[PATCH] iio: humidity: am2315: fix negative temperature decoding

[PATCH] iio: humidity: am2315: fix negative temperature decoding

[Vidhu Sarwal]

The AM2315/AM2320 report temperature as a sign-magnitude value,
with bit 15 as the sign and bits 14-0 as the magnitude.
am2315_read_data() stores the raw value directly in an s16, causing
negative temperatures to be interpreted as two's complement. Both the
read_raw and the triggered buffer paths are affected.

Decode the sign bit before storing the temperature value. Positive
readings are unaffected.

Fixes: 366a3270c1c5 ("iio: humidity: Add support for AM2315")
Signed-off-by: Vidhu Sarwal <vidhu.linux@gmail.com>
---
Other implementations [1] [2] for the same sensor have hit this bug when the
sensor was placed in a freezer, although neither is kernel code.
 [1] https://github.com/adafruit/Adafruit_CircuitPython_AM2320/issues/1
 [2] https://github.com/letscontrolit/ESPEasy/issues/307

 drivers/iio/humidity/am2315.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/iio/humidity/am2315.c b/drivers/iio/humidity/am2315.c
index f29baa251f9f..31c4f52a0108 100644
--- a/drivers/iio/humidity/am2315.c
+++ b/drivers/iio/humidity/am2315.c
@@ -113,6 +113,7 @@ static int am2315_read_data(struct am2315_data *data,
 	 */
 	u8 rx_buf[8];
 	u16 crc;
+	u16 temp_raw;
 
 	/* First wake up the device. */
 	am2315_ping(data->client);
@@ -144,8 +145,15 @@ static int am2315_read_data(struct am2315_data *data,
 
 	sensor_data->hum_data = (rx_buf[AM2315_HUM_OFFSET] << 8) |
 				 rx_buf[AM2315_HUM_OFFSET + 1];
-	sensor_data->temp_data = (rx_buf[AM2315_TEMP_OFFSET] << 8) |
-				  rx_buf[AM2315_TEMP_OFFSET + 1];
+
+	/*
+	 * Temperature is sign-magnitude, not two's complement:
+	 * bit 15 is the sign, bits 14-0 are the magnitude (x0.1 degC).
+	 */
+	temp_raw = (rx_buf[AM2315_TEMP_OFFSET] << 8) |
+		   rx_buf[AM2315_TEMP_OFFSET + 1];
+	sensor_data->temp_data = (temp_raw & 0x8000) ?
+				  -(temp_raw & 0x7fff) : temp_raw;
 
 	return ret;
 
-- 
2.53.0

Re: [PATCH] iio: humidity: am2315: fix negative temperature decoding

[Andy Shevchenko]

On Mon, Sep 28, 2026 at 5:25 PM Vidhu Sarwal <vidhu.linux@gmail.com> wrote:
>
> The AM2315/AM2320 report temperature as a sign-magnitude value,
> with bit 15 as the sign and bits 14-0 as the magnitude.
> am2315_read_data() stores the raw value directly in an s16, causing
> negative temperatures to be interpreted as two's complement. Both the
> read_raw and the triggered buffer paths are affected.
>
> Decode the sign bit before storing the temperature value. Positive
> readings are unaffected.

...

> -       sensor_data->temp_data = (rx_buf[AM2315_TEMP_OFFSET] << 8) |
> -                                 rx_buf[AM2315_TEMP_OFFSET + 1];
> +
> +       /*
> +        * Temperature is sign-magnitude, not two's complement:
> +        * bit 15 is the sign, bits 14-0 are the magnitude (x0.1 degC).
> +        */
> +       temp_raw = (rx_buf[AM2315_TEMP_OFFSET] << 8) |
> +                  rx_buf[AM2315_TEMP_OFFSET + 1];
> +       sensor_data->temp_data = (temp_raw & 0x8000) ?
> +                                 -(temp_raw & 0x7fff) : temp_raw;

Can you use sign_extend32()?

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] iio: humidity: am2315: fix negative temperature decoding

[Vidhu Sarwal]

Hi Andy,
On Mon, Sep 28, 2026 at 9:01 PM Andy Shevchenko
<andy.shevchenko@gmail.com> wrote:

> Can you use sign_extend32()?

I checked whether sign_extend32() can be used here, but I don't
think itcan be applied since as per datasheet [1], The temperature value
is sign-magnitude such that bit 15 is the sign and bits 14 to 0
contain the magnitude. For eg., 0x8065 represents -101.
sign_extend32(temp_raw, 15) would instead interpret 0x8065 as a
two's-complement value, giving us -32667.

So I think the sign bit needs to be decoded explicitly by masking
bit 15 and negating the magnitude when set.

[1] https://cdn-shop.adafruit.com/datasheets/AM2315.pdf

Regards,
Vidhu Sarwal

Re: [PATCH] iio: humidity: am2315: fix negative temperature decoding

[Andy Shevchenko]

On Tue, Sep 29, 2026 at 02:37:41AM +0530, Vidhu Sarwal wrote:
> On Mon, Sep 28, 2026 at 9:01 PM Andy Shevchenko
> <andy.shevchenko@gmail.com> wrote:

...

> > Can you use sign_extend32()?

> I checked whether sign_extend32() can be used here, but I don't
> think itcan be applied since as per datasheet [1], The temperature value
> is sign-magnitude such that bit 15 is the sign and bits 14 to 0
> contain the magnitude. For eg., 0x8065 represents -101.
> sign_extend32(temp_raw, 15) would instead interpret 0x8065 as a
> two's-complement value, giving us -32667.
> 
> So I think the sign bit needs to be decoded explicitly by masking
> bit 15 and negating the magnitude when set.

Oh, okay, we (IIO community) already discussed the possibility to have a helper
like sign_extend32() for this type of cases, but it appears not so worth it,
IIRC.

> [1] https://cdn-shop.adafruit.com/datasheets/AM2315.pdf

-- 
With Best Regards,
Andy Shevchenko