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); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 61 | //---------------------------------- |
| 62 | // PCH_BMC_THERMTRIP function related definition |
| 63 | //---------------------------------- |
| 64 | // GPIO Lines and Event Descriptors |
| 65 | static gpiod::line pchThermtripLine; |
| 66 | static boost::asio::posix::stream_descriptor pchThermtripEvent(io); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 67 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 68 | static void cpuIERRLog() |
| 69 | { |
| 70 | sd_journal_send("MESSAGE=HostError: IERR", "PRIORITY=%i", LOG_INFO, |
| 71 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 72 | "REDFISH_MESSAGE_ARGS=%s", "IERR", NULL); |
| 73 | } |
| 74 | |
| 75 | static void cpuIERRLog(const int cpuNum) |
| 76 | { |
| 77 | std::string msg = "IERR on CPU " + std::to_string(cpuNum + 1); |
| 78 | |
| 79 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 80 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 81 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 82 | } |
| 83 | |
| 84 | static void cpuIERRLog(const int cpuNum, const std::string& type) |
| 85 | { |
| 86 | std::string msg = type + " IERR on CPU " + std::to_string(cpuNum + 1); |
| 87 | |
| 88 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 89 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 90 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 91 | } |
| 92 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 93 | static void cpuERRXLog(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 94 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 95 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout"; |
| 96 | |
| 97 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 98 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 99 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 100 | } |
| 101 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 102 | static void cpuERRXLog(const int errPin, const int cpuNum) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 103 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 104 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout on CPU " + |
| 105 | std::to_string(cpuNum + 1); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 106 | |
| 107 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 108 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 109 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 110 | } |
| 111 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 112 | static void smiTimeoutLog() |
| 113 | { |
| 114 | sd_journal_send("MESSAGE=HostError: SMI Timeout", "PRIORITY=%i", LOG_INFO, |
| 115 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 116 | "REDFISH_MESSAGE_ARGS=%s", "SMI Timeout", NULL); |
| 117 | } |
| 118 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 119 | static void ssbThermTripLog() |
| 120 | { |
| 121 | sd_journal_send("MESSAGE=HostError: SSB thermal trip", "PRIORITY=%i", |
| 122 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 123 | "OpenBMC.0.1.SsbThermalTrip", NULL); |
| 124 | } |
| 125 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 126 | static void initializeErrorState(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 127 | static void initializeHostState() |
| 128 | { |
| 129 | conn->async_method_call( |
| 130 | [](boost::system::error_code ec, |
| 131 | const std::variant<std::string>& property) { |
| 132 | if (ec) |
| 133 | { |
| 134 | return; |
| 135 | } |
| 136 | const std::string* state = std::get_if<std::string>(&property); |
| 137 | if (state == nullptr) |
| 138 | { |
| 139 | std::cerr << "Unable to read host state value\n"; |
| 140 | return; |
| 141 | } |
| 142 | hostOff = *state == "xyz.openbmc_project.State.Host.HostState.Off"; |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 143 | // If the system is on, initialize the error state |
| 144 | if (!hostOff) |
| 145 | { |
| 146 | initializeErrorState(); |
| 147 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 148 | }, |
| 149 | "xyz.openbmc_project.State.Host", "/xyz/openbmc_project/state/host0", |
| 150 | "org.freedesktop.DBus.Properties", "Get", |
| 151 | "xyz.openbmc_project.State.Host", "CurrentHostState"); |
| 152 | } |
| 153 | |
| 154 | static std::shared_ptr<sdbusplus::bus::match::match> startHostStateMonitor() |
| 155 | { |
| 156 | return std::make_shared<sdbusplus::bus::match::match>( |
| 157 | *conn, |
| 158 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 159 | "member='PropertiesChanged',arg0namespace='xyz.openbmc_project.State." |
