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; |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame^] | 30 | static std::shared_ptr<sdbusplus::asio::dbus_interface> hostErrorTimeoutIface; |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 31 | |
| 32 | static bool hostOff = true; |
| 33 | |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame^] | 34 | static size_t caterrTimeoutMs = 2000; |
| 35 | const static constexpr size_t caterrTimeoutMsMax = 600000; // 10 minutes maximum |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 36 | const static constexpr size_t errTimeoutMs = 90000; |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 37 | const static constexpr size_t smiTimeoutMs = 90000; |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 38 | const static constexpr size_t crashdumpTimeoutS = 300; |
| 39 | |
| 40 | // Timers |
| 41 | // Timer for CATERR asserted |
| 42 | static boost::asio::steady_timer caterrAssertTimer(io); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 43 | // Timer for ERR0 asserted |
| 44 | static boost::asio::steady_timer err0AssertTimer(io); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 45 | // Timer for ERR1 asserted |
| 46 | static boost::asio::steady_timer err1AssertTimer(io); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 47 | // Timer for ERR2 asserted |
| 48 | static boost::asio::steady_timer err2AssertTimer(io); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 49 | // Timer for SMI asserted |
| 50 | static boost::asio::steady_timer smiAssertTimer(io); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 51 | |
| 52 | // GPIO Lines and Event Descriptors |
| 53 | static gpiod::line caterrLine; |
| 54 | static boost::asio::posix::stream_descriptor caterrEvent(io); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 55 | static gpiod::line err0Line; |
| 56 | static boost::asio::posix::stream_descriptor err0Event(io); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 57 | static gpiod::line err1Line; |
| 58 | static boost::asio::posix::stream_descriptor err1Event(io); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 59 | static gpiod::line err2Line; |
| 60 | static boost::asio::posix::stream_descriptor err2Event(io); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 61 | static gpiod::line smiLine; |
| 62 | static boost::asio::posix::stream_descriptor smiEvent(io); |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 63 | static gpiod::line cpu1FIVRFaultLine; |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 64 | static gpiod::line cpu1ThermtripLine; |
| 65 | static boost::asio::posix::stream_descriptor cpu1ThermtripEvent(io); |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 66 | static gpiod::line cpu2FIVRFaultLine; |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 67 | static gpiod::line cpu2ThermtripLine; |
| 68 | static boost::asio::posix::stream_descriptor cpu2ThermtripEvent(io); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 69 | static gpiod::line cpu1VRHotLine; |
| 70 | static boost::asio::posix::stream_descriptor cpu1VRHotEvent(io); |
| 71 | static gpiod::line cpu2VRHotLine; |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 72 | static boost::asio::posix::stream_descriptor cpu1MemABCDVRHotEvent(io); |
| 73 | static gpiod::line cpu1MemEFGHVRHotLine; |
| 74 | static boost::asio::posix::stream_descriptor cpu1MemEFGHVRHotEvent(io); |
| 75 | static gpiod::line cpu2MemABCDVRHotLine; |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 76 | static boost::asio::posix::stream_descriptor cpu2VRHotEvent(io); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 77 | static gpiod::line cpu1MemABCDVRHotLine; |
| 78 | static boost::asio::posix::stream_descriptor cpu2MemABCDVRHotEvent(io); |
| 79 | static gpiod::line cpu2MemEFGHVRHotLine; |
| 80 | static boost::asio::posix::stream_descriptor cpu2MemEFGHVRHotEvent(io); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 81 | //---------------------------------- |
| 82 | // PCH_BMC_THERMTRIP function related definition |
| 83 | //---------------------------------- |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 84 | static gpiod::line pchThermtripLine; |
