arm64: dts: qcom: sm6125: add TSENS die temperature sensing
Subsystems (5)
cover arm64: dts: qcom: sm6125: add TSENS die temperature sensing
SM6125 has a TSENS v2 block that mainline does not describe, so the SoC has no die temperature sensing: on a stock kernel the only thermal zones on an SM6125 board are the handful backed by PMIC ADC channels.
The block needs no new driver code. Its register and interrupt layout is identical to SM6115's and, being v2, it self-calibrates and needs no qfprom nvmem cells, so it binds through the existing qcom,tsens-v2 fallback. Patch 1 documents the SoC-specific compatible; patch 2 adds the controller and the thermal zones.
Sensor-to-zone mapping was taken from the downstream device tree for the Xiaomi Mi A3 (laurel_sprout) and cross-checked against sm6115.dtsi, which agrees on every sensor the two have in common (2, 3, 4, 13, 14 and 15). Sensor 5 is not referenced downstream and is left out, so 15 of the 16 sensors get zones.
No cooling devices are referenced, since SM6125 has no GPU node upstream yet. The zones report temperature and carry passive and critical trips; a cooling map for gpu-thermal can follow alongside the GPU node.
Tested on a Xiaomi Mi A3: tsens0 binds at 4411000.thermal-sensor with no dmesg output, the thermal zone count goes from 5 to 20, and the 15 new die zones read 37.7-39.6 degrees Celsius at idle. Under a 660 s glmark2-es2-drm soak the hottest zone reached 52.2 degrees Celsius against its 85 degrees Celsius passive trip.
Both SM6125 boards (laurel-sprout and ginkgo) compile with no new dtc
warnings: pristine master emits 7 unique_unit_address warnings on the geniqup nodes, and the series emits the same 7.
I have no other SM6125 board to test on, so the zone naming beyond the sensors shared with SM6115 rests on the downstream device tree.
Roman Linev (2):
dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
arm64: dts: qcom: sm6125: add TSENS and thermal zones
.../bindings/thermal/qcom-tsens.yaml | 1 + arch/arm64/boot/dts/qcom/sm6125.dtsi | 299 ++++++++++++++++++ 2 files changed, 300 insertions(+)
Signature · 1 line
2.55.0
PATCH 1/2 dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
SM6125 has a TSENS v2 block with 16 sensors. Its register and interrupt layout is identical to SM6115's, and it needs no calibration cells because v2 self-calibrates, so it is generic apart from the SoC-specific compatible.
Document it in the v2 list, bound by the qcom,tsens-v2 fallback.
--- Documentation/devicetree/bindings/thermal/qcom-tsens.yaml | 1 + 1 file changed, 1 insertion(+)
Documentation/devicetree/bindings/thermal/qcom-tsens.yaml +1 -0
@@ -78,6 +78,7 @@ properties: - qcom,sdm845-tsens - qcom,shikra-tsens - qcom,sm6115-tsens + - qcom,sm6125-tsens - qcom,sm6350-tsens - qcom,sm6375-tsens - qcom,sm8150-tsens
Signature · 1 line
2.55.0
Re: [PATCH 1/2] dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
Quoted 11 lines · Roman Linev
SM6125 has a TSENS v2 block with 16 sensors. Its register and interrupt layout is identical to SM6115's, and it needs no calibration cells because v2 self-calibrates, so it is generic apart from the SoC-specific compatible.
Document it in the v2 list, bound by the qcom,tsens-v2 fallback.
Signed-off-by Roman Linev <[email protected]>--- Documentation/devicetree/bindings/thermal/qcom-tsens.yaml | 1 + 1 file changed, 1 insertion(+)
Best regards,
Krzysztof
PATCH 2/2 arm64: dts: qcom: sm6125: add TSENS and thermal zones
In reply to Roman LinevSM6125 has a TSENS v2 controller with 16 sensors, at 0x04411000 (TM) and 0x04410000 (SROT), with SPI 275 as "uplow" and SPI 190 as "critical". The register and interrupt layout is identical to SM6115's, and being v2 the block self-calibrates, so it needs no qfprom nvmem cells and no new driver code: it binds through the existing qcom,tsens-v2 fallback.
