Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1 | /* |
| 2 | // Copyright (c) 2019 Intel Corporation |
| 3 | // |
| 4 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | // you may not use this file except in compliance with the License. |
| 6 | // You may obtain a copy of the License at |
| 7 | // |
| 8 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | // |
| 10 | // Unless required by applicable law or agreed to in writing, software |
| 11 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | // See the License for the specific language governing permissions and |
| 14 | // limitations under the License. |
| 15 | */ |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 16 | #include <peci.h> |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 17 | #include <systemd/sd-journal.h> |
| 18 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 19 | #include <bitset> |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 20 | #include <boost/asio/posix/stream_descriptor.hpp> |
| 21 | #include <gpiod.hpp> |
| 22 | #include <iostream> |
| 23 | #include <sdbusplus/asio/object_server.hpp> |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 24 | #include <variant> |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 25 | |
| 26 | namespace host_error_monitor |
| 27 | { |
| 28 | static boost::asio::io_service io; |
| 29 | static std::shared_ptr<sdbusplus::asio::connection> conn; |
| 30 | |
| 31 | static bool hostOff = true; |
| 32 | |
| 33 | const static constexpr size_t caterrTimeoutMs = 2000; |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 34 | const static constexpr size_t errTimeoutMs = 90000; |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 35 | const static constexpr size_t smiTimeoutMs = 90000; |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 36 | const static constexpr size_t crashdumpTimeoutS = 300; |
| 37 | |
| 38 | // Timers |
| 39 | // Timer for CATERR asserted |
| 40 | static boost::asio::steady_timer caterrAssertTimer(io); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 41 | // Timer for ERR0 asserted |
| 42 | static boost::asio::steady_timer err0AssertTimer(io); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 43 | // Timer for ERR1 asserted |
| 44 | static boost::asio::steady_timer err1AssertTimer(io); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 45 | // Timer for ERR2 asserted |
| 46 | static boost::asio::steady_timer err2AssertTimer(io); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 47 | // Timer for SMI asserted |
| 48 | static boost::asio::steady_timer smiAssertTimer(io); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 49 | |
| 50 | // GPIO Lines and Event Descriptors |
| 51 | static gpiod::line caterrLine; |
| 52 | static boost::asio::posix::stream_descriptor caterrEvent(io); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 53 | static gpiod::line err0Line; |
| 54 | static boost::asio::posix::stream_descriptor err0Event(io); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 55 | static gpiod::line err1Line; |
| 56 | static boost::asio::posix::stream_descriptor err1Event(io); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 57 | static gpiod::line err2Line; |
| 58 | static boost::asio::posix::stream_descriptor err2Event(io); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 59 | static gpiod::line smiLine; |
| 60 | static boost::asio::posix::stream_descriptor smiEvent(io); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 61 | static gpiod::line cpu1ThermtripLine; |
| 62 | static boost::asio::posix::stream_descriptor cpu1ThermtripEvent(io); |
| 63 | static gpiod::line cpu2ThermtripLine; |
| 64 | static boost::asio::posix::stream_descriptor cpu2ThermtripEvent(io); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 65 | static gpiod::line cpu1VRHotLine; |
| 66 | static boost::asio::posix::stream_descriptor cpu1VRHotEvent(io); |
| 67 | static gpiod::line cpu2VRHotLine; |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 68 | static boost::asio::posix::stream_descriptor cpu1MemABCDVRHotEvent(io); |
| 69 | static gpiod::line cpu1MemEFGHVRHotLine; |
| 70 | static boost::asio::posix::stream_descriptor cpu1MemEFGHVRHotEvent(io); |
| 71 | static gpiod::line cpu2MemABCDVRHotLine; |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 72 | static boost::asio::posix::stream_descriptor cpu2VRHotEvent(io); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 73 | static gpiod::line cpu1MemABCDVRHotLine; |
| 74 | static boost::asio::posix::stream_descriptor cpu2MemABCDVRHotEvent(io); |
| 75 | static gpiod::line cpu2MemEFGHVRHotLine; |
| 76 | static boost::asio::posix::stream_descriptor cpu2MemEFGHVRHotEvent(io); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 77 | //---------------------------------- |
| 78 | // PCH_BMC_THERMTRIP function related definition |
| 79 | //---------------------------------- |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 80 | static gpiod::line pchThermtripLine; |
| 81 | static boost::asio::posix::stream_descriptor pchThermtripEvent(io); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 82 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 83 | static void cpuIERRLog() |
| 84 | { |
| 85 | sd_journal_send("MESSAGE=HostError: IERR", "PRIORITY=%i", LOG_INFO, |
| 86 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 87 | "REDFISH_MESSAGE_ARGS=%s", "IERR", NULL); |
| 88 | } |
| 89 | |
| 90 | static void cpuIERRLog(const int cpuNum) |
| 91 | { |
| 92 | std::string msg = "IERR on CPU " + std::to_string(cpuNum + 1); |
| 93 | |
| 94 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 95 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 96 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 97 | } |
| 98 | |
| 99 | static void cpuIERRLog(const int cpuNum, const std::string& type) |
| 100 | { |
| 101 | std::string msg = type + " IERR on CPU " + std::to_string(cpuNum + 1); |
| 102 | |
| 103 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 104 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 105 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 106 | } |
| 107 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 108 | static void cpuERRXLog(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 109 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 110 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout"; |
| 111 | |