| 85 | static boost::asio::posix::stream_descriptor pchThermtripEvent(io); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 86 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 87 | static void cpuIERRLog() |
| 88 | { |
| 89 | sd_journal_send("MESSAGE=HostError: IERR", "PRIORITY=%i", LOG_INFO, |
| 90 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 91 | "REDFISH_MESSAGE_ARGS=%s", "IERR", NULL); |
| 92 | } |
| 93 | |
| 94 | static void cpuIERRLog(const int cpuNum) |
| 95 | { |
| 96 | std::string msg = "IERR on CPU " + std::to_string(cpuNum + 1); |
| 97 | |
| 98 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 99 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 100 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 101 | } |
| 102 | |
| 103 | static void cpuIERRLog(const int cpuNum, const std::string& type) |
| 104 | { |
| 105 | std::string msg = type + " IERR on CPU " + std::to_string(cpuNum + 1); |
| 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 | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 112 | static void cpuERRXLog(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 113 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 114 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout"; |
| 115 | |
| 116 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 117 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 118 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 119 | } |
| 120 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 121 | static void cpuERRXLog(const int errPin, const int cpuNum) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 122 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 123 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout on CPU " + |
| 124 | std::to_string(cpuNum + 1); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 125 | |
| 126 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 127 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 128 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 129 | } |
| 130 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 131 | static void smiTimeoutLog() |
| 132 | { |
| 133 | sd_journal_send("MESSAGE=HostError: SMI Timeout", "PRIORITY=%i", LOG_INFO, |
| 134 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 135 | "REDFISH_MESSAGE_ARGS=%s", "SMI Timeout", NULL); |
| 136 | } |
| 137 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 138 | static void cpuBootFIVRFaultLog(const int cpuNum) |
| 139 | { |
| 140 | std::string msg = "Boot FIVR Fault on CPU " + std::to_string(cpuNum); |
| 141 | |
| 142 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 143 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 144 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 145 | } |
| 146 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 147 | static void cpuThermTripLog(const int cpuNum) |
| 148 | { |
| 149 | std::string msg = "CPU " + std::to_string(cpuNum) + " thermal trip"; |
| 150 | |
| 151 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 152 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 153 | "OpenBMC.0.1.CPUThermalTrip", "REDFISH_MESSAGE_ARGS=%d", |
| 154 | cpuNum, NULL); |
| 155 | } |
| 156 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 157 | static void cpuVRHotLog(const std::string& vr) |
| 158 | { |
| 159 | std::string msg = vr + " Voltage Regulator Overheated."; |
| 160 | |
| 161 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 162 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 163 | "OpenBMC.0.1.VoltageRegulatorOverheated", |
| 164 | "REDFISH_MESSAGE_ARGS=%s", vr.c_str(), NULL); |
| 165 | } |
| 166 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 167 | static void ssbThermTripLog() |
| 168 | { |
| 169 | sd_journal_send("MESSAGE=HostError: SSB thermal trip", "PRIORITY=%i", |
| 170 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 171 | "OpenBMC.0.1.SsbThermalTrip", NULL); |
| 172 | } |
| 173 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 174 | static void initializeErrorState(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 175 | static void initializeHostState() |
