Title: ALSA: hda/ca0132 - Sound BlasterX AE-5 Plus (1102:0191): center, rear-right and LFE randomly silent after codec init

Title: ALSA: hda/ca0132 - Sound BlasterX AE-5 Plus (1102:0191): center, rear-right and LFE randomly silent after codec init

[Michael Harrell]

Title: ALSA: hda/ca0132 - Sound BlasterX AE-5 Plus (1102:0191): center, 
rear-right and LFE randomly silent after codec init

Hardware: Creative Sound BlasterX AE-5 Plus, PCI 1102:0012, subsystem 
1102:0191
   (driver reports "CA0132: picked fixup for PCI SSID 1102:0191" and 
identifies it
   as "Sound BlasterX AE-5"). Analog 5.1 to a Logitech Z906 (3x 3.5mm).
   AMD Ryzen (Granite Ridge) board, card is sole device behind bridge 
00:02.2.
Kernel: 7.2.7-200.fc44.x86_64 (Fedora 44), alsa-firmware installed,
   "ca0132 DSP downloaded and running" every time.
Works correctly under Windows 11 on the same machine and cables.

Problem:
After each codec initialisation (boot, driver unbind/bind, PCI bus reset,
or PCI remove+rescan), analog 5.1 output comes up in one of two states at
random:
   - good: all six channels correct
   - bad:  only FL, FR and RL audible; FC, RR and LFE silent
Roughly 3 of 8 initialisations were good in testing. Nothing in software
distinguishes the two states:
   - All ALSA controls identical (alsactl store diff): Output Select = 
Speakers,
     Surround Channel Config = 5.1, Center/LFE/Surround switches on,
     HP/Speaker Auto Detect off, Bass Redirection off, Full-Range 
Front/Rear on.
   - /proc/asound/cardN/codec#1 dumps identical apart from master amp 
volume.
   - Pin configs identical.
   - GPIO IO[1] data bit differs between some good and bad dumps, but 
was also
     observed as 1 in a bad state, so it is not a reliable indicator.
   - Result is independent of whether a plug is present in the headphone 
jack.
Toggling Output Select or Surround Channel Config after init (which re-runs
ca0132_alt_select_out / AE-5 ca0113 setup) sometimes turns a bad state into
a good one, but not reliably.

Separate but related: with runtime PM enabled (snd_hda_intel power_save=10,
set by tuned "balanced"), a good state is lost after the codec runtime-
suspends and resumes. Keeping the codec at power/control=on prevents that.
This suggests the AE-5 output setup (ca0113 MMIO / DAC) is not fully
re-applied on resume, and may be racy on initial setup.

Happy to test patches or provide codec dumps, alsactl state files and
kernel logs for good and bad states (collected on several boots).


Thanks,

Michael