Hi Pierre-Louis, On 8/16/21 3:31 PM, Pierre-Louis Bossart wrote: > Hi Hans, > I am a bit confused by the use of acpi_dev_add_driver_gpios(). I understand admittedly my solution there is a bit hacky. > You call > it from the dailink .init function, and you call > acpi_dev_remove_driver_gpios() from the .exit function. > >> @@ -1172,9 +1250,60 @@ static int byt_rt5640_init(struct snd_soc_pcm_runtime *runtime) >> snd_soc_component_set_jack(component, &priv->jack, NULL); >> } >> >> + if (byt_rt5640_quirk & BYT_RT5640_JD_HP_ELITEP_1000G2) { >> + ret = snd_soc_card_jack_new(card, "Headset", >> + SND_JACK_HEADSET, >> + &priv->jack, rt5640_pins, >> + ARRAY_SIZE(rt5640_pins)); >> + if (ret) >> + return ret; >> + >> + ret = snd_soc_card_jack_new(card, "Headset 2", >> + SND_JACK_HEADSET, >> + &priv->jack2, rt5640_pins2, >> + ARRAY_SIZE(rt5640_pins2)); >> + if (ret) >> + return ret; >> + >> + acpi_dev_add_driver_gpios(ACPI_COMPANION(priv->codec_dev), >> + byt_rt5640_hp_elitepad_1000g2_gpios); >> + >> + rt5640_jack_gpio.data = priv; >> + rt5640_jack_gpio.gpiod_dev = priv->codec_dev; >> + rt5640_jack_gpio.jack_status_check = byt_rt5640_hp_elitepad_1000g2_jack1_check; >> + ret = snd_soc_jack_add_gpios(&priv->jack, 1, &rt5640_jack_gpio); >> + if (ret) { >> + acpi_dev_remove_driver_gpios(ACPI_COMPANION(priv->codec_dev)); >> + return ret; >> + } >> + >> + rt5640_set_ovcd_params(component); >> + rt5640_jack2_gpio.data = component; >> + rt5640_jack2_gpio.gpiod_dev = priv->codec_dev; >> + rt5640_jack2_gpio.jack_status_check = byt_rt5640_hp_elitepad_1000g2_jack2_check; >> + ret = snd_soc_jack_add_gpios(&priv->jack2, 1, &rt5640_jack2_gpio); >> + if (ret) { >> + snd_soc_jack_free_gpios(&priv->jack, 1, &rt5640_jack_gpio); >> + acpi_dev_remove_driver_gpios(ACPI_COMPANION(priv->codec_dev)); >> + return ret; >> + } >> + } >> + >> return 0; >> } >> >> +static void byt_rt5640_exit(struct snd_soc_pcm_runtime *runtime) >> +{ >> + struct snd_soc_card *card = runtime->card; >> + struct byt_rt5640_private *priv = snd_soc_card_get_drvdata(card); >> + >> + if (byt_rt5640_quirk & BYT_RT5640_JD_HP_ELITEP_1000G2) { >> + snd_soc_jack_free_gpios(&priv->jack2, 1, &rt5640_jack2_gpio); >> + snd_soc_jack_free_gpios(&priv->jack, 1, &rt5640_jack_gpio); >> + acpi_dev_remove_driver_gpios(ACPI_COMPANION(priv->codec_dev)); >> + } >> +} > > so far so good, but you also add/remove gpios in the machine driver probe Ack. >> @@ -1490,6 +1620,20 @@ static int snd_byt_rt5640_mc_probe(struct platform_device *pdev) >> return -EPROBE_DEFER; >> priv->codec_dev = get_device(codec_dev); >> >> + if (byt_rt5640_quirk & BYT_RT5640_JD_HP_ELITEP_1000G2) { >> + acpi_dev_add_driver_gpios(ACPI_COMPANION(priv->codec_dev), >> + byt_rt5640_hp_elitepad_1000g2_gpios); So this second add here (which runs first, so it really is the first add) >> + priv->hsmic_detect = devm_fwnode_gpiod_get(&pdev->dev, codec_dev->fwnode, >> + "headset-mic-detect", GPIOD_IN, >> + "headset-mic-detect"); Is to make sure this call can succeed. >> + acpi_dev_remove_driver_gpios(ACPI_COMPANION(priv->codec_dev)); And this is to not have to add yet an other error-exit path which does this to the probe() function. By simply always removing the lookup here after the gpiod_get() we keep the error-exit paths as they were, rather then making them more complicated. But I guess things won't be so bad wrt err-exit-path complexity as for them to really be a problem, so if you prefer I can also: 1. Remove the second acpi_dev_add_driver_gpios + acpi_dev_remove_driver_gpios pair from the dai_link .init/.exit. 2. Remove the acpi_dev_remove_driver_gpios(ACPI_COMPANION(priv->codec_dev) call here moving it to snd_byt_rt5640_mc_remove() 3. Introduce a acpi_dev_remove_driver_gpios() remove in the error-exit paths of snd_byt_rt5640_mc_probe where necessary. >> + if (IS_ERR(priv->hsmic_detect)) { >> + ret_val = PTR_ERR(priv->hsmic_detect); >> + dev_err_probe(&pdev->dev, ret_val, "getting hsmic-detect GPIO\n"); >> + goto err_device; >> + } >> + } > Does this part need to be part of the machine driver probe, or could it > be moved in the dailink .init function? The idea here is that the gpiod_get may fail with -EPROBE_DEFER and then I want to fail as early as possible, so right in the probe function. This is also why the error is logged with dev_err_probe() which does not log anything for EPROBE_DEFER as retval. > Is this because you wanted to use devm_ function? No, I did consider adding the gpiod_get() for priv->hsmic_detect to the dai_link .init function and then I would just use a normal get, combined with an explicit _put in the dailink exit. I put this gpiod_get in the platform_driver probe to handle EPROBE_DEFER early on, rather then having it happen deep inside the devm_snd_soc_register_card() call-graph (when it calls the dailink .init). I would prefer to keep the gpiod_get inside snd_byt_rt5640_mc_probe for this reason, but as mentioned I can removed the second acpi_dev_add_driver_gpios + acpi_dev_remove_driver_gpios call pair from the dai_link .init/.exit. Please let me know how you want to proceed with this. Regards, Hans