SteamOS 3.9 (Navi 31/33 HDMI Audio) RIRB Desync Causes Cyclic Hardware DMA Loop & Deafening AC3 Whine

Start by reading the reported RIRB diagnostics and the suggested entry points sound/pci/hda/hda_intel.c and hda_controller.c. The report proposes several possible recovery paths but does not identify a confirmed root cause or a specific change to make. Done would require establishing the failure and validating a safe recovery for the affected HDMI audio hardware.

Hello! I've been hunting down a bug over the past few weeks and finally nabbed it. Occasionally, and randomly it seems, my TV will suddenly emit a deafening buzzing sound. It seems to be an issue with the AMD Navi controller in the Steam Machine.

The rest of this markdown was generated by my local troubleshooting agent.

---

• Category: Audio / AMDGPU / Kernel Driver (snd_hda_intel, amdgpu)
• System: Valve Steam Machine / Neptune Station (DenCube)

---

1. System Environment & Hardware Specifications

Component	Specification
OS	SteamOS 3.9.5 (Build ID: 20261023.100, Codename: holo)
Kernel	7.2.7-valve2-2-neptune-72-g31504582a073 #1 SMP PREEMPT_DYNAMIC (x86_64)
GPU	AMD Navi 33 [Radeon Graphics] ([1002:7481], rev c7)
GPU Subsystem	Valve Software Device ([1e44:1717])
Audio Controller	AMD Navi 31 HDMI/DP Audio ([1002:ab30], PCI 0000:03:00.1)
Audio Driver	snd_hda_intel (bound to amdgpu via amdgpu_dm_audio_component_bind_ops)
Sound Server	PipeWire 1.6.8 / WirePlumber 1.6.8
Default Audio Profile	Digital Surround 5.1 (HDMI/AC3) (alsa_output.pci-0000_03_00.1.hdmi-ac3-surround via ALSA a52 plugin)
Connected Display	Vizio M65-E0 4K HDR TV via HDMI (HDMI-A-1)

---

2. Problem Description

Randomly during idle or media playback (often days or hours apart), the audio output over HDMI suddenly degenerates into a loud, high-pitched digital whine/buzzing screech.

Key Characteristics:

1. Immune to Software Controls: Muting audio, reducing system volume to 0%, or closing all media applications does not stop or alter the buzzing sound.
2. ALSA Streams Already Closed: Inspecting /proc/asound/card0/pcm*/sub*/status confirms all PCM streams are in state closed. Userspace is not actively feeding any audio frames.
3. Hardware Loop: The AMD HDMI audio controller continues transmitting a cyclic loop of corrupted DMA buffer frames at the hardware/silicon level.
4. Resolution via Link Renegotiation: The noise stops immediately if:
   • The connected TV is powered off and on, or
   • The HDMI input on the TV is switched back and forth, or
   • A low-level PCI unbind/bind is issued against the HDMI audio controller (echo 0000:03:00.1 > /sys/bus/pci/drivers/snd_hda_intel/unbind && echo 0000:03:00.1 > bind).

---

3. Kernel Diagnostic Logs & Root Cause Analysis

A. The Incident in Kernel Logs (dmesg / journalctl -k)

At the exact onset of the whine (e.g., 08:43:10 and 11:41:25), the kernel ring buffer logs a cascade of RIRB (Response Inbound Ring Buffer) response failures from snd_hda_intel:

Nov 05 11:41:25 DenCube kernel: snd_hdac_bus_update_rirb: 71 callbacks suppressed
Nov 05 11:41:25 DenCube kernel: snd_hda_intel 0000:03:00.1: spurious response 0x0:0x0, last cmd=0x220000
Nov 05 11:41:25 DenCube kernel: snd_hda_intel 0000:03:00.1: spurious response 0x0:0x0, last cmd=0x220000
Nov 05 11:41:25 DenCube kernel: snd_hda_intel 0000:03:00.1: spurious response 0x0:0x0, last cmd=0x1f0500
Nov 05 11:41:25 DenCube kernel: snd_hda_intel 0000:03:00.1: spurious response 0x233:0x0, last cmd=0x1f0500

