Re: [RFC PATCH] ALSA: hda/intel: Disable power saving on ASUS GU605CX

[Takashi Iwai]

On Sun, 06 Sep 2026 22:24:39 +0200,
Suzu wrote:
> 
> On the ASUS ROG Zephyrus G16 GU605CX, the ALC285 analog headphone
> output produces a loud pop about ten seconds after playback stops
> with power_save=10. This was reproduced with iLoud Micro Monitor
> powered speakers connected to the 3.5 mm jack. Setting power_save=0
> eliminated the idle pops in repeated pause/resume tests on
> 7.2.3-arch1-2, with power_save_controller left enabled.
> 
> Extend the power-save denylist check to cover the tested Intel HDA
> controller (8086:7728, subsystem 1043:1034) on this board. Check the
> DMI board vendor and name as well, since GU605CW also uses subsystem
> 1043:1034 and has not been tested for this issue. Keep the existing
> pm_blacklist override behavior.
> 
> Disabling idle power saving can increase idle power consumption on
> the matched controller. A separate pop during shutdown was still
> observed with power_save=0; this change does not address it.
> 
> Link: https://github.com/omacom/omarchy/pull/10498
> Assisted-by: LLM
> Signed-off-by: Suzu <suzu1dev@xxxxxxxxx>
> ---
> Validation completed on the affected GU605CX:
> - Built and booted the complete patched 7.3-rc1 kernel; the native
>   kernel build log contains no warnings.
> - Verified that the loaded HDA module's srcversion matches the built
>   artifact for 7.3.0-rc1-audio-pop-test.
> - Confirmed power_save=10, power_save_controller=Y and pm_blacklist=-1,
>   with the expected 1043:1034 power-save denylist message.
> - In five samples spanning 40 seconds, the Intel analog PipeWire sink
>   was SUSPENDED while the ALC285 remained runtime-active with
>   power/control=on. This was sampled observation, not continuous tracing.
> - Intel and NVIDIA HDMI codecs were observed suspended with
>   power/control=auto; the NVIDIA HDA controller was also suspended.
> - After booting this kernel and receiving the pause/resume test
>   instructions, the user confirmed that the popping was no longer
>   audible. The exact number of cycles and pause durations were not
>   separately reported for this patched-kernel test.
> - The earlier power_save=0 workaround also passed human pause/resume
>   listening tests on 7.2.3-arch1-2, with controller power saving enabled.
> 
> Headphone switching, internal speakers and suspend/resume checks have
> not been reported for this patched kernel. Boot-time popping remains
> unverified. A shutdown pop was previously observed with power_save=0;
> this change does not address that shutdown path. No reliable measurement
> of additional idle power consumption is available.
> 
> Test-build context: an Arch-derived configuration with debug information
> and BTF disabled was used. NVIDIA 610.57.04 test modules required a
> separate local adaptation for the 7.3 dmem cgroup registration API.
> Those modules built and initialized; their objtool return warnings also
> occurred in the production-driver build. The NVIDIA adaptation is not
> part of this audio patch. Full graphics regression testing was not done.
> 
> AI assistance: I asked OpenAI Codex to investigate the idle pops
> and prepare a GU605CX-specific fix. It assisted with the patch,
> changelog and build/runtime checks. I performed the listening
> tests; results are described above. I booted the patched kernel
> and confirmed that the idle pop was no longer audible.
> 
> No introducing commit has been identified, so no Fixes tag is proposed.
> 
>  sound/hda/controllers/intel.c | 22 +++++++++++++++++-----
>  1 file changed, 17 insertions(+), 5 deletions(-)
> 
> diff --git a/sound/hda/controllers/intel.c b/sound/hda/controllers/intel.c
> index 24015c73a..6da1d7ee5 100644
> --- a/sound/hda/controllers/intel.c
> +++ b/sound/hda/controllers/intel.c
> @@ -2328,18 +2328,30 @@ static const struct snd_pci_quirk power_save_denylist[] = {
>  	{}
>  };
>  
> +static bool azx_is_power_save_denylisted(struct pci_dev *pci)
> +{
> +	if (snd_pci_quirk_lookup(pci, power_save_denylist))
> +		return true;
> +
> +	/* GU605CW shares the subsystem ID; only GU605CX was tested. */
> +	return pci->vendor == PCI_VENDOR_ID_INTEL &&
> +	       pci->device == PCI_DEVICE_ID_INTEL_HDA_ARL &&
> +	       pci->subsystem_vendor == 0x1043 &&
> +	       pci->subsystem_device == 0x1034 &&
> +	       dmi_match(DMI_BOARD_VENDOR, "ASUSTeK COMPUTER INC.") &&
> +	       dmi_match(DMI_BOARD_NAME, "GU605CX");
> +}
> +
>  static void set_default_power_save(struct azx *chip)
>  {
>  	struct hda_intel *hda = container_of(chip, struct hda_intel, chip);
>  	int val = power_save;
>  
>  	if (pm_blacklist < 0) {
> -		const struct snd_pci_quirk *q;
> -
> -		q = snd_pci_quirk_lookup(chip->pci, power_save_denylist);
> -		if (q && val) {
> +		if (val && azx_is_power_save_denylisted(chip->pci)) {
>  			dev_info(chip->card->dev, "device %04x:%04x is on the power_save denylist, forcing power_save to 0\n",
> -				 q->subvendor, q->subdevice);
> +				 chip->pci->subsystem_vendor,
> +				 chip->pci->subsystem_device);
>  			val = 0;
>  			hda->runtime_pm_disabled = 1;
>  		}
> 
> base-commit: 45b5beb60bf7bd41c55ef17f6f2d28b351fad6c0

First of all, note that the power-save deny-list is a last resort.
Often better to tweak the existing quirks than disabling the runtime
PM.

And, even if we go for the pm deny-list, you can just add PCI SSID.
Both models are very likely same configurations.

Last but not least, you must provide a full real name (or a known
identity) in Signed-off-by tag.  It's a legal requirement.  The From
tag should also be the same, too.


thanks,

Takashi