Add support for hostless/dailess pipelines #10329

ranj063 · 2025-10-25T22:28:05Z

Preparatory patches with updates to the tone module and topology support for the tone and pipeline direction attribute.

Copilot

Pull Request Overview

This PR adds support for hostless pipelines using a tone generator as an example. The changes enable pipelines to run without a host connection by introducing direction awareness and implementing the tone generator as a modular component.

Key changes:

Implemented pipeline direction tracking in IPC4 to support hostless operation
Refactored tone generator from legacy component to module adapter architecture
Added configuration support for tone generator in SDW platform topologies

Reviewed Changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tools/topology/topology2/platform/intel/sdw-amp-generic.conf	Adds tone generator pipeline configuration with PCM definition and routing
tools/topology/topology2/include/pipelines/virtual-siggen-mixin-playback.conf	Defines new virtual signal generator to mixer pipeline for tone playback
tools/topology/topology2/include/components/virtual.conf	Adds stream_name attribute to virtual widget component
tools/topology/topology2/include/components/siggen.conf	New signal generator component definition with configuration attributes
tools/topology/topology2/cavs-sdw.conf	Includes new pipeline and adds SPK_TONE_PLAYBACK configuration flag
src/ipc/ipc4/helper.c	Extracts and stores pipeline direction from IPC extension data
src/include/sof/audio/pipeline.h	Adds direction tracking fields to pipeline structure
src/include/ipc4/pipeline.h	Defines direction bitfields in IPC4 pipeline extension structure
src/audio/tone.toml	Adds module configuration manifest for tone generator
src/audio/tone.c	Refactors tone generator to module adapter interface with sink API
src/audio/module_adapter/module_adapter.c	Propagates pipeline direction to module devices

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

Copilot · 2025-10-25T22:28:52Z

tools/topology/topology2/include/components/siggen.conf

+# A generic siggen widget. All attributes defined herein are namespaced
+# by alsatplg to "Object.Widget.siggen.N.attribute_name"
+#
+# Usage: this component can be used by declaring int a parent object. i.e.


Corrected spelling of 'int' to 'in'.

Suggested change

# Usage: this component can be used by declaring int a parent object. i.e.

# Usage: this component can be used by declaring in a parent object. i.e.

Copilot · 2025-10-25T22:28:52Z

tools/topology/topology2/include/components/siggen.conf

+	# Attribute categories
+	attributes {
+		#
+		# The sighen widget name would be constructed using the index and instance attributes.


Corrected spelling of 'sighen' to 'siggen'.

Suggested change

# The sighen widget name would be constructed using the index and instance attributes.

# The siggen widget name would be constructed using the index and instance attributes.

Copilot · 2025-10-25T22:28:52Z

src/audio/tone.c

+	int ret;
+
+	/* tone generator only ever has 1 sink */
+	output_frames = sink_get_free_frames(sinks[0]);


Inconsistent array indexing. Line 395 uses sinks[0] while line 389 assigns sink = sinks[0]. Consider using the sink variable consistently throughout the function (e.g., sink_get_free_frames(sink)).

kv2019i

One question on design.

kv2019i · 2025-10-27T08:41:55Z

src/audio/module_adapter/module_adapter.c

+		if (dev->pipeline->direction_set) {
+			dev->direction_set = true;
+			dev->direction = dev->pipeline->direction;
+		}


Hmm, ipc_comp_connect can propagates direction both ways, so shouldn't it be possible to set component directions based on "DAI copier"? The direction is passed to DAI copier already from host "get_gateway_direction(node_id.f.dma_type)", so direction is known.

@kv2019i ipc_comp_connect checks both sides but the trigger is always propagated from source->sink. So if there's no host copier, the tone pipeline wont have a source direction to hinge upon

@ranj063 Right but can we extend this? Seems FW has enough information to do the right thing without any new information from host (even if it doesn't do it now).

@kv2019i I dont want to extend it at all, in fact Its quite a convoluted logic having to walk through the entire pipeline just to set the component's direction. This should be simplified

@ranj063 Couldn't this module be converted into some virtual/fake DAI and replace the host copier with it?

