nvidia-gpu: add voltage sensor
This commit introduces a voltage sensor for the GPU.
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/voltage_NVIDIA_GB200_GPU_0_Voltage_0
{
"@odata.id": "/redfish/v1/Chassis/NVIDIA_GB200_1/Sensors/voltage_NVIDIA_GB200_GPU_0_Voltage_0",
"@odata.type": "#Sensor.v1_2_0.Sensor",
"Id": "voltage_NVIDIA_GB200_GPU_0_Voltage_0",
"Name": "NVIDIA GB200 GPU 0 Voltage 0",
"Reading": 0.735,
"ReadingRangeMax": 1.8446744073709552e+16,
"ReadingRangeMin": 0.0,
"ReadingType": "Voltage",
"ReadingUnits": "V",
"Status": {
"Health": "OK",
"State": "Enabled"
}
}%
```
Change-Id: I3d98f3d7c11221a42460c6f8420c927c1b1711b2
Signed-off-by: Harshit Aghera <haghera@nvidia.com>
diff --git a/src/nvidia-gpu/NvidiaGpuDevice.cpp b/src/nvidia-gpu/NvidiaGpuDevice.cpp
index 5f3e83b..e966d57 100644
--- a/src/nvidia-gpu/NvidiaGpuDevice.cpp
+++ b/src/nvidia-gpu/NvidiaGpuDevice.cpp
@@ -17,6 +17,7 @@
#include <NvidiaGpuEnergySensor.hpp>
#include <NvidiaGpuPowerSensor.hpp>
#include <NvidiaGpuThresholds.hpp>
+#include <NvidiaGpuVoltageSensor.hpp>
#include <boost/asio/io_context.hpp>
#include <phosphor-logging/lg2.hpp>
#include <sdbusplus/asio/connection.hpp>
@@ -83,6 +84,10 @@
conn, mctpRequester, name + "_Energy_0", path, eid, gpuEnergySensorId,
objectServer, std::vector<thresholds::Threshold>{});
+ voltageSensor = std::make_shared<NvidiaGpuVoltageSensor>(
+ conn, mctpRequester, name + "_Voltage_0", path, eid, gpuVoltageSensorId,
+ objectServer, std::vector<thresholds::Threshold>{});
+
lg2::info("Added GPU {NAME} Sensors with chassis path: {PATH}.", "NAME",
name, "PATH", path);
@@ -98,6 +103,7 @@
}
powerSensor->update();
energySensor->update();
+ voltageSensor->update();
waitTimer.expires_after(std::chrono::milliseconds(sensorPollMs));
waitTimer.async_wait([this](const boost::system::error_code& ec) {
diff --git a/src/nvidia-gpu/NvidiaGpuDevice.hpp b/src/nvidia-gpu/NvidiaGpuDevice.hpp
index 0a85b9d..6cfed42 100644
--- a/src/nvidia-gpu/NvidiaGpuDevice.hpp
+++ b/src/nvidia-gpu/NvidiaGpuDevice.hpp
@@ -12,6 +12,7 @@
#include "NvidiaGpuSensor.hpp"
#include <NvidiaGpuEnergySensor.hpp>
+#include <NvidiaGpuVoltageSensor.hpp>
#include <boost/asio/io_context.hpp>
#include <boost/asio/steady_timer.hpp>
#include <sdbusplus/asio/connection.hpp>
@@ -58,6 +59,7 @@
std::shared_ptr<NvidiaGpuTempSensor> tLimitSensor;
std::shared_ptr<NvidiaGpuPowerSensor> powerSensor;
std::shared_ptr<NvidiaGpuEnergySensor> energySensor;
+ std::shared_ptr<NvidiaGpuVoltageSensor> voltageSensor;
SensorConfigs configs;
diff --git a/src/nvidia-gpu/NvidiaGpuMctpVdm.cpp b/src/nvidia-gpu/NvidiaGpuMctpVdm.cpp
index b70e0c2..efde488 100644
--- a/src/nvidia-gpu/NvidiaGpuMctpVdm.cpp
+++ b/src/nvidia-gpu/NvidiaGpuMctpVdm.cpp
@@ -349,5 +349,69 @@
return 0;
}
+
+int encodeGetVoltageRequest(uint8_t instanceId, uint8_t sensorId,
+ std::span<uint8_t> buf)
+{
