blob: e966d57a2242a58fc3c7bebb32d3fd89ebd900ba [file] [log] [blame]
Harshit Agherafa2a5b92025-05-22 11:35:39 +05301/*
2 * SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION &
3 * AFFILIATES. All rights reserved.
4 * SPDX-License-Identifier: Apache-2.0
5 */
6
7#include "NvidiaGpuDevice.hpp"
8
9#include "NvidiaDeviceDiscovery.hpp"
10#include "NvidiaGpuSensor.hpp"
11#include "Thresholds.hpp"
12#include "Utils.hpp"
13
14#include <bits/basic_string.h>
15
16#include <MctpRequester.hpp>
Harshit Aghera128c91d2025-05-27 14:20:24 +053017#include <NvidiaGpuEnergySensor.hpp>
Harshit Agherac8dab722025-05-08 15:57:42 +053018#include <NvidiaGpuPowerSensor.hpp>
Harshit Agherac20108d2025-05-07 16:20:16 +053019#include <NvidiaGpuThresholds.hpp>
Harshit Agherab55847f2025-05-27 14:53:56 +053020#include <NvidiaGpuVoltageSensor.hpp>
Harshit Agherafa2a5b92025-05-22 11:35:39 +053021#include <boost/asio/io_context.hpp>
22#include <phosphor-logging/lg2.hpp>
23#include <sdbusplus/asio/connection.hpp>
24#include <sdbusplus/asio/object_server.hpp>
25
26#include <chrono>
27#include <cstdint>
28#include <memory>
29#include <string>
Harshit Agherac20108d2025-05-07 16:20:16 +053030#include <utility>
Harshit Agherafa2a5b92025-05-22 11:35:39 +053031#include <vector>
32
33GpuDevice::GpuDevice(const SensorConfigs& configs, const std::string& name,
34 const std::string& path,
35 const std::shared_ptr<sdbusplus::asio::connection>& conn,
36 uint8_t eid, boost::asio::io_context& io,
37 mctp::MctpRequester& mctpRequester,
38 sdbusplus::asio::object_server& objectServer) :
39 eid(eid), sensorPollMs(std::chrono::milliseconds{configs.pollRate}),
40 waitTimer(io, std::chrono::steady_clock::duration(0)),
41 mctpRequester(mctpRequester), conn(conn), objectServer(objectServer),
42 configs(configs), name(escapeName(name)), path(path)
43{
44 makeSensors();
45}
46
47void GpuDevice::makeSensors()
48{
49 tempSensor = std::make_shared<NvidiaGpuTempSensor>(
Harshit Aghera0e1718c2025-05-05 12:26:35 +053050 conn, mctpRequester, name + "_TEMP_0", path, eid, gpuTempSensorId,
51 objectServer, std::vector<thresholds::Threshold>{});
52
Harshit Agherac20108d2025-05-07 16:20:16 +053053 readThermalParameters(
54 eid,
55 std::vector<gpuThresholdId>{gpuTLimitWarnringThresholdId,
56 gpuTLimitCriticalThresholdId,
57 gpuTLimitHardshutDownThresholdId},
58 mctpRequester, [this](uint8_t rc, std::vector<int32_t> thresholds) {
59 std::vector<thresholds::Threshold> tLimitThresholds{};
60 if (rc == 0)
61 {
62 tLimitThresholds = {
63 thresholds::Threshold{thresholds::Level::WARNING,
64 thresholds::Direction::LOW,
65 static_cast<double>(thresholds[0])},
66 thresholds::Threshold{thresholds::Level::CRITICAL,
67 thresholds::Direction::LOW,
68 static_cast<double>(thresholds[1])},
69 thresholds::Threshold{thresholds::Level::HARDSHUTDOWN,
70 thresholds::Direction::LOW,
71 static_cast<double>(thresholds[2])}};
72 }
73
74 tLimitSensor = std::make_shared<NvidiaGpuTempSensor>(
75 conn, mctpRequester, name + "_TEMP_1", path, eid,
76 gpuTLimitSensorId, objectServer, std::move(tLimitThresholds));
77 });
Harshit Agherafa2a5b92025-05-22 11:35:39 +053078
Harshit Agherac8dab722025-05-08 15:57:42 +053079 powerSensor = std::make_shared<NvidiaGpuPowerSensor>(
80 conn, mctpRequester, name + "_Power_0", path, eid, gpuPowerSensorId,
81 objectServer, std::vector<thresholds::Threshold>{});
82
Harshit Aghera128c91d2025-05-27 14:20:24 +053083 energySensor = std::make_shared<NvidiaGpuEnergySensor>(
84 conn, mctpRequester, name + "_Energy_0", path, eid, gpuEnergySensorId,
85 objectServer, std::vector<thresholds::Threshold>{});
86
Harshit Agherab55847f2025-05-27 14:53:56 +053087 voltageSensor = std::make_shared<NvidiaGpuVoltageSensor>(
88 conn, mctpRequester, name + "_Voltage_0", path, eid, gpuVoltageSensorId,
89 objectServer, std::vector<thresholds::Threshold>{});
90
Harshit Agherafa2a5b92025-05-22 11:35:39 +053091 lg2::info("Added GPU {NAME} Sensors with chassis path: {PATH}.", "NAME",
92 name, "PATH", path);
93
94 read();
95}
96
97void GpuDevice::read()
98{
99 tempSensor->update();
Harshit Agherac20108d2025-05-07 16:20:16 +0530100 if (tLimitSensor)
101 {
102 tLimitSensor->update();
103 }
Harshit Agherac8dab722025-05-08 15:57:42 +0530104 powerSensor->update();
Harshit Aghera128c91d2025-05-27 14:20:24 +0530105 energySensor->update();
Harshit Agherab55847f2025-05-27 14:53:56 +0530106 voltageSensor->update();
Harshit Agherafa2a5b92025-05-22 11:35:39 +0530107
108 waitTimer.expires_after(std::chrono::milliseconds(sensorPollMs));
109 waitTimer.async_wait([this](const boost::system::error_code& ec) {
110 if (ec)
111 {
112 return;
113 }
114 read();
115 });
116}