sndio output not working on FreeBSD

When was it last seen working for you? It works for me (without sndiod).

I added some more error messages in svn trunk. Can you build from that and re-test? Configure is generated just with autoreconf -iv, with the autotools present, of course.

When was it last seen working for you?

Don't recall offhand, will test later.

It works for me (without sndiod).

Hm, now when you mentioned sndiod, I noticed that it starts working for me after starting sndiod. When I stop it, things stop working again.

I added some more error messages in svn trunk. Can you build from that and re-test?

High Performance MPEG 1.0/2.0/2.5 Audio Player for Layers 1, 2 and 3
        version 1.28.0; written and copyright by Michael Hipp and others
        free software (LGPL) without any warranty but with best wishes
Decoder: generic
Trying output module: sndio, device: <nil>
Module dir search relative to: /home/novel/code/mpg123/trunk/src/.libs
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/../lib/mpg123
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/plugins
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/libout123/modules/.libs
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/libout123/modules
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs
Module dir: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs
Module path: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs/output_sndio.so
Note: sndio is the last output option... showing you any error messages now.
[src/libout123/modules/sndio.c:open_sndio():79] error: parameter setup  failure
[src/libout123/libout123.c:check_output_module():984] error: Module 'sndio' device open failed.
[src/libout123/libout123.c:out123_open():485] error: Found no driver out of [sndio] working with device <default>.
main: [src/mpg123.c:check_fatal_output():332] error: out123 error 3: failure loading driver module

I guess [src/libout123/modules/sndio.c:open_sndio():79] error: parameter setup failure is what we're looking for?

I'm not sure if that's the issue that if it works with sndiod running, I guess I need to change --with-default-audio value from sndio to sndio,oss to keep the old behavior when mpg123 falls back to oss?

Yes, that seems to be the issue. Your hardware does not support signed 16 bit encoding, by any chance?

Do you get more information when setting SIO_DEBUG=2 in the environment?

Regarding --with-default-audio … just use --with-audio=sndio,oss. If you don't set --with-default-audio, it's the same.

Your hardware does not support signed 16 bit encoding, by any chance?

How can I check that?

Do you get more information when setting SIO_DEBUG=2 in the environment?

Looks like the right flag for my setup is SNDIO_DEBUG and it does provide some additional info:

Module path: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs/output_sndio.so
Note: sndio is the last output option... showing you any error messages now.
_aucat_open: host= unit=0 devnum=0 opt=default
/tmp/sndio-1001/sock0: No such file or directory
/tmp/sndio/sock0: No such file or directory
sio_oss_setpar: CHANNELS: Invalid argument
[src/libout123/modules/sndio.c:open_sndio():79] error: parameter setup  failure

Regarding --with-default-audio … just use --with-audio=sndio,oss. If you don't set --with-default-audio, it's the same.

Great, thanks for the info.

Please have a go with the current https://mpg123.org/snapshot/ . I had to fix up the whole format detection stuff. On my system, it correctly sees the hardware support now (only 44.1 and 48 kHz stereo).

I've tried the latest svn version (r4946), assuming it's not older than the snapshot.
The problem still remains:

(10:33) novel@kloomba:~/code/mpg123/trunk/src %> SNDIO_DEBUG=2 ./mpg123 -o sndio -vv ~/test.mp3
High Performance MPEG 1.0/2.0/2.5 Audio Player for Layers 1, 2 and 3
        version 1.28.0; written and copyright by Michael Hipp and others
        free software (LGPL) without any warranty but with best wishes
Decoder: generic
Trying output module: sndio, device: <nil>
Module dir search relative to: /home/novel/code/mpg123/trunk/src/.libs
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/../lib/mpg123
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/plugins
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/libout123/modules/.libs
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/libout123/modules
Looking for module dir: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs
Module dir: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs
Module path: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs/output_sndio.so
Note: sndio is the last output option... showing you any error messages now.
_aucat_open: host= unit=0 devnum=0 opt=default
/tmp/sndio-1001/sock0: No such file or directory
/tmp/sndio/sock0: No such file or directory
sio_oss_setpar: CHANNELS: Invalid argument
[src/libout123/modules/sndio.c:open_sndio():127] error: parameter setup  failure
[src/libout123/libout123.c:check_output_module():984] error: Module 'sndio' device open failed.
[src/libout123/libout123.c:out123_open():485] error: Found no driver out of [sndio] working with device <default>.
main: [src/mpg123.c:check_fatal_output():332] error: out123 error 3: failure loading driver module
(10:34) novel@kloomba:~/code/mpg123/trunk/src %> 