| 112 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 113 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 114 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 115 | } |
| 116 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 117 | static void cpuERRXLog(const int errPin, const int cpuNum) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 118 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 119 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout on CPU " + |
| 120 | std::to_string(cpuNum + 1); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 121 | |
| 122 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 123 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 124 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 125 | } |
| 126 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 127 | static void smiTimeoutLog() |
| 128 | { |
| 129 | sd_journal_send("MESSAGE=HostError: SMI Timeout", "PRIORITY=%i", LOG_INFO, |
| 130 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 131 | "REDFISH_MESSAGE_ARGS=%s", "SMI Timeout", NULL); |
| 132 | } |
| 133 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 134 | static void cpuThermTripLog(const int cpuNum) |
| 135 | { |
| 136 | std::string msg = "CPU " + std::to_string(cpuNum) + " thermal trip"; |
| 137 | |
| 138 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 139 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 140 | "OpenBMC.0.1.CPUThermalTrip", "REDFISH_MESSAGE_ARGS=%d", |
| 141 | cpuNum, NULL); |
| 142 | } |
| 143 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 144 | static void cpuVRHotLog(const std::string& vr) |
| 145 | { |
| 146 | std::string msg = vr + " Voltage Regulator Overheated."; |
| 147 | |
| 148 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 149 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 150 | "OpenBMC.0.1.VoltageRegulatorOverheated", |
| 151 | "REDFISH_MESSAGE_ARGS=%s", vr.c_str(), NULL); |
| 152 | } |
| 153 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 154 | static void ssbThermTripLog() |
| 155 | { |
| 156 | sd_journal_send("MESSAGE=HostError: SSB thermal trip", "PRIORITY=%i", |
| 157 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 158 | "OpenBMC.0.1.SsbThermalTrip", NULL); |
| 159 | } |
| 160 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 161 | static void initializeErrorState(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 162 | static void initializeHostState() |
| 163 | { |
| 164 | conn->async_method_call( |
| 165 | [](boost::system::error_code ec, |
| 166 | const std::variant<std::string>& property) { |
| 167 | if (ec) |
| 168 | { |
| 169 | return; |
| 170 | } |
| 171 | const std::string* state = std::get_if<std::string>(&property); |
| 172 | if (state == nullptr) |
| 173 | { |
| 174 | std::cerr << "Unable to read host state value\n"; |
| 175 | return; |
| 176 | } |
| 177 | hostOff = *state == "xyz.openbmc_project.State.Host.HostState.Off"; |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 178 | // If the system is on, initialize the error state |
| 179 | if (!hostOff) |
| 180 | { |
| 181 | initializeErrorState(); |
| 182 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 183 | }, |
| 184 | "xyz.openbmc_project.State.Host", "/xyz/openbmc_project/state/host0", |
| 185 | "org.freedesktop.DBus.Properties", "Get", |
| 186 | "xyz.openbmc_project.State.Host", "CurrentHostState"); |
| 187 | } |
| 188 | |
| 189 | static std::shared_ptr<sdbusplus::bus::match::match> startHostStateMonitor() |
| 190 | { |
| 191 | return std::make_shared<sdbusplus::bus::match::match>( |
| 192 | *conn, |
| 193 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 194 | "member='PropertiesChanged',arg0namespace='xyz.openbmc_project.State." |
| 195 | "Host'", |
| 196 | [](sdbusplus::message::message& msg) { |
| 197 | std::string interfaceName; |
| 198 | boost::container::flat_map<std::string, std::variant<std::string>> |
| 199 | propertiesChanged; |
| 200 | std::string state; |
| 201 | try |
| 202 | { |
| 203 | msg.read(interfaceName, propertiesChanged); |
| 204 | state = |
| 205 | std::get<std::string>(propertiesChanged.begin()->second); |
| 206 | } |
| 207 | catch (std::exception& e) |
| 208 | { |
| 209 | std::cerr << "Unable to read host state\n"; |
| 210 | return; |
| 211 | } |
| 212 | hostOff = state == "xyz.openbmc_project.State.Host.HostState.Off"; |
| 213 | |
| 214 | // No host events should fire while off, so cancel any pending |
| 215 | // timers |
| 216 | if (hostOff) |
| 217 | { |
| 218 | caterrAssertTimer.cancel(); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 219 | err0AssertTimer.cancel(); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 220 | err1AssertTimer.cancel(); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 221 | err2AssertTimer.cancel(); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 222 | smiAssertTimer.cancel(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 223 | } |
| 224 | }); |
| 225 | } |
| 226 | |
| 227 | static bool requestGPIOEvents( |
| 228 | const std::string& name, const std::function<void()>& handler, |
| 229 | gpiod::line& gpioLine, |
| 230 | boost::asio::posix::stream_descriptor& gpioEventDescriptor) |
| 231 | { |
| 232 | // Find the GPIO line |
| 233 | gpioLine = gpiod::find_line(name); |
| 234 | if (!gpioLine) |
| 235 | { |
| 236 | std::cerr << "Failed to find the " << name << " line\n"; |
| 237 | return false; |
| 238 | } |
| 239 | |
| 240 | try |
| 241 | { |
| 242 | gpioLine.request( |
| 243 | {"host-error-monitor", gpiod::line_request::EVENT_BOTH_EDGES}); |
| 244 | } |
| 245 | catch (std::exception&) |
| 246 | { |
| 247 | std::cerr << "Failed to request events for " << name << "\n"; |
| 248 | return false; |
| 249 | } |
| 250 | |
| 251 | int gpioLineFd = gpioLine.event_get_fd(); |
| 252 | if (gpioLineFd < 0) |
| 253 | { |
| 254 | std::cerr << "Failed to get " << name << " fd\n"; |
| 255 | return false; |
| 256 | } |
| 257 | |
| 258 | gpioEventDescriptor.assign(gpioLineFd); |
| 259 | |
| 260 | gpioEventDescriptor.async_wait( |
| 261 | boost::asio::posix::stream_descriptor::wait_read, |
| 262 | [&name, handler](const boost::system::error_code ec) { |
| 263 | if (ec) |
| 264 | { |
| 265 | std::cerr << name << " fd handler error: " << ec.message() |
| 266 | << "\n"; |
| 267 | return; |
| 268 | } |
| 269 | handler(); |
| 270 | }); |
| 271 | return true; |
| 272 | } |
| 273 | |
| 274 | static void startPowerCycle() |
| 275 | { |
| 276 | conn->async_method_call( |
| 277 | [](boost::system::error_code ec) { |