Add the controller and thermal zones for the 15 sensors that are used. The sensor-to-zone mapping comes from the downstream device tree for Xiaomi Mi A3 (laurel_sprout), cross-checked against mainline sm6115.dtsi, which agrees on sensors 2, 3, 4, 13, 14 and 15. Sensor 5 is not referenced downstream and is left out.
No cooling devices are wired up: SM6125 has no GPU node upstream yet, so there is nothing to reference. The zones report temperature and carry passive and critical trips; cooling maps can be added alongside the consumers when those land.
Tested on Xiaomi Mi A3 (sm6125): tsens0 binds at 4411000.thermal-sensor with no dmesg output, the thermal zone count goes from 5 to 20, and the 15 new die zones read 37.7-39.6 degrees Celsius at idle. Under a 660 s glmark2-es2-drm soak the hottest zone reached 52.2 degrees Celsius, well below its 85 degrees Celsius passive trip.
Both SM6125 boards still build with no new dtc warnings.
--- arch/arm64/boot/dts/qcom/sm6125.dtsi | 299 +++++++++++++++++++++++++++ 1 file changed, 299 insertions(+)
arch/arm64/boot/dts/qcom/sm6125.dtsi +299 -0
@@ -719,6 +719,17 @@ spmi_bus: spmi@1c40000 { #interrupt-cells = <4>; }; + tsens0: thermal-sensor@4411000 { + compatible = "qcom,sm6125-tsens", "qcom,tsens-v2"; + reg = <0x0 0x04411000 0x0 0x1ff>, /* TM */ + <0x0 0x04410000 0x0 0x8>; /* SROT */ + #qcom,sensors = <16>; + interrupts = <GIC_SPI 275 IRQ_TYPE_LEVEL_HIGH>, + <GIC_SPI 190 IRQ_TYPE_LEVEL_HIGH>; + interrupt-names = "uplow", "critical"; + #thermal-sensor-cells = <1>; + }; + rpm_msg_ram: sram@45f0000 { compatible = "qcom,rpm-msg-ram"; reg = <0x0 0x045f0000 0x0 0x7000>;
@@ -1634,6 +1645,294 @@ intc: interrupt-controller@f200000 { }; }; + thermal-zones { + /* + * Sensor-to-zone mapping verified against the downstream + * laurel_sprout device tree (tsens@4410000, sensors 0..15) and + * cross-checked with mainline sm6115.dtsi, which agrees on + * sensors 2, 3, 4, 13, 14 and 15. Sensor 5 is unused by the + * downstream tree and is therefore left out. + * cpu4..cpu7 are the downstream cpu-1-{0..3} (cluster 1) sensors. + */ + aoss0-thermal { + thermal-sensors = <&tsens0 0>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cdsp-thermal { + thermal-sensors = <&tsens0 1>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + wlan-thermal { + thermal-sensors = <&tsens0 2>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + camera-thermal { + thermal-sensors = <&tsens0 3>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + video-thermal { + thermal-sensors = <&tsens0 4>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpuss0-thermal { + thermal-sensors = <&tsens0 6>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpuss1-thermal { + thermal-sensors = <&tsens0 7>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpuss2-thermal { + thermal-sensors = <&tsens0 8>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpu4-thermal { + thermal-sensors = <&tsens0 9>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpu5-thermal { + thermal-sensors = <&tsens0 10>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpu6-thermal { + thermal-sensors = <&tsens0 11>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + cpu7-thermal { + thermal-sensors = <&tsens0 12>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + modem0-thermal { + thermal-sensors = <&tsens0 13>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + display-thermal { + thermal-sensors = <&tsens0 14>; + + trips { + trip-point0 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point1 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + + gpu-thermal { + polling-delay-passive = <250>; + + thermal-sensors = <&tsens0 15>; + + trips { + trip-point0 { + temperature = <85000>; + hysteresis = <1000>; + type = "passive"; + }; + + trip-point1 { + temperature = <95000>; + hysteresis = <2000>; + type = "passive"; + }; + + trip-point2 { + temperature = <115000>; + hysteresis = <1000>; + type = "critical"; + }; + }; + }; + }; + timer { compatible = "arm,armv8-timer"; interrupts = <GIC_PPI 1 IRQ_TYPE_LEVEL_LOW>,
Signature · 1 line
2.55.0
Re: [PATCH 2/2] arm64: dts: qcom: sm6125: add TSENS and thermal zones
Quoted 12 lines · Roman Linev
SM6125 has a TSENS v2 controller with 16 sensors, at
0x04411000(TM) and0x04410000(SROT), with SPI 275 as "uplow" and SPI 190 as "critical". The register and interrupt layout is identical to SM6115's, and being v2 the block self-calibrates, so it needs no qfprom nvmem cells and no new driver code: it binds through the existing qcom,tsens-v2 fallback.Add the controller and thermal zones for the 15 sensors that are used. The sensor-to-zone mapping comes from the downstream device tree for Xiaomi Mi A3 (
laurel_sprout), cross-checked against mainlinesm6115.dtsi, which agrees on sensors 2, 3, 4, 13, 14 and 15. Sensor 5 is not referenced downstream and is left out.
