Re: [PATCH] aplay: Support setting timestamp

[Takashi Sakamoto <o-takashi@sakamocchi.jp>]

Hi,

On Thu, Jun 16, 2022 at 12:00:22PM +0200, Pavel Hofman wrote:
> Dne 16. 06. 22 v 10:13 Takashi Sakamoto napsal(a):
> > 
> > On Thu, Jun 16, 2022 at 08:54:26AM +0200, Pavel Hofman wrote:
> > > To allow enabling timestamp and specify its type, a new option
> > > --tstamp-type=TYPE is added. Recognized values are none (default),
> > > gettimeofday, monotonic, monotonic-raw.
> > > 
> > > Signed-off-by: Pavel Hofman <pavel.hofman@ivitera.com>
> > > ---
> > >   aplay/aplay.1 |  4 ++++
> > >   aplay/aplay.c | 32 ++++++++++++++++++++++++++++++++
> > >   2 files changed, 36 insertions(+)
> > I prefer the idea to work for timestamp feature defined in ALSA PCM
> > interface, while I have a mixed feeling to integrate `aplay` tool, since
> > I have an intension to obsolete the tool with `axfer` tool with more
> > robust design with command argument compatibility (as much as possible).
> > 
> > This is not so strong request but would I ask you to work for `axfer` tool
> > instead of `aplay`? Then, it's preferable that the name of command
> > argument is decided with enough care of all of timestamp feature in ALSA
> > PCM interface, since we have two categories of timestamps at least; e.g.
> > system timestamp and audio timestamp. As long as I know, they possibly use
> > different clock sources, thus these two timestamps have different levels
> > of clock id, I think.
> > 
> > Of course, it's a loose accord in the community to obsolete `aplay`, and
> > it's easy to decide to continue aplay integration. (I'm not in leading
> > place of the project.) I'll be a bit happy if people take care of axfer
> > tool as well.
> 
> Thanks for your input. I use aplay in my project and needed to have tstamps
> enabled in proc status files for my simple code which calculates relative
> samplerate between capture and playback soundcards (one of them having rate
> adjustable - audio gadget, snd-aloop)
> https://mailman.alsa-project.org/pipermail/alsa-devel/2022-June/201647.html

I had read your message, then replied to the patch.

I'm not at so strong objection to your patch, however if I'm allowed to
note my opinion honestly, it surely brings an issue in a point of
application design. In short, the purpose of the patch is just for the
case to retrieve the history of PCM buffer pointer position and system
time stamp via procfs node for PCM substream of aplay for analysis,
therefore it's not for aplay itself.

As you know, aplay program has no code to process time stamp except for
the case of XRUN detection. I can easily imagine the future that the new
command line option is rarely used, except at your laboratory or office.


In my opinion, the better practice for your case is to add the way to
configure parameters of PCM substream for time stamp operation; e.g. add
kobject parameter to ALSA PCM device for PCM substream (please avoid to
hack procfs node since it's ancient way in unregulated world).

Although it's probably and technically possible, it has side effect to
user processes of existent ALSA PCM applications such as PipeWire plugins
and PulseAudio modules when the applications voluntarily process time
stamp for any purpose.

> . The existing aplay did not have this feature, so I added it and submitted
> the patch. I did not know aplay was planned to be obsoleted, it seems to
> receive a healthy stream of patches.
 
> As of the tstamp terminology - what command option would be more appropriate
> instead? Thanks a lot,

It's a kind of sophisticated work, I think.

For your information, the design of several kinds of time stamp in ALSA
PCM interface (from my memo written 2020) is below:

 - system time stamp
    - Available for several purposes
        - trigger (struct snd_pcm_status.trigger_tstamp)
           - the record of time stamp at start/stop/suspend/resume/pause
             of PCM substream.
        - current (struct snd_pcm_status.tstamp)
           - the record of the latest time stamp at updating hwptr, at
             xrun and reset of PCM substream.
        - driver (struct snd_pcm_status.driver_tstamp)
           - the record of the latest time stamp when the driver operates
             PCM substream at both hardIRQ/softIRQ and process contexts
    - Multiple levels of clock_id
       - monotinic time
       - monotonic raw time
       - real time (default, gettimeofday)
    - The sampling timing at hardIRQ context is invocation of hardIRQ
      handler instead of the time of actual interrupt request, thus
      includes time jitter due to CPU-level IRQ-mask.

 - audio time stamp (struct snd_pcm_status.audio_tstamp,
                     struct snd_pcm_mmap_status.audio_tstamp)
    - timespec compensated for audio frame granularity
    - Available when audio-related hardware supports specific function
      to record time of DMA transmission or audio link.
    - Currently, implemented for Intel HDA driver in Intel CPU (Skylate
      generation or later) with Global Time Stamp (GTS) register.
    - Multiple levels
       - default
           - computed just from hwptr
       - link audio time
       - link absolute audio time
       - link estimated audio time
       - link synchronized autio time
    - For detail, please refer to
      `Documentation/sound/designs/timestamping.rst`.


Cheers

Takashi Sakamoto