+ if (buf.size() < sizeof(GetVoltageRequest))
+ {
+ return EINVAL;
+ }
+
+ auto* msg = reinterpret_cast<GetVoltageRequest*>(buf.data());
+
+ ocp::accelerator_management::BindingPciVidInfo header{};
+ header.ocp_accelerator_management_msg_type =
+ static_cast<uint8_t>(ocp::accelerator_management::MessageType::REQUEST);
+ header.instance_id = instanceId &
+ ocp::accelerator_management::instanceIdBitMask;
+ header.msg_type = static_cast<uint8_t>(MessageType::PLATFORM_ENVIRONMENTAL);
+
+ auto rc = packHeader(header, msg->hdr.msgHdr.hdr);
+
+ if (rc != 0)
+ {
+ return rc;
+ }
+
+ msg->hdr.command =
+ static_cast<uint8_t>(PlatformEnvironmentalCommands::GET_VOLTAGE);
+ msg->hdr.data_size = sizeof(sensorId);
+ msg->sensor_id = sensorId;
+
+ return 0;
+}
+
+int decodeGetVoltageResponse(std::span<const uint8_t> buf,
+ ocp::accelerator_management::CompletionCode& cc,
+ uint16_t& reasonCode, uint32_t& voltage)
+{
+ auto rc =
+ ocp::accelerator_management::decodeReasonCodeAndCC(buf, cc, reasonCode);
+
+ if (rc != 0 || cc != ocp::accelerator_management::CompletionCode::SUCCESS)
+ {
+ return rc;
+ }
+
+ if (buf.size() < sizeof(GetVoltageResponse))
+ {
+ return EINVAL;
+ }
+
+ const auto* response =
+ reinterpret_cast<const GetVoltageResponse*>(buf.data());
+
+ const uint16_t dataSize = le16toh(response->hdr.data_size);
+
+ if (dataSize != sizeof(uint32_t))
+ {
+ return EINVAL;
+ }
+
+ voltage = le32toh(response->voltage);
+
+ return 0;
+}
// NOLINTEND(cppcoreguidelines-pro-type-reinterpret-cast)
} // namespace gpu
diff --git a/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp b/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
index 10a3efc..e378cfc 100644
--- a/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
+++ b/src/nvidia-gpu/NvidiaGpuMctpVdm.hpp
@@ -33,6 +33,7 @@
READ_THERMAL_PARAMETERS = 0x02,
GET_CURRENT_POWER_DRAW = 0x03,
GET_CURRENT_ENERGY_COUNTER = 0x06,
+ GET_VOLTAGE = 0x0F,
};
enum class DeviceIdentification : uint8_t
@@ -71,6 +72,8 @@
using GetCurrentEnergyCounterRequest = GetNumericSensorReadingRequest;
+using GetVoltageRequest = GetNumericSensorReadingRequest;
+
struct GetTemperatureReadingResponse
{
ocp::accelerator_management::CommonResponse hdr;
@@ -95,6 +98,12 @@
uint64_t energy;
} __attribute__((packed));
+struct GetVoltageResponse
+{
+ ocp::accelerator_management::CommonResponse hdr;
+ uint32_t voltage;
+} __attribute__((packed));
+
int packHeader(const ocp::accelerator_management::BindingPciVidInfo& hdr,
ocp::accelerator_management::BindingPciVid& msg);
@@ -138,4 +147,11 @@
std::span<const uint8_t> buf,
ocp::accelerator_management::CompletionCode& cc, uint16_t& reasonCode,
uint64_t& energy);
+
+int encodeGetVoltageRequest(uint8_t instanceId, uint8_t sensorId,
+ std::span<uint8_t> buf);
+
+int decodeGetVoltageResponse(std::span<const uint8_t> buf,
+ ocp::accelerator_management::CompletionCode& cc,
