Bug 207407

Created attachment 288667 [details]
alsa-info.sh output on 5.6.6 when all four speakers are working

Created attachment 288669 [details]
alsa-info.sh output on 5.4.8 when all speakers are working (without SOF)

If it's a regression around 5.4.10, the most suspected one is the backport of commit d2cd795c4ece1a24fda170c35eeb4f17d9826cbb.

Could you try to revert it?

It is not necessary to revert this.. Just boot with:

options snd-intel-dspcfg dsp-driver=1
options snd-hda-intel model=thinkpad

The problem is that in this case the bottom speakers (controlled with Bass volume) are always at max (DAC without volume control at NID 0x06).

Could you also try to mute the Speaker in alsamixer (keep volume)?

(In reply to Jaroslav Kysela from comment #4)
> It is not necessary to revert this.. Just boot with:
> 
> options snd-intel-dspcfg dsp-driver=1
> options snd-hda-intel model=thinkpad
> 
> The problem is that in this case the bottom speakers (controlled with Bass
> volume) are always at max (DAC without volume control at NID 0x06).
> 
> Could you also try to mute the Speaker in alsamixer (keep volume)?

With these parameters, all speakers work properly (just like on 5.4.8). Volume control can be done by adding the following to /usr/share/pulseaudio/alsa-mixer/paths/analog-output.conf.common:

[Element Master]
switch = mute
volume = ignore

However, without SOF, the microphone array isn't detected. I'll try to revert commit d2cd795c4ece1a24fda170c35eeb4f17d9826cbb and report back.

Created attachment 288677 [details]
alsa-info output with commit d2cd795c reverted

Comment on attachment 288677 [details]
alsa-info output with commit d2cd795c reverted

I reverted commit d2cd795c4ece1a24fda170c35eeb4f17d9826cbb on kernel/git/stable/linux.git (linux-5.6.y branch), and built the kernel from source. Now internal speakers and mics work properly, but volume control does not work.

I attached the output of alsa-info.sh on this kernel, if it helps.

Could you compare, if you have the same result with the unmodified kernel when you mute (in alsamixer) the Speaker (not Bass Speaker)?

With unmodified kernel 5.6.6 and PulseAudio 13.0, speakers work correctly when I change 'Speaker' volume to 100 and mute 'Speaker' in alsamixer. With PulseAudio 13.99.1, speakers are muted when 'Speaker' is muted in alsamixer.

Created attachment 288681 [details]
Codec diagram

Here is the codec diagram for X1 7th. Basically, the current kernels try to use DAC NID 0x02 for both woofer (PIN NID 17h) and tweeter (PIN NID 14h). The DAC NID 0x02 has the volume control, because DAC NID 0x06 (used in previous kernels) was intended for S/PDIF and has no volume control.

I probably see similar thing on my test X1, but I cannot say that woofers (bottom) does not work. If woofers are on max, it's difficult to distinguish the sound source. It seems to me that tweeters (top speakers) are a bit low and when everything is unmuted, the sound looks a bit more balanced but weaker. Unfortunately, I don't have Windows here, so I cannot test this settings.

If you do hear a difference, then the Windows driver does something which Linux does not. I see equalizer and ClassD amp on the path. This part of hardware is undocumented and we can probably reverse engineer the communication from the Windows driver (e.g. https://github.com/Conmanx360/QemuHDADump).

I am doing another tests today with my X1 (without Windows) and all speakers are working correctly.

Kernel: 5.5.7

Firmware:

# fwupdmgr upgrade
No upgrades for Thunderbolt Controller: current version is 44.00: 44.00=same
No upgrades for System Firmware: current version is 0.1.30: 0.1.30=same, 0.1.29=older, 0.1.27=older, 0.1.26=older, 0.1.25=older
No upgrades for Touchpad: current version is 1.3.3013337: 1.3.3013337=same
No upgrades for UEFI Device Firmware: current version is 192.47.1524: 192.47.1524=same
No upgrades for UEFI Device Firmware: current version is 0.1.15: 0.1.15=same, 0.1.14=older, 0.1.13=older, 0.1.12=older, 0.1.10=older

Test procedure

- disable and kill pulseaudio (I usually rename /usr/bin/pulseaudio)
- run speaker-test (or any other app which directly use the speakers)
  - speaker-test -D hw:0 -c 2 -d 10000
- alsamixer
  - Master to 100%, unmute
  - Speaker - play with volume and mute/unmute
  - Bass Speaker - play with mute/unmute

Speaker mute switch correctly turns on/off tweeters on top and the Bass Speaker switch correctly turns on/off the bottom woofers.