BTW, my sound device info is:

(10:34) novel@kloomba:~/code/mpg123/trunk/src %> cat /dev/sndstat 
Installed devices:
pcm0: <Realtek ALC887 (Rear Analog 7.1/2.0)> (play/rec) default
pcm1: <Realtek ALC887 (Front Analog)> (play/rec)
pcm2: <Realtek ALC887 (Onboard Digital)> (play)
pcm3: <Intel Kaby Lake (HDMI/DP 8ch)> (play)
No devices installed from userspace.
(10:34) novel@kloomba:~/code/mpg123/trunk/src %> 

OK, added one more debugging line. This is weird. Please do a debug build again and compare with my output of

SNDIO_DEBUG=2 src/mpg123 -o sndio -vv src/tests/sweep.mp3

It prints the following:

Module path: /home/novel/code/mpg123/trunk/src/.libs/../libout123/modules/.libs/output_sndio.so
[src/libout123/module.c:open_module_here():148] debug: Module symbol: mpg123_output_module_info
Note: sndio is the last output option... showing you any error messages now.
_aucat_open: host= unit=0 devnum=0 opt=default
/tmp/sndio-1001/sock0: No such file or directory
/tmp/sndio/sock0: No such file or directory
sio_oss_setpar: CHANNELS: Invalid argument
[src/libout123/modules/sndio.c:open_sndio():128] error: parameter setup  failure
[src/libout123/libout123.c:check_output_module():978] debug: ao->open() = -1
[src/libout123/libout123.c:check_output_module():984] error: Module 'sndio' device open failed.

I can see that in your setup it's probing soundcards, and doesn't seem to happen in my case.

Yes. Where I see sound cards, you see only that line

sio_oss_setpar: CHANNELS: Invalid argument

At that point, all we do is this:

    struct sio_hdl *hdl;
    struct sio_par par;

    hdl = sio_open(ao->device /* NULL is fine */, SIO_PLAY, 0);
    if (hdl == NULL)
    {
        error("Got nothing from sio_open(). ");
        return -1;
    }

    sio_initpar(&par);

    par.le = SIO_LE_NATIVE;

    if (!sio_setpar(hdl, &par) || !sio_getpar(hdl, &par)
        || par.le != SIO_LE_NATIVE )
    {
        if(!AOQUIET)
            error("parameter setup  failure");
        sio_close(hdl);
        return -1;
    }

Could it be that your hardware does not support native endian format?

We got conversion switches for that. But we'd have to communicate that to the main mpg123 process so that it instructs libmpg123 to swap endianess. Alternatively, we can do the swap also in the module, having it depend on libsyn123 for that or just hacking it.

Can you confirm your platform being big/little endian, along with the sample format the device expects? You could start with the example above and check the par.le setting.

Can you confirm your platform being big/little endian

I believe it's little endian, as it's amd64 (to be specific, Intel i5-8600).

... along with the sample format the device expects?

I wasn't able to figure out how to check that.

You could start with the example above and check the par.le setting.

I've updated the code this way:

@@ -124,6 +125,7 @@
        if (!sio_setpar(hdl, &par) || !sio_getpar(hdl, &par)
                || par.le != SIO_LE_NATIVE )
        {
+               mdebug("par.le = %d (%d)\n", par.le, SIO_LE_NATIVE);
                if(!AOQUIET)
                        error("parameter setup  failure");
                sio_close(hdl);

It prints:

[src/libout123/modules/sndio.c:open_sndio():128] debug: par.le = 1 (1)

Also I've added one more debug bit:

@@ -120,10 +121,12 @@
                sio_close(hdl);
                return -1;
        }
+       printf("channels = %d\n", par.pchan);

And it prints:

channels = -1

Looks like the error message is coming from the following sndio code:

        if (ioctl(hdl->fd, SNDCTL_DSP_CHANNELS, &hdl->chan) == -1) {
                DPERROR("sio_oss_setpar: CHANNELS");
                hdl->sio.eof = 1;
                return 0;
        }

I've updated mpg123's sndio.c to set par.pchan to "2" (instead of -1) and it works.

Looks like it's hitting this code while doing this ioctl: https://cgit.freebsd.org/src/tree/sys/dev/sound/pcm/dsp.c#n1478

I presume the sio_setpar(hdl, &par) is failing right away. Could take that conditonal apart andcheck that first.

