Re: [PATCH 3/3] mfd: mt6397-core: Add mt6397 AUXADC support

On 9/15/26 11:24 PM, sashiko-bot@kernel.org wrote:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] iio: mt6397-auxadc: hardware state leak on error paths
--

commit e17dd59587f106380c3ede8aa27c118b33350f88
Author: Ryan Brue [off-list ref]

mfd: mt6397-core: Add mt6397 AUXADC support

The mt6397 PMIC includes an AUXADC. Register the AUXADC in the mt6397
devices array to allow the corresponding driver to probe using compatible
string.

quoted

diff --git a/drivers/mfd/mt6397-core.c b/drivers/mfd/mt6397-core.c
index 9ec9519965887..5d5f816492aaa 100644
--- a/drivers/mfd/mt6397-core.c
+++ b/drivers/mfd/mt6397-core.c

@@ -261,6 +261,9 @@ static const struct mfd_cell mt6359_devs[] = {
  
  static const struct mfd_cell mt6397_devs[] = {
  	{
+		.name = "mt6397-auxadc",
+		.of_compatible = "mediatek,mt6397-auxadc",

[Severity: Medium]
Enabling this device exposes a hardware state leak on error paths in the
mt6397-auxadc driver.

In mt6397_auxadc_isense_enable(), if the second regmap write fails, does the
function return without unwinding the first write, leaving
MT6397_CHR_CON16_ADCIN_VSEN_EN incorrectly enabled?

drivers/iio/adc/mt6397-auxadc.c:mt6397_auxadc_isense_enable() {
     ...
	ret = regmap_update_bits(map, MT6397_CHR_CON16,
				 MT6397_CHR_CON16_ADCIN_SEL,
				 MT6397_CHR_CON16_ADCIN_VSEN_EN);
	if (ret)
		return ret;

	return regmap_set_bits(map, MT6397_AUXADC_CON14,
			       MT6397_AUXADC_CON14_CH0_NORM_SEL |
			       MT6397_AUXADC_CON14_CH0_LBAT_SEL);
}

Similarly, during teardown in mt6397_auxadc_isense_disable(), the use of the
?: operator short-circuits the sequence.

If the first regmap_clear_bits() fails, will it abandon the cleanup and skip
disabling MT6397_CHR_CON16_ADCIN_SEL?

drivers/iio/adc/mt6397-auxadc.c:mt6397_auxadc_isense_disable() {
     ...
	ret = regmap_clear_bits(map, MT6397_AUXADC_CON14,
				MT6397_AUXADC_CON14_CH0_NORM_SEL |
				MT6397_AUXADC_CON14_CH0_LBAT_SEL);

	return ret ?: regmap_clear_bits(map, MT6397_CHR_CON16,
					MT6397_CHR_CON16_ADCIN_SEL);
}

A similar issue appears in mt6397_auxadc_battemp_bias().

Does this function lack a rollback path if the second or third regmap
operation fails, permanently leaking previously modified register states?

drivers/iio/adc/mt6397-auxadc.c:mt6397_auxadc_battemp_bias() {
     ...
	if (on) {
		ret = regmap_set_bits(map, MT6397_AUXADC_CON0,
				      MT6397_AUXADC_CON0_BUF_PWD_ON);
		if (ret)
			return ret;
		ret = regmap_set_bits(map, MT6397_AUXADC_CON0,
				      MT6397_AUXADC_CON0_BUF_PWD_B);
		if (ret)
			return ret;
		return regmap_set_bits(map, MT6397_CHR_CON7,
				       MT6397_CHR_CON7_BATON_TDET_EN);
	}
     ...
}

Good find, exiting early could leave the select bits set on CON14, or the battery detect bias current or input buffer on in CON0/CON7. Will be fixed in v2.