Note that pulseaudio automatically turns off the Master when you mute Speaker, so it's not possible to test this with pulseaudio.

I did the same tests on my system (booted directly to Linux from poweroff). All speakers make sound and controls in alsamixer do work correctly, as you described.

My test procedure and results

- disabled and killed pulseaudio by renaming /usr/bin/pulseaudio
- Played song with VLC Media Player
  - set Audio > Output module to 'ALSA audio output'
  - set Audio > Device to 'sof-hda-dsp'
  - Played Faded by Alan Walker (has a wide range of frequencies)
- In alsamixer:
  - Master volume set to 100%, unmuteed
  - Speaker muted and Bass Speaker unmuted: Speaker and Master volume controls overall sound volume; sound is very similar to Windows
  - Speaker and Bass Speaker unmuted: all speakers work; volume is noticeably weaker;
  - Speaker unmuted and Bass Speaker muted: sound is very quiet; tremble only, no bass

I also observed the speakers' behavior on Windows. The majority of sound come from the bottom woofers, the top-firing tweeters are on but barely noticeable. To me, sound on Windows is not that different from the "Speaker muted and Bass Speaker unmuted" scenario.

I think the difference between Speaker muted and Speaker unmuted largely depends on preference. As many desktop Linux users come from Windows, I guess we should stay close to the sound quality on Windows? Let me know what you think.

FYI, I suspect that the undocumented parts of the codec diagram have something to do with the "Dolby Atmos Speaker System" and proprietary Dolby technology.

My firmware and kernel  information

- Kernel: 5.5.7

- Firmware:

# fwupdmgr get-devices
├─Thunderbolt Controller:
│     Current version:     44.00
│     Vendor:              Lenovo (TBT:0x0109)
├─Embedded Controller:
│     Current version:     0.1.15
│     Minimum Version:     0.1.15
│     Vendor:              DMI:LENOVO
├─Intel Management Engine:
│     Current version:     192.49.1534
│     Minimum Version:     0.0.1
│     Vendor:              DMI:LENOVO
├─System Firmware:
│     Current version:     0.1.30
│     Minimum Version:     0.0.1
│     Vendor:              LENOVO (DMI:LENOVO)
├─Touchpad:
│     Current version:     1.3.3013337
│     Bootloader Version:  8.0.0
│     Vendor:              Synaptics (HIDRAW:0x06CB)
├─UEFI Device Firmware:
│     Current version:     16777224
│     Minimum Version:     1
├─UEFI Device Firmware:
│     Current version:     4784136
│     Minimum Version:     4784136
│     Vendor:              DMI:LENOVO

Perhaps the Linux HDA driver should use separate 0x02 and 0x03 outputs like Windows driver does to maintain the volume controls separate. But there is a problem to map this in UCM (two volume controls with the preset offset).

I also got an X1C7 here (currently running Linux Mint 19.3/18.04LTS with 5.3.0-40 kernel. I currently have all 4 speakers working with SOF 1.4.2. Used this script to enable it https://gist.github.com/hamidzr/dd81e429dc86f4327ded7a2030e7d7d9
It breaky my audio though on 5.3.41+ kernels and only the tweeters work there. Anything i can do to help? I'm not so into the details of alsa but open for any instructions for testing!