+ uint16_t& reasonCode, uint32_t& voltage);
} // namespace gpu
diff --git a/src/nvidia-gpu/NvidiaGpuVoltageSensor.cpp b/src/nvidia-gpu/NvidiaGpuVoltageSensor.cpp
new file mode 100644
index 0000000..0a968f9
--- /dev/null
+++ b/src/nvidia-gpu/NvidiaGpuVoltageSensor.cpp
@@ -0,0 +1,138 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION &
+ * AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "NvidiaGpuVoltageSensor.hpp"
+
+#include "SensorPaths.hpp"
+#include "Thresholds.hpp"
+#include "Utils.hpp"
+#include "sensor.hpp"
+
+#include <bits/basic_string.h>
+
+#include <MctpRequester.hpp>
+#include <NvidiaDeviceDiscovery.hpp>
+#include <NvidiaGpuMctpVdm.hpp>
+#include <OcpMctpVdm.hpp>
+#include <phosphor-logging/lg2.hpp>
+#include <sdbusplus/asio/connection.hpp>
+#include <sdbusplus/asio/object_server.hpp>
+
+#include <cstddef>
+#include <cstdint>
+#include <functional>
+#include <limits>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+using namespace std::literals;
+
+// Reading from the device is in millivolts and unit set on the dbus is volts.
+static constexpr double gpuVoltageSensorMaxReading =
+ std::numeric_limits<uint64_t>::max() / 1000.0;
+static constexpr double gpuVoltageSensorMinReading =
+ std::numeric_limits<uint64_t>::min();
+
+NvidiaGpuVoltageSensor::NvidiaGpuVoltageSensor(
+ std::shared_ptr<sdbusplus::asio::connection>& conn,
+ mctp::MctpRequester& mctpRequester, const std::string& name,
+ const std::string& sensorConfiguration, const uint8_t eid, uint8_t sensorId,
+ sdbusplus::asio::object_server& objectServer,
+ std::vector<thresholds::Threshold>&& thresholdData) :
+ Sensor(escapeName(name), std::move(thresholdData), sensorConfiguration,
+ "energy", false, true, gpuVoltageSensorMaxReading,
+ gpuVoltageSensorMinReading, conn),
+ eid(eid), sensorId{sensorId}, mctpRequester(mctpRequester),
+ objectServer(objectServer)
+{
+ std::string dbusPath = sensorPathPrefix + "voltage/"s + escapeName(name);
+
+ sensorInterface = objectServer.add_interface(
+ dbusPath, "xyz.openbmc_project.Sensor.Value");
+
+ for (const auto& threshold : thresholds)
+ {
+ std::string interface = thresholds::getInterface(threshold.level);
+ thresholdInterfaces[static_cast<size_t>(threshold.level)] =
+ objectServer.add_interface(dbusPath, interface);
+ }
+
+ association = objectServer.add_interface(dbusPath, association::interface);
+
+ // Sensor values are only updated when the difference between the new and
+ // previous value exceeds the hysteresisPublish threshold. This threshold
+ // defaults to ((max - min) * 0.0001). Since this sensor lacks defined
+ // min/max values, theoretical limits are used instead, creating a large
+ // hysteresisPublish value that blocks D-Bus updates. Setting
+ // hysteresisPublish to 0 forces all sensor value changes to be published
+ // to D-Bus.
