Alsa-Devel Archive on lore.kernel.org
 help / color / Atom feed
From: Takashi Iwai <tiwai@suse.de>
To: Vinod Koul <vkoul@kernel.org>
Cc: alsa-devel@alsa-project.org, pilsun.jang@samsung.com,
	khw0178.kim@samsung.com, lgirdwood@gmail.com, kimty@samsung.com,
	s47.kang@samsung.com, tiwai@suse.com,
	'Pierre-Louis Bossart' <pierre-louis.bossart@linux.intel.com>,
	hmseo@samsung.com, Gyeongtaek Lee <gt82.lee@samsung.com>,
	tkjung@samsung.com
Subject: Re: [PATCH] ALSA: compress: allow pause and resume during draining
Date: Mon, 12 Oct 2020 09:01:39 +0200
Message-ID: <s5hk0vwq65o.wl-tiwai@suse.de> (raw)
In-Reply-To: <20201012052525.GH2968@vkoul-mobl>

On Mon, 12 Oct 2020 07:25:25 +0200,
Vinod Koul wrote:
> 
> Hi Takashi, Jaroslav,
> 
> On 10-10-20, 11:08, Takashi Iwai wrote:
> > On Fri, 09 Oct 2020 19:43:40 +0200,
> > Jaroslav Kysela wrote:
> > > 
> > > Dne 09. 10. 20 v 17:13 Takashi Iwai napsal(a):
> > > > On Thu, 08 Oct 2020 11:49:24 +0200,
> > > > Gyeongtaek Lee wrote:
> > > >>
> > > >> On 10/06/20 11:57 PM, Pierre-Louis Bossart wrote:
> > > >>>> The SM in kernel might be bit more convoluted so was wondering if we can
> > > >>>> handle this in userland. The changelog for this patch says that for
> > > >>>> test case was sending whole file, surely that is not an optimal approach.
> > > >>>
> > > >>> It's rather common to have to deal with very small files, even with PCM, 
> > > >>> e.g. for notifications. It's actually a classic test case that exposes 
> > > >>> design issues in drivers, where e.g. the last part of the notification 
> > > >>> is not played.
> > > >>>
> > > >>>> Should we allow folks to send whole file to kernel and then issue
> > > >>>> partial drain?
> > > >>>
> > > >>> I don't think there should be a conceptual limitation here. If the 
> > > >>> userspace knows that the last part of the file is smaller than a 
> > > >>> fragment it should be able to issue a drain (or partial drain if it's a 
> > > >>> gapless solution).
> > > >>>
> > > >>> However now that I think of it, I am not sure what happens if the file 
> > > >>> is smaller than a fragment. That may very well be a limitation in the 
> > > >>> design.
> > > >>>
> > > >> Thanks for the comments.
> > > >>
> > > >> Actually, problem can be occurred with big file also.
> > > >> Application usually requests draining after sending last frame.
> > > >> If user clicks pause button after draining, pause will be failed
> > > >> and the file just be played until end.
> > > >> If application stop and start playback for this case, 
> > > >> start of last frame will be heard again because stop sets state to SETUP,
> > > >> and write is needed to set the state to PREPARED for start.
> > > >> If bitrate of the file is low, time stamp will be reversed and be heard weird.
> > > >> I also hope this problem can be handled in userspace easily but I couldn't find a way for now.
> > > >>
> > > >> I think that this is the time that I should share fixed patch following the comments to help the discussion.
> > > >> Following opinions are added to the patch.
> > > >> 1. it's be much nicer to have a new state - Takashi
> > > > 
> > > > Well, it wasn't me; I'm not against the new state *iff* it would end
> > > > up with cleaner code.  Admittedly, the new state can be more
> > > > "consistent" regarding the state transition.  If we allow the PAUSE
> > > > state during DRAINING, it'll lead to multiple states after resuming
> > > > the pause.
> > > > 
> > > >> 2. We can add this state to asound.h so the user space can be updated. - Jaroslav
> > > >> 3. don't forget to increase the SNDRV_COMPRESS_VERSION - Jaroslav
> > > >>
> > > >> I'm bit wondering whether it is good to increase SNDRV_COMPRESS_VERSION
> > > >> with a change in asound.h not in compress_offload.h.
> > > >> Should I increase SNDRV_PCM_VERSION also?
> > > > 
> > > > Yes, if we really add the PCM state, it's definitely needed.
> > > > 
> > > >> And what happened if I request back-porting a patch which changes VERSION to stables?
> > > > 
> > > > If we introduce the new change, it must be safe to the old kernels,
> > > > too.  The problem is only about the compatibility of the user-space
> > > > program, not about the kernel.
> > > > 
> > > > 
> > > > HOWEVER: I'm still concerned by the addition of a new PCM state.
> > > > Jaroslav suggested two steps approach, (1) first add the state only in
> > > > the uapi header, then use (2) the new state actually.  But, this
> > > > doesn't help much, simply because the step 1 won't catch real bugs.
> > > > 
> > > > Even if we add a new state and change the SNDRV_PCM_STATE_LAST, I
> > > > guess most of code can be compiled fine.  So, at the step 1, no one
> > > > notices it and bothered, either.  But, at the step 2, you'll hit a
> > > > problem.
> > > > 
> > > > Adding a new state is something like to add a new color to the traffic
> > > > signal.  In some countries, the car turning right at a crossing
> > > > doesn't have to stop at a red signal.  Suppose that we want to control
> > > > it, and change the international rule by introducing a new color (say
> > > > magenta) signal to stop the car turning right.  That'll be a big
> > > > confusion because most drivers are trained for only red, green and
> > > > yellow signals.
> > > > 
> > > > Similarly, if we add a new PCM state, every program code that deals
> > > > with the PCM state may be confused by the new state.  It has to be
> > > > reviewed and corrected manually, because it's no syntax problem the
> > > > compiler may catch.
> > > 
> > > If there is a handshake between both side, this problem is gone. We can just
> > > add another flag / ioctl / whatever to activate the new behaviour.
> > 
> > That's another tricky part.  We do have already some handshake in
> > alsa-lib to determine the supported protocol.  However, if a code in
> > question is outside that influence, we can't ensure that all belonging
> > components understand the new one.  e.g. if a program uses an
> > intermediate library, it's free from alsa-lib changes.  Or, imagine
> > some plugin.
> > 
> > If this were a change of the API function, we may have a better
> > control.  We may provide different versioned symbols in the worst
> > case, too.  But, an enum is essentially hard-coded, so we have no
> > direct influence once after it's compiled.
> 
> So what if we add another state but keep it in kernel (hidden from
> userspace)...?