B. Hardware DMA Loop & AC3 Bitstream Desynchronization

1. AC-3 Bitstream Encapsulation: SteamOS leverages the ALSA a52 plugin profile (hdmi-ac3-surround) to stream 5.1 surround over 2-channel HDMI PCM carriers by wrapping compressed Dolby Digital AC-3 bitstreams into IEC61937 frames.
2. Interrupt Stall & DMA Ring Free-Wheel: When the HD-Audio RIRB bus controller desynchronizes or drops interrupt callbacks, the hardware DMA ring buffer stalls or loses sync with the codec command pipeline. The hardware FIFO in the Navi 31/33 audio controller freezes in a cyclic loop, repeating the existing buffer memory.
3. TV Fallback to Raw PCM Decoding: Because the stream preamble / HDMI Audio InfoFrame is corrupted or missing, the receiver/TV DSP fails to recognize the payload as an AC-3 bitstream and treats the incoming compressed packets as uncompressed Linear PCM. Playing compressed Dolby Digital packet data as raw 16/24-bit PCM generates a deafening digital whine/buzz.

C. Proof via Hardware PCI Rebind

Issuing an unbind/rebind on the PCI function instantly clears the buzz without touching the TV:

echo 0000:03:00.1 > /sys/bus/pci/drivers/snd_hda_intel/unbind
sleep 1
echo 0000:03:00.1 > /sys/bus/pci/drivers/snd_hda_intel/bind

The resulting kernel output demonstrates the complete controller reset:

Nov 05 11:47:31 DenCube kernel: snd_hdac_bus_update_rirb: 30 callbacks suppressed
Nov 05 11:47:32 DenCube kernel: snd_hda_intel 0000:03:00.1: Force to non-snoop mode
Nov 05 11:47:32 DenCube kernel: snd_hda_intel 0000:03:00.1: bound 0000:03:00.0 (ops amdgpu_dm_audio_component_bind_ops [amdgpu])
Nov 05 11:47:32 DenCube kernel: input: HDA ATI HDMI HDMI/DP,pcm=3 as /devices/.../sound/card0/input41

---

4. Evaluated Workarounds & Limitations

Approach	Result	Assessment
power_save=0 / power_save_controller=N	Ineffective	Disabling runtime PM does not prevent RIRB timeouts; error reproduced within hours.
single_cmd=1	Not Recommended	Forces synchronous polling (udelay), breaks unsolicited events (headphone jack sensing, HDMI hotplugging), and introduces micro-stutters during gaming.
Switch to Stereo LPCM	Effective, but loses 5.1	Switching from hdmi-ac3-surround to uncompressed hdmi-stereo prevents compressed bitstream decoding errors, but limits output to 2.0 audio.
Automated Watchdog (PCI Rebind)	Proven Workaround	A background daemon detecting snd_hdac_bus_update_rirb and cycling 0000:03:00.1 clears the issue automatically in ~500ms.

---

5. Suggested Upstream Kernel / Driver Resolutions

1. Hardware DMA FIFO Reset on RIRB Stall:
   In sound/pci/hda/hda_intel.c / hda_controller.c, when a cascade of spurious response or snd_hdac_bus_update_rirb callbacks are suppressed, the driver should trigger an internal DMA engine reset and mute the stream rather than allowing the DMA FIFO to free-wheel in an infinite loop.
2. AMDGPU DM Audio Component Re-Handshake:
   Improve the error recovery between amdgpu_dm_audio_component_bind_ops and snd_hda_intel on Navi 31/33 architectures when RIRB timeouts occur, initiating an automatic stream re-initialization rather than requiring an external HPD pulse.
3. Fallback Stream Muting in PipeWire / ALSA A52:
   When the hardware reports bus errors or buffer stalls on an a52 compressed sink, immediately mute or zero the buffer to protect user hearing and equipment from raw PCM playback.