Add support for hostless pipelines #5571

ranj063 · 2025-10-25T01:26:24Z

This PR introduces the implementation of hostless pipelines in the SOF
driver by leveraging virtual widgets for DAPM. Hostless pipelines are used
in scenarios where audio data does not originate from or terminate at the
host, such as internal DSP processing pipelines or feedback loops.

The implementation includes:

Support for virtual widgets in the topology to represent hostless pipelines.
These virtual widgets are solely for the purpose of propagating the
hw_params confirguation to the codec in the absence of a AIF component
Virtual widgets are treated as checkpoints to stop propagating widget
setup/free in the SOF driver. This patch updates the logic to the
widget setup and free routines to handle virtual widgets and the rest of
pipeline downstream from a virtual widget.

An example of a hostless pipeline in topology is:

Virtual input widget -> ToneGen -> mixin -> mixout -> gain -> DAI widget

This change enables the SOF driver to handle pipelines that are entirely
internal to the DSP, improving support for use cases like sidetone,
amplifier feedback, and other DSP-only processing scenarios.

Add a couple of bitfields in the IPC message extension for pipelines to include the direction. This will be useful to set the direction for modules belonging to hostless pipelines. Signed-off-by: Ranjani Sridharan <[email protected]>

Add support for the siggen type widgets in order to support hostless pipelines with the tone generator module in the firmware. Signed-off-by: Ranjani Sridharan <[email protected]>

In the case of hostless pipelines, there are no AIF-type widgets and the tone generator is the source of the pipeline. For such pipelines, skip setting the host config when setting the hw params. Signed-off-by: Ranjani Sridharan <[email protected]>

This patch introduces the implementation of hostless pipelines in the SOF driver by leveraging virtual widgets for DAPM. Hostless pipelines are used in scenarios where audio data does not originate from or terminate at the host, such as internal DSP processing pipelines or feedback loops. The implementation includes: - Support for virtual widgets in the topology to represent hostless pipelines. These virtual widgets are solely for the purpose of propagating the hw_params confirguation to the codec in the absence of an AIF component. - Virtual widgets are treated as checkpoints to stop propagating widget setup/free in the SOF driver. This patch updates the logic to the widget setup and free routines to handle virtual widgets and the rest of pipeline downstream from a virtual widget. An example of a hostless pipeline in topology is: Virtual input widget -> ToneGen -> mixin -> mixout -> gain -> DAI widget This change enables the SOF driver to handle pipelines that are entirely internal to the DSP, improving support for use cases like sidetone, amplifier feedback, and other DSP-only processing scenarios. Signed-off-by: Ranjani Sridharan <[email protected]>

ujfalusi

This only adds support for one type of use of a siggen, right?
Can you share some topology pictures how this suppose to look like?

In my mind the siggen is a DAI or an AIF, depending on the pipeline.

when it is an 'AIF':
siggen -> mixin -> mixout -> gain -> DAI widget

when it is a 'DAI':
siggen -> mixin -> mixout -> gain -> AIF widget

you can use the same module to generate audio to be played out on speaker, but you can use it to generate tone to be captured by host.
You can use the generated signal to be mixed with other audio and routed out to DAI or AIF.

From the tone generator's pow it does not matter if it is on capture or playback stream, doe it? It will generate signal to it's output and then that will traverse around in the path to the destination...

I also guess we can have a tone generator which emits 0 and does not hing else (special case of a tone I guess)

Another thing which would be a natural companion of tone generator is a /dev/null or black hole DAI/AIF which just eats up audio as it is receiving it..

This hostless playback can be enabled with a flip of a kcontrol? Or how this will work?

ujfalusi · 2025-10-29T06:53:33Z

include/sound/sof/ipc4/header.h

 #define SOF_IPC4_GLB_PIPE_EXT_CORE_ID(x)	((x) << SOF_IPC4_GLB_PIPE_EXT_CORE_ID_SHIFT)

+#define SOF_IPC4_GLB_PIPE_EXT_DIRECTION_SET_SHIFT	24
+#define SOF_IPC4_GLB_PIPE_EXT_DIRECTION_SET(x)	((x) << SOF_IPC4_GLB_PIPE_EXT_DIRECTION_SET_SHIFT)


SOF_IPC4_GLB_PIPE_EXT_DIRECTION_VALID ?

ujfalusi · 2025-10-29T07:05:45Z

sound/soc/sof/ipc4-topology.c


+	/* set pipeline direction */
+	pipeline->msg.extension |= SOF_IPC4_GLB_PIPE_EXT_DIRECTION_SET(0x1);
+	pipeline->msg.extension |= SOF_IPC4_GLB_PIPE_EXT_DIRECTION(dir);