| 160 | "Host'", |
| 161 | [](sdbusplus::message::message& msg) { |
| 162 | std::string interfaceName; |
| 163 | boost::container::flat_map<std::string, std::variant<std::string>> |
| 164 | propertiesChanged; |
| 165 | std::string state; |
| 166 | try |
| 167 | { |
| 168 | msg.read(interfaceName, propertiesChanged); |
| 169 | state = |
| 170 | std::get<std::string>(propertiesChanged.begin()->second); |
| 171 | } |
| 172 | catch (std::exception& e) |
| 173 | { |
| 174 | std::cerr << "Unable to read host state\n"; |
| 175 | return; |
| 176 | } |
| 177 | hostOff = state == "xyz.openbmc_project.State.Host.HostState.Off"; |
| 178 | |
| 179 | // No host events should fire while off, so cancel any pending |
| 180 | // timers |
| 181 | if (hostOff) |
| 182 | { |
| 183 | caterrAssertTimer.cancel(); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 184 | err0AssertTimer.cancel(); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame^] | 185 | err1AssertTimer.cancel(); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 186 | err2AssertTimer.cancel(); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 187 | smiAssertTimer.cancel(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 188 | } |
| 189 | }); |
| 190 | } |
| 191 | |
| 192 | static bool requestGPIOEvents( |
| 193 | const std::string& name, const std::function<void()>& handler, |
| 194 | gpiod::line& gpioLine, |
| 195 | boost::asio::posix::stream_descriptor& gpioEventDescriptor) |
| 196 | { |
| 197 | // Find the GPIO line |
| 198 | gpioLine = gpiod::find_line(name); |
| 199 | if (!gpioLine) |
| 200 | { |
| 201 | std::cerr << "Failed to find the " << name << " line\n"; |
| 202 | return false; |
| 203 | } |
| 204 | |
| 205 | try |
| 206 | { |
| 207 | gpioLine.request( |
| 208 | {"host-error-monitor", gpiod::line_request::EVENT_BOTH_EDGES}); |
| 209 | } |
| 210 | catch (std::exception&) |
| 211 | { |
| 212 | std::cerr << "Failed to request events for " << name << "\n"; |
| 213 | return false; |
| 214 | } |
| 215 | |
| 216 | int gpioLineFd = gpioLine.event_get_fd(); |
| 217 | if (gpioLineFd < 0) |
| 218 | { |
| 219 | std::cerr << "Failed to get " << name << " fd\n"; |
| 220 | return false; |
| 221 | } |
| 222 | |
| 223 | gpioEventDescriptor.assign(gpioLineFd); |
| 224 | |
| 225 | gpioEventDescriptor.async_wait( |
| 226 | boost::asio::posix::stream_descriptor::wait_read, |
| 227 | [&name, handler](const boost::system::error_code ec) { |
| 228 | if (ec) |
| 229 | { |
| 230 | std::cerr << name << " fd handler error: " << ec.message() |
| 231 | << "\n"; |
| 232 | return; |
| 233 | } |
| 234 | handler(); |
| 235 | }); |
| 236 | return true; |
| 237 | } |
| 238 | |
| 239 | static void startPowerCycle() |
| 240 | { |
| 241 | conn->async_method_call( |
| 242 | [](boost::system::error_code ec) { |
| 243 | if (ec) |
| 244 | { |
| 245 | std::cerr << "failed to set Chassis State\n"; |
| 246 | } |
| 247 | }, |
| 248 | "xyz.openbmc_project.State.Chassis", |
| 249 | "/xyz/openbmc_project/state/chassis0", |
| 250 | "org.freedesktop.DBus.Properties", "Set", |
| 251 | "xyz.openbmc_project.State.Chassis", "RequestedPowerTransition", |
| 252 | std::variant<std::string>{ |
| 253 | "xyz.openbmc_project.State.Chassis.Transition.PowerCycle"}); |
| 254 | } |
| 255 | |
| 256 | static void startCrashdumpAndRecovery(bool recoverSystem) |
| 257 | { |
| 258 | std::cout << "Starting crashdump\n"; |
| 259 | static std::shared_ptr<sdbusplus::bus::match::match> crashdumpCompleteMatch; |
| 260 | static boost::asio::steady_timer crashdumpTimer(io); |
| 261 | |
| 262 | crashdumpCompleteMatch = std::make_shared<sdbusplus::bus::match::match>( |
| 263 | *conn, |
| 264 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 265 | "member='PropertiesChanged',arg0namespace='com.intel.crashdump'", |
| 266 | [recoverSystem](sdbusplus::message::message& msg) { |
| 267 | crashdumpTimer.cancel(); |
| 268 | std::cout << "Crashdump completed\n"; |
| 269 | if (recoverSystem) |
| 270 | { |
| 271 | std::cout << "Recovering the system\n"; |
| 272 | startPowerCycle(); |
| 273 | } |
| 274 | crashdumpCompleteMatch.reset(); |
| 275 | }); |
| 276 | |
| 277 | crashdumpTimer.expires_after(std::chrono::seconds(crashdumpTimeoutS)); |
| 278 | crashdumpTimer.async_wait([](const boost::system::error_code ec) { |
| 279 | if (ec) |
| 280 | { |
| 281 | // operation_aborted is expected if timer is canceled |
| 282 | if (ec != boost::asio::error::operation_aborted) |
| 283 | { |
| 284 | std::cerr << "Crashdump async_wait failed: " << ec.message() |
| 285 | << "\n"; |
| 286 | } |