| 176 | { |
| 177 | conn->async_method_call( |
| 178 | [](boost::system::error_code ec, |
| 179 | const std::variant<std::string>& property) { |
| 180 | if (ec) |
| 181 | { |
| 182 | return; |
| 183 | } |
| 184 | const std::string* state = std::get_if<std::string>(&property); |
| 185 | if (state == nullptr) |
| 186 | { |
| 187 | std::cerr << "Unable to read host state value\n"; |
| 188 | return; |
| 189 | } |
| 190 | hostOff = *state == "xyz.openbmc_project.State.Host.HostState.Off"; |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 191 | // If the system is on, initialize the error state |
| 192 | if (!hostOff) |
| 193 | { |
| 194 | initializeErrorState(); |
| 195 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 196 | }, |
| 197 | "xyz.openbmc_project.State.Host", "/xyz/openbmc_project/state/host0", |
| 198 | "org.freedesktop.DBus.Properties", "Get", |
| 199 | "xyz.openbmc_project.State.Host", "CurrentHostState"); |
| 200 | } |
| 201 | |
| 202 | static std::shared_ptr<sdbusplus::bus::match::match> startHostStateMonitor() |
| 203 | { |
| 204 | return std::make_shared<sdbusplus::bus::match::match>( |
| 205 | *conn, |
| 206 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 207 | "member='PropertiesChanged',arg0namespace='xyz.openbmc_project.State." |
| 208 | "Host'", |
| 209 | [](sdbusplus::message::message& msg) { |
| 210 | std::string interfaceName; |
| 211 | boost::container::flat_map<std::string, std::variant<std::string>> |
| 212 | propertiesChanged; |
| 213 | std::string state; |
| 214 | try |
| 215 | { |
| 216 | msg.read(interfaceName, propertiesChanged); |
| 217 | state = |
| 218 | std::get<std::string>(propertiesChanged.begin()->second); |
| 219 | } |
| 220 | catch (std::exception& e) |
| 221 | { |
| 222 | std::cerr << "Unable to read host state\n"; |
| 223 | return; |
| 224 | } |
| 225 | hostOff = state == "xyz.openbmc_project.State.Host.HostState.Off"; |
| 226 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 227 | if (hostOff) |
| 228 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 229 | // No host events should fire while off, so cancel any pending |
| 230 | // timers |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 231 | caterrAssertTimer.cancel(); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 232 | err0AssertTimer.cancel(); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 233 | err1AssertTimer.cancel(); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 234 | err2AssertTimer.cancel(); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 235 | smiAssertTimer.cancel(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 236 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 237 | else |
| 238 | { |
| 239 | // Handle any initial errors when the host turns on |
| 240 | initializeErrorState(); |
| 241 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 242 | }); |
| 243 | } |
| 244 | |
| 245 | static bool requestGPIOEvents( |
| 246 | const std::string& name, const std::function<void()>& handler, |
| 247 | gpiod::line& gpioLine, |
| 248 | boost::asio::posix::stream_descriptor& gpioEventDescriptor) |
| 249 | { |
| 250 | // Find the GPIO line |
| 251 | gpioLine = gpiod::find_line(name); |
| 252 | if (!gpioLine) |
| 253 | { |
| 254 | std::cerr << "Failed to find the " << name << " line\n"; |
| 255 | return false; |
| 256 | } |
| 257 | |
| 258 | try |
| 259 | { |
| 260 | gpioLine.request( |
| 261 | {"host-error-monitor", gpiod::line_request::EVENT_BOTH_EDGES}); |
| 262 | } |
| 263 | catch (std::exception&) |
| 264 | { |
| 265 | std::cerr << "Failed to request events for " << name << "\n"; |
| 266 | return false; |
| 267 | } |
| 268 | |
| 269 | int gpioLineFd = gpioLine.event_get_fd(); |
| 270 | if (gpioLineFd < 0) |
| 271 | { |
| 272 | std::cerr << "Failed to get " << name << " fd\n"; |
| 273 | return false; |
| 274 | } |
| 275 | |
| 276 | gpioEventDescriptor.assign(gpioLineFd); |
| 277 | |
| 278 | gpioEventDescriptor.async_wait( |
| 279 | boost::asio::posix::stream_descriptor::wait_read, |
| 280 | [&name, handler](const boost::system::error_code ec) { |