@ranj063 Ack. I guess this is more inline with the IPC4 philosophy of letting host provide all programming. It does still feel a bit redundant to me. The whole notion of direction is a bit redundant in describing a graph when you already provide have a description of a graph and how nodes are connected. It seems only entities that have some special functionality around playback/capture are the copier/endpoint modules and for these there's already a direction flag passed (e.g. in copier config blob). So if "direction" is only related to the endpoints, looking up the direction by traversin the graph topology, does not seem so convoluted.

tmleman · 2025-10-27T09:38:47Z

src/include/ipc4/pipeline.h

 			uint32_t core_id        : 4;
-			uint32_t rsvd2          : 6;
+			uint32_t direction_set	: 1;
+			uint32_t direction	: 1;
+			uint32_t rsvd2          : 4;
 			uint32_t _reserved_2    : 2;


Changes in these structures require prior synchronization and acceptance from the architecture side.

@tmleman ack. How do I go about this?

lyakh · 2025-10-27T13:10:59Z

src/include/sof/audio/pipeline.h

 	uint32_t time_domain;	/**< scheduling time domain */
 	uint32_t attributes;   /**< pipeline attributes from IPC extension msg/ */
+	uint32_t direction_set;	/**< flag indicating if direction set from IPC extension msg? */
+	uint32_t direction;	/**< pipeline direction from IPC extension msg */


lyakh · 2025-10-27T13:18:08Z

src/audio/tone.c

-	dev->ipc_config = *config;
+	comp_info(dev, "tone_new()");

 	cd = rzalloc(SOF_MEM_FLAG_USER, sizeof(*cd));


mod_alloc() while at it to save @jsarha some work :-)

lyakh · 2025-10-27T13:25:00Z

src/audio/tone.c

-static SHARED_DATA struct comp_driver_info comp_tone_info = {
-	.drv = &comp_tone,
-};
+#if CONFIG_COMP_TONE_MODULE


it isn't tristate yet, it's boolean, LLEXT isn't supported for the tone generator yet, no need for this branch yet

lyakh · 2025-10-27T13:25:35Z

src/audio/tone.toml

@@ -0,0 +1,21 @@
+#ifndef LOAD_TYPE


this file is probably unneeded yet - nobody includes it

lgirdwood · 2025-10-27T16:53:05Z

tools/topology/topology2/include/components/virtual.conf

+	#
+	## Stream name - maps to the DAPM widget's stream name
+	#
+	DefineAttribute."stream_name" {


Should this be virtual_stream_name ? i.e. if if for virtual usage only we best name/state that so its used correctly.

The need to pass direction was discussed, I don't have strong opposition on either path, just wanted to check the options have been considered.

ranj063 · 2025-11-21T00:33:03Z

@kv2019i @singalsu Ive dropped the extension of the pipeline IPC message and used the pin logic in the tone module to set the direction now.

lyakh · 2025-11-21T07:21:28Z

src/audio/tone.c

-static void tone_s32_default(struct comp_dev *dev, struct audio_stream *sink,
-			     uint32_t frames)
+static int tone_s32_passthrough(struct processing_module *mod, struct sof_sink *sink,
+				struct sof_source *source)


why does a tone generator have to support pass-through? Was it also possible with IPC3?

The tone generator operates as silence generator in this case, when there is no playback happening. Silence is a kind of audio signal too.

@lyakh with IPC3 we didnt have firmware based echo reference feature. This new addition to do passthrough is required for supporting that with IPC4 for SDW topologies.

The tone generator operates as silence generator in this case, when there is no playback happening. Silence is a kind of audio signal too.

@singalsu interesting, thanks. I think we also have ad-hoc silence generation in some components. Is it planned to remove those and use "tone" in all cases?

I'd assume those silence generator capabilities remain like now. E.g. mixin/mixout operation is defined by IPC4. While with this component with no Windows OS dependence we can define it as we need.

lgirdwood

@ranj063 feel free to fix the spelling in a subsequent PR if CI passes

singalsu · 2025-11-21T17:15:46Z

src/audio/tone.c

-
-static void tone_s32_default(struct comp_dev *dev, struct audio_stream *sink,
-			     uint32_t frames)
+static int tone_s32_passthrough(struct processing_module *mod, struct sof_sink *sink,


Can we use for passthrough the helper source_to_sink_copy(). There's an example in

sof/src/audio/template/template.c

Line 85 in 9a3ae61

source_to_sink_copy(source, sink, true, frames * cd->frame_bytes);

Can we do this as a future optimzation @singalsu

dbaluta · 2025-11-24T07:36:07Z

@ranj063 I wonder how difficult for your change would be to keep the support for IPC3. We want to add this feature to our release too but IMX doesnt support IPC4 yet.

src/audio/tone.c

ranj063 · 2025-12-03T16:35:04Z

@ranj063 I wonder how difficult for your change would be to keep the support for IPC3. We want to add this feature to our release too but IMX doesnt support IPC4 yet.