| 287 | std::cout << "Crashdump timer canceled\n"; |
| 288 | return; |
| 289 | } |
| 290 | std::cerr << "Crashdump failed to complete before timeout\n"; |
| 291 | crashdumpCompleteMatch.reset(); |
| 292 | }); |
| 293 | |
| 294 | conn->async_method_call( |
| 295 | [](boost::system::error_code ec) { |
| 296 | if (ec) |
| 297 | { |
| 298 | std::cerr << "failed to start Crashdump\n"; |
| 299 | crashdumpTimer.cancel(); |
| 300 | crashdumpCompleteMatch.reset(); |
| 301 | } |
| 302 | }, |
| 303 | "com.intel.crashdump", "/com/intel/crashdump", |
| 304 | "com.intel.crashdump.Stored", "GenerateStoredLog"); |
| 305 | } |
| 306 | |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 307 | static void incrementCPUErrorCount(int cpuNum) |
| 308 | { |
| 309 | std::string propertyName = "ErrorCountCPU" + std::to_string(cpuNum + 1); |
| 310 | |
| 311 | // Get the current count |
| 312 | conn->async_method_call( |
| 313 | [propertyName](boost::system::error_code ec, |
| 314 | const std::variant<uint8_t>& property) { |
| 315 | if (ec) |
| 316 | { |
| 317 | std::cerr << "Failed to read " << propertyName << ": " |
| 318 | << ec.message() << "\n"; |
| 319 | return; |
| 320 | } |
| 321 | const uint8_t* errorCountVariant = std::get_if<uint8_t>(&property); |
| 322 | if (errorCountVariant == nullptr) |
| 323 | { |
| 324 | std::cerr << propertyName << " invalid\n"; |
| 325 | return; |
| 326 | } |
| 327 | uint8_t errorCount = *errorCountVariant; |
| 328 | if (errorCount == std::numeric_limits<uint8_t>::max()) |
| 329 | { |
| 330 | std::cerr << "Maximum error count reached\n"; |
| 331 | return; |
| 332 | } |
| 333 | // Increment the count |
| 334 | errorCount++; |
| 335 | conn->async_method_call( |
| 336 | [propertyName](boost::system::error_code ec) { |
| 337 | if (ec) |
| 338 | { |
| 339 | std::cerr << "Failed to set " << propertyName << ": " |
| 340 | << ec.message() << "\n"; |
| 341 | } |
| 342 | }, |
| 343 | "xyz.openbmc_project.Settings", |
| 344 | "/xyz/openbmc_project/control/processor_error_config", |
| 345 | "org.freedesktop.DBus.Properties", "Set", |
| 346 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName, |
| 347 | std::variant<uint8_t>{errorCount}); |
| 348 | }, |
| 349 | "xyz.openbmc_project.Settings", |
| 350 | "/xyz/openbmc_project/control/processor_error_config", |
| 351 | "org.freedesktop.DBus.Properties", "Get", |
| 352 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName); |
| 353 | } |
| 354 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 355 | static bool checkIERRCPUs() |
| 356 | { |
| 357 | bool cpuIERRFound = false; |
| 358 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 359 | cpu++, addr++) |
| 360 | { |
| 361 | uint8_t cc = 0; |
| 362 | CPUModel model{}; |
| 363 | uint8_t stepping = 0; |
| 364 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 365 | { |
| 366 | std::cerr << "Cannot get CPUID!\n"; |
| 367 | continue; |
| 368 | } |
| 369 | |
| 370 | switch (model) |
| 371 | { |
| 372 | case skx: |
| 373 | { |
| 374 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 375 | // that caused the IERR |
| 376 | uint32_t mcaErrSrcLog = 0; |
| 377 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 378 | &cc) != PECI_CC_SUCCESS) |
| 379 | { |
| 380 | continue; |
| 381 | } |
| 382 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 383 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 384 | { |
| 385 | // TODO: Light the CPU fault LED? |
| 386 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 387 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 388 | // Next check if it's a CPU/VR mismatch by reading the |
| 389 | // IA32_MC4_STATUS MSR (0x411) |
| 390 | uint64_t mc4Status = 0; |
| 391 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 392 | PECI_CC_SUCCESS) |
| 393 | { |
| 394 | continue; |
| 395 | } |
| 396 | // Check MSEC bits 31:24 for |
| 397 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 398 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 399 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 400 | if ((mc4Status & (0x40 << 24)) || |
| 401 | (mc4Status & (0x42 << 24)) || |
| 402 | (mc4Status & (0x43 << 24))) |
| 403 | { |
| 404 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 405 | continue; |
| 406 | } |
| 407 | |
| 408 | // Next check if it's a Core FIVR fault by looking for a |
| 409 | // non-zero value of CORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 410 | // 80h) |
| 411 | uint32_t coreFIVRErrLog = 0; |
| 412 | if (peci_RdPCIConfigLocal( |
| 413 | addr, 1, 30, 2, 0x80, sizeof(uint32_t), |
| 414 | (uint8_t*)&coreFIVRErrLog, &cc) != PECI_CC_SUCCESS) |
