffer *buf = sof_static_buffer_get(r->buffer_id);
+		if (!buf)
+			continue;
+
+		if (r->src_comp_id == comp_id) {
+			struct sof_source *source = audio_buffer_get_source(&buf->audio_buffer);
+
+			if (source && source_get_comp_state(source) != COMP_STATE_NOT_EXIST)
+				return true;
+		}
+
+		if (r->sink_comp_id == comp_id) {
+			struct sof_sink *sink = audio_buffer_get_sink(&buf->audio_buffer);
+
+			if (sink && sink_get_state(sink) != COMP_STATE_NOT_EXIST)
+				return true;
+		}
+	}


lets inline comment the block and decisions here.

Done in 80c0fb7 — added inline comments documenting both the IPC4 processing-module path (mod->num_of_sources / mod->num_of_sinks) and the endpoint component fallback (source_get_comp_state() / sink_get_state()).

lgirdwood · 2026-11-04T19:14:44Z

+			if (src) {
+				bu.primary.r.module_id = IPC4_MOD_ID(dev_comp_id(src));
+				bu.primary.r.instance_id = IPC4_INST_ID(dev_comp_id(src));
+			}
+			if (sink) {
+				bu.extension.r.dst_module_id = IPC4_MOD_ID(dev_comp_id(sink));
+				bu.extension.r.dst_instance_id = IPC4_INST_ID(dev_comp_id(sink));
+			}
+
+			if (src && buf_sink) {
+				struct bind_info unbind_src = {
+					.ipc4_data = &bu,
+					.bind_type = COMP_BIND_TYPE_SINK,
+					.sink = buf_sink,
+				};
+				comp_unbind(src, &unbind_src);
+				pipeline_disconnect(src, buf, PPL_CONN_DIR_COMP_TO_BUFFER);
+			}
+			if (sink && buf_source) {
+				struct bind_info unbind_sink = {
+					.ipc4_data = &bu,
+					.bind_type = COMP_BIND_TYPE_SOURCE,
+					.source = buf_source,
+				};
+				comp_unbind(sink, &unbind_sink);
+				pipeline_disconnect(sink, buf, PPL_CONN_DIR_BUFFER_TO_COMP);
+			}
+
+			audio_buffer_free(&buf->audio_buffer);


can we inline comment each phase here.

Done in 80c0fb7 — added inline comments for each teardown phase in sof_static_topology_unwind() (Phases 1–5: endpoint resolution, queue recalculation, comp_unbind(), pipeline_disconnect() + audio_buffer_free(), and IPC registry + component/pipeline destruction).

lgirdwood · 2026-11-04T19:16:18Z

+			if (cdev && dev_comp_type(cdev) == SOF_COMP_MODULE_ADAPTER) {
+				struct processing_module *mod = comp_mod(cdev);
+
+				if (mod) {
+					struct ipc4_base_module_cfg *bcfg = &mod->priv.cfg.base_cfg;
+					uint32_t period_us = cdev->period ? cdev->period : 1000;
+					uint32_t frames = (rate * period_us) / 1000000;
+					uint32_t frame_bytes = (bcfg->audio_fmt.depth / 8) *
+							       bcfg->audio_fmt.channels_count;
+
+					if (frames == 0)
+						frames = 48;
+					bcfg->audio_fmt.sampling_frequency = rate;
+					if (frame_bytes) {
+						bcfg->ibs = frames * frame_bytes;
+						if (!cdesc->caps.sink_rate)
+							bcfg->obs = bcfg->ibs;
+					}
+				}
+			}


again, lets comment phases and assumptions

Done in 80c0fb7 — added inline comments documenting each phase and assumption in the module base_cfg update loop, including upstream fixed-output rate converter propagation via sof_static_resolve_comp_in_rate().

lgirdwood · 2026-11-04T19:16:48Z

+					struct sof_sink *sink =
+						audio_buffer_get_sink(&buf->audio_buffer);
+					struct sof_source *source =
+						audio_buffer_get_source(&buf->audio_buffer);
+					enum sof_ipc_frame fmt = source ?
+						source_get_frm_fmt(source) : SOF_IPC_FRAME_S16_LE;
+					uint32_t channels = source ?
+						source_get_channels(source) : 2;
+
 					init_buffer_params(buf, src_cdesc->direction,
-							   audio_stream_get_frm_fmt(&buf->stream),
-							   rate,
-							   audio_stream_get_channels(&buf->stream));
+							   fmt, rate, channels);
+
+					struct comp_dev *src_dev =
+						sof_static_comp_get(route->src_comp_id);
+					struct comp_dev *sink_dev =
+						sof_static_comp_get(route->sink_comp_id);
+
+					if (sink && src_dev &&
+					    dev_comp_type(src_dev) == SOF_COMP_MODULE_ADAPTER) {
+						struct processing_module *src_mod =
+							comp_mod(src_dev);
+
+						if (src_mod && src_mod->priv.cfg.base_cfg.obs)
+							sink_set_min_free_space(sink,
+										src_mod->priv.cfg.base_cfg.obs);
+					}
+					if (source && sink_dev &&
+					    dev_comp_type(sink_dev) == SOF_COMP_MODULE_ADAPTER) {
+						struct processing_module *sink_mod =
+							comp_mod(sink_dev);
+
+						if (sink_mod && sink_mod->priv.cfg.base_cfg.ibs)
+							source_set_min_available(source,
+										 sink_mod->priv.cfg.base_cfg.ibs);
+					}


