commit 5e7deccd14dcac790028a6641291cc019c1c4e52
author Harshit Aghera <haghera@nvidia.com> Wed May 07 16:20:16 2025 +0530
committer Ed Tanous <ed@tanous.net> Thu Jul 10 15:01:22 2025 +0000
tree 84807211fb4ed2e7959c95f4c8c39b410e3bb3b3
parent ba138dae62cfd96571372a3e22317dc57ab72c80

nvidia-gpu: add thresholds support to TLimit

This patch introduces support for retrieving GPU TLimit thresholds
directly from the GPU device. TLimit Temperature represents the
difference in degrees Celsius between the current GPU temperature and
the initial throttle threshold. The patch also enables the extraction of
three critical throttle thresholds — Warning Low, Critical Low, and Hard
Shutdown Low — from the GPU hardware.

Tested: Build an image for gb200nvl-obmc machine with the following
patches cherry picked. This patches are needed to enable the mctp stack.

https://gerrit.openbmc.org/c/openbmc/openbmc/+/79422

```
$ curl -s -k -u 'root:0penBmc' https://10.137.203.137/redfish/v1/Chassis/NVIDIA_GB200_1/Sensors/temperature_NVIDIA_GB200_GPU_0_TEMP_1
{
  "@odata.id": "/redfish/v1/Chassis/NVIDIA_GB200_1/Sensors/temperature_NVIDIA_GB200_GPU_0_TEMP_1",
  "@odata.type": "#Sensor.v1_2_0.Sensor",
  "Id": "temperature_NVIDIA_GB200_GPU_0_TEMP_1",
  "Name": "NVIDIA GB200 GPU 0 TEMP 1",
  "Reading": 57.3984375,
  "ReadingRangeMax": 127.0,
  "ReadingRangeMin": -128.0,
  "ReadingType": "Temperature",
  "ReadingUnits": "Cel",
  "Status": {
    "Health": "OK",
    "State": "Enabled"
  },
  "Thresholds": {
    "LowerCaution": {
      "Reading": 0.0
    },
    "LowerCritical": {
      "Reading": 0.0
    },
    "LowerFatal": {
      "Reading": 0.0
    }
  }
}%
```

Change-Id: I6f2ff2652ce9246287f9bd63c4297d9ad3229963
Signed-off-by: Harshit Aghera <haghera@nvidia.com>

src/nvidia-gpu/NvidiaGpuMctpVdm.hpp

diff --git a/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp b/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
index ce3b393..f7c78b8 100644
--- a/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
+++ b/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
@@ -30,6 +30,7 @@
 enum class PlatformEnvironmentalCommands : uint8_t
 {
     GET_TEMPERATURE_READING = 0x00,
+    READ_THERMAL_PARAMETERS = 0x02,
 };
 
 enum class DeviceIdentification : uint8_t
@@ -57,12 +58,20 @@
 
 using GetTemperatureReadingRequest = GetNumericSensorReadingRequest;
 
+using ReadThermalParametersRequest = GetNumericSensorReadingRequest;
+
 struct GetTemperatureReadingResponse
 {
     ocp::accelerator_management::CommonResponse hdr;
     int32_t reading;
 } __attribute__((packed));
 
+struct ReadThermalParametersResponse
+{
+    ocp::accelerator_management::CommonResponse hdr;
+    int32_t threshold;
+} __attribute__((packed));
+
 int packHeader(const ocp::accelerator_management::BindingPciVidInfo& hdr,
                ocp::accelerator_management::BindingPciVid& msg);
 
@@ -82,4 +91,12 @@
     ocp::accelerator_management::CompletionCode& cc, uint16_t& reasonCode,
     double& temperatureReading);
 
+int encodeReadThermalParametersRequest(uint8_t instanceId, uint8_t sensorId,
+                                       std::span<uint8_t> buf);
+
+int decodeReadThermalParametersResponse(
+    std::span<const uint8_t> buf,
+    ocp::accelerator_management::CompletionCode& cc, uint16_t& reasonCode,
+    int32_t& threshold);
+
 } // namespace gpu