Skip to main content

<alp/temperature.h> — On-module and Die Temperature

New in v0.17.0. Two handle-less reads in signed integer milli-degrees Celsius: the SoM's on-module ambient sensor, and the SoC's own die (junction) temperature.

#include <alp/temperature.h>

Quick example​

int32_t milli_c;

if (alp_temperature_read_milli_c(&milli_c) == ALP_OK) {
printf("SoM ambient: %d mC\n", (int)milli_c);
}

if (alp_temperature_read_die_milli_c(&milli_c) == ALP_OK) {
printf("SoC die: %d mC\n", (int)milli_c);
}

There is nothing to open, hold or select: the on-module manifest carries at most one temperature sensor per SoM, so the API takes no handle and no index.

Functions​

CallReads
alp_temperature_read_milli_c(int32_t *milli_c)The on-module ambient-temperature sensor (e.g. the TI TMP112 on E1M-AEN).
alp_temperature_read_die_milli_c(int32_t *milli_c)The SoC die (junction) temperature. When the SoC has several die sensors, the hottest is reported — the figure a throttling or protection decision cares about.

Both write *milli_c only on ALP_OK and leave it untouched on any error.

The contract is units and sign only. The fitted part's resolution (TMP112: 62.5 m°C/LSB) is informational, so a future SoM with a different part changes nothing a caller can observe except the reading's granularity.

Two different quantities, kept apart on purpose​

The die temperature reads far above ambient under load. The two are never folded together, so a SoM can have either, both or neither. The ambient sensor is also kept out of <alp/hw_info.h>: hw_info is boot-time identity, this is polled telemetry.

Presence of the ambient sensor is a SoM fact — it comes from the SoM preset's on-module device population — not a SoC fact, so it is not exposed through ALP_HAS(), which is generated per SoC and would wrongly claim the sensor for every SoM sharing that die.

Return codes​

Codealp_temperature_read_milli_calp_temperature_read_die_milli_c
ALP_OKValid read.Valid read.
ALP_ERR_INVALmilli_c is NULL.milli_c is NULL.
ALP_ERR_NOSUPPORTThe build reads no on-module sensor (every non-AEN target today — V2N / V2M modules carry an on-module TMP112 at BRD_I2C 0x40, but the SDK does not read it through this call yet), or a sensor is declared but no driver bound it.No thermal source: baremetal, a Zephyr tree with no die-temp0 node (or no driver bound to it), or a Linux kernel with no matching thermal zone.
ALP_ERR_NOT_READYA driver bound the sensor but it did not come up (e.g. it NACKs at boot).The Zephyr device bound but did not come up.
ALP_ERR_IOTransfer fault while reading.Transfer fault (Zephyr), or matching Linux zones exist but none could be read.

Treat ALP_ERR_NOSUPPORT as "this build has no such sensor" and degrade quietly; it is the expected answer on many targets today.

Coverage in v0.17.0​

TargetAmbient (read_milli_c)Die (read_die_milli_c)
E1M-AEN, ZephyrYes — binds the metadata-emitted alp-temp0 devicetree alias through the upstream Zephyr sensor API (needs CONFIG_SENSOR).Through the upstream die-temp0 devicetree alias (SENSOR_CHAN_DIE_TEMP). The SDK's SoM board trees do not declare it, so it answers only when a SoC or board tree provides one; otherwise ALP_ERR_NOSUPPORT.
E1M-X V2N / V2M, LinuxALP_ERR_NOSUPPORTYes — the hottest /sys/class/thermal/thermal_zone*/temp whose type begins with cpu-thermal (zones are matched by type, never by index). The kernel owns the sensor hardware.
BaremetalALP_ERR_NOSUPPORTALP_ERR_NOSUPPORT

On AEN the ambient read goes through the Zephyr sensor driver, never by opening the on-module I2C bus directly: an independent open at a different bitrate would race your own handle on that bus.

ABI status​

[ABI-EXPERIMENTAL] — new in v0.17.0.

See also​

Questions about this page? Email contact@alplab.ai