| 278 | if (ec) |
| 279 | { |
| 280 | std::cerr << "failed to set Chassis State\n"; |
| 281 | } |
| 282 | }, |
| 283 | "xyz.openbmc_project.State.Chassis", |
| 284 | "/xyz/openbmc_project/state/chassis0", |
| 285 | "org.freedesktop.DBus.Properties", "Set", |
| 286 | "xyz.openbmc_project.State.Chassis", "RequestedPowerTransition", |
| 287 | std::variant<std::string>{ |
| 288 | "xyz.openbmc_project.State.Chassis.Transition.PowerCycle"}); |
| 289 | } |
| 290 | |
| 291 | static void startCrashdumpAndRecovery(bool recoverSystem) |
| 292 | { |
| 293 | std::cout << "Starting crashdump\n"; |
| 294 | static std::shared_ptr<sdbusplus::bus::match::match> crashdumpCompleteMatch; |
| 295 | static boost::asio::steady_timer crashdumpTimer(io); |
| 296 | |
| 297 | crashdumpCompleteMatch = std::make_shared<sdbusplus::bus::match::match>( |
| 298 | *conn, |
| 299 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 300 | "member='PropertiesChanged',arg0namespace='com.intel.crashdump'", |
| 301 | [recoverSystem](sdbusplus::message::message& msg) { |
| 302 | crashdumpTimer.cancel(); |
| 303 | std::cout << "Crashdump completed\n"; |
| 304 | if (recoverSystem) |
| 305 | { |
| 306 | std::cout << "Recovering the system\n"; |
| 307 | startPowerCycle(); |
| 308 | } |
| 309 | crashdumpCompleteMatch.reset(); |
| 310 | }); |
| 311 | |
| 312 | crashdumpTimer.expires_after(std::chrono::seconds(crashdumpTimeoutS)); |
| 313 | crashdumpTimer.async_wait([](const boost::system::error_code ec) { |
| 314 | if (ec) |
| 315 | { |
| 316 | // operation_aborted is expected if timer is canceled |
| 317 | if (ec != boost::asio::error::operation_aborted) |
| 318 | { |
| 319 | std::cerr << "Crashdump async_wait failed: " << ec.message() |
| 320 | << "\n"; |
| 321 | } |
| 322 | std::cout << "Crashdump timer canceled\n"; |
| 323 | return; |
| 324 | } |
| 325 | std::cerr << "Crashdump failed to complete before timeout\n"; |
| 326 | crashdumpCompleteMatch.reset(); |
| 327 | }); |
| 328 | |
| 329 | conn->async_method_call( |
| 330 | [](boost::system::error_code ec) { |
| 331 | if (ec) |
| 332 | { |
| 333 | std::cerr << "failed to start Crashdump\n"; |
| 334 | crashdumpTimer.cancel(); |
| 335 | crashdumpCompleteMatch.reset(); |
| 336 | } |
| 337 | }, |
| 338 | "com.intel.crashdump", "/com/intel/crashdump", |
| 339 | "com.intel.crashdump.Stored", "GenerateStoredLog"); |
| 340 | } |
| 341 | |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 342 | static void incrementCPUErrorCount(int cpuNum) |
| 343 | { |
| 344 | std::string propertyName = "ErrorCountCPU" + std::to_string(cpuNum + 1); |
| 345 | |
| 346 | // Get the current count |
| 347 | conn->async_method_call( |
| 348 | [propertyName](boost::system::error_code ec, |
| 349 | const std::variant<uint8_t>& property) { |
| 350 | if (ec) |
| 351 | { |
| 352 | std::cerr << "Failed to read " << propertyName << ": " |
| 353 | << ec.message() << "\n"; |
| 354 | return; |
| 355 | } |
| 356 | const uint8_t* errorCountVariant = std::get_if<uint8_t>(&property); |
| 357 | if (errorCountVariant == nullptr) |
| 358 | { |
| 359 | std::cerr << propertyName << " invalid\n"; |
| 360 | return; |
| 361 | } |
| 362 | uint8_t errorCount = *errorCountVariant; |
| 363 | if (errorCount == std::numeric_limits<uint8_t>::max()) |
| 364 | { |
| 365 | std::cerr << "Maximum error count reached\n"; |
| 366 | return; |
| 367 | } |
| 368 | // Increment the count |
| 369 | errorCount++; |
| 370 | conn->async_method_call( |
| 371 | [propertyName](boost::system::error_code ec) { |
| 372 | if (ec) |
| 373 | { |
| 374 | std::cerr << "Failed to set " << propertyName << ": " |
| 375 | << ec.message() << "\n"; |
| 376 | } |
| 377 | }, |
| 378 | "xyz.openbmc_project.Settings", |
| 379 | "/xyz/openbmc_project/control/processor_error_config", |
| 380 | "org.freedesktop.DBus.Properties", "Set", |
| 381 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName, |
| 382 | std::variant<uint8_t>{errorCount}); |
| 383 | }, |
| 384 | "xyz.openbmc_project.Settings", |
| 385 | "/xyz/openbmc_project/control/processor_error_config", |
| 386 | "org.freedesktop.DBus.Properties", "Get", |
| 387 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName); |
| 388 | } |
| 389 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 390 | static bool checkIERRCPUs() |
| 391 | { |
| 392 | bool cpuIERRFound = false; |
| 393 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 394 | cpu++, addr++) |
| 395 | { |
| 396 | uint8_t cc = 0; |
| 397 | CPUModel model{}; |
| 398 | uint8_t stepping = 0; |
| 399 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 400 | { |
| 401 | std::cerr << "Cannot get CPUID!\n"; |
| 402 | continue; |
| 403 | } |
| 404 | |
| 405 | switch (model) |
| 406 | { |
| 407 | case skx: |
| 408 | { |
| 409 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 410 | // that caused the IERR |
| 411 | uint32_t mcaErrSrcLog = 0; |
| 412 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 413 | &cc) != PECI_CC_SUCCESS) |
| 414 | { |
| 415 | continue; |
| 416 | } |
| 417 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 418 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 419 | { |
| 420 | // TODO: Light the CPU fault LED? |
| 421 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 422 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 423 | // Next check if it's a CPU/VR mismatch by reading the |
| 424 | // IA32_MC4_STATUS MSR (0x411) |
| 425 | uint64_t mc4Status = 0; |
| 426 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 427 | PECI_CC_SUCCESS) |
| 428 | { |
| 429 | continue; |
| 430 | } |
| 431 | // Check MSEC bits 31:24 for |
| 432 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 433 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 434 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 435 | if ((mc4Status & (0x40 << 24)) || |
| 436 | (mc4Status & (0x42 << 24)) || |
| 437 | (mc4Status & (0x43 << 24))) |
| 438 | { |
| 439 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 440 | continue; |
| 441 | } |
| 442 | |
| 443 | // Next check if it's a Core FIVR fault by looking for a |
| 444 | // non-zero value of CORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 445 | // 80h) |
| 446 | uint32_t coreFIVRErrLog = 0; |
| 447 | if (peci_RdPCIConfigLocal( |
| 448 | addr, 1, 30, 2, 0x80, sizeof(uint32_t), |
| 449 | (uint8_t*)&coreFIVRErrLog, &cc) != PECI_CC_SUCCESS) |
| 450 | { |
| 451 | continue; |
| 452 | } |
| 453 | if (coreFIVRErrLog) |
| 454 | { |