Document what dir value is what direction, I assume playback is 0, capture is 1?
and to be safe with the integer:

pipeline->msg.extension |= SOF_IPC4_GLB_PIPE_EXT_DIRECTION(!!dir);

Another way to put this could be to
#define SOF_IPC4_GLB_PIPE_EXT_DIR_MASK GENMASK(25, 24)
#define SOF_IPC4_GLB_PIPE_EXT_DIR_SET(dir) ((dir + 1) << 24)

so:
0 means it is not provided
1 means it is playback direction
2 means capture
3 for later use

ujfalusi · 2025-10-29T07:57:49Z

sound/soc/sof/sof-audio.c

 #include "ops.h"

+/*
+ * Check is a DAI widget is an aggregated DAI. Aggregated DAI's have names ending in numbers


Check if a DAI widget...

ujfalusi · 2025-10-29T08:08:41Z

sound/soc/sof/sof-audio.c

+{
+	return (WIDGET_IS_DAI(swidget->id) &&
+	       isdigit(swidget->widget->name[strlen(swidget->widget->name) - 1]) &&
+	       swidget->widget->name[strlen(swidget->widget->name) - 1] != '0');


can you give some example how this looks like in real life?
dai-copier.HDA.iDisp1.playback - a read DAI
dai-copier.HDA.iDisp1.playback[1-9] - a fake DAI
?

What are these 'aggregate'? I have hard time to put this into context..

ujfalusi · 2025-10-29T08:13:38Z

sound/soc/sof/sof-audio.c

+		goto sink_unprepare;
+
+	/* skip aggregated DAIs */
+	if (!swidget || is_aggregated_dai(swidget))


I think this is going to cause change in behavior?

Previously we just returned in case of virtual_widget(), now we will proceed to sink_unprepare

Now we will return instead of proceeding to sink_unprepare if swidget is NULL.

Are you sure of these changes will not have side effects?

ujfalusi · 2025-10-29T08:14:26Z

sound/soc/sof/sof-audio.c

+	/* skip aggregated DAIs */
+	if (!swidget || is_aggregated_dai(swidget))
+		return 0;
+


Right, I think you are doing this deliberately...

ujfalusi · 2025-10-29T08:21:05Z

sound/soc/sof/sof-audio.c

+		return snd_sof_find_swidget_sname_type(scomp, pcm_name, snd_soc_dapm_input);
+	}
+
+	return snd_sof_find_swidget_sname_type(scomp, pcm_name, snd_soc_dapm_aif_out);


Should we look for snd_soc_dapm_output if aif_out is not present?

ujfalusi · 2025-10-29T08:22:08Z

sound/soc/sof/sof-audio.c

 		int i;

+		/* skip aggregated DAIs */
+		if (!swidget || is_aggregated_dai(swidget))


you are inside of if (swidget) {

ujfalusi · 2025-10-29T08:23:50Z

sound/soc/sof/sof-audio.c

+		    widget->id != snd_soc_dapm_input))
 			continue;

 		/* starting widget for capture is DAI type */


Similar change for capture?

ujfalusi · 2025-10-29T09:17:19Z

sound/soc/sof/sof-audio.c

+	if (!swidget || is_aggregated_dai(swidget))
+		return 0;
+
+	err = sof_widget_free(sdev, widget->dobj.private);


you can use swidget as parameter.

singalsu · 2025-10-29T09:46:48Z

Another thing which would be a natural companion of tone generator is a /dev/null or black hole DAI/AIF which just eats up audio as it is receiving it..

We'd need this kind of black hole component also for audio signal analysis, e.g. SPL meter that updates it's ALSA control for measured sound level. It reads and uses the source samples but produces no other output than occasional update of a control with notification to user space.

ranj063 · 2025-12-02T01:55:20Z

replaced with #5609

ranj063 requested review from bardliao, dbaluta, kv2019i, lgirdwood, lyakh, plbossart and ujfalusi as code owners October 25, 2025 01:26

ranj063 added 3 commits October 24, 2025 20:08

ASoC: SOF: ipc4-topology: Add support for siggen widget

fed3d97

Add support for the siggen type widgets in order to support hostless pipelines with the tone generator module in the firmware. Signed-off-by: Ranjani Sridharan <[email protected]>

ranj063 force-pushed the sof-dev-101525 branch from 0720941 to fb66474 Compare October 25, 2025 03:08

ujfalusi reviewed Oct 29, 2025

View reviewed changes

ranj063 closed this Dec 2, 2025

Add support for hostless pipelines #5571

Add support for hostless pipelines #5571

Uh oh!

Conversation

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

Uh oh!

Uh oh!

ujfalusi left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

singalsu commented Oct 29, 2025

Uh oh!

ranj063 commented Dec 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ranj063 commented Oct 25, 2025 •

edited

Loading

ujfalusi left a comment •

edited

Loading