XO-1.75: Scratch & Etoys freeze when playing audio

[bert]

To reproduce, launch Etoys, click the red car icon on the top-right, click the yellow exclamation mark button next to "Car make sound: croak".

Etoys freezes, and fills its log file with gazillions of entries reading:

snd_pcm_writei returned -11

Build 19 / q4b08

[Quozl]

Etoys freeze when playing data is probably related to #10831, if it only occurs on XO-1.75 and not on XO-1 not XO-1.5.

[Quozl]

Please test in os6.

[greenfeld]

Still broken in os7; same -11 error.

[erikos]

Scratch has the same issue and the same error:

- launch Scratch

- on the left select 'Sound' as category

- click on 'play sound [meow]'

sound_Start: snd_pcm_hw_params: Invalid argument
snd_pcm_writei returned -11

[bert]

(In #11454) Most likely another symptom of #11193

[saadia]

I usually reproduce it in Scratch, by clicking the Sound toolbox, and then "play sound (meow)". The snd_pcm_writei error appears in ~olpc/.sugar/default/logs/edu.mit.media.ScratchActivity-1.log

I tried debugging this in sound/core/pcm_native.c. At one point (after turning on the RULES_DEBUG flag in that file) I thought it had to do with the RATE hw param but I'm not sure.

The code from the Squeak sound driver that is making the snd_pcm_write_i is here: http://squeakvm.org/cgi-bin/viewcvs.cgi/trunk/platforms/unix/vm-sound-ALSA/sqUnixSoundALSA.c?rev=2168&view=auto

[corbet]

11 is EAGAIN. There seems to be a clear bug in the squeak code; it is asking for non-blocking I/O, but it is not prepared for an EAGAIN return.

That said, the reason it's getting EAGAIN is because something is clogging up in the sound driver. If I turn off non-blocking behavior, it just wedges. Figuring that one out seems like a logical next step...

[Quozl]

Tested, kernel e3f6aac does fix.

[saadia]

One thing to note is there are some warnings on the console: [ 124.063794] snd_pcm_update_hw_ptr0: 110 callbacks suppressed [ 129.133711] snd_pcm_update_hw_ptr0: 118 callbacks suppressed [ 134.189602] snd_pcm_update_hw_ptr0: 119 callbacks suppressed [ 139.245962] snd_pcm_update_hw_ptr0: 119 callbacks suppressed [ 144.303130] snd_pcm_update_hw_ptr0: 116 callbacks suppressed [ 149.318714] snd_pcm_update_hw_ptr0: 118 callbacks suppressed [ 154.375691] snd_pcm_update_hw_ptr0: 118 callbacks suppressed [ 159.431626] snd_pcm_update_hw_ptr0: 119 callbacks suppressed [ 164.487347] snd_pcm_update_hw_ptr0: 118 callbacks suppressed

[godiard]

I am trying to test the proposal of changes to squeak by Jonathan Corbet

with os12 and os11 (without changes in the kernel),

to see if we can solve the problem from the side of squeak.

os12 + scratch without changes, playing sound works, a little crispy, recording gives a error: "sound StartRecording: snd_pcm_hw_params: Invalid argument"

os12 + scratch with changes (*), playing sound works ok, recording show a error: "unable to find control Capture, 0"

os11 + scratch with changes, scratch frozen like without changes.

Attached is a diff with the changes I did (I hope I understood Jon proposal) and the original code.

A quick note: in the code we have

snd(pcm_nonblock(output_handle, 1), "sound_Start: snd_pcm_nonblock"); snd(async_add_pcm_handler(&output_handler, output_handle, output_callback, 0), "soundStart: snd_add_pcm_handler");

I have replaced 1 by 0 [1], but I don't know what to do with async_add_pcm_handler

[godiard]

Derek O'Connell, squeak developer said:

Hi all,

I last looked at ALSA code over three years ago for the Nokia tablets and elected to bypass ALSA altogether, primarily because it and the PCM lower level code was not handling EINTR/EAGAIN properly. The resulting Squeak plug-in was later reworked to create the PulseAudio plug-in. So I can't help other than to suggest using the Piano Keyboard morph to test since that will make any excessive latency obvious where-as simply playing a single sound will not.

[martin.langhoff]

On the 11.3.1 stream, OS12 and later have a kernel-side workaround, so please test-in-build.

We are working with upstreams to get the right fix into the activities too -- see #11500 for eToys and #11501 for Scratch.

[greenfeld]

We no longer freeze eToys or Scratch when playing audio in 11.3.1 os18. However the sample sounds in Scratch play back a bit scratchy, and recording audio in Scratch still does not work.

[godiard]

Sam: if you close this ticket, please open one about sound quality and another about recording.