Repository navigation
[BUG] pause resume PID **** had non-zero exit status when checking multiple-pause/resume #8770
Description
Describe the bug
This issue happens when checking the multiple-pause/resume case, it shows "pause resume PID **** had non-zero exit status" in console log, but there's no error in kernel log or mtrace. The reproduction rate is low about 20%. This issue can be reproduced on both MTL and LNL NOCODEC platforms.
Inner test ID:
LNLM_RVP_NOCODEC: 37146
MTLP_RVP_NOCODEC: 37016
To Reproduce
~/sof-test/test-case/multiple-pause-resume.sh -r 50
Reproduction Rate
20%
Environment
- Branch name and commit hash of the 2 repositories: sof (firmware/topology) and linux (kernel driver).
- Kernel: {0ba5acc6d6e4}
- SOF: {16b53b423c75}
- Name of the topology file
- Topology: {sof-lnl-nocodec.tplg and sof-mtl-nocodec.tplg}
- Name of the platform(s) on which the bug is observed.
- Platform: {LNL-NOCODEC and MTL-NOCODEC}
Screenshots or console output
=== PAUSE ===
(50/50) pcm'Port1' cmd'aplay' id'9': Wait for 32 ms before resume
2024-01-21 23:46:26 UTC [REMOTE_INFO] pipeline: Deepbuffer Port0 with aplay
2024-01-21 23:46:26 UTC [REMOTE_INFO] pipeline: Port1 with aplay
2024-01-21 23:46:26 UTC [REMOTE_INFO] Check expect exit status
declare -- cmd="journalctl_cmd --since=@1705880315"
2024-01-21 23:46:26 UTC [REMOTE_ERROR] pause resume PID 65117 had non-zero exit status
2024-01-21 23:46:26 UTC [REMOTE_INFO] Starting func_exit_handler(1)
Activity
marc-hb commentedon Jan 25, 2024
https://sof-ci.01.org/sofpr/PR8759/build2309/devicetest/index.html?model=TGLU_RVP_NOCODEC-ipc4&testcase=multiple-pause-resume-50
https://sof-ci.01.org/linuxpr/PR4793/build1164/devicetest/index.html?model=MTLP_RVP_NOCODEC&testcase=multiple-pause-resume-50
https://sof-ci.01.org/sofpr/PR8803/build2374/devicetest/index.html?model=MTLP_RVP_NOCODEC&testcase=multiple-pause-resume-50
https://sof-ci.01.org/sofpr/PR8850/build2646/devicetest/index.html?model=MTLP_RVP_NOCODEC&testcase=multiple-pause-resume-50
https://sof-ci.01.org/sofpr/PR8853/build2655/devicetest/index.html?model=MTLP_RVP_NOCODEC&testcase=multiple-pause-resume-50
kv2019i commentedon Feb 2, 2024
@serhiy-katsyuba-intel @abonislawski @lyakh @teburd I think I found the reason why the DMA stops moving. The GEN bit is not set for some reason. When the problem occurs, I can see that bit26 of DGCS is not set (added debug to call the existing intel_adsp_hda_dbg() macro on error:
So "dgcs: 0x80810000" is wrong and explains why data is not moving and we get flood of "no bytes to copy" messages. Without GEN set, the DMA cannot write to DSP memory.
It seems with multiple cores using the host DMA engine, at least writes to DGCS are not immediately reflected in the read back values. Based on this, I made a few fix attempt pushed at https://github.com/kv2019i/zephyr/commits/202402-hdadma-gen-bit-fix-attempts but these don't completely solve the problem.
Even if I added a 10ms busyloop to poll for GEN bit to be set, it is read back as 0, so something is blocking the setting.
test-case/multiple-pipeline.sh: ignore aplay/arecord early exit
kv2019i commentedon Mar 6, 2024
This will be fixed by zephyrproject-rtos/zephyr#69480 . Assigning to @serhiy-katsyuba-intel for follow-up as you are the fix author. Let's not close this SOF issue until we have the fix integrated to SOF main.
kv2019i commentedon Apr 2, 2024
Closing as fix merged. Please reopen if seen again.