Really bad sound output

[asdplayer]

I compiled and installed succesfully your kernel, but I got problems at runtime.
I used localmodconfig, and everything else is working. I also followed the
instructions given in this repository. I attached a dmesg and my kernel config that I used (I've renamed it to .txt just to upload it).
cx2072x_dmesg_while_using.log
cx2072x_kernel_config_4.19.4.txt

I tried to reproduce audio, but there are some problems.
Everything was also checked against a working USB soundcard, the Behringer UMC404HD, to rule out the most obvious issues in user software.

=== Play mp3/m4a/ogg/wav using Audacious player, to ALSA:
plays at 2x speed (and 2x pitch), with some noise, like gross aliasing.
The card is reported twice (this is normal I think), as:
-> sysdefault:CARD=chtcx2072x
-> usbstream:CARD=chtcx2072x
selecting the latter results in an error: "ALSA error: snd_pcm_open failed: Invalid argument." Why does it say "usbstream" anyway? The same is reported by aplay -L (list pcm outputs)

=== Play using Audacity editor, to ALSA:
plays at normal speed and pitch.
(There is consistent lagging of the UI when changing play/capture settings from the toolbars, while the soundcard is disabled and reinitialized many (~10) times, but this is fault of Audacity).

=== Play mp3/m4a/ogg/wav/video using Parole video player, which uses gstreamer, which uses PulseAudio:
There is major distortion, but playback speed is OK. See below for an hypothesis.

=== Play mp3/m4a/ogg/wav/video using VLC player, to ALSA:
Audio is played at 2x speed and 2x pitch, but is sinchronized to real time. So, some length of audio is played, then there is a void, then it restarts... At random intervals, maybe 3 gaps a second. The play time is therefore not shortened.
Using Pulseaudio backend instead of direct ALSA, VLC yelds the same exact result obtained from Parole.

=== Play test signals with kokkinizita's jaaa, through JACK, 1024 sample buffer:
I played various test signals.
Playing a 200 Hz sine:

• There is a small high frequency distortion byproduct; there is maximum signal output about -42 dB (just under 8-bit noise level?), more than that produces harsh distorsion out of the soundcard. This seems like the wrong endianness is applied to the audio signal, so a small signal stays mainly in the center byte and is represented correctly; the least significant byte is over-represented resulting fundamentally in 1-bit error noise magnified by a 16-bit shift, and if the MSB is used, the signal is completely messed up.
• 400 Hz output waveform is measured out of the soundcard (I used a tuner app and the Spectroid app on my mobile, and my ear roughly confirms), when no distorsion is happening. This seems like the codec is set for 96 kHz, while applications believe to play at 48 kHz (because they are so told) (but effectively they are requested twice the data per "real" time unit).
• The 2nd (right) digital channel outputs sound to both channels of headphones and speakers. Playing the sample on the 1st (left) digital channel results in almost no sound, as if I were hearing just an analog crosstalk. (the 2nd channel is right, while the 1st is left, I think.)
  === Lowering buffer size to 256 samples or 128 samples:
  There is another type of problem in the signal output, I can't surely tell what that is. It's like the buffer cannot be less than 512 samples, and it is filled with 0's or random values if I choose a smaller buffer size. Sounds from audio players seem played at the right speed now, I'm not sure about pitch. It seems like the signal is mixed with another frequency that could be a byproduct of the incorrect handling of buffers. I have done guesswork measuring waveforms with a mobile phone, so it's just "wild" guesswork.

=== Playing various sounds through JACK, with synths and audio players:
Every application that outputs over around -45 dB of digital signal results is harsh distorsion out of the soundcard.
Some applications kind of work, they reproduce audio at 2x speed and 2x pitch, and output gets distorted over about -45 dB; they are: jaaa, synth_v1, audacious (which also has a JACK backend), yoshimi.
Some applications don't work, and sometimes wreck the JACK instance so that the JACK server is running, but no sound can be reproduced anymore by any application: mixxx, fluidsynth.
Bristol crashes 3 seconds after started, but that is kind of expected, as the code is old and unstable.
VLC (which has a JACK backend, too) sounds as before, pitch and speed are 2x, except that now intervals with and without sounds are equally spaced and have the same duration, like 50%/50%.

JACK is configured for "playback only", because I could not start it in "duplex" mode (so I only have output, and not input + output).
If I try to start JACK in duplex mode, it says something unhelpful like "overall operation failed", and in the dmesg log appear hundreds of these lines:
[ 177.914611] intel_sst_acpi 808622A8:00: sst: Busy wait failed, cant send this msg
This is a good topic for another report maybe.