| 415 | { |
| 416 | continue; |
| 417 | } |
| 418 | if (coreFIVRErrLog) |
| 419 | { |
| 420 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 421 | continue; |
| 422 | } |
| 423 | |
| 424 | // Next check if it's an Uncore FIVR fault by looking for a |
| 425 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 426 | // 84h) |
| 427 | uint32_t uncoreFIVRErrLog = 0; |
| 428 | if (peci_RdPCIConfigLocal(addr, 1, 30, 2, 0x84, |
| 429 | sizeof(uint32_t), |
| 430 | (uint8_t*)&uncoreFIVRErrLog, |
| 431 | &cc) != PECI_CC_SUCCESS) |
| 432 | { |
| 433 | continue; |
| 434 | } |
| 435 | if (uncoreFIVRErrLog) |
| 436 | { |
| 437 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 438 | continue; |
| 439 | } |
| 440 | |
| 441 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 442 | // both zero, but MSEC bits 31:24 have either |
| 443 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 444 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 445 | // uncore FIVR fault |
| 446 | if (!coreFIVRErrLog && !uncoreFIVRErrLog && |
| 447 | ((mc4Status & (0x51 << 24)) || |
| 448 | (mc4Status & (0x52 << 24)))) |
| 449 | { |
| 450 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 451 | continue; |
| 452 | } |
| 453 | cpuIERRLog(cpu); |
| 454 | } |
| 455 | break; |
| 456 | } |
| 457 | case icx: |
| 458 | { |
| 459 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 460 | // that caused the IERR |
| 461 | uint32_t mcaErrSrcLog = 0; |
| 462 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 463 | &cc) != PECI_CC_SUCCESS) |
| 464 | { |
| 465 | continue; |
| 466 | } |
| 467 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 468 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 469 | { |
| 470 | // TODO: Light the CPU fault LED? |
| 471 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 472 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 473 | // Next check if it's a CPU/VR mismatch by reading the |
| 474 | // IA32_MC4_STATUS MSR (0x411) |
| 475 | uint64_t mc4Status = 0; |
| 476 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 477 | PECI_CC_SUCCESS) |
| 478 | { |
| 479 | continue; |
| 480 | } |
| 481 | // TODO: Update MSEC/MSCOD_31_24 check |
| 482 | // Check MSEC bits 31:24 for |
| 483 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 484 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 485 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 486 | if ((mc4Status & (0x40 << 24)) || |
| 487 | (mc4Status & (0x42 << 24)) || |
| 488 | (mc4Status & (0x43 << 24))) |
| 489 | { |
| 490 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 491 | continue; |
| 492 | } |
| 493 | |
| 494 | // Next check if it's a Core FIVR fault by looking for a |
| 495 | // non-zero value of CORE_FIVR_ERR_LOG (B(31) D30 F2 offsets |
| 496 | // C0h and C4h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 497 | uint32_t coreFIVRErrLog0 = 0; |
| 498 | uint32_t coreFIVRErrLog1 = 0; |
| 499 | if (peci_RdEndPointConfigPciLocal( |
| 500 | addr, 0, 14, 30, 2, 0xC0, sizeof(uint32_t), |
| 501 | (uint8_t*)&coreFIVRErrLog0, &cc) != PECI_CC_SUCCESS) |
| 502 | { |
| 503 | continue; |
| 504 | } |
| 505 | if (peci_RdEndPointConfigPciLocal( |
| 506 | addr, 0, 14, 30, 2, 0xC4, sizeof(uint32_t), |
| 507 | (uint8_t*)&coreFIVRErrLog1, &cc) != PECI_CC_SUCCESS) |
| 508 | { |
| 509 | continue; |
| 510 | } |
| 511 | if (coreFIVRErrLog0 || coreFIVRErrLog1) |
| 512 | { |
| 513 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 514 | continue; |
| 515 | } |
| 516 | |
| 517 | // Next check if it's an Uncore FIVR fault by looking for a |
| 518 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(31) D30 F2 |
| 519 | // offset 84h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 520 | uint32_t uncoreFIVRErrLog = 0; |
| 521 | if (peci_RdEndPointConfigPciLocal( |
| 522 | addr, 0, 14, 30, 2, 0x84, sizeof(uint32_t), |
| 523 | (uint8_t*)&uncoreFIVRErrLog, |
| 524 | &cc) != PECI_CC_SUCCESS) |
| 525 | { |
| 526 | continue; |
| 527 | } |
| 528 | if (uncoreFIVRErrLog) |
| 529 | { |
| 530 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 531 | continue; |
| 532 | } |
| 533 | |
| 534 | // TODO: Update MSEC/MSCOD_31_24 check |
| 535 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 536 | // both zero, but MSEC bits 31:24 have either |
| 537 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 538 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 539 | // uncore FIVR fault |
| 540 | if (!coreFIVRErrLog0 && !coreFIVRErrLog1 && |
| 541 | !uncoreFIVRErrLog && |
| 542 | ((mc4Status & (0x51 << 24)) || |
| 543 | (mc4Status & (0x52 << 24)))) |