| 455 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 456 | continue; |
| 457 | } |
| 458 | |
| 459 | // Next check if it's an Uncore FIVR fault by looking for a |
| 460 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 461 | // 84h) |
| 462 | uint32_t uncoreFIVRErrLog = 0; |
| 463 | if (peci_RdPCIConfigLocal(addr, 1, 30, 2, 0x84, |
| 464 | sizeof(uint32_t), |
| 465 | (uint8_t*)&uncoreFIVRErrLog, |
| 466 | &cc) != PECI_CC_SUCCESS) |
| 467 | { |
| 468 | continue; |
| 469 | } |
| 470 | if (uncoreFIVRErrLog) |
| 471 | { |
| 472 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 473 | continue; |
| 474 | } |
| 475 | |
| 476 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 477 | // both zero, but MSEC bits 31:24 have either |
| 478 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 479 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 480 | // uncore FIVR fault |
| 481 | if (!coreFIVRErrLog && !uncoreFIVRErrLog && |
| 482 | ((mc4Status & (0x51 << 24)) || |
| 483 | (mc4Status & (0x52 << 24)))) |
| 484 | { |
| 485 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 486 | continue; |
| 487 | } |
| 488 | cpuIERRLog(cpu); |
| 489 | } |
| 490 | break; |
| 491 | } |
| 492 | case icx: |
| 493 | { |
| 494 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 495 | // that caused the IERR |
| 496 | uint32_t mcaErrSrcLog = 0; |
| 497 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 498 | &cc) != PECI_CC_SUCCESS) |
| 499 | { |
| 500 | continue; |
| 501 | } |
| 502 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 503 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 504 | { |
| 505 | // TODO: Light the CPU fault LED? |
| 506 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 507 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 508 | // Next check if it's a CPU/VR mismatch by reading the |
| 509 | // IA32_MC4_STATUS MSR (0x411) |
| 510 | uint64_t mc4Status = 0; |
| 511 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 512 | PECI_CC_SUCCESS) |
| 513 | { |
| 514 | continue; |
| 515 | } |
| 516 | // TODO: Update MSEC/MSCOD_31_24 check |
| 517 | // Check MSEC bits 31:24 for |
| 518 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 519 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 520 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 521 | if ((mc4Status & (0x40 << 24)) || |
| 522 | (mc4Status & (0x42 << 24)) || |
| 523 | (mc4Status & (0x43 << 24))) |
| 524 | { |
| 525 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 526 | continue; |
| 527 | } |
| 528 | |
| 529 | // Next check if it's a Core FIVR fault by looking for a |
| 530 | // non-zero value of CORE_FIVR_ERR_LOG (B(31) D30 F2 offsets |
| 531 | // C0h and C4h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 532 | uint32_t coreFIVRErrLog0 = 0; |
| 533 | uint32_t coreFIVRErrLog1 = 0; |
| 534 | if (peci_RdEndPointConfigPciLocal( |
| 535 | addr, 0, 14, 30, 2, 0xC0, sizeof(uint32_t), |
| 536 | (uint8_t*)&coreFIVRErrLog0, &cc) != PECI_CC_SUCCESS) |
| 537 | { |
| 538 | continue; |
| 539 | } |
| 540 | if (peci_RdEndPointConfigPciLocal( |
| 541 | addr, 0, 14, 30, 2, 0xC4, sizeof(uint32_t), |
| 542 | (uint8_t*)&coreFIVRErrLog1, &cc) != PECI_CC_SUCCESS) |
| 543 | { |
| 544 | continue; |
| 545 | } |
| 546 | if (coreFIVRErrLog0 || coreFIVRErrLog1) |
| 547 | { |
| 548 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 549 | continue; |
| 550 | } |
| 551 | |
| 552 | // Next check if it's an Uncore FIVR fault by looking for a |
| 553 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(31) D30 F2 |
| 554 | // offset 84h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 555 | uint32_t uncoreFIVRErrLog = 0; |
| 556 | if (peci_RdEndPointConfigPciLocal( |
| 557 | addr, 0, 14, 30, 2, 0x84, sizeof(uint32_t), |
| 558 | (uint8_t*)&uncoreFIVRErrLog, |
| 559 | &cc) != PECI_CC_SUCCESS) |
| 560 | { |
| 561 | continue; |
| 562 | } |
| 563 | if (uncoreFIVRErrLog) |
| 564 | { |
| 565 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 566 | continue; |
| 567 | } |
| 568 | |
| 569 | // TODO: Update MSEC/MSCOD_31_24 check |
| 570 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 571 | // both zero, but MSEC bits 31:24 have either |
| 572 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 573 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 574 | // uncore FIVR fault |
| 575 | if (!coreFIVRErrLog0 && !coreFIVRErrLog1 && |
| 576 | !uncoreFIVRErrLog && |
| 577 | ((mc4Status & (0x51 << 24)) || |
| 578 | (mc4Status & (0x52 << 24)))) |
| 579 | { |
| 580 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 581 | continue; |
| 582 | } |
| 583 | cpuIERRLog(cpu); |
| 584 | } |
| 585 | break; |
| 586 | } |
| 587 | } |
| 588 | } |
| 589 | return cpuIERRFound; |
| 590 | } |
| 591 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 592 | static void caterrAssertHandler() |
| 593 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 594 | caterrAssertTimer.expires_after(std::chrono::milliseconds(caterrTimeoutMs)); |
| 595 | caterrAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 596 | if (ec) |
| 597 | { |
| 598 | // operation_aborted is expected if timer is canceled |
| 599 | // before completion. |
| 600 | if (ec != boost::asio::error::operation_aborted) |
| 601 | { |
| 602 | std::cerr << "caterr timeout async_wait failed: " |
| 603 | << ec.message() << "\n"; |
| 604 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 605 | return; |
| 606 | } |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 607 | std::cerr << "CATERR asserted for " << std::to_string(caterrTimeoutMs) |
| 608 | << " ms\n"; |
| 609 | if (!checkIERRCPUs()) |
| 610 | { |
| 611 | cpuIERRLog(); |
| 612 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 613 | conn->async_method_call( |
| 614 | [](boost::system::error_code ec, |
| 615 | const std::variant<bool>& property) { |
| 616 | if (ec) |
| 617 | { |
| 618 | return; |
| 619 | } |
| 620 | const bool* reset = std::get_if<bool>(&property); |
| 621 | if (reset == nullptr) |
| 622 | { |
| 623 | std::cerr << "Unable to read reset on CATERR value\n"; |
| 624 | return; |
| 625 | } |
| 626 | startCrashdumpAndRecovery(*reset); |
| 627 | }, |
| 628 | "xyz.openbmc_project.Settings", |
| 629 | "/xyz/openbmc_project/control/processor_error_config", |
| 630 | "org.freedesktop.DBus.Properties", "Get", |
| 631 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnCATERR"); |
| 632 | }); |
| 633 | } |
| 634 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 635 | static void caterrHandler() |