| 281 | if (ec) |
| 282 | { |
| 283 | std::cerr << name << " fd handler error: " << ec.message() |
| 284 | << "\n"; |
| 285 | return; |
| 286 | } |
| 287 | handler(); |
| 288 | }); |
| 289 | return true; |
| 290 | } |
| 291 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 292 | static bool requestGPIOInput(const std::string& name, gpiod::line& gpioLine) |
| 293 | { |
| 294 | // Find the GPIO line |
| 295 | gpioLine = gpiod::find_line(name); |
| 296 | if (!gpioLine) |
| 297 | { |
| 298 | std::cerr << "Failed to find the " << name << " line.\n"; |
| 299 | return false; |
| 300 | } |
| 301 | |
| 302 | // Request GPIO input |
| 303 | try |
| 304 | { |
| 305 | gpioLine.request({__FUNCTION__, gpiod::line_request::DIRECTION_INPUT}); |
| 306 | } |
| 307 | catch (std::exception&) |
| 308 | { |
| 309 | std::cerr << "Failed to request " << name << " input\n"; |
| 310 | return false; |
| 311 | } |
| 312 | |
| 313 | return true; |
| 314 | } |
| 315 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 316 | static void startPowerCycle() |
| 317 | { |
| 318 | conn->async_method_call( |
| 319 | [](boost::system::error_code ec) { |
| 320 | if (ec) |
| 321 | { |
| 322 | std::cerr << "failed to set Chassis State\n"; |
| 323 | } |
| 324 | }, |
| 325 | "xyz.openbmc_project.State.Chassis", |
| 326 | "/xyz/openbmc_project/state/chassis0", |
| 327 | "org.freedesktop.DBus.Properties", "Set", |
| 328 | "xyz.openbmc_project.State.Chassis", "RequestedPowerTransition", |
| 329 | std::variant<std::string>{ |
| 330 | "xyz.openbmc_project.State.Chassis.Transition.PowerCycle"}); |
| 331 | } |
| 332 | |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 333 | static void startCrashdumpAndRecovery(bool recoverSystem, |
| 334 | const std::string& triggerType) |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 335 | { |
| 336 | std::cout << "Starting crashdump\n"; |
| 337 | static std::shared_ptr<sdbusplus::bus::match::match> crashdumpCompleteMatch; |
| 338 | static boost::asio::steady_timer crashdumpTimer(io); |
| 339 | |
| 340 | crashdumpCompleteMatch = std::make_shared<sdbusplus::bus::match::match>( |
| 341 | *conn, |
| 342 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 343 | "member='PropertiesChanged',arg0namespace='com.intel.crashdump'", |
| 344 | [recoverSystem](sdbusplus::message::message& msg) { |
| 345 | crashdumpTimer.cancel(); |
| 346 | std::cout << "Crashdump completed\n"; |
| 347 | if (recoverSystem) |
| 348 | { |
| 349 | std::cout << "Recovering the system\n"; |
| 350 | startPowerCycle(); |
| 351 | } |
| 352 | crashdumpCompleteMatch.reset(); |
| 353 | }); |
| 354 | |
| 355 | crashdumpTimer.expires_after(std::chrono::seconds(crashdumpTimeoutS)); |
| 356 | crashdumpTimer.async_wait([](const boost::system::error_code ec) { |
| 357 | if (ec) |
| 358 | { |
| 359 | // operation_aborted is expected if timer is canceled |
| 360 | if (ec != boost::asio::error::operation_aborted) |
| 361 | { |
| 362 | std::cerr << "Crashdump async_wait failed: " << ec.message() |
| 363 | << "\n"; |
| 364 | } |
| 365 | std::cout << "Crashdump timer canceled\n"; |
| 366 | return; |
| 367 | } |
| 368 | std::cerr << "Crashdump failed to complete before timeout\n"; |
| 369 | crashdumpCompleteMatch.reset(); |
| 370 | }); |
| 371 | |
| 372 | conn->async_method_call( |
| 373 | [](boost::system::error_code ec) { |
| 374 | if (ec) |
| 375 | { |
| 376 | std::cerr << "failed to start Crashdump\n"; |
| 377 | crashdumpTimer.cancel(); |
| 378 | crashdumpCompleteMatch.reset(); |
| 379 | } |
| 380 | }, |
| 381 | "com.intel.crashdump", "/com/intel/crashdump", |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 382 | "com.intel.crashdump.Stored", "GenerateStoredLog", triggerType); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 383 | } |
| 384 | |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 385 | static void incrementCPUErrorCount(int cpuNum) |
| 386 | { |
| 387 | std::string propertyName = "ErrorCountCPU" + std::to_string(cpuNum + 1); |
| 388 | |
| 389 | // Get the current count |
| 390 | conn->async_method_call( |
| 391 | [propertyName](boost::system::error_code ec, |