| 544 | { |
| 545 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 546 | continue; |
| 547 | } |
| 548 | cpuIERRLog(cpu); |
| 549 | } |
| 550 | break; |
| 551 | } |
| 552 | } |
| 553 | } |
| 554 | return cpuIERRFound; |
| 555 | } |
| 556 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 557 | static void caterrAssertHandler() |
| 558 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 559 | caterrAssertTimer.expires_after(std::chrono::milliseconds(caterrTimeoutMs)); |
| 560 | caterrAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 561 | if (ec) |
| 562 | { |
| 563 | // operation_aborted is expected if timer is canceled |
| 564 | // before completion. |
| 565 | if (ec != boost::asio::error::operation_aborted) |
| 566 | { |
| 567 | std::cerr << "caterr timeout async_wait failed: " |
| 568 | << ec.message() << "\n"; |
| 569 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 570 | return; |
| 571 | } |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 572 | std::cerr << "CATERR asserted for " << std::to_string(caterrTimeoutMs) |
| 573 | << " ms\n"; |
| 574 | if (!checkIERRCPUs()) |
| 575 | { |
| 576 | cpuIERRLog(); |
| 577 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 578 | conn->async_method_call( |
| 579 | [](boost::system::error_code ec, |
| 580 | const std::variant<bool>& property) { |
| 581 | if (ec) |
| 582 | { |
| 583 | return; |
| 584 | } |
| 585 | const bool* reset = std::get_if<bool>(&property); |
| 586 | if (reset == nullptr) |
| 587 | { |
| 588 | std::cerr << "Unable to read reset on CATERR value\n"; |
| 589 | return; |
| 590 | } |
| 591 | startCrashdumpAndRecovery(*reset); |
| 592 | }, |
| 593 | "xyz.openbmc_project.Settings", |
| 594 | "/xyz/openbmc_project/control/processor_error_config", |
| 595 | "org.freedesktop.DBus.Properties", "Get", |
| 596 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnCATERR"); |
| 597 | }); |
| 598 | } |
| 599 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 600 | static void caterrHandler() |
| 601 | { |
| 602 | if (!hostOff) |
| 603 | { |
| 604 | gpiod::line_event gpioLineEvent = caterrLine.event_read(); |
| 605 | |
| 606 | bool caterr = |
| 607 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 608 | if (caterr) |
| 609 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 610 | caterrAssertHandler(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 611 | } |
| 612 | else |
| 613 | { |
| 614 | caterrAssertTimer.cancel(); |
| 615 | } |
| 616 | } |
| 617 | caterrEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 618 | [](const boost::system::error_code ec) { |
| 619 | if (ec) |
| 620 | { |
| 621 | std::cerr << "caterr handler error: " |
| 622 | << ec.message() << "\n"; |
| 623 | return; |
| 624 | } |
| 625 | caterrHandler(); |
| 626 | }); |
| 627 | } |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 628 | static void pchThermtripHandler() |
| 629 | { |
| 630 | if (!hostOff) |
| 631 | { |
| 632 | gpiod::line_event gpioLineEvent = pchThermtripLine.event_read(); |
| 633 | |
| 634 | bool pchThermtrip = |
| 635 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 636 | if (pchThermtrip) |
| 637 | { |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 638 | ssbThermTripLog(); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 639 | } |
| 640 | } |
| 641 | pchThermtripEvent.async_wait( |
| 642 | boost::asio::posix::stream_descriptor::wait_read, |
| 643 | [](const boost::system::error_code ec) { |
| 644 | if (ec) |
| 645 | { |
| 646 | std::cerr << "PCH Thermal trip handler error: " << ec.message() |
| 647 | << "\n"; |
| 648 | return; |
| 649 | } |
| 650 | pchThermtripHandler(); |
| 651 | }); |
| 652 | } |
| 653 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 654 | static std::bitset<MAX_CPUS> checkERRPinCPUs(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 655 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 656 | int errPinSts = (1 << errPin); |
| 657 | std::bitset<MAX_CPUS> errPinCPUs = 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 658 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 659 | cpu++, addr++) |
| 660 | { |
| 661 | if (peci_Ping(addr) == PECI_CC_SUCCESS) |
| 662 | { |
| 663 | uint8_t cc = 0; |
| 664 | CPUModel model{}; |
| 665 | uint8_t stepping = 0; |
| 666 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 667 | { |
| 668 | std::cerr << "Cannot get CPUID!\n"; |
| 669 | continue; |
| 670 | } |
| 671 | |
| 672 | switch (model) |
| 673 | { |
| 674 | case skx: |
| 675 | { |