Quoted 4 lines
No cooling devices are wired up: SM6125 has no GPU node upstream yet, so there is nothing to reference. The zones report temperature and carry passive and critical trips; cooling maps can be added alongside the consumers when those land.
CPUs are good candidates too
[...]
arch/arm64/boot/dts/qcom/sm6125.dtsi spmi_bus: spmi@1c40000 { · line 725
Quoted 4 lines
arch/arm64/boot/dts/qcom/sm6125.dtsi +4
-0+ tsens0: thermal-sensor@4411000 { + compatible = "qcom,sm6125-tsens", "qcom,tsens-v2"; + reg = <0x0 0x04411000 0x0 0x1ff>, /* TM */ + <0x0 0x04410000 0x0 0x8>; /* SROT */
Let's drop the comments and make the regions 0x1000-long
[...]
arch/arm64/boot/dts/qcom/sm6125.dtsi intc: interrupt-controller@f200000 { · line 1651
Quoted 4 lines
arch/arm64/boot/dts/qcom/sm6125.dtsi +4
-0+ thermal-zones { + /* + * Sensor-to-zone mapping verified against the downstream + * laurel_sprout device tree (tsens@4410000, sensors 0..15) and
Very good
arch/arm64/boot/dts/qcom/sm6125.dtsi intc: interrupt-controller@f200000 { · line 1652
Quoted 1 line
arch/arm64/boot/dts/qcom/sm6125.dtsi +1
-0+ * cross-checked with mainline sm6115.dtsi, which agrees on
This is a different SoC, you shouldn't rely on that
arch/arm64/boot/dts/qcom/sm6125.dtsi intc: interrupt-controller@f200000 { · line 1654
Quoted 2 lines
arch/arm64/boot/dts/qcom/sm6125.dtsi +2
-0+ * sensors 2, 3, 4, 13, 14 and 15. Sensor 5 is unused by the + * downstream tree and is therefore left out.
and the docs confirm that it's unused
I think this comment should be dropped, these details seem more fitting for a commit message
Konrad
Re: [PATCH 2/2] arm64: dts: qcom: sm6125: add TSENS and thermal zones
In reply to Roman LinevQuoted 23 lines · Roman Linev
SM6125 has a TSENS v2 controller with 16 sensors, at
0x04411000(TM) and0x04410000(SROT), with SPI 275 as "uplow" and SPI 190 as "critical". The register and interrupt layout is identical to SM6115's, and being v2 the block self-calibrates, so it needs no qfprom nvmem cells and no new driver code: it binds through the existing qcom,tsens-v2 fallback.Add the controller and thermal zones for the 15 sensors that are used. The sensor-to-zone mapping comes from the downstream device tree for Xiaomi Mi A3 (
laurel_sprout), cross-checked against mainlinesm6115.dtsi, which agrees on sensors 2, 3, 4, 13, 14 and 15. Sensor 5 is not referenced downstream and is left out.No cooling devices are wired up: SM6125 has no GPU node upstream yet, so there is nothing to reference. The zones report temperature and carry passive and critical trips; cooling maps can be added alongside the consumers when those land.
Tested on Xiaomi Mi A3 (sm6125): tsens0 binds at 4411000.thermal-sensor with no dmesg output, the thermal zone count goes from 5 to 20, and the 15 new die zones read 37.7-39.6 degrees Celsius at idle. Under a 660 s glmark2-es2-drm soak the hottest zone reached 52.2 degrees Celsius, well below its 85 degrees Celsius passive trip.
Quoted 3 lines
Both SM6125 boards still build with no new dtc warnings.