+ hysteresisPublish = 0;
+
+ setInitialProperties(sensor_paths::unitVolts);
+}
+
+NvidiaGpuVoltageSensor::~NvidiaGpuVoltageSensor()
+{
+ for (const auto& iface : thresholdInterfaces)
+ {
+ objectServer.remove_interface(iface);
+ }
+ objectServer.remove_interface(sensorInterface);
+ objectServer.remove_interface(association);
+}
+
+void NvidiaGpuVoltageSensor::checkThresholds()
+{
+ thresholds::checkThresholds(this);
+}
+
+void NvidiaGpuVoltageSensor::processResponse(int sendRecvMsgResult)
+{
+ if (sendRecvMsgResult != 0)
+ {
+ lg2::error(
+ "Error updating Voltage Sensor: sending message over MCTP failed, rc={RC}",
+ "RC", sendRecvMsgResult);
+ return;
+ }
+
+ ocp::accelerator_management::CompletionCode cc{};
+ uint16_t reasonCode = 0;
+ uint32_t voltageValue = 0;
+
+ auto rc =
+ gpu::decodeGetVoltageResponse(response, cc, reasonCode, voltageValue);
+
+ if (rc != 0 || cc != ocp::accelerator_management::CompletionCode::SUCCESS)
+ {
+ lg2::error(
+ "Error updating Voltage Sensor: decode failed, rc={RC}, cc={CC}, reasonCode={RESC}",
+ "RC", rc, "CC", cc, "RESC", reasonCode);
+ return;
+ }
+
+ // Reading from the device is in microvolts and unit set on the dbus
+ // is volts.
+ updateValue(voltageValue / 1000000.0);
+}
+
+void NvidiaGpuVoltageSensor::update()
+{
+ auto rc = gpu::encodeGetVoltageRequest(0, sensorId, request);
+
+ if (rc != 0)
+ {
+ lg2::error("Error updating Voltage Sensor: encode failed, rc={RC}",
+ "RC", rc);
+ return;
+ }
+
+ mctpRequester.sendRecvMsg(
+ eid, request, response,
+ [this](int sendRecvMsgResult) { processResponse(sendRecvMsgResult); });
+}
diff --git a/src/nvidia-gpu/NvidiaGpuVoltageSensor.hpp b/src/nvidia-gpu/NvidiaGpuVoltageSensor.hpp
new file mode 100644
index 0000000..a8c1038
--- /dev/null
+++ b/src/nvidia-gpu/NvidiaGpuVoltageSensor.hpp
@@ -0,0 +1,57 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION &
+ * AFFILIATES. All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#pragma once
+
+#include "MctpRequester.hpp"
+#include "Thresholds.hpp"
+#include "sensor.hpp"
+
+#include <NvidiaGpuMctpVdm.hpp>
+#include <sdbusplus/asio/connection.hpp>
+#include <sdbusplus/asio/object_server.hpp>
+
+#include <array>
+#include <cstdint>
+#include <memory>
+#include <string>
+#include <vector>
+
+constexpr uint8_t gpuVoltageSensorId{0};
+
+struct NvidiaGpuVoltageSensor : public Sensor
+{
+ public:
+ NvidiaGpuVoltageSensor(
+ std::shared_ptr<sdbusplus::asio::connection>& conn,
+ mctp::MctpRequester& mctpRequester, const std::string& name,
+ const std::string& sensorConfiguration, uint8_t eid, uint8_t sensorId,
+ sdbusplus::asio::object_server& objectServer,
+ std::vector<thresholds::Threshold>&& thresholdData);
+
+ ~NvidiaGpuVoltageSensor() override;
+
+ void checkThresholds() override;
+
+ void update();
+
+ private:
+ void processResponse(int sendRecvMsgResult);
+
+ uint8_t eid{};
+
+ uint8_t sensorId;
+
+ std::shared_ptr<sdbusplus::asio::connection> conn;
+
+ mctp::MctpRequester& mctpRequester;
+
+ sdbusplus::asio::object_server& objectServer;
+