| 676 | // 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] | 677 | // the ERRx (B(0) D8 F0 offset 210h) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 678 | uint32_t errpinsts = 0; |
| 679 | if (peci_RdPCIConfigLocal( |
| 680 | addr, 0, 8, 0, 0x210, sizeof(uint32_t), |
| 681 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 682 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 683 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 684 | } |
| 685 | break; |
| 686 | } |
| 687 | case icx: |
| 688 | { |
| 689 | // 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] | 690 | // 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] | 691 | // accessed on PECI as bus 13) |
| 692 | uint32_t errpinsts = 0; |
| 693 | if (peci_RdEndPointConfigPciLocal( |
| 694 | addr, 0, 13, 0, 3, 0x274, sizeof(uint32_t), |
| 695 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 696 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 697 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 698 | } |
| 699 | break; |
| 700 | } |
| 701 | } |
| 702 | } |
| 703 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 704 | return errPinCPUs; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 705 | } |
| 706 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 707 | static void errXAssertHandler(const int errPin, |
| 708 | boost::asio::steady_timer& errXAssertTimer) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 709 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 710 | // ERRx status is not guaranteed through the timeout, so save which |
| 711 | // CPUs have it asserted |
| 712 | std::bitset<MAX_CPUS> errPinCPUs = checkERRPinCPUs(errPin); |
| 713 | errXAssertTimer.expires_after(std::chrono::milliseconds(errTimeoutMs)); |
| 714 | errXAssertTimer.async_wait([errPin, errPinCPUs]( |
| 715 | const boost::system::error_code ec) { |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 716 | if (ec) |
| 717 | { |
| 718 | // operation_aborted is expected if timer is canceled before |
| 719 | // completion. |
| 720 | if (ec != boost::asio::error::operation_aborted) |
| 721 | { |
| 722 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 723 | << "\n"; |
| 724 | } |
| 725 | return; |
| 726 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 727 | std::cerr << "ERR" << std::to_string(errPin) << " asserted for " |
| 728 | << std::to_string(errTimeoutMs) << " ms\n"; |
| 729 | if (errPinCPUs.count()) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 730 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 731 | for (int i = 0; i < errPinCPUs.size(); i++) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 732 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 733 | if (errPinCPUs[i]) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 734 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 735 | cpuERRXLog(errPin, i); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 736 | } |
| 737 | } |
| 738 | } |
| 739 | else |
| 740 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 741 | cpuERRXLog(errPin); |
| 742 | } |
| 743 | }); |
| 744 | } |
| 745 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 746 | static void err0AssertHandler() |
| 747 | { |
| 748 | // Handle the standard ERR0 detection and logging |
| 749 | const static constexpr int err0 = 0; |
| 750 | errXAssertHandler(err0, err0AssertTimer); |
| 751 | } |
| 752 | |
| 753 | static void err0Handler() |
| 754 | { |
| 755 | if (!hostOff) |
| 756 | { |
| 757 | gpiod::line_event gpioLineEvent = err0Line.event_read(); |
| 758 | |
| 759 | bool err0 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 760 | if (err0) |
| 761 | { |
| 762 | err0AssertHandler(); |
| 763 | } |
| 764 | else |
| 765 | { |
| 766 | err0AssertTimer.cancel(); |
| 767 | } |
| 768 | } |
| 769 | err0Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 770 | [](const boost::system::error_code ec) { |
| 771 | if (ec) |
| 772 | { |
| 773 | std::cerr |
| 774 | << "err0 handler error: " << ec.message() |
| 775 | << "\n"; |
| 776 | return; |
| 777 | } |
| 778 | err0Handler(); |
| 779 | }); |
| 780 | } |
| 781 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame^] | 782 | static void err1AssertHandler() |
| 783 | { |
| 784 | // Handle the standard ERR1 detection and logging |
| 785 | const static constexpr int err1 = 1; |
| 786 | errXAssertHandler(err1, err1AssertTimer); |
| 787 | } |
| 788 | |
| 789 | static void err1Handler() |
| 790 | { |
| 791 | if (!hostOff) |
| 792 | { |
| 793 | gpiod::line_event gpioLineEvent = err1Line.event_read(); |
| 794 | |
| 795 | bool err1 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 796 | if (err1) |
| 797 | { |
