IPC: introduce new IPC message service and rework the Intel Audio DSP host IPC driver #91606

dcpleung · 2025-06-13T20:57:43Z

This introduces a new IPC message service which is based on the idea of exchanging messages via IPC. This is in contrast to the other IPC service which is based on exchanging data as raw byte arrays.

The first customer of this new API is the Intel Audio DSP host IPC driver. It is reworked to be a backend to the IPC message service. This is the first step (of many) in moving host IPC communication in SOF (Sound Open Firmware) to utilizing a more generic API.

kv2019i

@dcpleung this looks very good and the old-on-new layer will make it easy to integrate this. I'll try to test this tomorrow with SOF.

Link back to SOF ticket thesofproject/sof#9697

kv2019i · 2025-07-01T16:35:56Z

soc/intel/intel_adsp/Kconfig

+	  Deprecated config for IPC. Will be removed in the future.
+
+config INTEL_ADSP_IPC_OLD_INTERFACE
+	bool "Expose old interface for the IPC"


This is old-on-top-of-new is really cool and eases the transition!

I skimmed through the SOF code and I think it is possible to apply similar changes in common/ipc.c to utilize the new interface.

lgirdwood

LGTM.

kv2019i

Did a test run with SOF on a ACE1.5 device (and using the old interface so no changes to SOF side yet) and seems to be working great. It worked so good I had to increase IPC_SERVICE log level a bit and verify it is really being used :) ... and it was.

lyakh · 2025-07-04T10:43:59Z

include/zephyr/ipc/ipc_msg_service.h

+	/**
+	 * @brief New packet arrived.
+	 *
+	 * This callback is called when new data is received.


should we mention that it can be called in interrupt context?

Added a note in the doxygen.

lyakh · 2025-07-04T10:49:03Z

subsys/ipc/ipc_msg_service/backends/ipc_msg_intel_adsp_ipc.c

+		if (ept_cfg->cb.event != NULL) {
+			int ret = ept_cfg->cb.event(INTEL_ADSP_IPC_EVT_DONE, NULL, ept_cfg->priv);
+
+			if (ret == INTEL_ADSP_IPC_EVT_RET_EXT_COMPELTE) {


looks like a typo "COMPLETE"

lyakh · 2025-07-04T11:05:04Z

tests/boards/intel_adsp/smoke/src/cpus.c

@@ -120,7 +120,7 @@ static void core_smoke(void *arg)
 	       clk_ratios[cpu] / 1000, clk_ratios[cpu] % 1000);

 	for (int i = 0; i < cpu; i++) {
-		int32_t diff = MAX(1, abs(clk_ratios[i] - clk_ratios[cpu]));
+		int32_t diff = MAX(1, (uint32_t)llabs((int64_t)clk_ratios[i] - clk_ratios[cpu]));


would a simpler abs((int32_t)(clk_ratios[i] - clk_ratios[cpu])) be enough?

Clock is not fast enough to need the signed bit. So we can do that to avoid 64-bit math.

This removes the IPM test from the intel_adsp smoke test. The IPM driver is not being used by SOF and is there simply for this test. The host IPC functionality is also being tested via test_host_ipc. The additional usage of mailbox in the test does not provide much value since the mailbox is shared memory between the DSP and the host. We would have bigger problems if the shared memory does not work properly. Removing the test is in preparation to removing the CAVS host IPM driver. That is also a preparation to rewrite the host IPC driver to the IPC API, and there is no need to maintain another host IPC driver that is not used by SOF. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This IPM driver is not being used by SOF and is there simply for a test which does not provide much additional value compared to the existing host IPC tests. That IPM specific test has been removed so there is no need to keep this driver in the tree. This is in preparation to rewrite the host IPC driver to utilize the IPC API. There is no need to maintain another driver that is not being used in SOF. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

The SoC specific IPC driver is for host IPC, and not IDC (which is between CPUs). So there is no need to use the IDC devicetree binding to enable the kconfig. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This introduces a new IPC mechanism: IPC message service, where IPC transactions are centered around messages instead of raw byte streams. This provides a bit more clarity in the code as this shows what type of message is being sent. This also allows backend to select the delivery mechanism based on the message type for supported hardware. This models after the existing IPC service and backend. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This adds a few basic tests for the IPC message service. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This reworks the host IPC driver as a backend of the newly introduced IPC message service. This is the first step to rework IPC in SOF (Sound Open Firmware) into using a more generic IPC API instead of a SoC specific one. For now, it keeps the old interface to maintain usability as it is going to be a multiple process to rework IPC over there. Also, the structure of the new IPC backend resembles the SoC specific driver to make it easier to compare between them at this first iteration. Future optimizations will probably be needed once we start modifying the SOF side to utilize the new IPC message service API. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

The IPC is only needed for old hardware earlier than CAVS 2.5. They are no longer supported in the tree as their board configs have been removed. So there is no need to do IPC anymore. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

On ACE, the CPU halting requires the cooperation of the PM subsys to set the CPU to SOFT_OFF. The mechanism has not been implemented in the smoke test. So skip the halting test on ACE for now. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

