soundwire: add runtime BPT support by bardliao · Pull Request #5604 · thesofproject/linux

bardliao · 2025-11-26T08:28:25Z

The driver may want to use BPT stream to download the firmware while an audio stream is running. This series calculates the available columns in a SoundWire frame and use them for the BPT stream. The streams are on on a first-come, first-served basis but will reserve 2 columns for the audio stream if possible. That should apply to most cases.

Copilot

Pull request overview

This PR implements runtime BPT (Bulk Point-to-Point) stream support for SoundWire, enabling firmware downloads while audio streams are active. The changes calculate available columns in SoundWire frames and allocate them to BPT streams on a first-come, first-served basis, with a reservation of 2 columns for audio streams when possible.

Key changes:

Modified bandwidth calculation to exclude BPT streams and use remaining available bandwidth
Added bpt_hstop field to track BPT stream column allocation
Updated frame column allocation logic to dynamically reserve space for audio streams

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
include/linux/soundwire/sdw.h	Added `bpt_hstop` field to track BPT stream column boundaries
drivers/soundwire/stream.c	Excluded BPT streams from bandwidth calculations during prepare/deprepare
drivers/soundwire/intel_ace2x.c	Updated to use `bpt_hstop` instead of total columns for BPT buffer calculations
drivers/soundwire/generic_bandwidth_allocation.c	Implemented dynamic column allocation logic with BPT stream filtering and reservation

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

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

bardliao · 2026-06-26T12:34:23Z

 restore_params:
 	memcpy(&bus->params, &params, sizeof(params));
+	bus->bpt_hstop = prev_bpt_hstop;
 	return ret;


This should belong to another PR. I will fix it later.

The restore_params block in _sdw_prepare_stream() was restoring only the last bus iterated when an error occurred. For multi-link streams with multiple buses, this left other buses in a partially reconfigured state. Fix this by: 1. Adding bpt_hstop to sdw_bus to track the BPT stream hstop value. 2. Adding saved_params and saved_bpt_hstop to sdw_master_runtime to store per-bus backup state. 3. Pre-saving all buses' params and bpt_hstop before the modify loop. 4. Restoring all buses from their per-bus saved state in restore_params. Addresses review comment: #5604 (comment)

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

The lane_used_bandwidth variable store the used bandwidth of each lane. It should belong to the sdw_bus_params struct instead of in the sdw_bus struct directly. Besides, _sdw_prepare_stream() will restore the bus->params, and the lane_used_bandwidth values will not be restored if it is not in the params. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

Restore_params restores bus->params and bus->bpt_hstop from single backups that are overwritten in the first master_list loop. If failure happens later (e.g. do_bank_switch() or port prepare in the second loop), the current bus may not match the saved snapshot, so multi-link streams can restore the wrong state into the wrong bus and leave other buses partially reconfigured. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

We should get as much as bandwidth for the BPT stream. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

The DP0 (BPT) params are computed in sdw_compute_dp0_port_params(). We should exclude the BPT stream when calculating the audio streams. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

We just need to conunt the audio stream bandwidth and BRA stream will use the remaining bandwidth. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

colume The existing code assumes the column number will not change, but it could change if curr_dr_freq changes. Calculate the new column number before checking the bandwidth to make the checking be more accurate. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

To allow BPT and audio stream work simultaneously, we need to record the hstop of the BPT stream. And use column bpt_hstop + 1 to the last column for audio streams. No function changed since bus->params.bpt_hstop is set to bus->params.col - 1 for now. Will update bus->audio_stream_hstart in the follow up commit. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

Update bus->params.bpt_hstop to record the hstop of the BPT stream. And return -EAGAIN when there is no bandwidth for the BPT stream. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

When a BPT stream is running, we should subtract the columns that is used by the BPT stream in bandwidth calculation. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

Now the SoundWire BPT stream and the audio stream can share the SoundWire bus bandwidth. However, it is still not allowed to have more than 1 BPT stream running simultaneously. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

Copilot

Pull request overview

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

bardliao · 2026-07-02T06:34:29Z

 	struct sdw_master_runtime *m_rt;
 	struct sdw_bus *bus;
 	struct sdw_master_prop *prop;
-	struct sdw_bus_params params;
 	int ret;



ret = do_bank_switch(stream); is always called. In other words, ret will never be uninitialized

bardliao · 2026-07-16T01:36:08Z

@charleskeepax @simontrimmer @rfvirgil Could you take a look?

plbossart

The problem is the same as in other BPT/BRA PRs, all the logic is driven by the use of non-audio bandwidth at a given time without clarifying what happens if another audio stream starts while BPT/BRA is running.
Returning -EBUSY for an audio stream is a big no-no IMHO.