| 798 | err1AssertHandler(); |
| 799 | } |
| 800 | else |
| 801 | { |
| 802 | err1AssertTimer.cancel(); |
| 803 | } |
| 804 | } |
| 805 | err1Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 806 | [](const boost::system::error_code ec) { |
| 807 | if (ec) |
| 808 | { |
| 809 | std::cerr |
| 810 | << "err1 handler error: " << ec.message() |
| 811 | << "\n"; |
| 812 | return; |
| 813 | } |
| 814 | err1Handler(); |
| 815 | }); |
| 816 | } |
| 817 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 818 | static void err2AssertHandler() |
| 819 | { |
| 820 | // Handle the standard ERR2 detection and logging |
| 821 | const static constexpr int err2 = 2; |
| 822 | errXAssertHandler(err2, err2AssertTimer); |
| 823 | // Also handle reset for ERR2 |
| 824 | err2AssertTimer.async_wait([](const boost::system::error_code ec) { |
| 825 | if (ec) |
| 826 | { |
| 827 | // operation_aborted is expected if timer is canceled before |
| 828 | // completion. |
| 829 | if (ec != boost::asio::error::operation_aborted) |
| 830 | { |
| 831 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 832 | << "\n"; |
| 833 | } |
| 834 | return; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 835 | } |
| 836 | conn->async_method_call( |
| 837 | [](boost::system::error_code ec, |
| 838 | const std::variant<bool>& property) { |
| 839 | if (ec) |
| 840 | { |
| 841 | return; |
| 842 | } |
| 843 | const bool* reset = std::get_if<bool>(&property); |
| 844 | if (reset == nullptr) |
| 845 | { |
| 846 | std::cerr << "Unable to read reset on ERR2 value\n"; |
| 847 | return; |
| 848 | } |
| 849 | startCrashdumpAndRecovery(*reset); |
| 850 | }, |
| 851 | "xyz.openbmc_project.Settings", |
| 852 | "/xyz/openbmc_project/control/processor_error_config", |
| 853 | "org.freedesktop.DBus.Properties", "Get", |
| 854 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnERR2"); |
| 855 | }); |
| 856 | } |
| 857 | |
| 858 | static void err2Handler() |
| 859 | { |
| 860 | if (!hostOff) |
| 861 | { |
| 862 | gpiod::line_event gpioLineEvent = err2Line.event_read(); |
| 863 | |
| 864 | bool err2 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 865 | if (err2) |
| 866 | { |
| 867 | err2AssertHandler(); |
| 868 | } |
| 869 | else |
| 870 | { |
| 871 | err2AssertTimer.cancel(); |
| 872 | } |
| 873 | } |
| 874 | err2Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 875 | [](const boost::system::error_code ec) { |
| 876 | if (ec) |
| 877 | { |
| 878 | std::cerr |
| 879 | << "err2 handler error: " << ec.message() |
| 880 | << "\n"; |
| 881 | return; |
| 882 | } |
| 883 | err2Handler(); |
| 884 | }); |
| 885 | } |
| 886 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 887 | static void smiAssertHandler() |
| 888 | { |
| 889 | smiAssertTimer.expires_after(std::chrono::milliseconds(smiTimeoutMs)); |
| 890 | smiAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 891 | if (ec) |
| 892 | { |
| 893 | // operation_aborted is expected if timer is canceled before |
| 894 | // completion. |
| 895 | if (ec != boost::asio::error::operation_aborted) |
| 896 | { |
| 897 | std::cerr << "smi timeout async_wait failed: " << ec.message() |
| 898 | << "\n"; |
| 899 | } |
| 900 | return; |
| 901 | } |
| 902 | std::cerr << "SMI asserted for " << std::to_string(smiTimeoutMs) |
| 903 | << " ms\n"; |
| 904 | smiTimeoutLog(); |
| 905 | conn->async_method_call( |
| 906 | [](boost::system::error_code ec, |
| 907 | const std::variant<bool>& property) { |
| 908 | if (ec) |
| 909 | { |
| 910 | return; |
| 911 | } |
| 912 | const bool* reset = std::get_if<bool>(&property); |
| 913 | if (reset == nullptr) |
| 914 | { |
| 915 | std::cerr << "Unable to read reset on SMI value\n"; |
| 916 | return; |
| 917 | } |
| 918 | startCrashdumpAndRecovery(*reset); |
| 919 | }, |
| 920 | "xyz.openbmc_project.Settings", |
| 921 | "/xyz/openbmc_project/control/bmc_reset_disables", |
| 922 | "org.freedesktop.DBus.Properties", "Get", |
| 923 | "xyz.openbmc_project.Control.ResetDisables", "ResetOnSMI"); |
| 924 | }); |
| 925 | } |
| 926 | |
| 927 | static void smiHandler() |
| 928 | { |
| 929 | if (!hostOff) |
| 930 | { |
| 931 | gpiod::line_event gpioLineEvent = smiLine.event_read(); |
| 932 | |
| 933 | bool smi = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 934 | if (smi) |
| 935 | { |
| 936 | smiAssertHandler(); |
| 937 | } |
| 938 | else |
| 939 | { |
| 940 | smiAssertTimer.cancel(); |
| 941 | } |
| 942 | } |
| 943 | smiEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 944 | [](const boost::system::error_code ec) { |
| 945 | if (ec) |
| 946 | { |
| 947 | std::cerr |
