Re: [PATCH 2/2] iio: adc: qcom-pm8xxx-xoadc: fix incorrect calibration values
From: Antony Kurniawan Soemardi
Date: Thu Oct 30 2025 - 13:05:28 EST
On 10/28/2025 4:44 PM, Konrad Dybcio wrote:
On 10/27/25 6:29 PM, Antony Kurniawan Soemardi wrote:
On msm8960 phones, the XOADC driver was using incorrect calibrationThe PM8921 datasheet suggests a single valid value of BIT(5)=1, BIT(6)=0
values:
absolute calibration dx = 625000 uV, dy = 4 units
ratiometric calibration dx = 1800, dy = -29041 units
As a result, reading from the IIO bus returned unexpected results:
in_voltage_7 (USB_VBUS): 0
in_voltage_10 (125V): 0
The issue was caused by not setting the ratiometric scale (amux_ip_rsv)
from the predefined channels. Additionally, the downstream code always
set the ADC_ARB_USRP_DIG_PARAM register to PM8XXX_ADC_ARB_ANA_DIG [1].
That value does not include the SEL_SHIFT0 and SEL_SHIFT1 bits. Enabling
those bits caused calibration errors too, so they were removed.
With these fixes, calibration now uses the correct values:
absolute calibration dx = 625000 uV, dy = 6307 units
ratiometric calibration dx = 1800, dy = 18249 units
Reading from the IIO bus now returns expected results:
in_voltage_7 (USB_VBUS): 4973836
in_voltage_10 (125V): 1249405
[1] https://github.com/LineageOS/android_kernel_sony_msm8960t/blob/93319b1e5aa343ec1c1aabcb028c5e88c7df7c01/drivers/hwmon/pm8xxx-adc.c#L407-L408
Signed-off-by: Antony Kurniawan Soemardi <linux@xxxxxxxxxxxxxx>
---
drivers/iio/adc/qcom-pm8xxx-xoadc.c | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)
diff --git a/drivers/iio/adc/qcom-pm8xxx-xoadc.c b/drivers/iio/adc/qcom-pm8xxx-xoadc.c
index 8555f34036fb13c41ac720dc02c1dc39876e9198..a53d361456ec36b66d258041877bd96ab37838c4 100644
--- a/drivers/iio/adc/qcom-pm8xxx-xoadc.c
+++ b/drivers/iio/adc/qcom-pm8xxx-xoadc.c
@@ -503,10 +503,11 @@ static int pm8xxx_read_channel_rsv(struct pm8xxx_xoadc *adc,
goto unlock;
/* Decimation factor */
- ret = regmap_write(adc->map, ADC_ARB_USRP_DIG_PARAM,
- ADC_ARB_USRP_DIG_PARAM_SEL_SHIFT0 |
- ADC_ARB_USRP_DIG_PARAM_SEL_SHIFT1 |
- ch->decimation << ADC_DIG_PARAM_DEC_SHIFT);
+ ret = regmap_update_bits(adc->map,
+ ADC_ARB_USRP_DIG_PARAM,
+ ADC_ARB_USRP_DIG_PARAM_DEC_RATE0 |
+ ADC_ARB_USRP_DIG_PARAM_DEC_RATE1,
for a "1K" (1/1024?) ratio, although a comment in this file suggests
BIT(5)|BIT(6) is also valid and corresponds to 1/4096.. I wouldn't be
surprised if that is the case
The previously set bits are a field called DECI_SEL but are otherwise left
undescribed
So, do you think we can leave the BIT(0) and BIT(1) as is? I have a feeling
that if they aren't set, these changes might prevent the APQ8060 Dragonboard
from reading the cm3605 sensor? or maybe not?
I mean this one, since the driver was originally tested on that board
https://github.com/torvalds/linux/blob/e53642b87a4f4b03a8d7e5f8507fc3cd0c595ea6/arch/arm/boot/dts/qcom/qcom-apq8060-dragonboard.dts#L67-L79
Hope this is useful
Konrad
Thanks,
Antony K. S.