+ std::array<uint8_t, sizeof(gpu::GetVoltageRequest)> request{};
+
+ std::array<uint8_t, sizeof(gpu::GetVoltageResponse)> response{};
+};
diff --git a/src/nvidia-gpu/meson.build b/src/nvidia-gpu/meson.build
index 11aaab3..39ce871 100644
--- a/src/nvidia-gpu/meson.build
+++ b/src/nvidia-gpu/meson.build
@@ -8,6 +8,7 @@
'NvidiaGpuSensor.cpp',
'NvidiaGpuSensorMain.cpp',
'NvidiaGpuThresholds.cpp',
+ 'NvidiaGpuVoltageSensor.cpp',
'OcpMctpVdm.cpp',
)
diff --git a/src/nvidia-gpu/tests/NvidiaGpuSensorTest.cpp b/src/nvidia-gpu/tests/NvidiaGpuSensorTest.cpp
index 2c02d3b..d293ca5 100644
--- a/src/nvidia-gpu/tests/NvidiaGpuSensorTest.cpp
+++ b/src/nvidia-gpu/tests/NvidiaGpuSensorTest.cpp
@@ -947,6 +947,153 @@
EXPECT_EQ(result, EINVAL); // Should indicate error for invalid data size
}
+
+// Tests for GpuMctpVdm::encodeGetVoltageRequest function
+TEST_F(GpuMctpVdmTests, EncodeGetVoltageRequestSuccess)
+{
+ const uint8_t instanceId = 8;
+ const uint8_t sensorId = 4;
+ std::array<uint8_t, sizeof(gpu::GetVoltageRequest)> buf{};
+
+ int result = gpu::encodeGetVoltageRequest(instanceId, sensorId, buf);
+
+ EXPECT_EQ(result, 0);
+
+ gpu::GetVoltageRequest request{};
+ std::memcpy(&request, buf.data(), sizeof(request));
+
+ EXPECT_EQ(request.hdr.msgHdr.hdr.pci_vendor_id,
+ htobe16(gpu::nvidiaPciVendorId));
+ EXPECT_EQ(request.hdr.msgHdr.hdr.instance_id &
+ ocp::accelerator_management::instanceIdBitMask,
+ instanceId & ocp::accelerator_management::instanceIdBitMask);
+ EXPECT_NE(request.hdr.msgHdr.hdr.instance_id &
+ ocp::accelerator_management::requestBitMask,
+ 0);
+ EXPECT_EQ(request.hdr.msgHdr.hdr.ocp_accelerator_management_msg_type,
+ static_cast<uint8_t>(gpu::MessageType::PLATFORM_ENVIRONMENTAL));
+
+ // Verify request data
+ EXPECT_EQ(
+ request.hdr.command,
+ static_cast<uint8_t>(gpu::PlatformEnvironmentalCommands::GET_VOLTAGE));
+ EXPECT_EQ(request.hdr.data_size, sizeof(sensorId));
+ EXPECT_EQ(request.sensor_id, sensorId);
+}
+
+// Tests for GpuMctpVdm::decodeGetVoltageResponse function
+TEST_F(GpuMctpVdmTests, DecodeGetVoltageResponseSuccess)
+{
+ // Create a mock successful response
+ std::array<uint8_t, sizeof(gpu::GetVoltageResponse)> buf{};
+
+ gpu::GetVoltageResponse response{};
+ ocp::accelerator_management::BindingPciVidInfo headerInfo{};
+ headerInfo.ocp_accelerator_management_msg_type = static_cast<uint8_t>(
+ ocp::accelerator_management::MessageType::RESPONSE);
+ headerInfo.instance_id = 8;
+ headerInfo.msg_type =
+ static_cast<uint8_t>(gpu::MessageType::PLATFORM_ENVIRONMENTAL);
+
+ gpu::packHeader(headerInfo, response.hdr.msgHdr.hdr);
+
+ // Populate response data
+ response.hdr.command =
+ static_cast<uint8_t>(gpu::PlatformEnvironmentalCommands::GET_VOLTAGE);
+ response.hdr.completion_code = static_cast<uint8_t>(
+ ocp::accelerator_management::CompletionCode::SUCCESS);
+ response.hdr.reserved = 0;
+ response.hdr.data_size = htole16(sizeof(uint32_t));
+
+ // Set a voltage value of 12.5V (12.5 * 1000 = 12500 mV)
+ response.voltage = htole32(12500);
+
+ std::memcpy(buf.data(), &response, sizeof(response));