@dbaluta I've left the ipc3 version as is now and created a new implementation for IPC4

lyakh · 2025-12-04T09:36:20Z

src/audio/tone/tone-ipc4.c

+			n_wrap_dest = output_end - output_pos;
+
+			/* Process until wrap or completed n */
+			n_min = (n < n_wrap_dest) ? n : n_wrap_dest;


here a MIN() would fit perfectly

lyakh · 2025-12-04T09:36:55Z

src/audio/tone/tone-ipc4.c

+	/* sg->w is angle in Q4.28 radians format, sin() returns Q1.31 */
+	/* sg->a is amplitude as Q1.31 */
+	sine =
+		q_mults_32x32(sin_fixed_32b(sg->w), sg->a,


merge with previous line please

lyakh · 2025-12-04T09:37:27Z

src/audio/tone/tone-ipc4.c

+
+	if (sg->mute)
+		return 0;
+	else


else not needed

lyakh · 2025-12-04T09:38:22Z

src/audio/tone/tone-ipc4.c

+	int64_t w;
+	/* sg->w is angle in Q4.28 radians format, sin() returns Q1.31 */
+	/* sg->a is amplitude as Q1.31 */
+	sine =


I think it's checkpatch that has a habit of complaining about missing empty lines between variable declarations and code, not sure if comments make it happy though

lyakh · 2025-12-04T09:40:31Z

src/audio/tone/tone-ipc4.c

+	sg->f = (f > f_max) ? f_max : f;
+	/* Q16 x Q31 -> Q28 */
+	w_tmp = q_multsr_32x32(sg->f, sg->c, Q_SHIFT_BITS_64(16, 31, 28));
+	w_tmp = (w_tmp > PI_Q4_28) ? PI_Q4_28 : w_tmp; /* Limit to pi Q4.28 */


lyakh · 2025-12-04T09:41:14Z

src/audio/tone/tone-ipc4.c

+	int i;
+
+	sg->a_target = a;
+	sg->a = (sg->ramp_step > sg->a_target) ? sg->a_target : sg->ramp_step;


lyakh · 2025-12-04T09:41:45Z

src/audio/tone/tone-ipc4.c

+	 */
+	for (i = 0; i < TONE_NUM_FS; i++) {
+		if (fs == tone_fs_list[i])
+			idx = i;


lyakh · 2025-12-04T09:42:44Z

src/audio/tone-ipc4.c

+
+	comp_info(dev, "tone_new()");
+
+	cd = rzalloc(SOF_MEM_FLAG_USER, sizeof(*cd));


mod_alloc()

lyakh · 2025-12-04T09:45:53Z

src/audio/tone/tone-ipc4.c

+	struct comp_data *cd = module_get_private_data(mod);
+	int i;
+
+	comp_info(mod->dev, "tone_reset()");


please check and remove all function names from logs, also please try to reduce the number of *_info() level logs, e.g. this one doesn't seem to be really important

@lyakh all of these comments above are on existing code. I will fix the mod_alloc() but Im going to leave the rest for now for the sake of consistency with the xisting code in tone.c. We can clean it up later

lyakh · 2025-12-04T09:47:36Z

src/audio/tone/tone-ipc4.c

+};
+
+#if CONFIG_COMP_TONE_MODULE
+/* modular: llext dynamic link */


again, unless you're making it tristate, this isn't needed, as well as the .toml file below.

@lyakh @singalsu requested to keep the toml file as he is testing with it

@ranj063 then do you maybe have the remaining code for making it a real LLEXT? Not much is left - "tristate" in Kconfig, a couple of simple cmake changes, that can be copied from plentiful examples?