That's fine, then it's just a kernel's business, and it should be
determined which one makes the code better.

But, there are things to be considered, though:

- SNDRV_PCM_STATE_* is defined as snd_pcm_state_t with __bitwise.
  This indicates that the type has to be defined in that way
  explicitly.

- Having a value over SNDRV_PCM_STATE_LAST internally is hackish.

> Right now tinycompress does not make use of PCM streams, kernel handles
> these. I am not aware of any other implementation.
> 
> So if the scope if within compress then it might work...

Yes.  But currently the API uses SND_PCM_* even for the compress
stuff.  Changing this value means to have influence on PCM, even if
PCM stuff doesn't use it yet.  (At least you'd need to increase
SND_PCM_STATE_LAST, for example.)

That said, if we want to change only for compress API by assuming that
the impact must be negligible, the first step would be to move from
SND_PCM_STATE_* to the own state, SND_COMPRESS_STATE_*.  The values
should be compatible, but this has to be changed at first.  Then you
can introduce a new value there.


thanks,

Takashi

  reply index

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <CGME20200929084051epcas2p35fb2228ed1bdfce6a7ddf5b37c944823@epcas2p3.samsung.com>
2020-09-29  8:40 ` Gyeongtaek Lee
2020-09-29  8:54   ` Takashi Iwai
2020-09-29  9:17     ` Gyeongtaek Lee
2020-09-29 14:00       ` Pierre-Louis Bossart
2020-10-01 10:29     ` Vinod Koul
2020-10-01 15:28       ` Pierre-Louis Bossart
2020-10-06  6:21         ` Vinod Koul
2020-10-06 14:57           ` Pierre-Louis Bossart
2020-10-08  9:49             ` Gyeongtaek Lee
2020-10-09 15:13               ` Takashi Iwai
2020-10-09 17:43                 ` Jaroslav Kysela
2020-10-10  9:08                   ` Takashi Iwai
2020-10-12  5:25                     ` Vinod Koul
2020-10-12  7:01                       ` Takashi Iwai [this message]
2020-10-12 12:24                         ` Vinod Koul
2020-10-12 13:29                           ` Jaroslav Kysela
2020-10-12 13:55                             ` Vinod Koul
2020-10-12 14:10                               ` Jaroslav Kysela
2020-10-12 14:21                                 ` Takashi Iwai
2020-10-12 14:46                                   ` Jaroslav Kysela
2020-10-12 14:59                                     ` Takashi Iwai
2020-10-15 10:47                                       ` Gyeongtaek Lee
2020-10-20  5:23                                         ` Gyeongtaek Lee
2020-10-26  9:18                                         ` Gyeongtaek Lee
2020-10-26 17:01                                           ` Takashi Iwai
2020-10-27  1:56                                             ` Gyeongtaek Lee
     [not found] <CGME20200928105009epcas2p4a65d50d9d09800281395a490d1844ef3@epcas2p4.samsung.com>
2020-09-28 10:50 ` Gyeongtaek Lee
2020-09-28 11:13   ` Jaroslav Kysela
2020-09-28 14:34     ` Pierre-Louis Bossart
2020-09-29  1:51       ` Gyeongtaek Lee
2020-09-29  7:12         ` Takashi Iwai
2020-09-29 17:27           ` Jaroslav Kysela
2020-09-30  9:35             ` Takashi Iwai
2020-09-30  9:57               ` Jaroslav Kysela
2020-09-30 10:33                 ` Takashi Iwai
2020-09-30 11:23                   ` Jaroslav Kysela
2020-10-01 10:35               ` Vinod Koul

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=s5hk0vwq65o.wl-tiwai@suse.de \
    --to=tiwai@suse.de \
    --cc=alsa-devel@alsa-project.org \
    --cc=gt82.lee@samsung.com \
    --cc=hmseo@samsung.com \
    --cc=khw0178.kim@samsung.com \
    --cc=kimty@samsung.com \
    --cc=lgirdwood@gmail.com \
    --cc=pierre-louis.bossart@linux.intel.com \
    --cc=pilsun.jang@samsung.com \
    --cc=s47.kang@samsung.com \
    --cc=tiwai@suse.com \
    --cc=tkjung@samsung.com \
    --cc=vkoul@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Alsa-Devel Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/alsa-devel/0 alsa-devel/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 alsa-devel alsa-devel/ https://lore.kernel.org/alsa-devel \
		alsa-devel@alsa-project.org
	public-inbox-index alsa-devel

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.alsa-project.alsa-devel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git