| 636 | { |
| 637 | if (!hostOff) |
| 638 | { |
| 639 | gpiod::line_event gpioLineEvent = caterrLine.event_read(); |
| 640 | |
| 641 | bool caterr = |
| 642 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 643 | if (caterr) |
| 644 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 645 | caterrAssertHandler(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 646 | } |
| 647 | else |
| 648 | { |
| 649 | caterrAssertTimer.cancel(); |
| 650 | } |
| 651 | } |
| 652 | caterrEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 653 | [](const boost::system::error_code ec) { |
| 654 | if (ec) |
| 655 | { |
| 656 | std::cerr << "caterr handler error: " |
| 657 | << ec.message() << "\n"; |
| 658 | return; |
| 659 | } |
| 660 | caterrHandler(); |
| 661 | }); |
| 662 | } |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 663 | |
| 664 | static void cpu1ThermtripHandler() |
| 665 | { |
| 666 | if (!hostOff) |
| 667 | { |
| 668 | gpiod::line_event gpioLineEvent = cpu1ThermtripLine.event_read(); |
| 669 | |
| 670 | bool cpu1Thermtrip = |
| 671 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 672 | if (cpu1Thermtrip) |
| 673 | { |
| 674 | cpuThermTripLog(1); |
| 675 | } |
| 676 | } |
| 677 | cpu1ThermtripEvent.async_wait( |
| 678 | boost::asio::posix::stream_descriptor::wait_read, |
| 679 | [](const boost::system::error_code ec) { |
| 680 | if (ec) |
| 681 | { |
| 682 | std::cerr << "CPU 1 Thermtrip handler error: " << ec.message() |
| 683 | << "\n"; |
| 684 | return; |
| 685 | } |
| 686 | cpu1ThermtripHandler(); |
| 687 | }); |
| 688 | } |
| 689 | |
| 690 | static void cpu2ThermtripHandler() |
| 691 | { |
| 692 | if (!hostOff) |
| 693 | { |
| 694 | gpiod::line_event gpioLineEvent = cpu2ThermtripLine.event_read(); |
| 695 | |
| 696 | bool cpu2Thermtrip = |
| 697 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 698 | if (cpu2Thermtrip) |
| 699 | { |
| 700 | cpuThermTripLog(2); |
| 701 | } |
| 702 | } |
| 703 | cpu2ThermtripEvent.async_wait( |
| 704 | boost::asio::posix::stream_descriptor::wait_read, |
| 705 | [](const boost::system::error_code ec) { |
| 706 | if (ec) |
| 707 | { |
| 708 | std::cerr << "CPU 2 Thermtrip handler error: " << ec.message() |
| 709 | << "\n"; |
| 710 | return; |
| 711 | } |
| 712 | cpu2ThermtripHandler(); |
| 713 | }); |
| 714 | } |
| 715 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 716 | static void cpu1VRHotHandler() |
| 717 | { |
| 718 | if (!hostOff) |
| 719 | { |
| 720 | gpiod::line_event gpioLineEvent = cpu1VRHotLine.event_read(); |
| 721 | |
| 722 | bool cpu1VRHot = |
| 723 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 724 | if (cpu1VRHot) |
| 725 | { |
| 726 | cpuVRHotLog("CPU 1"); |
| 727 | } |
| 728 | } |
| 729 | cpu1VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 730 | [](const boost::system::error_code ec) { |
| 731 | if (ec) |
| 732 | { |
| 733 | std::cerr << "CPU 1 VRHot handler error: " |
| 734 | << ec.message() << "\n"; |
| 735 | return; |
| 736 | } |
| 737 | cpu1VRHotHandler(); |
| 738 | }); |
| 739 | } |
| 740 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 741 | static void cpu1MemABCDVRHotHandler() |
| 742 | { |
| 743 | if (!hostOff) |
| 744 | { |
| 745 | gpiod::line_event gpioLineEvent = cpu1MemABCDVRHotLine.event_read(); |
| 746 | |
| 747 | bool cpu1MemABCDVRHot = |
| 748 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 749 | if (cpu1MemABCDVRHot) |
| 750 | { |
| 751 | cpuVRHotLog("CPU 1 Memory ABCD"); |
| 752 | } |
| 753 | } |
| 754 | cpu1MemABCDVRHotEvent.async_wait( |
| 755 | boost::asio::posix::stream_descriptor::wait_read, |
| 756 | [](const boost::system::error_code ec) { |
| 757 | if (ec) |
| 758 | { |
| 759 | std::cerr << "CPU 1 Memory ABCD VRHot handler error: " |
| 760 | << ec.message() << "\n"; |
| 761 | return; |
| 762 | } |
| 763 | cpu1MemABCDVRHotHandler(); |
| 764 | }); |
| 765 | } |
| 766 | |
| 767 | static void cpu1MemEFGHVRHotHandler() |
| 768 | { |
| 769 | if (!hostOff) |
| 770 | { |
| 771 | gpiod::line_event gpioLineEvent = cpu1MemEFGHVRHotLine.event_read(); |
| 772 | |
| 773 | bool cpu1MemEFGHVRHot = |
| 774 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 775 | if (cpu1MemEFGHVRHot) |
| 776 | { |
| 777 | cpuVRHotLog("CPU 1 Memory EFGH"); |
| 778 | } |
| 779 | } |
| 780 | cpu1MemEFGHVRHotEvent.async_wait( |
| 781 | boost::asio::posix::stream_descriptor::wait_read, |
| 782 | [](const boost::system::error_code ec) { |
| 783 | if (ec) |
| 784 | { |
| 785 | std::cerr << "CPU 1 Memory EFGH VRHot handler error: " |
| 786 | << ec.message() << "\n"; |
| 787 | return; |
| 788 | } |
| 789 | cpu1MemEFGHVRHotHandler(); |
| 790 | }); |
| 791 | } |
| 792 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 793 | static void cpu2VRHotHandler() |
| 794 | { |
| 795 | if (!hostOff) |
| 796 | { |
| 797 | gpiod::line_event gpioLineEvent = cpu2VRHotLine.event_read(); |
| 798 | |
| 799 | bool cpu2VRHot = |
| 800 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 801 | if (cpu2VRHot) |
| 802 | { |
| 803 | cpuVRHotLog("CPU 2"); |
| 804 | } |
| 805 | } |
| 806 | cpu2VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 807 | [](const boost::system::error_code ec) { |
| 808 | if (ec) |
| 809 | { |
| 810 | std::cerr << "CPU 2 VRHot handler error: " |
| 811 | << ec.message() << "\n"; |
| 812 | return; |
| 813 | } |
| 814 | cpu2VRHotHandler(); |
| 815 | }); |
| 816 | } |
| 817 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 818 | static void cpu2MemABCDVRHotHandler() |
| 819 | { |
| 820 | if (!hostOff) |
| 821 | { |
| 822 | gpiod::line_event gpioLineEvent = cpu2MemABCDVRHotLine.event_read(); |
| 823 | |
| 824 | bool cpu2MemABCDVRHot = |
| 825 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 826 | if (cpu2MemABCDVRHot) |
| 827 | { |
| 828 | cpuVRHotLog("CPU 2 Memory ABCD"); |
| 829 | } |
| 830 | } |
| 831 | cpu2MemABCDVRHotEvent.async_wait( |
| 832 | boost::asio::posix::stream_descriptor::wait_read, |
| 833 | [](const boost::system::error_code ec) { |
| 834 | if (ec) |
| 835 | { |
| 836 | std::cerr << "CPU 2 Memory ABCD VRHot handler error: " |
| 837 | << ec.message() << "\n"; |
| 838 | return; |
| 839 | } |
| 840 | cpu2MemABCDVRHotHandler(); |
| 841 | }); |
| 842 | } |
| 843 | |
| 844 | static void cpu2MemEFGHVRHotHandler() |
| 845 | { |
| 846 | if (!hostOff) |
| 847 | { |
| 848 | gpiod::line_event gpioLineEvent = cpu2MemEFGHVRHotLine.event_read(); |
| 849 | |
| 850 | bool cpu2MemEFGHVRHot = |
