Sound regression snd_hda_intel: PCI rescan fix

Fixing snd_hda_intel Regression: PCI Rescan Restores Audio on Old Laptop

After a kernel update on my Emachines E732 with Intel 5 Series/3400 Series HDA (chip-ID 8086:3b56), audio vanished. aplay -l reports no soundcards found. lspci spots the device at 00:1b.0, inxi -Axxz confirms the snd_hda_intel driver, but journalctl is flooded with errors: hdaudio hdaudioC0D0: no AFG or MFG node found across all nodes.

dmesg tells the real story: during initialization, snd_hda_intel 0000:00:1b.0: CORB reset timeout#2, CORBRP = 65535, followed by no codecs initialized. CORB (Command Output Ring Buffer) is the ring buffer for codec commands. A value of 0xFFFF signals a hung bus or inaccessible hardware—reading from an uninitialized register.

BAR on cold boot: mem 0xd4400000-0xd4403fff 64bit. Reloading the module (modprobe -r/-a snd_hda_intel) repeats the error with fresh timestamps.

Ineffective Quick Fixes

Standard tweaks fell flat:

• User-space tools: Reinstalling PipeWire/WirePlumber, tweaking alsamixer—useless without ALSA devices.
• Module options: options snd-hda-intel enable=1 power_save=0 power_save_controller=N model=auto in /etc/modprobe.d/alsa-fix.conf.
• GRUB params: pci=nocrs pci=nomsi pcie_aspm=off snd_hda_intel.power_save=0 snd_hda_intel.dma_mask=0.
• Hot-reload: modprobe snd_hda_intel msi=0 dsp_driver=1 probe_mask=1.

The issue stems from early initialization: LTS kernels or LiveCDs worked due to different module load order or BIOS state.

[   17.492159] snd_hda_intel 0000:00:1b.0: CORB reset timeout#2, CORBRP = 65535
[   17.492380] snd_hda_intel 0000:00:1b.0: no codecs initialized

The Fix: PCI Hot-Rescan via sysfs

Key insight: Cold boot assigns a bad BAR, but hot-rescan on a warmed-up system works. Here's the sequence:

• echo 1 | sudo tee /sys/bus/pci/devices/0000:00:1b.0/remove
• sleep 2
• echo 1 | sudo tee /sys/bus/pci/rescan

Fresh dmesg output:

[  473.969035] pci 0000:00:1b.0: [8086:3b56] ... 
[  473.969102] pci 0000:00:1b.0: BAR 0 [mem 0x00000000-0x00003fff 64bit]
[  473.969982] pci 0000:00:1b.0: BAR 0 [mem 0x23c000000-0x23c003fff 64bit]: assigned
[  474.095498] input: HDA Intel MID Mic as /devices/pci0000:00/0000:00:1b.0/sound/card0/input24
[  474.095675] input: HDA Intel MID Headphone as /devices/pci0000:00/0000:00:1b.0/sound/card0/input25

Audio is back: sound card in /dev/snd, PipeWire detects devices. The culprit? BIOS hands out bogus params early on; rescan after full PCI init assigns the correct BAR.

Automating for Everyday Use

For a production systemd script (post-boot):

#!/bin/bash
PCI_ID="0000:00:1b.0"
echo 1 > /sys/bus/pci/devices/$PCI_ID/remove
sleep 2
echo 1 > /sys/bus/pci/rescan
# Verify
if aplay -l | grep -q "card 0"; then
    systemctl --user restart pipewire pipewire-pulse wireplumber
fi

Drop it in /etc/systemd/system/pci-audio-rescan.service with After=multi-user.target.

Diagnosing Similar Regressions

For snd_hda_intel on legacy Intel HDA:

• Grab logs: dmesg | grep '00:1b.0', journalctl -b -p err | grep hda.
• Check BAR: lspci -vv -s 00:1b.0 | grep BAR.
• Test rescan: Use the sysfs method above.
• Compare kernels: LTS vs mainline—regression in PCI alloc or power management.
• BIOS/UEFI: Update if available, though it rarely helps on 2010-era chipsets.

This hits older hardware hard after kernel 6.8+: PCI enumeration and ASPM changes.

Key Takeaways

• CORBRP=0xFFFF flags a stalled HDA controller, not BIOS/codec issues.
• PCI rescan via sysfs sidesteps cold boot timeouts by reassigning BAR.
• LTS kernels succeed thanks to timing: modules load post-BIOS handoff.
• Automation: systemd service ensures reliability on daily drivers.
• Diagnosis: Focus on dmesg over user-space; zero in on PCI BAR and CORB/RIRB.

— Editorial Team