But anyway, SIO_LE_NATIVE is true, so sndio also thinks your platform is little endian. That shouldn't be it … We put nothing into par. The system is free to choose.

What's with that OSS layer you have? Can you shnow -vv output for -o oss? Do you have native OSS or a mixing/converting layer in between (I remember that from OSS 3 vs 4)?

Maybe some FreeBSD people can shed some light on this.

This doesn't make sense. The call where it is failing is the one to figure out the default format for the device. I'm following the sndio man page here:

    •   Initialize a sio_par structure using sio_initpar() and fill it with the desired parameters.  Then call sio_setpar() to request the device to use them.  Parameters left unset in
         the sio_par structure will be set to device-specific defaults.

I thought sio_initpar() intiializes the structure for 'unset' values. But maybe it's relying on things being zeroed beforehand. Can you check if it works properly when setting par.pchan=0 ?

Then … it should also work when you do another svn up, as I am doing explicit zeroing of the structure now (thought sio_initpar() would do that, buggy?).

svn up version doesn't work still. Right after sio_initpar() it prints -1 (actually I noticed that I use "%d" while pchan is unsigned, if I use "%u", it prints 4294967295.
par.pchan = 0 right after sio_initpar() makes things work again.

BTW, for sndio-1.7.0 which I'm using implementation looks like this:

#define SIO_PAR_MAGIC   0x83b905a4

void
sio_initpar(struct sio_par *par)
{
        memset(par, 0xff, sizeof(struct sio_par));
        par->__magic = SIO_PAR_MAGIC;
}

And in sio_oss_setpar() it seems the only thing it does about channels is:

        if (hdl->sio.mode & SIO_PLAY)
                hdl->chan = par->pchan;
        else if (hdl->sio.mode & SIO_REC)
                hdl->chan = par->rchan;

Oh, OK. So 'unset' doesn't mean zero, it means all bits set. But actually, it does mean zero for you in practice. Let's take the attached example code. It does this on Ubuntu:

$ !gcc
gcc -o sndiotest sndiotest.c -lsndio
$ pasuspender -- ./sndiotest 
initpar channels: 4294967295
opened
setpar successful
default format: 2 ch, 48000 Hz

For your box, you probably need to change pchan to zero before setting parameters.

Doesn't that awfully look like a bug in FreeBSD/sndio to you? Can you report that?

Setting pchan=0 seems to work for both you an me, but is it correct? The documentation suggests that zero means 'unset', while the code does 0xff on the bytes. I dunno what to make of this.

I updated svn to set pchan to zero for querying mode. I guess we'll have to roll with that. But at least the documentation should be fixed.

Asked sndio here: https://sndio.org/arch/0138.html

This is now fixed in sndio git: https://github.com/ratchov/sndio/commit/aac840d9f91c52a1dc3dcbdbfc3c628a80a717a1.

Looks like the issue can be closed now. Not sure if the "par.pchan = 0" workaround should be kept or just wait while users eventually move to a newer sndio version that has that issue fixed.

Thanks

Looks like pchan=0 is still working as before. So I am inclined to keep the workaround for current systems, but will note the time when it was fixed.

Does upstream have an opinion about this? Are they adding pchan=0 to the docs as equivalent to unset?

Are you still talking about upstream as sndio or FreeBSD OSS?

From what I can see, there's no special treatment of pchan == 0 in sndio, only comparison with the trap value (~0U).

In OSS it looks like the set channels ioctl just writes back current number of channels if the supplied value is 0: https://cgit.freebsd.org/src/tree/sys/dev/sound/pcm/dsp.c#n1515, which appears to be safe enough.

Oh, right. This is only an issue in the OSS part of sndio. They did not handle their default unset value.

Thing is … channel count 0 could trip up other backends of sndio, right? So things work only by accident here, as oss handles 0 channels to mean unset itself.

Maybe we can try a workaround that does not potentially break other systems. What about this:

1. Try defaults.
2. If even defaults fail, try 2 channels as most likely guess.
3. If even that fails, we got true trouble.

That should work, I guess. It looks like we can also use sio_getcap() to make sure 2 channels are supported.

Hm. The whole point was ‘give me some default preferred format’. So if that fails … use the getformats stuff. OK, we could call the caps, determine supported channel counts, pick 2 if present, 1 as fallback … or an even higher number if it's a 7 channel interface?

It would then be up to us to decide on the default count to use. What's reasonable?

1. 2 channel stereo as the norma
2. otherwise maximum supported count

Or always the maximum count? Like a 5.1 system will give you 6 channels as default format?