EDIT:
My computer is the Asus E200H, the soundcard is the CX20723.
And some typo fixes.
And another thing: If you see the need, in a week's time I can attach an oscilloscope to the headphone output to diagnose further; by now I don't have time. Let me know!

[heikomat]

Oh damn, you've gone way more in-depth than i ever have, nice work!
I'd be very interested in at least reproducing your findings, but i have only ever worked with debian-based distributions. Getting this to work on arch-based distributions would be really nice though.

this guy seems to have been successfull at using this kernel, but he probably used kernel version 4.17, and some things have changed since then.

I see your kernel-version is 4.19.4. The newset version i merged is 4.19. Did you apply the fixes of my cx2072x branch yourself?

Could you write up something like a reproduction manual? This way i might see if something is odd or missing, and could try it myself.

[asdplayer]

I'm writing the complete process of compilation I followed. Basically I edit Arch linux build scripts (PKGBUILD files), either the ones used to compile official packages or user-provided ones in the AUR.
I made patches from your branch of the kernel. I don't have enough free space on my E200 to pull two source trees at the same time, so, instead of using diff and learning git, I copied and pasted from the Files changed tab of this page in your repository into patches I modified by hand. They output some warning messages when applied, but otherwise work (How unexpected!). I checked each file by hand afterwards.
By the way I also wonder why is there a difference in a seemingly unrelated file: drivers/pwm/pwm-lpss.c.
Having both the modified tree AND the patches, I tried various kernels, and I obtained the same results on all of them.

• 4.19.1, official source tarball, my ugly cx2072x patches, -rt patches, some other patches for Arch and bfq-mq scheduler.
• 4.19, from your repository, with the same patches (but without the need to use my ugly cx2072x patches).
• 4.19.4, Arch Linux kernel Git repo and the cx2072x patches. No 'native optimizations' option was available on this one, so I chose 'Intel Atom' instead.

All these kernels used make localmodconfig, then I configured them further with make menuconfig to enable native compiler optimization and enable the two options for the cx2072x. And to check every thing I know.
The realtime kernels also had many security features disabled, including memory layout randomization and even features I don't entirely understand.
Instead, while configuring the kernel from the official Arch repo, I left everything as default (other than enabling cx2072x and compile optimization for the Atom architecture), so almost all sensible security features are compiled in.
Before the week ends I'll have more storage space, so I'll try to compile another kernel, based only on this repository, and see if it works.
I also tried to read the sources of the cx2072x driver, but I don't understand many things and I don't have proper documentation of the codec. For example it would be good to be able to use the inbuilt equalizer, or the 192 kHz sample rate, but I'm not ready to code yet.
Thank you for the time you spend on this, I'll post the complete compilation steps soon.

[heikomat]

The specific (seemingly unrelated) function you mentioned was not orignally added by me.
When i started this patched kernel, i based it on Fixes from tiwai. he had 3 different branches that fixed different things reagarding cherrytrail.

The specific commit including the reasoning for adding the function was this one

I'll later check if it is really no longer needed, and if so, remove it

[7twin]

@heikomat @asdplayer
Just saw this - never got a notification for it. I did indeed not have it on 4.19 yet, as on my 4.19 setup I didn't yet have need for sound.

4.20 is supposed to come soon too, so I would be definitely interested in how to get it to work with either 4.19.12 or 4.20.

Thanks for doing all this work for getting sound working!

[heikomat]

@7twin nice to see you here :)
I'll merge 4.20 tomorrow (in about 10-12 hours from now) and make the regular debian/ubuntu build

[heikomat]

As promised, 4.20 is merged and the installer-script updated

[7twin]

Thanks, well done! will try to test it asap!

[7twin]

@heikomat can confirm it working, thanks! the only thing I faced was having to rename: /net/netfilter/xt_rateest.ko to /net/netfilter/xt_RATEEST.ko else during make modules_install it would complain about not being able to stat that file.

I did get two warnings after the successful install about xt_rateest_put together with another one possibly not existing, but haven't seen anything break (yet), so not sure if that is anything to be worried about and possibly just the regular warning output.

[heikomat]

@7twin what are the steps you took to build the kernel?
Maybe we can dockerize the process and add a download for arch in future versions

[7twin]

@heikomat good that you mentioned that, reminded me that I anyway wanted to write down the process of getting it to work on arch for a long time, but kept forgetting it, here it is: https://github.com/7twin/arch_sound_e200ha

I am terrible at docker though, so can't help much there, hopefully somebody can pick up those instructions and create a working docker image.

[heikomat]

Thanks for that!
I do a lot with docker, so i'll give it a try