The 4th CPU test tries to gauge tight loop performance. However, it has been observed occasionally that the calculated would go over just a bit. The threshold was 3, and the observed ratio has been bouncing between 2.9 and 3.1. So up the threshold a little bit to account for that. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

Fix complier warning about using abs() with unsigned value. So typecast it into signed 32-bit value first before feeding it to abs(). The clock on DSP is not fast enough to need 64-bit math for the extra one signed bit as 32-bit value. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This adds ACE boards to the allowed platforms so they can be tested. Note that this only adds boards that I have tested to run correctly with the tests. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This updates the Intel Audio DSP smoke test to use the new IPC API based on IPC message service. The old API is still being tested for now until all users of the old API has moved to the new one. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

This adds documentation for the newly introduced IPC message service APIs. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

sonarqubecloud · 2025-07-19T01:20:11Z

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
1.3% Duplication on New Code

See analysis details on SonarQube Cloud

dcpleung requested review from kv2019i and nashif June 13, 2025 21:00

dcpleung force-pushed the intel_adsp/rework_ipc branch 9 times, most recently from cb760a0 to 241d4be Compare June 20, 2025 16:39

dcpleung force-pushed the intel_adsp/rework_ipc branch from 241d4be to 4900a7c Compare June 27, 2025 22:00

nashif added area: IPC Inter-Process Communication Architecture Review Discussion in the Architecture WG required labels Jun 30, 2025

github-project-automation bot added this to Architecture Review Jun 30, 2025

github-project-automation bot moved this to Todo in Architecture Review Jun 30, 2025

kv2019i reviewed Jul 1, 2025

View reviewed changes

kv2019i mentioned this pull request Jul 1, 2025

[FEATURE][Zephyr] replace mailbox.h with a IPC abstraction thesofproject/sof#9697

Open

dcpleung force-pushed the intel_adsp/rework_ipc branch from 4900a7c to b4c4720 Compare July 1, 2025 22:42

lgirdwood previously approved these changes Jul 2, 2025

View reviewed changes

kv2019i previously approved these changes Jul 2, 2025

View reviewed changes

dcpleung dismissed stale reviews from kv2019i and lgirdwood via 2223636 July 2, 2025 16:55

lyakh mentioned this pull request Jul 4, 2025

ipc: rework Zephyr IPC interfaces thesofproject/sof#10089

Draft

lyakh reviewed Jul 4, 2025

View reviewed changes

dcpleung force-pushed the intel_adsp/rework_ipc branch 4 times, most recently from 3331b27 to 190f98c Compare July 18, 2025 18:02

dcpleung added 12 commits July 18, 2025 17:39

soc: intel_adsp: remove IDC dt default for CONFIG_INTEL_ADSP_IPC

831c70e

The SoC specific IPC driver is for host IPC, and not IDC (which is between CPUs). So there is no need to use the IDC devicetree binding to enable the kconfig. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

tests: ipc: add tests for IPC message service

b579639

This adds a few basic tests for the IPC message service. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

tests: boards: intel_adsp/smoke: adds ACE boards

3d2ef4a

This adds ACE boards to the allowed platforms so they can be tested. Note that this only adds boards that I have tested to run correctly with the tests. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

doc: add bits about IPC message service

fd6d222

This adds documentation for the newly introduced IPC message service APIs. Signed-off-by: Daniel Leung <daniel.leung@intel.com>

dcpleung force-pushed the intel_adsp/rework_ipc branch from 190f98c to fd6d222 Compare July 19, 2025 00:55

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

IPC: introduce new IPC message service and rework the Intel Audio DSP host IPC driver #91606

IPC: introduce new IPC message service and rework the Intel Audio DSP host IPC driver #91606

dcpleung commented Jun 13, 2025

Uh oh!

kv2019i left a comment •

edited

Loading

Uh oh!

kv2019i Jul 1, 2025

Uh oh!

dcpleung Jul 1, 2025

Uh oh!

lgirdwood left a comment

Uh oh!

kv2019i left a comment

Uh oh!

lyakh Jul 4, 2025

Uh oh!

dcpleung Jul 18, 2025

Uh oh!

lyakh Jul 4, 2025

Uh oh!

dcpleung Jul 18, 2025

Uh oh!

lyakh Jul 4, 2025

Uh oh!

dcpleung Jul 17, 2025

Uh oh!

sonarqubecloud bot commented Jul 19, 2025

Uh oh!

Uh oh!

IPC: introduce new IPC message service and rework the Intel Audio DSP host IPC driver #91606

Are you sure you want to change the base?

IPC: introduce new IPC message service and rework the Intel Audio DSP host IPC driver #91606

Conversation

dcpleung commented Jun 13, 2025

Uh oh!

kv2019i 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!

lgirdwood left a comment

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!

sonarqubecloud bot commented Jul 19, 2025

Quality Gate passed

Uh oh!

Uh oh!

kv2019i left a comment •

edited

Loading