[Sound-open-firmware] The issue of XRUN happened in Kernel and PANIC in firmware

[Liam Girdwood]

On Wed, 2018-06-13 at 16:08 +0800, zhigangw wrote:
> Hello All:
> 
>          When we use the "aplay" tool to play the audio wave file, a 
> XRUN (under-run) would happen.
> 
> It is high possibility on my APL-GPMRB platform (100%).
> 
> Which is detected by Linux kernel in snd_pcm_update_state() function.
> 
> This means the "hw_ptr" reaches "appl_ptr".
> 
>  From the log, before XRUN happened, the "hw_ptr" is updated by DSP side 
> (through the IPC) without any problem.
> 
> The "appl_ptr" is updated with about 8ms interval, but in some case, the 
> "appl_ptr" is not updated in time.
> 
> The "appl_ptr" is updated by main thread (in __snd_pcm_lib_xfer() function).
> 
> Then the XRUN happens!
> 
> 
> Below is the call-stack of the "aplay" tool. The *safe_read()* function 
> reads the audio data into local buffer from wave file.
> 
> I add the timestamp log before and after this function. Usually the 
> delta time should be very small (less than 1ms).
> 
> But sometimes, the safe_read() will take more than 50ms!  Note: It reads 
> the audio data successfully at last.


FS latency. Does this happen if you play /dev/zero ?

> 
> In this time, the underrun issue will be detected by kernel.
> 
>              playback_go()
> *safe_read()*
>                      pcm_write()
>                              writei_func()
>                                      snd_pcm_writei()
>                                              _snd_pcm_writei()
> pcm->fast_ops->writei()
> snd_pcm_hw_writei()
>      ioctl(fd, SNDRV_PCM_IOCTL_WRITEI_FRAMES, &xferi)
> 
> 
> When the XRUN is detected by kernel, the recover process will start.
> 
> the kernel will send STOP command to DSP by IPC, then send the START 
> command to DSP.
> 
> During the recover process, our DSP will hits panic!!! This possibility 
> is very high on my platform (100%).
> 

This panic is a separate issue.

> 
> Summary:
> 
>      1) From the dump data of the DSP(when panic happens),
> 
>          DSP panic happened in the wait_for_completion_timeout() function.
> 
>          When it is waken up by other interrupt (slept by waiti 
> instruction), it hits crash when loading a data into register.
> 
>          I checked there is no stack-overflow in this time.
> 
>          I do not understand the log: Exception Cause: 
> LoadStorePIFDataErrorCause, Synchronous PIF data error during LoadStore 
> access.
> 
>          Do you have any idea?

This means you are doing IO with non existent memory region. Check your
pointers. 

Liam

> 
> 
>      2) For the "aplay" tool's issue. the safe_read() function wastes 
> too much time sometimes.
> 
>          I have no idea how to debug this issue. or we should report 
> this issue to any other team?
> 
> 
> thanks
> 
> br
> 
> ~zhigang
> 
> 
> 
> 
> 
> 
> _______________________________________________
> Sound-open-firmware mailing list
> Sound-open-firmware at alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/sound-open-firmware