Signed-off-by Roman Linev <[email protected]>
Arm64 DT patches, against torvalds/master.
Changes in v2:
- wire CPU cooling: #cooling-cells on all eight CPUs and per-CPU passive cooling maps for the cluster-1 CPUs, cpu4..cpu7 [Konrad]
- make the TM/SROT register regions 0x1000 long and drop the inline comments [Konrad]
- drop the sensor-mapping comment from the DTS; the provenance now lives in the commit message [Konrad]
Patch 1/2 is unchanged and carries Krzysztof's Acked-by. Abel's Reviewed-by was given on the v1 content of 2/2; it is not carried onto the reworked patch.
v1: https://lore.kernel.org/r/[email protected]/
Patches inlined below (also against torvalds/master).
[PATCH v2 1/2] dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
SM6125 has a TSENS v2 block with 16 sensors. Its register and interrupt layout is identical to SM6115's, and it needs no calibration cells because v2 self-calibrates, so it is generic apart from the SoC-specific compatible.
Document it in the v2 list, bound by the qcom,tsens-v2 fallback.
--- Documentation/devicetree/bindings/thermal/qcom-tsens.yaml | 1 + 1 file changed, 1 insertion(+)
Documentation/devicetree/bindings/thermal/qcom-tsens.yaml +1 -0
@@ -79,6 +79,7 @@ properties: - qcom,sdm845-tsens - qcom,shikra-tsens - qcom,sm6115-tsens + - qcom,sm6125-tsens - qcom,sm6350-tsens - qcom,sm6375-tsens - qcom,sm8150-tsens
Signature · 33 lines
2.55.0
[PATCH v2 2/2] arm64: dts: qcom: sm6125: add TSENS and thermal zones
SM6125 has a TSENS v2 controller with 16 sensors, at 0x04411000 (TM) and
0x04410000 (SROT), with SPI 275 as "uplow" and SPI 190 as "critical".
The register and interrupt layout is identical to SM6115's, and being v2
the block self-calibrates, so it needs no qfprom nvmem cells and no new
driver code: it binds through the existing qcom,tsens-v2 fallback.
Add the controller and thermal zones for the 15 sensors that are used.
The sensor-to-zone mapping comes from the downstream device tree for
Xiaomi Mi A3 (laurel_sprout); the binding documents sensor 5 as unused,
and it is left out. cpu4..cpu7 are the downstream cpu-1-{0..3} cluster
sensors: they get #cooling-cells and per-CPU passive cooling maps, which
stay inert until a cpufreq driver registers cooling devices. The other
zones report temperature and carry passive and critical trips; GPU
cooling can be mapped alongside the GPU node when it lands.
Tested on Xiaomi Mi A3 (sm6125): tsens0 binds at 4411000.thermal-sensor
with no dmesg output, the thermal zone count goes from 5 to 20, and the
15 new die zones read 37.7-39.6 degrees Celsius at idle. Under a 660 s
glmark2-es2-drm soak the hottest zone reached 52.2 degrees Celsius, well
below its 85 degrees Celsius passive trip.
Both SM6125 boards still build with no new dtc warnings.
Signed-off-by: Roman Linev <[email protected]>
---
arch/arm64/boot/dts/qcom/sm6125.dtsi | 336 +++++++++++++++++++++++++++
1 file changed, 336 insertions(+)arch/arm64/boot/dts/qcom/sm6125.dtsi +336 -0
402 lines not on this page · raw message
Signature · 1 line
2.55.0
Quoted 41 lines · Roman Linev
SM6125 has a TSENS v2 block that mainline does not describe, so the SoC has no die temperature sensing: on a stock kernel the only thermal zones on an SM6125 board are the handful backed by PMIC ADC channels.
The block needs no new driver code. Its register and interrupt layout is identical to SM6115's and, being v2, it self-calibrates and needs no qfprom nvmem cells, so it binds through the existing qcom,tsens-v2 fallback. Patch 1 documents the SoC-specific compatible; patch 2 adds the controller and the thermal zones.
Sensor-to-zone mapping was taken from the downstream device tree for the Xiaomi Mi A3 (
laurel_sprout) and cross-checked againstsm6115.dtsi, which agrees on every sensor the two have in common (2, 3, 4, 13, 14 and 15). Sensor 5 is not referenced downstream and is left out, so 15 of the 16 sensors get zones.No cooling devices are referenced, since SM6125 has no GPU node upstream yet. The zones report temperature and carry passive and critical trips; a cooling map for gpu-thermal can follow alongside the GPU node.