lets inline comment each phase

Done in 80c0fb7 — added inline comments for each phase of the route buffer update loop (effective rate resolution, PCM parameter refresh, and period threshold recalculation).

In sof_static_topology_init(), Stage 7 calls sof_static_topology_init_controls() which dispatches to sof_static_kcontrol_set() to set initial default control values. However, sof_static_kcontrol_set() checks for an active topology via: if (!s_active_topo) return -ENODEV; Because s_active_topo was previously assigned only after Stage 7 succeeded, all initial kcontrol configurations failed with -ENODEV (-19). Publish s_active_topo prior to running Stage 7 so control lookups succeed, and ensure s_active_topo is reset to NULL in err_unwind if initialization fails. Signed-off-by: Liam Girdwood <lgirdwood@gmail.com>

… prepare Update the static pipeline loader to use the SOF sink/source and audio_buffer APIs instead of legacy comp_buffer / audio_stream accesses, and skip unrouted components during Stage 6 prepare: 1. In init_buffer_params(), configure stream parameters, frame/valid formats, alignment constants, min_free_space / min_available, and buffer reset via sink_set_params(), source_set_params(), and audio_buffer_reset(). 2. Register instantiated static components in ipc_get()->comp_list so IPC4 module bind/unbind callbacks (such as mixout_bind()) can resolve connected endpoints via ipc4_get_comp_dev(), with matching list cleanup in sof_static_topology_unwind(). 3. In sof_static_topology_init_routes() (Stage 4), validate per-component source/sink queue indices against CONFIG_MODULE_MAX_CONNECTIONS, IPC4 queue bitfield limits, and module max_sinks/max_sources before binding buffer sof_sink and sof_source interfaces via comp_bind(). Correspondingly call comp_unbind(), pipeline_disconnect(), and audio_buffer_free() in sof_static_topology_unwind(). 4. In sof_static_topology_prepare_comps() (Stage 6), check whether components have bound sinks/sources via sof_static_comp_has_sink_source() (checking mod->num_of_sources / mod->num_of_sinks for module adapter components and source_get_comp_state() / sink_get_state() for endpoint components) instead of inspecting dev->bsource_list and dev->bsink_list. 5. In sof_static_pipeline_set_sample_rate(), propagate fixed converter output rates (caps.sink_rate) downstream via sof_static_resolve_comp_in_rate(), query buffer format and channels via source_get_frm_fmt() and source_get_channels(), update IPC4 module base_cfg rate/IBS/OBS, and refresh sink/source thresholds. Signed-off-by: Liam Girdwood <lgirdwood@gmail.com>

lgirdwood force-pushed the topic/static-pipeline-fixes branch from 9084313 to a0838bc Compare October 15, 2026 18:25

lgirdwood changed the title ~~audio: pipeline: fix IPC3 support and control init in static pipeline~~ audio: pipeline: fix control init order and unrouted component prepare in static pipeline Oct 15, 2026

softwarecki reviewed Oct 29, 2026

View reviewed changes

lgirdwood force-pushed the topic/static-pipeline-fixes branch from a0838bc to 6cd4592 Compare November 1, 2026 20:37

lgirdwood marked this pull request as ready for review November 4, 2026 16:28

lgirdwood requested review from dbaluta, kv2019i, lbetlej, mmaka1 and plbossart as code owners November 4, 2026 16:28

Copilot AI balanced review requested due to automatic review settings November 4, 2026 16:28

Copilot started reviewing on behalf of lgirdwood November 4, 2026 16:28

View session

Copilot AI reviewed Nov 4, 2026

View reviewed changes

lgirdwood commented Nov 4, 2026

View reviewed changes

lgirdwood added 2 commits November 4, 2026 19:35

lgirdwood force-pushed the topic/static-pipeline-fixes branch from 6cd4592 to 80c0fb7 Compare November 4, 2026 19:39

kv2019i approved these changes Nov 6, 2026

View reviewed changes

Conversation

lgirdwood commented Oct 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

intel-sofci commented Nov 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR 11315: test results

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

lgirdwood commented Oct 15, 2026 •

edited

Loading

intel-sofci commented Nov 4, 2026 •

edited

Loading