| 851 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 852 | if (cpu2MemEFGHVRHot) |
| 853 | { |
| 854 | cpuVRHotLog("CPU 2 Memory EFGH"); |
| 855 | } |
| 856 | } |
| 857 | cpu2MemEFGHVRHotEvent.async_wait( |
| 858 | boost::asio::posix::stream_descriptor::wait_read, |
| 859 | [](const boost::system::error_code ec) { |
| 860 | if (ec) |
| 861 | { |
| 862 | std::cerr << "CPU 2 Memory EFGH VRHot handler error: " |
| 863 | << ec.message() << "\n"; |
| 864 | return; |
| 865 | } |
| 866 | cpu2MemEFGHVRHotHandler(); |
| 867 | }); |
| 868 | } |
| 869 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 870 | static void pchThermtripHandler() |
| 871 | { |
| 872 | if (!hostOff) |
| 873 | { |
| 874 | gpiod::line_event gpioLineEvent = pchThermtripLine.event_read(); |
| 875 | |
| 876 | bool pchThermtrip = |
| 877 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 878 | if (pchThermtrip) |
| 879 | { |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 880 | ssbThermTripLog(); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 881 | } |
| 882 | } |
| 883 | pchThermtripEvent.async_wait( |
| 884 | boost::asio::posix::stream_descriptor::wait_read, |
| 885 | [](const boost::system::error_code ec) { |
| 886 | if (ec) |
| 887 | { |
| 888 | std::cerr << "PCH Thermal trip handler error: " << ec.message() |
| 889 | << "\n"; |
| 890 | return; |
| 891 | } |
| 892 | pchThermtripHandler(); |
| 893 | }); |
| 894 | } |
| 895 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 896 | static std::bitset<MAX_CPUS> checkERRPinCPUs(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 897 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 898 | int errPinSts = (1 << errPin); |
| 899 | std::bitset<MAX_CPUS> errPinCPUs = 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 900 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 901 | cpu++, addr++) |
| 902 | { |
| 903 | if (peci_Ping(addr) == PECI_CC_SUCCESS) |
| 904 | { |
| 905 | uint8_t cc = 0; |
| 906 | CPUModel model{}; |
| 907 | uint8_t stepping = 0; |
| 908 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 909 | { |
| 910 | std::cerr << "Cannot get CPUID!\n"; |
| 911 | continue; |
| 912 | } |
| 913 | |
| 914 | switch (model) |
| 915 | { |
| 916 | case skx: |
| 917 | { |
| 918 | // Check the ERRPINSTS to see if this is the CPU that caused |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 919 | // the ERRx (B(0) D8 F0 offset 210h) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 920 | uint32_t errpinsts = 0; |
| 921 | if (peci_RdPCIConfigLocal( |
| 922 | addr, 0, 8, 0, 0x210, sizeof(uint32_t), |
| 923 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 924 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 925 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 926 | } |
| 927 | break; |
| 928 | } |
| 929 | case icx: |
| 930 | { |
| 931 | // Check the ERRPINSTS to see if this is the CPU that caused |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 932 | // the ERRx (B(30) D0 F3 offset 274h) (Note: Bus 30 is |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 933 | // accessed on PECI as bus 13) |
| 934 | uint32_t errpinsts = 0; |
| 935 | if (peci_RdEndPointConfigPciLocal( |
| 936 | addr, 0, 13, 0, 3, 0x274, sizeof(uint32_t), |
| 937 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 938 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 939 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 940 | } |
| 941 | break; |
| 942 | } |
| 943 | } |
| 944 | } |
| 945 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 946 | return errPinCPUs; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 947 | } |
| 948 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 949 | static void errXAssertHandler(const int errPin, |
| 950 | boost::asio::steady_timer& errXAssertTimer) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 951 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 952 | // ERRx status is not guaranteed through the timeout, so save which |
| 953 | // CPUs have it asserted |
| 954 | std::bitset<MAX_CPUS> errPinCPUs = checkERRPinCPUs(errPin); |
| 955 | errXAssertTimer.expires_after(std::chrono::milliseconds(errTimeoutMs)); |
| 956 | errXAssertTimer.async_wait([errPin, errPinCPUs]( |
| 957 | const boost::system::error_code ec) { |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 958 | if (ec) |
| 959 | { |
| 960 | // operation_aborted is expected if timer is canceled before |
| 961 | // completion. |
| 962 | if (ec != boost::asio::error::operation_aborted) |
| 963 | { |
| 964 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 965 | << "\n"; |
| 966 | } |
| 967 | return; |
| 968 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 969 | std::cerr << "ERR" << std::to_string(errPin) << " asserted for " |
| 970 | << std::to_string(errTimeoutMs) << " ms\n"; |
| 971 | if (errPinCPUs.count()) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 972 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 973 | for (int i = 0; i < errPinCPUs.size(); i++) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 974 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 975 | if (errPinCPUs[i]) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 976 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 977 | cpuERRXLog(errPin, i); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 978 | } |
| 979 | } |
| 980 | } |
| 981 | else |
| 982 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 983 | cpuERRXLog(errPin); |
| 984 | } |
| 985 | }); |
| 986 | } |
| 987 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 988 | static void err0AssertHandler() |
| 989 | { |
| 990 | // Handle the standard ERR0 detection and logging |
| 991 | const static constexpr int err0 = 0; |
| 992 | errXAssertHandler(err0, err0AssertTimer); |
| 993 | } |
| 994 | |
| 995 | static void err0Handler() |
| 996 | { |
| 997 | if (!hostOff) |
| 998 | { |
| 999 | gpiod::line_event gpioLineEvent = err0Line.event_read(); |
| 1000 | |
| 1001 | bool err0 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1002 | if (err0) |
| 1003 | { |
| 1004 | err0AssertHandler(); |
| 1005 | } |
| 1006 | else |
| 1007 | { |
| 1008 | err0AssertTimer.cancel(); |
| 1009 | } |
| 1010 | } |
| 1011 | err0Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1012 | [](const boost::system::error_code ec) { |
| 1013 | if (ec) |
| 1014 | { |
| 1015 | std::cerr |
| 1016 | << "err0 handler error: " << ec.message() |
| 1017 | << "\n"; |
| 1018 | return; |