| 948 | << "smi handler error: " << ec.message() |
| 949 | << "\n"; |
| 950 | return; |
| 951 | } |
| 952 | smiHandler(); |
| 953 | }); |
| 954 | } |
| 955 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 956 | static void initializeErrorState() |
| 957 | { |
| 958 | // Handle CPU_CATERR if it's asserted now |
| 959 | if (caterrLine.get_value() == 0) |
| 960 | { |
| 961 | caterrAssertHandler(); |
| 962 | } |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 963 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 964 | // Handle CPU_ERR0 if it's asserted now |
| 965 | if (err0Line.get_value() == 0) |
| 966 | { |
| 967 | err0AssertHandler(); |
| 968 | } |
| 969 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame^] | 970 | // Handle CPU_ERR1 if it's asserted now |
| 971 | if (err1Line.get_value() == 0) |
| 972 | { |
| 973 | err1AssertHandler(); |
| 974 | } |
| 975 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 976 | // Handle CPU_ERR2 if it's asserted now |
| 977 | if (err2Line.get_value() == 0) |
| 978 | { |
| 979 | err2AssertHandler(); |
| 980 | } |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 981 | |
| 982 | // Handle SMI if it's asserted now |
| 983 | if (smiLine.get_value() == 0) |
| 984 | { |
| 985 | smiAssertHandler(); |
| 986 | } |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 987 | |
| 988 | // Handle PCH_BMC_THERMTRIP if it's asserted now |
| 989 | if (pchThermtripLine.get_value() == 0) |
| 990 | { |
| 991 | ssbThermTripLog(); |
| 992 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 993 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 994 | } // namespace host_error_monitor |
| 995 | |
| 996 | int main(int argc, char* argv[]) |
| 997 | { |
| 998 | // setup connection to dbus |
| 999 | host_error_monitor::conn = |
| 1000 | std::make_shared<sdbusplus::asio::connection>(host_error_monitor::io); |
| 1001 | |
| 1002 | // Host Error Monitor Object |
| 1003 | host_error_monitor::conn->request_name( |
| 1004 | "xyz.openbmc_project.HostErrorMonitor"); |
| 1005 | sdbusplus::asio::object_server server = |
| 1006 | sdbusplus::asio::object_server(host_error_monitor::conn); |
| 1007 | |
| 1008 | // Start tracking host state |
| 1009 | std::shared_ptr<sdbusplus::bus::match::match> hostStateMonitor = |
| 1010 | host_error_monitor::startHostStateMonitor(); |
| 1011 | |
| 1012 | // Initialize the host state |
| 1013 | host_error_monitor::initializeHostState(); |
| 1014 | |
| 1015 | // Request CPU_CATERR GPIO events |
| 1016 | if (!host_error_monitor::requestGPIOEvents( |
| 1017 | "CPU_CATERR", host_error_monitor::caterrHandler, |
| 1018 | host_error_monitor::caterrLine, host_error_monitor::caterrEvent)) |
| 1019 | { |
| 1020 | return -1; |
| 1021 | } |
| 1022 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1023 | // Request CPU_ERR0 GPIO events |
| 1024 | if (!host_error_monitor::requestGPIOEvents( |
| 1025 | "CPU_ERR0", host_error_monitor::err0Handler, |
| 1026 | host_error_monitor::err0Line, host_error_monitor::err0Event)) |
| 1027 | { |
| 1028 | return -1; |
| 1029 | } |
| 1030 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame^] | 1031 | // Request CPU_ERR1 GPIO events |
| 1032 | if (!host_error_monitor::requestGPIOEvents( |
| 1033 | "CPU_ERR1", host_error_monitor::err1Handler, |
| 1034 | host_error_monitor::err1Line, host_error_monitor::err1Event)) |
| 1035 | { |
| 1036 | return -1; |
| 1037 | } |
| 1038 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1039 | // Request CPU_ERR2 GPIO events |
| 1040 | if (!host_error_monitor::requestGPIOEvents( |
| 1041 | "CPU_ERR2", host_error_monitor::err2Handler, |
| 1042 | host_error_monitor::err2Line, host_error_monitor::err2Event)) |
| 1043 | { |
| 1044 | return -1; |
| 1045 | } |
| 1046 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1047 | // Request SMI GPIO events |
| 1048 | if (!host_error_monitor::requestGPIOEvents( |
| 1049 | "SMI", host_error_monitor::smiHandler, host_error_monitor::smiLine, |
| 1050 | host_error_monitor::smiEvent)) |
| 1051 | { |
| 1052 | return -1; |
| 1053 | } |
| 1054 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1055 | // Request PCH_BMC_THERMTRIP GPIO events |
| 1056 | if (!host_error_monitor::requestGPIOEvents( |
| 1057 | "PCH_BMC_THERMTRIP", host_error_monitor::pchThermtripHandler, |
| 1058 | host_error_monitor::pchThermtripLine, |
| 1059 | host_error_monitor::pchThermtripEvent)) |
| 1060 | { |
| 1061 | return -1; |
| 1062 | } |
| 1063 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1064 | host_error_monitor::io.run(); |
| 1065 | |
| 1066 | return 0; |
| 1067 | } |