[BUG] Stream interrupted during pause and resume operations

[kkarask]

Describe the bug While pipelines connected to SdW interface are working pause and resume operation are done on playback stream. Capture is interrupted during stream operation.

Topology

  pipe_plb
                +------------------------+
+----------+    | +------+   +---------+ |     +-------+      +------+
| Host plb |----->|Buffer|-->| Dai ALH | |---->| PDI A | ---> | DP A |
+----------+    | +------+   +---------+ |     +-------+      +------+
                +------------------------+                       ||
                                                                 ||
                pipe_cap                                         ||
                +------------------------+                       ||
+----------+    | +------+   +---------+ |     +-------+      +------+
| Host cap |<-----|Buffer|<--| Dai ALH | |<----| PDI B |<---- | DP B |
+----------+    | +------+   +---------+ |     +-------+      +------+
                +------------------------+

To Reproduce Run test 16_07_TestSdwAlhLoopbackStreamOperations16000Hz24b32b4ch with Diagnostic Driver with option: --playback-iterations=75

Reproduction Rate Sporadic, 1/75

Environment 1) Firmware branch name and commit

• SOF: Branch: main, Hash: db8ee55b528a71ae0f6825803d967ea13e2a3769

2) Name of the platform(s) on which the bug is observed.

• Platform: TGL RVP

Screenshots or console output

Logs: TGL_16_07.zip

[lgirdwood]

@plbossart is this intended ? I assume a shared clock or similar ?

[plbossart]

Sorry, unable to comment without more details: a) I have no idea if the loopback is done at the bus level (as we do with the mockup) or internally in the SoundWire IP. b) I have no idea what the screenshots represent.

One point though: SoundWire streams are unidirectional. They have a life of their own.

[lgirdwood]

@kkarask are you able to check the diagnostic driver code and answer a) and b) above. Thanks !

[kkarask]

We don't use codec in that test. We use internal loopback just before output. When pipeline is playing it writes to a certain place in the frame and reads from exactly the same place from that frame. Both Data Port A and Data Port B indicates the same place. It is located before physical bus.

Screenshots represent pause and resume stream. In that test we perform 5x pause/resume actions on playback stream in 1 second intervals. During that operation stream was unexpectedly ended.

[lgirdwood]

@kkarask thanks, btw are you using the Linux soundwire drivers as part of this test or does the slim driver wholly program all host DSP and soundwire IP ?

[plbossart]

Sorry, I don't understand how this works. We've only used bus-level feedback in our tests. the loopback 'before physical bus' isn't a capability I am familiar with.

And I don't understand what exactly the scenario is: SoundWire does not have the concept of pause. Do you perform a bank switch to enable/disable channels for the pause on playback, while leaving the capture stream as is?

[XiaoyunWu6666]

on sh-tglu-sku0a3e-sdw-01 , I tried to aplay with Jack Out pipeline , pause resume during aplay while Dhw:0,1 Jack in arecording. No errors happened. on sh-tglu-rvp-sdw-01, I tried to aplay with jack out pipeline , pause resume during aplay while Jack in arecording . No errors happened either.

Environment Kernel Branch: topic/sof-dev Kernel Commit: bed17efc SOF Branch: main SOF Commit: 8fa5ff579351

[mengdonglin]

@XiaoyunWu6666 Thanks. @kkarask @bkokoszx test in Linux shows that for 2 independent soundwire streams, pausing the playback stream will not impact the capture stream.

[plbossart]

@mengdonglin @XiaoyunWu6666 you need to use the Jack Out and Jack In devices (hw:0,0 for playback and hw:0,1 for capture) Your other experiments use different devices on different links, that's not quite what the firmware team are testing.

[XiaoyunWu6666]

@plbossart @mengdonglin @keqiaozhang thanks for reminding . I used the Jack Out t and Jack In devices instead , pausing the playback stream did not impact the capture stream. Test result above was updated

[kkarask]

@lgirdwood Slim driver is just a "proxy" which passes information on. All logic about configuration is in python tests.

@plbossart FW doesn't have knowledge about SDW and doesn't pause SoundWire. Only DMA is paused by FW. Loopback is on ALH side and is configured internally by the test.

[antonirx]

Reproduced on TGL 5594ad3512830feaa478677718036ffabf51a454