Add a new implementation for the tone modeule for IPC4 with no kcontrol support as of now. This will be added later. Signed-off-by: Ranjani Sridharan <[email protected]>

Create a new conf file for the siggen module. Signed-off-by: Ranjani Sridharan <[email protected]>

Add 2 new tokens for pipeline direction and atttributes for the pipeline component. This change is required to implement the firmware based echo reference feature to inform the kernel of the pipeline direction in order for it to make a decision on which pipelines to set up when the echo reference capture pipeline is set up. The direction_valid attribute is required for backward-compatibility in the kernel driver for older topologies where the direction is not set. Signed-off-by: Ranjani Sridharan <[email protected]>

Copilot AI review requested due to automatic review settings October 25, 2025 22:28

ranj063 requested review from RanderWang, dbaluta, jsarha, jxstelter, kv2019i, lbetlej, lgirdwood, marcinszkudlinski, mmaka1, pblaszko, plbossart and singalsu as code owners October 25, 2025 22:28

Copilot AI reviewed Oct 25, 2025

View reviewed changes

kv2019i previously requested changes Oct 27, 2025

View reviewed changes

tmleman requested changes Oct 27, 2025

View reviewed changes

lyakh reviewed Oct 27, 2025

View reviewed changes

lgirdwood reviewed Oct 27, 2025

View reviewed changes

ranj063 force-pushed the fix/hostless_pipelines branch from 89e64f3 to 2ee7695 Compare November 21, 2025 00:31

ranj063 changed the title ~~Add support for hostless pipelines (ex: tone generator)~~ Add support for hostless/dailess pipelines Nov 21, 2025

ranj063 requested review from kv2019i, lgirdwood and lyakh November 21, 2025 00:32

lyakh reviewed Nov 21, 2025

View reviewed changes

lgirdwood approved these changes Nov 21, 2025

View reviewed changes

ranj063 force-pushed the fix/hostless_pipelines branch 2 times, most recently from a6f5ed5 to f4bbd3a Compare November 21, 2025 15:38

singalsu mentioned this pull request Nov 21, 2025

Tools: Topology: Add echo reference for SDW speaker and jack #10387

Draft

singalsu reviewed Nov 21, 2025

View reviewed changes

singalsu reviewed Nov 24, 2025

View reviewed changes

src/audio/tone.c Show resolved Hide resolved

singalsu reviewed Nov 24, 2025

View reviewed changes

src/audio/tone.c Outdated Show resolved Hide resolved

singalsu mentioned this pull request Dec 2, 2025

Add support for hostless and DAIless pipelines thesofproject/linux#5609

Open

ranj063 force-pushed the fix/hostless_pipelines branch from f4bbd3a to 99f1100 Compare December 3, 2025 16:32

ranj063 force-pushed the fix/hostless_pipelines branch from 99f1100 to a94e307 Compare December 3, 2025 19:22

lyakh reviewed Dec 4, 2025

View reviewed changes

ranj063 added 3 commits December 7, 2025 21:22

audio: tone: Add a new implementation for IPC4

7aac26d

Add a new implementation for the tone modeule for IPC4 with no kcontrol support as of now. This will be added later. Signed-off-by: Ranjani Sridharan <[email protected]>

topology2: components: siggen: Add new module

a8e314a

Create a new conf file for the siggen module. Signed-off-by: Ranjani Sridharan <[email protected]>

ranj063 force-pushed the fix/hostless_pipelines branch from a94e307 to 876c06c Compare December 8, 2025 05:23

	# Usage: this component can be used by declaring int a parent object. i.e.
	# Usage: this component can be used by declaring in a parent object. i.e.

	# The sighen widget name would be constructed using the index and instance attributes.
	# The siggen widget name would be constructed using the index and instance attributes.


		comp_info(dev, "tone_new()");

		cd = rzalloc(SOF_MEM_FLAG_USER, sizeof(*cd));

Add support for hostless/dailess pipelines #10329

Are you sure you want to change the base?

Add support for hostless/dailess pipelines #10329

Conversation

ranj063 commented Oct 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Copilot AI Oct 25, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 25, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 25, 2025

Choose a reason for hiding this comment

Uh oh!

kv2019i left a comment

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!

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!

ranj063 commented Nov 21, 2025

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!

lgirdwood left a comment

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!

dbaluta commented Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ranj063 commented Dec 3, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ranj063 commented Oct 25, 2025 •

edited

Loading

dbaluta commented Nov 24, 2025 •

edited

Loading