Tested on a Xiaomi Mi A3: tsens0 binds at 4411000.thermal-sensor with no dmesg output, the thermal zone count goes from 5 to 20, and the 15 new die zones read 37.7-39.6 degrees Celsius at idle. Under a 660 s glmark2-es2-drm soak the hottest zone reached 52.2 degrees Celsius against its 85 degrees Celsius passive trip.
Both SM6125 boards (laurel-sprout and ginkgo) compile with no new dtc
warnings: pristine master emits 7unique_unit_addresswarnings on the geniqup nodes, and the series emits the same 7.I have no other SM6125 board to test on, so the zone naming beyond the sensors shared with SM6115 rests on the downstream device tree.
Roman Linev (2):
dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
arm64: dts: qcom: sm6125: add TSENS and thermal zones.../bindings/thermal/qcom-tsens.yaml | 1 + arch/arm64/boot/dts/qcom/sm6125.dtsi | 299 ++++++++++++++++++ 2 files changed, 300 insertions(+)
Apologies for the noise — three copies of this v2 went out by accident (only the last one, with the full patches inline, is operative; the two earlier posts are incomplete drafts). Please disregard those and reply to this thread.
Konrad: v2 addresses your 2026-08-25 review asks (0x1000 TM/SROT regions, CPU cooling wired, comments moved to the commit message, no sm6115 cross-check reliance).
Quoted 46 lines · Roman Linev
Arm64 DT patches, against torvalds/master.
Changes in v2:
- wire CPU cooling: #cooling-cells on all eight CPUs and per-CPU passive cooling maps for the cluster-1 CPUs, cpu4..cpu7 [Konrad]
- make the TM/SROT register regions0x1000long and drop the inline comments [Konrad]
- drop the sensor-mapping comment from the DTS; the provenance now lives in the commit message [Konrad]Patch 1/2 is unchanged and carries Krzysztof's Acked-by. Abel's Reviewed-by was given on the v1 content of 2/2; it is not carried onto the reworked patch.
v1: https://lore.kernel.org/r/[email protected]/
Patches inlined below (also against torvalds/master).
[PATCH v2 1/2] dt-bindings: thermal: qcom-tsens: add qcom,sm6125-tsens
SM6125 has a TSENS v2 block with 16 sensors. Its register and interrupt layout is identical to SM6115's, and it needs no calibration cells because v2 self-calibrates, so it is generic apart from the SoC-specific compatible.
Document it in the v2 list, bound by the qcom,tsens-v2 fallback.
Signed-off-by Roman Linev <[email protected]>--- Documentation/devicetree/bindings/thermal/qcom-tsens.yaml | 1 + 1 file changed, 1 insertion(+)Documentation/devicetree/bindings/thermal/qcom-tsens.yaml +1
-0@@ -79,6 +79,7 @@ properties: - qcom,sdm845-tsens - qcom,shikra-tsens - qcom,sm6115-tsens + - qcom,sm6125-tsens - qcom,sm6350-tsens - qcom,sm6375-tsens - qcom,sm8150-tsens
Hi Roman,
Quoted 3 lines · Roman Linev
cpu4..cpu7 are the downstream cpu-1-{0..3} cluster
sensors: they get #cooling-cells and per-CPU passive cooling maps, which stay inert until a cpufreq driver registers cooling devices.
I have been running the same TSENS description on a Mi A3 (laurel-sprout), derived independently from the downstream trinket DT: same TM/SROT addresses, 16 sensors, SPI 275/190 and the same sensor map. It currently runs on next-20261113 and the zones read sane values.
I also have cpufreq-hw and CPU idle states for SM6125 working on that phone, which is the piece your cooling maps are waiting for. I just sent them as "[PATCH 0/3] arm64: dts: qcom: sm6125: Add cpufreq-hw and CPU idle states" (you are on Cc) and left #cooling-cells out so that it does not conflict with your patch 2/2.
I could only find v2 as a reply inside the v1 thread, without reviews since. Did I miss a standalone posting? If not, it may be worth resending it as its own series so that it gets picked up. Either way I will test the latest version on my phone and reply with a Tested-by.
Thanks,
Majid