So I just found another solution on ArchWiki. Doing the following makes the speakers behave the same as they do in Windows:
  - Apply this patch to '/usr/share/alsa/ucm2/sof-hda-dsp/HiFi.conf'

--- /usr/share/alsa/ucm2/sof-hda-dsp/HiFi.conf.old
+++ /usr/share/alsa/ucm2/sof-hda-dsp/HiFi.conf
@@ -20,9 +20,9 @@ SectionDevice."Headphones1" {
 	Value {
 		PlaybackPriority 200
 		PlaybackPCM "hw:${CardId}"
-		PlaybackMixerElem "Headphone"
+		PlaybackMixerElem "Master"
 		PlaybackMasterElem "Master"
-		PlaybackVolume "Headphone Playback Volume"
+		PlaybackVolume "Master Playback Volume"
 		PlaybackSwitch "Headphone Playback Switch"
 		If.jack {
 			Condition {
@@ -72,9 +72,9 @@ SectionDevice."Speaker" {
 	Value {
 		PlaybackPriority 100
 		PlaybackPCM "hw:${CardId}"
-		PlaybackMixerElem "Speaker"
+		PlaybackMixerElem "Master"
 		PlaybackMasterElem "Master"
-		PlaybackVolume "Speaker Playback Volume"
+		PlaybackVolume "Master Playback Volume"
 		PlaybackSwitch "Speaker Playback Switch"
 	}
 }

  - Run 'hda-verb /dev/snd/hwC0D0 0x17 SET_CONNECT_SEL 1' on startup

References:
  - https://wiki.archlinux.org/index.php/Talk:Lenovo_ThinkPad_X1_Carbon_(Gen_7)#New_audio_situation
  - https://gist.github.com/hamidzr/dd81e429dc86f4327ded7a2030e7d7d9#gistcomment-3335517
  - https://gist.github.com/hamidzr/dd81e429dc86f4327ded7a2030e7d7d9#gistcomment-3345062

That doesn't produce pleasing output on my X1 Yoga gen 4. The volume level is much too high and most of the volume adjustments can only be made between 1% and 5%. The front speakers do almost all the work while the rear speakers are practically inaudible.

Correction: the proposed modifications to HiFi.conf alone definitely produce better sound for me, but additionally running that hda-verb command ruins it and makes the front speakers much too loud. However without doing that, it's much improved with the HiFi.conf changes!

Hi Nate,

On my X1 Carbon 7th (not the Yoga, but hardware should be identical), patching HiFi.conf alone doesn't seem to make any difference. Patching HiFi.conf AND running that hda-verb command does help.

By the way, what version of PulseAudio and alsa-lib are you running? A few weeks ago, I found that running alsa-lib 1.2.3.1 and PulseAudio 13.99.1 (available in Ubuntu 20.04 and Fedora 32 repos) works much better than older versions of PulseAudio (because PulseAudio 13.99.1 correctly handles ucm2).

P.S. KDE Plasma user here and I LOVE it. Thanks for your work!

(In reply to Nate Graham from comment #18)
> Correction: the proposed modifications to HiFi.conf alone definitely produce
> better sound for me, but additionally running that hda-verb command ruins it
> and makes the front speakers much too loud. However without doing that, it's
> much improved with the HiFi.conf changes!

I'm using the latest PulseAudio built from git master, plus these patches installed on top:
- https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/271
- https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/merge_requests/295


I'm so glad you're enjoying Plasma! As a ThinkPad X1 user, you might appreciate https://invent.kde.org/plasma/plasma-pa/-/merge_requests/5 :)

A bit explanation 'hda-verb /dev/snd/hwC0D0 0x17 SET_CONNECT_SEL 1' is routing the bass/bottom speakers PIN 0x17 to DAC node 0x03 (this is the way I suggested in comment #14 and appaerently it's the routing which Windows use).

The UCM config change is wrong and it is not necessary (no volume range change).

The problem with this config / workaround is that the bass/bottom speaker is controlled with the 'Headphone' volume control (tweeters and bass speakers are separated). I need to find a way to describe in UCM, so pulseaudio can do the right volume control (busy with other UCM things right now). But - it's in my queue.

*** Bug 208133 has been marked as a duplicate of this bug. ***

(In reply to Jaroslav Kysela from comment #21)
> I need to find a way to describe in UCM, so pulseaudio can
> do the right volume control (busy with other UCM things right now). But -
> it's in my queue.

Thanks Jaroslav! That's great to hear.

https://lore.kernel.org/alsa-devel/3e999b99-0ce7-68b8-a923-07ba2f4d798d@perex.cz/T/#t

On kernel 5.8.12 with SOF firmware, the issue seems to be resolved. Big thanks to everyone involved!

Can confirm. This is fixed for me too (X1 Yoga gen 3) with kernel 5.8.12+ and sof-firmware.

Great job!