Catcrafts/milos-linux
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ASoC: codecs: wcd9378: keep the TX SoundWire bus out of clock-stop

Browse source 
The SDCA function engine (the SmartMIC/SmartJACK/SmartAMP sequencer
machinery activated by the FUNC_ACT class-load) dies when the TX
SoundWire bus enters clock-stop. All its registers keep their values,
so a regcache sync on resume restores nothing visible - the PDE simply
never services power-state requests again: PDE11_ACT_PS stays in PS3,
SEQ_TXn_STAT stays at pwr_dn_rdy, and even the TXn_VALID_CFG_OVR /
TXn_SEQ_TRIGGER_OVR sequencer overrides and a TX0_SEQ_SOFT_RST pulse
are ignored. Re-toggling FUNC_ACT (a real 0->1 edge on the bus) does
not revive it either; only a full codec reset does. The result was
capture recording pure digital silence: the whole DPCM -> CDC-DMA ->
TX macro -> SoundWire transport ran, but the ADC never powered.

Hold a runtime PM reference on the TX slave for as long as the codec
is bound, so the bus never clock-stops. This matches the downstream
stack, which marks the TX SoundWire master 'qcom,is-always-on' - with
full documentation available, Qualcomm made the same trade-off.

Also perform the class-load activation with plain writes instead of
update_bits so the 0->1 activation edge always reaches the hardware
regardless of regcache state.

Verified on the FP6: from a fresh boot, repeated captures across what
were previously bus suspend/resume cycles now power the sequencer every
time (PDE11 reaches PS0) and record live mic signal instead of zeros.

Assisted-by: Claude:claude-fable-5
Signed-off-by: Jorijn van der Graaf <jorijnvdgraaf@catcrafts.net>
This commit is contained in:
Jorijn van der Graaf 2026-07-06 00:58:51 +02:00
commit 5694c4a5f3
1 changed files with 36 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

50
sound/soc/codecs/wcd9378.c
View file

@ -165,28 +165,33 @@ static void wcd9378_class_load(struct snd_soc_component *component)
{ {
int i; int i;
snd_soc_component_update_bits(component, WCD9378_SMP_AMP_FUNC_ACT, /*
0x01, 0x01); * Plain writes, not update_bits, so the 0->1 activation edge
* always reaches the hardware regardless of regcache state.
* The engine boots from this edge only on a freshly reset
* codec; once it dies (bus clock-stop) no register write
* revives it, see the TX bus PM hold in wcd9378_bind().
*/
snd_soc_component_write(component, WCD9378_SMP_AMP_FUNC_ACT, 0x00);
snd_soc_component_write(component, WCD9378_SMP_AMP_FUNC_ACT, 0x01);
usleep_range(20000, 20010); usleep_range(20000, 20010);
snd_soc_component_update_bits(component, WCD9378_SMP_AMP_FUNC_STAT, snd_soc_component_write(component, WCD9378_SMP_AMP_FUNC_STAT, 0xff);
0xff, 0xff);
snd_soc_component_update_bits(component, WCD9378_SMP_JACK_FUNC_ACT, snd_soc_component_write(component, WCD9378_SMP_JACK_FUNC_ACT, 0x00);
0x01, 0x01); snd_soc_component_write(component, WCD9378_SMP_JACK_FUNC_ACT, 0x01);
usleep_range(30000, 30010); usleep_range(30000, 30010);
snd_soc_component_update_bits(component, WCD9378_CMT_GRP_MASK, snd_soc_component_update_bits(component, WCD9378_CMT_GRP_MASK,
0xff, 0x02); 0xff, 0x02);
snd_soc_component_update_bits(component, WCD9378_SMP_JACK_FUNC_STAT, snd_soc_component_write(component, WCD9378_SMP_JACK_FUNC_STAT, 0xff);
0xff, 0xff);
for (i = 0; i < 3; i++) { for (i = 0; i < 3; i++) {
snd_soc_component_update_bits(component, snd_soc_component_write(component,
WCD9378_SMP_MIC_FUNC_ACT(i), WCD9378_SMP_MIC_FUNC_ACT(i), 0x00);
0x01, 0x01); snd_soc_component_write(component,
WCD9378_SMP_MIC_FUNC_ACT(i), 0x01);
usleep_range(5000, 5010); usleep_range(5000, 5010);
snd_soc_component_update_bits(component, snd_soc_component_write(component,
WCD9378_SMP_MIC_FUNC_STAT(i), WCD9378_SMP_MIC_FUNC_STAT(i), 0xff);
0xff, 0xff);
} }
} }
@ -1219,10 +1224,26 @@ static int wcd9378_bind(struct device *dev)
goto err_remove_link3; goto err_remove_link3;
} }
/*
* The SDCA function engine dies when the TX bus enters clock-stop
* and only a codec reset revives it — registers keep their values
* so a regcache sync or a FUNC_ACT re-toggle does not help. The
* downstream stack sidesteps the same problem by marking the TX
* SoundWire master "qcom,is-always-on"; do the equivalent and
* keep the TX slave (and thus its bus) runtime-active while the
* codec is bound.
*/
ret = pm_runtime_resume_and_get(wcd9378->txdev);
if (ret < 0 && ret != -EACCES) {
dev_err(dev, "could not resume TX device\n");
goto err_remove_link3;
}
ret = snd_soc_register_component(dev, &soc_codec_dev_wcd9378, ret = snd_soc_register_component(dev, &soc_codec_dev_wcd9378,
wcd9378_dais, ARRAY_SIZE(wcd9378_dais)); wcd9378_dais, ARRAY_SIZE(wcd9378_dais));
if (ret) { if (ret) {
dev_err(dev, "Codec registration failed\n"); dev_err(dev, "Codec registration failed\n");
pm_runtime_put(wcd9378->txdev);
goto err_remove_link3; goto err_remove_link3;
} }
@ -1248,6 +1269,7 @@ static void wcd9378_unbind(struct device *dev)
struct wcd9378_priv *wcd9378 = dev_get_drvdata(dev); struct wcd9378_priv *wcd9378 = dev_get_drvdata(dev);
snd_soc_unregister_component(dev); snd_soc_unregister_component(dev);
pm_runtime_put(wcd9378->txdev);
device_link_remove(dev, wcd9378->txdev); device_link_remove(dev, wcd9378->txdev);
device_link_remove(dev, wcd9378->rxdev); device_link_remove(dev, wcd9378->rxdev);
device_link_remove(wcd9378->rxdev, wcd9378->txdev); device_link_remove(wcd9378->rxdev, wcd9378->txdev);