+
+ // Test decoding
+ ocp::accelerator_management::CompletionCode cc{};
+ uint16_t reasonCode{};
+ uint32_t voltage{};
+
+ int result = gpu::decodeGetVoltageResponse(buf, cc, reasonCode, voltage);
+
+ EXPECT_EQ(result, 0);
+ EXPECT_EQ(cc, ocp::accelerator_management::CompletionCode::SUCCESS);
+ EXPECT_EQ(reasonCode, 0);
+ EXPECT_EQ(voltage, 12500);
+}
+
+TEST_F(GpuMctpVdmTests, DecodeGetVoltageResponseError)
+{
+ std::array<uint8_t,
+ sizeof(ocp::accelerator_management::CommonNonSuccessResponse)>
+ buf{};
+
+ // Populate error response data
+ ocp::accelerator_management::CommonNonSuccessResponse errorResponse{};
+ ocp::accelerator_management::BindingPciVidInfo headerInfo{};
+ headerInfo.ocp_accelerator_management_msg_type = static_cast<uint8_t>(
+ ocp::accelerator_management::MessageType::RESPONSE);
+ headerInfo.instance_id = 8;
+ headerInfo.msg_type =
+ static_cast<uint8_t>(gpu::MessageType::PLATFORM_ENVIRONMENTAL);
+
+ gpu::packHeader(headerInfo, errorResponse.msgHdr.hdr);
+
+ errorResponse.command =
+ static_cast<uint8_t>(gpu::PlatformEnvironmentalCommands::GET_VOLTAGE);
+ errorResponse.completion_code = static_cast<uint8_t>(
+ ocp::accelerator_management::CompletionCode::ERR_NOT_READY);
+ errorResponse.reason_code = htole16(0x1234);
+
+ std::memcpy(buf.data(), &errorResponse, sizeof(errorResponse));
+
+ // Test decoding
+ ocp::accelerator_management::CompletionCode cc{};
+ uint16_t reasonCode{};
+ uint32_t voltage{};
+
+ int result = gpu::decodeGetVoltageResponse(buf, cc, reasonCode, voltage);
+
+ EXPECT_EQ(result, 0);
+ EXPECT_EQ(cc, ocp::accelerator_management::CompletionCode::ERR_NOT_READY);
+ EXPECT_EQ(reasonCode, 0x1234);
+}
+
+TEST_F(GpuMctpVdmTests, DecodeGetVoltageResponseInvalidSize)
+{
+ // Create a mock response with invalid data_size
+ std::array<uint8_t, sizeof(gpu::GetVoltageResponse)> buf{};
+
+ gpu::GetVoltageResponse response{};
+ ocp::accelerator_management::BindingPciVidInfo headerInfo{};
+ headerInfo.ocp_accelerator_management_msg_type = static_cast<uint8_t>(
+ ocp::accelerator_management::MessageType::RESPONSE);
+ headerInfo.instance_id = 8;
+ headerInfo.msg_type =
+ static_cast<uint8_t>(gpu::MessageType::PLATFORM_ENVIRONMENTAL);
+
+ gpu::packHeader(headerInfo, response.hdr.msgHdr.hdr);
+
+ response.hdr.command =
+ static_cast<uint8_t>(gpu::PlatformEnvironmentalCommands::GET_VOLTAGE);
+ response.hdr.completion_code = static_cast<uint8_t>(
+ ocp::accelerator_management::CompletionCode::SUCCESS);
+ response.hdr.reserved = 0;
+ response.hdr.data_size = htole16(2); // Invalid - should be sizeof(uint32_t)
+ response.voltage = htole32(12500);
+
+ std::memcpy(buf.data(), &response, sizeof(response));
+
+ // Test decoding
+ ocp::accelerator_management::CompletionCode cc{};
+ uint16_t reasonCode{};
+ uint32_t voltage{};
+
+ int result = gpu::decodeGetVoltageResponse(buf, cc, reasonCode, voltage);
+
+ EXPECT_EQ(result, EINVAL); // Should indicate error for invalid data size
+}
} // namespace gpu_mctp_tests
int main(int argc, char** argv)