I think we will need to model the fact that the BPT/BRA makes the most sense during initial and resume stages, which could avoid the need for these complicated concurrency policies.

plbossart · 2026-08-06T12:01:06Z

 continue;
 }
+
+ /* Don't count BPT stream bandwidth, it will use the remaining bandwidth */


that's a problem, if you allocated all the remaining bandwidth to BPT/BRA, then what happens if a new stream starts while BPT/BRA is on-going?
If you want to have concurrent audio/BPT streams, you need to have enough bandwidth for all, and make sure the 'remaining' non-audio part is still enough for BPT/BRA to be efficient compared to the command protocol.

I reserve 2 columns for the audio stream. I think it should be good enough for in practice.

bardliao · 2026-08-06T14:10:49Z

The problem is the same as in other BPT/BRA PRs, all the logic is driven by the use of non-audio bandwidth at a given time without clarifying what happens if another audio stream starts while BPT/BRA is running. Returning -EBUSY for an audio stream is a big no-no IMHO.

Yes, audio stream should always be at the first priority.

I think we will need to model the fact that the BPT/BRA makes the most sense during initial and resume stages, which could avoid the need for these complicated concurrency policies.

One scenario is that a codec and an amp are on the same bus and the amp needs to load the firmware before starting playback. When we switch from headphone to speaker while a stream is active, it is more efficient to download the firmware via BRA.
@charleskeepax Any input on this?

bardliao · 2026-08-10T09:17:58Z

@rfvirgil @simontrimmer FYI

rfvirgil · 2026-08-10T09:47:55Z

@plbossart

I think we will need to model the fact that the BPT/BRA makes the most sense during initial and resume stages, which could avoid the need for these complicated concurrency policies.

We can't limit BRA to only happening during boot and resume. There are real reasons why it can be needed at other times.

Apart from firmware download, the other big use-case is voice trigger.

for extracting faster-than-realtime voice trigger audio from buffers on the device.
for downloading voice trigger model data (our experience with Android voice trigger is that setting the trigger model can happen any time, it's not a boot-time-only operation)

Also the SDCA spec allows a device to request a "firmware" (really, any blob of data) at any time to satisfy a use-case.

bardliao requested review from charleskeepax, Copilot, shumingfan and ujfalusi November 26, 2025 08:28

CopilotAI reviewed Nov 26, 2025
View reviewed changes

Comment threadinclude/linux/soundwire/sdw.h Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated

bardliao force-pushed the runtime-bra branch 2 times, most recently from f0c9b03 to 7a5ec92Compare November 27, 2025 03:05

bardliao force-pushed the runtime-bra branch from 7a5ec92 to 2e1bb1eCompare June 25, 2026 11:48

bardliao requested a review from Copilot June 25, 2026 11:50

Copilotstarted reviewing on behalf of bardliao June 25, 2026 11:50 View session

CopilotAI reviewed Jun 25, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated

bardliao force-pushed the runtime-bra branch from 2e1bb1e to 88f70a7Compare June 25, 2026 14:33

bardliao requested a review from Copilot June 25, 2026 14:33

Copilotstarted reviewing on behalf of bardliao June 25, 2026 14:37 View session

CopilotAI reviewed Jun 25, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c
Comment threaddrivers/soundwire/stream.c

bardliao force-pushed the runtime-bra branch from 88f70a7 to 90601f0Compare June 26, 2026 08:19

bardliao requested a review from Copilot June 26, 2026 09:51

Copilotstarted reviewing on behalf of bardliao June 26, 2026 09:52 View session

CopilotAI reviewed Jun 26, 2026
View reviewed changes

CopilotAI mentioned this pull request Jun 26, 2026
soundwire: fix restore_params to recover all buses on prepare failure #5829
Closed