| 1019 | } |
| 1020 | err0Handler(); |
| 1021 | }); |
| 1022 | } |
| 1023 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1024 | static void err1AssertHandler() |
| 1025 | { |
| 1026 | // Handle the standard ERR1 detection and logging |
| 1027 | const static constexpr int err1 = 1; |
| 1028 | errXAssertHandler(err1, err1AssertTimer); |
| 1029 | } |
| 1030 | |
| 1031 | static void err1Handler() |
| 1032 | { |
| 1033 | if (!hostOff) |
| 1034 | { |
| 1035 | gpiod::line_event gpioLineEvent = err1Line.event_read(); |
| 1036 | |
| 1037 | bool err1 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1038 | if (err1) |
| 1039 | { |
| 1040 | err1AssertHandler(); |
| 1041 | } |
| 1042 | else |
| 1043 | { |
| 1044 | err1AssertTimer.cancel(); |
| 1045 | } |
| 1046 | } |
| 1047 | err1Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1048 | [](const boost::system::error_code ec) { |
| 1049 | if (ec) |
| 1050 | { |
| 1051 | std::cerr |
| 1052 | << "err1 handler error: " << ec.message() |
| 1053 | << "\n"; |
| 1054 | return; |
| 1055 | } |
| 1056 | err1Handler(); |
| 1057 | }); |
| 1058 | } |
| 1059 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1060 | static void err2AssertHandler() |
| 1061 | { |
| 1062 | // Handle the standard ERR2 detection and logging |
| 1063 | const static constexpr int err2 = 2; |
| 1064 | errXAssertHandler(err2, err2AssertTimer); |
| 1065 | // Also handle reset for ERR2 |
| 1066 | err2AssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1067 | if (ec) |
| 1068 | { |
| 1069 | // operation_aborted is expected if timer is canceled before |
| 1070 | // completion. |
| 1071 | if (ec != boost::asio::error::operation_aborted) |
| 1072 | { |
| 1073 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 1074 | << "\n"; |
| 1075 | } |
| 1076 | return; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1077 | } |
| 1078 | conn->async_method_call( |
| 1079 | [](boost::system::error_code ec, |
| 1080 | const std::variant<bool>& property) { |
| 1081 | if (ec) |
| 1082 | { |
| 1083 | return; |
| 1084 | } |
| 1085 | const bool* reset = std::get_if<bool>(&property); |
| 1086 | if (reset == nullptr) |
| 1087 | { |
| 1088 | std::cerr << "Unable to read reset on ERR2 value\n"; |
| 1089 | return; |
| 1090 | } |
| 1091 | startCrashdumpAndRecovery(*reset); |
| 1092 | }, |
| 1093 | "xyz.openbmc_project.Settings", |
| 1094 | "/xyz/openbmc_project/control/processor_error_config", |
| 1095 | "org.freedesktop.DBus.Properties", "Get", |
| 1096 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnERR2"); |
| 1097 | }); |
| 1098 | } |
| 1099 | |
| 1100 | static void err2Handler() |
| 1101 | { |
| 1102 | if (!hostOff) |
| 1103 | { |
| 1104 | gpiod::line_event gpioLineEvent = err2Line.event_read(); |
| 1105 | |
| 1106 | bool err2 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1107 | if (err2) |
| 1108 | { |
| 1109 | err2AssertHandler(); |
| 1110 | } |
| 1111 | else |
| 1112 | { |
| 1113 | err2AssertTimer.cancel(); |
| 1114 | } |
| 1115 | } |
| 1116 | err2Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1117 | [](const boost::system::error_code ec) { |
| 1118 | if (ec) |
| 1119 | { |
| 1120 | std::cerr |
| 1121 | << "err2 handler error: " << ec.message() |
| 1122 | << "\n"; |
| 1123 | return; |
| 1124 | } |
| 1125 | err2Handler(); |
| 1126 | }); |
| 1127 | } |
| 1128 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1129 | static void smiAssertHandler() |
| 1130 | { |
| 1131 | smiAssertTimer.expires_after(std::chrono::milliseconds(smiTimeoutMs)); |
| 1132 | smiAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1133 | if (ec) |
| 1134 | { |
| 1135 | // operation_aborted is expected if timer is canceled before |
| 1136 | // completion. |
| 1137 | if (ec != boost::asio::error::operation_aborted) |
| 1138 | { |
| 1139 | std::cerr << "smi timeout async_wait failed: " << ec.message() |
| 1140 | << "\n"; |
| 1141 | } |
| 1142 | return; |
| 1143 | } |
| 1144 | std::cerr << "SMI asserted for " << std::to_string(smiTimeoutMs) |
| 1145 | << " ms\n"; |
| 1146 | smiTimeoutLog(); |
| 1147 | conn->async_method_call( |
| 1148 | [](boost::system::error_code ec, |
| 1149 | const std::variant<bool>& property) { |
| 1150 | if (ec) |
| 1151 | { |
| 1152 | return; |
| 1153 | } |
| 1154 | const bool* reset = std::get_if<bool>(&property); |
| 1155 | if (reset == nullptr) |
| 1156 | { |
| 1157 | std::cerr << "Unable to read reset on SMI value\n"; |
| 1158 | return; |
| 1159 | } |
| 1160 | startCrashdumpAndRecovery(*reset); |
| 1161 | }, |
| 1162 | "xyz.openbmc_project.Settings", |
| 1163 | "/xyz/openbmc_project/control/bmc_reset_disables", |
| 1164 | "org.freedesktop.DBus.Properties", "Get", |
| 1165 | "xyz.openbmc_project.Control.ResetDisables", "ResetOnSMI"); |
| 1166 | }); |
| 1167 | } |
| 1168 | |
| 1169 | static void smiHandler() |
| 1170 | { |
| 1171 | if (!hostOff) |
| 1172 | { |
| 1173 | gpiod::line_event gpioLineEvent = smiLine.event_read(); |
| 1174 | |
| 1175 | bool smi = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1176 | if (smi) |
| 1177 | { |
| 1178 | smiAssertHandler(); |
| 1179 | } |
| 1180 | else |
| 1181 | { |
| 1182 | smiAssertTimer.cancel(); |
| 1183 | } |
| 1184 | } |
| 1185 | smiEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1186 | [](const boost::system::error_code ec) { |
| 1187 | if (ec) |
| 1188 | { |
| 1189 | std::cerr |
| 1190 | << "smi handler error: " << ec.message() |
| 1191 | << "\n"; |
| 1192 | return; |
| 1193 | } |
| 1194 | smiHandler(); |
| 1195 | }); |
| 1196 | } |
| 1197 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1198 | static void initializeErrorState() |
| 1199 | { |
| 1200 | // Handle CPU_CATERR if it's asserted now |
| 1201 | if (caterrLine.get_value() == 0) |
| 1202 | { |
| 1203 | caterrAssertHandler(); |
| 1204 | } |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1205 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1206 | // Handle CPU_ERR0 if it's asserted now |
| 1207 | if (err0Line.get_value() == 0) |
| 1208 | { |
| 1209 | err0AssertHandler(); |
| 1210 | } |
| 1211 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1212 | // Handle CPU_ERR1 if it's asserted now |
| 1213 | if (err1Line.get_value() == 0) |
| 1214 | { |
| 1215 | err1AssertHandler(); |
| 1216 | } |
| 1217 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1218 | // Handle CPU_ERR2 if it's asserted now |
| 1219 | if (err2Line.get_value() == 0) |
| 1220 | { |
| 1221 | err2AssertHandler(); |
