11 samples of SSP waveform point are error on APL with nocodec tplg

[thesofproject]

11 samples of SSP waveform point are error on APL with nocodec tplg

[zhuqingliang]

dmesg.log
rmbox.log

[lgirdwood]

How can you capture analog data if this is nocodec mode ? It seems like this should be using a topology for a codec.

[tlauda]

What does alsabat exactly do in terms of FW pipelines?
Does it create both render and capture pipelines and somehow loopbacks it back from SSP?
I'm trying to understand if this is for sure render problem or maybe a capture problem.

[keqiaozhang]

alsabat will create playback and capture pipeline, then play a specific sine wav and capture it through the loopback.
Today I used logic pro to capture the data from SDO port, look like the data is normal, so it's most like a capture issue.

[lgirdwood]

@keqiaozhang SSP capture FIFO is probably full of stale data. It could be cleared during trigger by reading the SSP FIFO directly using DSP reads.

[jocelyn-li]

Keqiao, please make sure it is FW issue or driver issue.

[tlauda]

@zhuqingliang You have used command run : alsabat -Dhw:0,0 -f dat -c2 -r 48000 -F 997, but screen from Audacity says it's 44.1 kHz. Which one is wrong?

[keqiaozhang]

Verified with latest code base on UP^2, found that the samples silent issue only happens in the first record and the number of the wrong samples is not fixed. After the first record, this issue cannot be reproduced. all the samples are correct. but this issue never happened with volume 16->16bit tplg, even in the first record.
Tested topology:

1. passthrough 16-16 nocodec tplg
2. volume 16-16 nocodec tplg
3. passthrough 32-32 nocodec tplg
4. volume 32-32 nocodec tplg

[lgirdwood]

@keqiaozhang, @tlauda ok, since this is first time only then we should read the SSP FIFO as part of the SSP driver init to clear out any stale samples. Should be a easy fix to try.

[tlauda]

Please verify on top of e69ebdd. I cannot reproduce this problem anymore.

[zhuqingliang]

We perform test with e69ebdd . There are a long period of blank bit at the head of bat.wav* . The reproduce percentage is about 20% to 30%.
reproduce steps ：
1.using test/test-ssp5-mclk-1-I2S-volume-s32le-s32le-48k-24576k-nocodec.tplg
2.reboot
3. run alsabat -Dhw:0,0 -f s32_le -r 48000 -c2 -F300 once boot up
4. audacity open /tmp/bat.wav.*

[lgirdwood]

@zhuqingliang have you verified this delay/race is not coming from alsabat/Linux userspace ?

[tlauda]

@zhuqingliang Can you provide dmesg log? I'm using the same topology and the same alsabat command, but I cannot reproduce this issue. Tried to reboot and run alsabat ~20 times.

[zhuqingliang]

If run alsabat several times , the issues will not happen again .

[keqiaozhang]

This issue is fixed by #111.