bardliao force-pushed the runtime-bra branch from 90601f0 to cf490b1Compare June 26, 2026 12:35

bardliao requested a review from Copilot June 26, 2026 12:39

Copilotstarted reviewing on behalf of bardliao June 26, 2026 12:41 View session

CopilotAI reviewed Jun 26, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c

bardliao force-pushed the runtime-bra branch from cf490b1 to 4e15f44Compare June 26, 2026 14:21

bardliao requested a review from Copilot June 26, 2026 14:22

Copilotstarted reviewing on behalf of bardliao June 26, 2026 14:23 View session

CopilotAI reviewed Jun 26, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c

bardliao force-pushed the runtime-bra branch from 4e15f44 to e8c6536Compare June 29, 2026 13:15

bardliao added 7 commits July 2, 2026 10:42

soundwire: use maximum sdw bus rate when BPT stream is running
7f1bbc3
We should get as much as bandwidth for the BPT stream. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

soundwire: generic_bandwidth_allocation: don't deal with BPT stream
534ecb4
The DP0 (BPT) params are computed in sdw_compute_dp0_port_params(). We should exclude the BPT stream when calculating the audio streams. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

soundwire: don't count BPT bandwidth
c8b36b8
We just need to conunt the audio stream bandwidth and BRA stream will use the remaining bandwidth. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

bardliao force-pushed the runtime-bra branch from 5d0215e to 3bbb52bCompare July 2, 2026 03:11

bardliao requested a review from Copilot July 2, 2026 03:14

Copilotstarted reviewing on behalf of bardliao July 2, 2026 03:14 View session

CopilotAI reviewed Jul 2, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated

bardliao force-pushed the runtime-bra branch from 3bbb52b to 94f9981Compare July 2, 2026 03:28

bardliao requested a review from Copilot July 2, 2026 03:28

Copilotstarted reviewing on behalf of bardliao July 2, 2026 03:29 View session

CopilotAI reviewed Jul 2, 2026
View reviewed changes

Comment threaddrivers/soundwire/generic_bandwidth_allocation.c
Comment threaddrivers/soundwire/generic_bandwidth_allocation.c Outdated

bardliao added 3 commits July 2, 2026 14:06

soundwire: update bpt_hstop
97d68a4
Update bus->params.bpt_hstop to record the hstop of the BPT stream. And return -EAGAIN when there is no bandwidth for the BPT stream. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

soundwire: subtract BPT columns in bandwidth calculation
e880ea3
When a BPT stream is running, we should subtract the columns that is used by the BPT stream in bandwidth calculation. Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>

bardliao force-pushed the runtime-bra branch from 94f9981 to b923d9fCompare July 2, 2026 06:11

bardliao requested a review from Copilot July 2, 2026 06:14

Copilotstarted reviewing on behalf of bardliao July 2, 2026 06:15 View session

CopilotAI reviewed Jul 2, 2026
View reviewed changes

bardliao marked this pull request as ready for review July 3, 2026 11:14

bardliao changed the title ~~[RFC] soundwire: add runtime BPT support~~soundwire: add runtime BPT supportJul 16, 2026

bardliao mentioned this pull request Aug 3, 2026
BRA/BPT Support for AMD Platforms(ACP 7.0/7.1/7.2 variants) and code fixes #5867
Open

plbossart reviewed Aug 6, 2026
View reviewed changes

Uh oh!

Conversation

bardliao commented Nov 26, 2025

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

bardliaoJun 26, 2026

Choose a reason for hiding this comment

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

CopilotAI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

bardliaoJul 2, 2026

Choose a reason for hiding this comment

Uh oh!

bardliao commented Jul 16, 2026

Uh oh!

plbossart left a comment

Choose a reason for hiding this comment

Uh oh!

plbossartAug 6, 2026

Choose a reason for hiding this comment

Uh oh!

bardliaoAug 6, 2026

Choose a reason for hiding this comment

Uh oh!

bardliao commented Aug 6, 2026