| 1222 | } |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1223 | |
| 1224 | // Handle SMI if it's asserted now |
| 1225 | if (smiLine.get_value() == 0) |
| 1226 | { |
| 1227 | smiAssertHandler(); |
| 1228 | } |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 1229 | |
| 1230 | // Handle PCH_BMC_THERMTRIP if it's asserted now |
| 1231 | if (pchThermtripLine.get_value() == 0) |
| 1232 | { |
| 1233 | ssbThermTripLog(); |
| 1234 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1235 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1236 | } // namespace host_error_monitor |
| 1237 | |
| 1238 | int main(int argc, char* argv[]) |
| 1239 | { |
| 1240 | // setup connection to dbus |
| 1241 | host_error_monitor::conn = |
| 1242 | std::make_shared<sdbusplus::asio::connection>(host_error_monitor::io); |
| 1243 | |
| 1244 | // Host Error Monitor Object |
| 1245 | host_error_monitor::conn->request_name( |
| 1246 | "xyz.openbmc_project.HostErrorMonitor"); |
| 1247 | sdbusplus::asio::object_server server = |
| 1248 | sdbusplus::asio::object_server(host_error_monitor::conn); |
| 1249 | |
| 1250 | // Start tracking host state |
| 1251 | std::shared_ptr<sdbusplus::bus::match::match> hostStateMonitor = |
| 1252 | host_error_monitor::startHostStateMonitor(); |
| 1253 | |
| 1254 | // Initialize the host state |
| 1255 | host_error_monitor::initializeHostState(); |
| 1256 | |
| 1257 | // Request CPU_CATERR GPIO events |
| 1258 | if (!host_error_monitor::requestGPIOEvents( |
| 1259 | "CPU_CATERR", host_error_monitor::caterrHandler, |
| 1260 | host_error_monitor::caterrLine, host_error_monitor::caterrEvent)) |
| 1261 | { |
| 1262 | return -1; |
| 1263 | } |
| 1264 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1265 | // Request CPU_ERR0 GPIO events |
| 1266 | if (!host_error_monitor::requestGPIOEvents( |
| 1267 | "CPU_ERR0", host_error_monitor::err0Handler, |
| 1268 | host_error_monitor::err0Line, host_error_monitor::err0Event)) |
| 1269 | { |
| 1270 | return -1; |
| 1271 | } |
| 1272 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1273 | // Request CPU_ERR1 GPIO events |
| 1274 | if (!host_error_monitor::requestGPIOEvents( |
| 1275 | "CPU_ERR1", host_error_monitor::err1Handler, |
| 1276 | host_error_monitor::err1Line, host_error_monitor::err1Event)) |
| 1277 | { |
| 1278 | return -1; |
| 1279 | } |
| 1280 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1281 | // Request CPU_ERR2 GPIO events |
| 1282 | if (!host_error_monitor::requestGPIOEvents( |
| 1283 | "CPU_ERR2", host_error_monitor::err2Handler, |
| 1284 | host_error_monitor::err2Line, host_error_monitor::err2Event)) |
| 1285 | { |
| 1286 | return -1; |
| 1287 | } |
| 1288 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1289 | // Request SMI GPIO events |
| 1290 | if (!host_error_monitor::requestGPIOEvents( |
| 1291 | "SMI", host_error_monitor::smiHandler, host_error_monitor::smiLine, |
| 1292 | host_error_monitor::smiEvent)) |
| 1293 | { |
| 1294 | return -1; |
| 1295 | } |
| 1296 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 1297 | // Request CPU1_THERMTRIP GPIO events |
| 1298 | if (!host_error_monitor::requestGPIOEvents( |
| 1299 | "CPU1_THERMTRIP", host_error_monitor::cpu1ThermtripHandler, |
| 1300 | host_error_monitor::cpu1ThermtripLine, |
| 1301 | host_error_monitor::cpu1ThermtripEvent)) |
| 1302 | { |
| 1303 | return -1; |
| 1304 | } |
| 1305 | |
| 1306 | // Request CPU2_THERMTRIP GPIO events |
| 1307 | if (!host_error_monitor::requestGPIOEvents( |
| 1308 | "CPU2_THERMTRIP", host_error_monitor::cpu2ThermtripHandler, |
| 1309 | host_error_monitor::cpu2ThermtripLine, |
| 1310 | host_error_monitor::cpu2ThermtripEvent)) |
| 1311 | { |
| 1312 | return -1; |
| 1313 | } |
| 1314 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1315 | // Request CPU1_VRHOT GPIO events |
| 1316 | if (!host_error_monitor::requestGPIOEvents( |
| 1317 | "CPU1_VRHOT", host_error_monitor::cpu1VRHotHandler, |
| 1318 | host_error_monitor::cpu1VRHotLine, |
| 1319 | host_error_monitor::cpu1VRHotEvent)) |
| 1320 | { |
| 1321 | return -1; |
| 1322 | } |
| 1323 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 1324 | // Request CPU1_MEM_ABCD_VRHOT GPIO events |
| 1325 | if (!host_error_monitor::requestGPIOEvents( |
| 1326 | "CPU1_MEM_ABCD_VRHOT", host_error_monitor::cpu1MemABCDVRHotHandler, |
| 1327 | host_error_monitor::cpu1MemABCDVRHotLine, |
| 1328 | host_error_monitor::cpu1MemABCDVRHotEvent)) |
| 1329 | { |
| 1330 | return -1; |
| 1331 | } |
| 1332 | |
| 1333 | // Request CPU1_MEM_EFGH_VRHOT GPIO events |
| 1334 | if (!host_error_monitor::requestGPIOEvents( |
| 1335 | "CPU1_MEM_EFGH_VRHOT", host_error_monitor::cpu1MemEFGHVRHotHandler, |
| 1336 | host_error_monitor::cpu1MemEFGHVRHotLine, |
| 1337 | host_error_monitor::cpu1MemEFGHVRHotEvent)) |
| 1338 | { |
| 1339 | return -1; |
| 1340 | } |
| 1341 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1342 | // Request CPU2_VRHOT GPIO events |
| 1343 | if (!host_error_monitor::requestGPIOEvents( |
| 1344 | "CPU2_VRHOT", host_error_monitor::cpu2VRHotHandler, |
| 1345 | host_error_monitor::cpu2VRHotLine, |
| 1346 | host_error_monitor::cpu2VRHotEvent)) |
| 1347 | { |
| 1348 | return -1; |
| 1349 | } |
| 1350 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame^] | 1351 | // Request CPU2_MEM_ABCD_VRHOT GPIO events |
| 1352 | if (!host_error_monitor::requestGPIOEvents( |
| 1353 | "CPU2_MEM_ABCD_VRHOT", host_error_monitor::cpu2MemABCDVRHotHandler, |
| 1354 | host_error_monitor::cpu2MemABCDVRHotLine, |
| 1355 | host_error_monitor::cpu2MemABCDVRHotEvent)) |
| 1356 | { |
| 1357 | return -1; |
| 1358 | } |
| 1359 | |
| 1360 | // Request CPU2_MEM_EFGH_VRHOT GPIO events |
| 1361 | if (!host_error_monitor::requestGPIOEvents( |
| 1362 | "CPU2_MEM_EFGH_VRHOT", host_error_monitor::cpu2MemEFGHVRHotHandler, |
| 1363 | host_error_monitor::cpu2MemEFGHVRHotLine, |
| 1364 | host_error_monitor::cpu2MemEFGHVRHotEvent)) |
| 1365 | { |
| 1366 | return -1; |
| 1367 | } |
| 1368 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1369 | // Request PCH_BMC_THERMTRIP GPIO events |
| 1370 | if (!host_error_monitor::requestGPIOEvents( |
| 1371 | "PCH_BMC_THERMTRIP", host_error_monitor::pchThermtripHandler, |
| 1372 | host_error_monitor::pchThermtripLine, |
| 1373 | host_error_monitor::pchThermtripEvent)) |
| 1374 | { |
| 1375 | return -1; |
| 1376 | } |
| 1377 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1378 | host_error_monitor::io.run(); |
| 1379 | |
| 1380 | return 0; |
| 1381 | } |