1 /*
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 "Inventory.hpp"
10 #include "NvidiaDeviceDiscovery.hpp"
11 #include "NvidiaGpuSensor.hpp"
12 #include "Thresholds.hpp"
13 #include "Utils.hpp"
14
15 #include <bits/basic_string.h>
16
17 #include <MctpRequester.hpp>
18 #include <NvidiaGpuEnergySensor.hpp>
19 #include <NvidiaGpuMctpVdm.hpp>
20 #include <NvidiaGpuPowerSensor.hpp>
21 #include <NvidiaGpuThresholds.hpp>
22 #include <NvidiaGpuVoltageSensor.hpp>
23 #include <boost/asio/io_context.hpp>
24 #include <phosphor-logging/lg2.hpp>
25 #include <sdbusplus/asio/connection.hpp>
26 #include <sdbusplus/asio/object_server.hpp>
27
28 #include <chrono>
29 #include <cstdint>
30 #include <functional>
31 #include <memory>
32 #include <string>
33 #include <utility>
34 #include <vector>
35
GpuDevice(const SensorConfigs & configs,const std::string & name,const std::string & path,const std::shared_ptr<sdbusplus::asio::connection> & conn,uint8_t eid,boost::asio::io_context & io,mctp::MctpRequester & mctpRequester,sdbusplus::asio::object_server & objectServer)36 GpuDevice::GpuDevice(const SensorConfigs& configs, const std::string& name,
37 const std::string& path,
38 const std::shared_ptr<sdbusplus::asio::connection>& conn,
39 uint8_t eid, boost::asio::io_context& io,
40 mctp::MctpRequester& mctpRequester,
41 sdbusplus::asio::object_server& objectServer) :
42 eid(eid), sensorPollMs(std::chrono::milliseconds{configs.pollRate}),
43 waitTimer(io, std::chrono::steady_clock::duration(0)),
44 mctpRequester(mctpRequester), conn(conn), objectServer(objectServer),
45 configs(configs), name(escapeName(name)), path(path)
46 {
47 inventory = std::make_shared<Inventory>(
48 conn, objectServer, name, mctpRequester,
49 gpu::DeviceIdentification::DEVICE_GPU, eid, io);
50 makeSensors();
51 }
52
makeSensors()53 void GpuDevice::makeSensors()
54 {
55 tempSensor = std::make_shared<NvidiaGpuTempSensor>(
56 conn, mctpRequester, name + "_TEMP_0", path, eid, gpuTempSensorId,
57 objectServer, std::vector<thresholds::Threshold>{});
58
59 readThermalParameters(
60 eid,
61 std::vector<gpuThresholdId>{gpuTLimitWarnringThresholdId,
62 gpuTLimitCriticalThresholdId,
63 gpuTLimitHardshutDownThresholdId},
64 mctpRequester,
65 std::bind_front(&GpuDevice::processTLimitThresholds, this));
66
67 dramTempSensor = std::make_shared<NvidiaGpuTempSensor>(
68 conn, mctpRequester, name + "_DRAM_0_TEMP_0", path, eid,
69 gpuDramTempSensorId, objectServer,
70 std::vector<thresholds::Threshold>{thresholds::Threshold{
71 thresholds::Level::CRITICAL, thresholds::Direction::HIGH, 95.0}});
72
73 powerSensor = std::make_shared<NvidiaGpuPowerSensor>(
74 conn, mctpRequester, name + "_Power_0", path, eid, gpuPowerSensorId,
75 objectServer, std::vector<thresholds::Threshold>{});
76
77 energySensor = std::make_shared<NvidiaGpuEnergySensor>(
78 conn, mctpRequester, name + "_Energy_0", path, eid, gpuEnergySensorId,
79 objectServer, std::vector<thresholds::Threshold>{});
80
81 voltageSensor = std::make_shared<NvidiaGpuVoltageSensor>(
82 conn, mctpRequester, name + "_Voltage_0", path, eid, gpuVoltageSensorId,
83 objectServer, std::vector<thresholds::Threshold>{});
84
85 lg2::info("Added GPU {NAME} Sensors with chassis path: {PATH}.", "NAME",
86 name, "PATH", path);
87
88 read();
89 }
90
processTLimitThresholds(uint8_t rc,const std::vector<int32_t> & thresholds)91 void GpuDevice::processTLimitThresholds(uint8_t rc,
92 const std::vector<int32_t>& thresholds)
93 {
94 std::vector<thresholds::Threshold> tLimitThresholds{};
95 if (rc == 0)
96 {
97 tLimitThresholds = {
98 thresholds::Threshold{thresholds::Level::WARNING,
99 thresholds::Direction::LOW,
100 static_cast<double>(thresholds[0])},
101 thresholds::Threshold{thresholds::Level::CRITICAL,
102 thresholds::Direction::LOW,
103 static_cast<double>(thresholds[1])},
104 thresholds::Threshold{thresholds::Level::HARDSHUTDOWN,
105 thresholds::Direction::LOW,
106 static_cast<double>(thresholds[2])}};
107 }
108
109 tLimitSensor = std::make_shared<NvidiaGpuTempSensor>(
110 conn, mctpRequester, name + "_TEMP_1", path, eid, gpuTLimitSensorId,
111 objectServer, std::move(tLimitThresholds));
112 }
113
read()114 void GpuDevice::read()
115 {
116 tempSensor->update();
117 if (tLimitSensor)
118 {
119 tLimitSensor->update();
120 }
121 dramTempSensor->update();
122 powerSensor->update();
123 energySensor->update();
124 voltageSensor->update();
125
126 waitTimer.expires_after(std::chrono::milliseconds(sensorPollMs));
127 waitTimer.async_wait([this](const boost::system::error_code& ec) {
128 if (ec)
129 {
130 return;
131 }
132 read();
133 });
134 }
135