| 392 | const std::variant<uint8_t>& property) { |
| 393 | if (ec) |
| 394 | { |
| 395 | std::cerr << "Failed to read " << propertyName << ": " |
| 396 | << ec.message() << "\n"; |
| 397 | return; |
| 398 | } |
| 399 | const uint8_t* errorCountVariant = std::get_if<uint8_t>(&property); |
| 400 | if (errorCountVariant == nullptr) |
| 401 | { |
| 402 | std::cerr << propertyName << " invalid\n"; |
| 403 | return; |
| 404 | } |
| 405 | uint8_t errorCount = *errorCountVariant; |
| 406 | if (errorCount == std::numeric_limits<uint8_t>::max()) |
| 407 | { |
| 408 | std::cerr << "Maximum error count reached\n"; |
| 409 | return; |
| 410 | } |
| 411 | // Increment the count |
| 412 | errorCount++; |
| 413 | conn->async_method_call( |
| 414 | [propertyName](boost::system::error_code ec) { |
| 415 | if (ec) |
| 416 | { |
| 417 | std::cerr << "Failed to set " << propertyName << ": " |
| 418 | << ec.message() << "\n"; |
| 419 | } |
| 420 | }, |
| 421 | "xyz.openbmc_project.Settings", |
| 422 | "/xyz/openbmc_project/control/processor_error_config", |
| 423 | "org.freedesktop.DBus.Properties", "Set", |
| 424 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName, |
| 425 | std::variant<uint8_t>{errorCount}); |
| 426 | }, |
| 427 | "xyz.openbmc_project.Settings", |
| 428 | "/xyz/openbmc_project/control/processor_error_config", |
| 429 | "org.freedesktop.DBus.Properties", "Get", |
| 430 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName); |
| 431 | } |
| 432 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 433 | static bool checkIERRCPUs() |
| 434 | { |
| 435 | bool cpuIERRFound = false; |
| 436 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 437 | cpu++, addr++) |
| 438 | { |
| 439 | uint8_t cc = 0; |
| 440 | CPUModel model{}; |
| 441 | uint8_t stepping = 0; |
| 442 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 443 | { |
| 444 | std::cerr << "Cannot get CPUID!\n"; |
| 445 | continue; |
| 446 | } |
| 447 | |
| 448 | switch (model) |
| 449 | { |
| 450 | case skx: |
| 451 | { |
| 452 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 453 | // that caused the IERR |
| 454 | uint32_t mcaErrSrcLog = 0; |
| 455 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 456 | &cc) != PECI_CC_SUCCESS) |
| 457 | { |
| 458 | continue; |
| 459 | } |
| 460 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 461 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 462 | { |
| 463 | // TODO: Light the CPU fault LED? |
| 464 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 465 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 466 | // Next check if it's a CPU/VR mismatch by reading the |
| 467 | // IA32_MC4_STATUS MSR (0x411) |
| 468 | uint64_t mc4Status = 0; |
| 469 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 470 | PECI_CC_SUCCESS) |
| 471 | { |
| 472 | continue; |
| 473 | } |
| 474 | // Check MSEC bits 31:24 for |
| 475 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 476 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 477 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 478 | if ((mc4Status & (0x40 << 24)) || |
| 479 | (mc4Status & (0x42 << 24)) || |
| 480 | (mc4Status & (0x43 << 24))) |
| 481 | { |
| 482 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 483 | continue; |
| 484 | } |
| 485 | |
| 486 | // Next check if it's a Core FIVR fault by looking for a |
| 487 | // non-zero value of CORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 488 | // 80h) |
| 489 | uint32_t coreFIVRErrLog = 0; |
| 490 | if (peci_RdPCIConfigLocal( |
| 491 | addr, 1, 30, 2, 0x80, sizeof(uint32_t), |
| 492 | (uint8_t*)&coreFIVRErrLog, &cc) != PECI_CC_SUCCESS) |
| 493 | { |
| 494 | continue; |
| 495 | } |
| 496 | if (coreFIVRErrLog) |
| 497 | { |
| 498 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 499 | continue; |
| 500 | } |
| 501 | |
| 502 | // Next check if it's an Uncore FIVR fault by looking for a |
| 503 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 504 | // 84h) |
| 505 | uint32_t uncoreFIVRErrLog = 0; |
| 506 | if (peci_RdPCIConfigLocal(addr, 1, 30, 2, 0x84, |
| 507 | sizeof(uint32_t), |
| 508 | (uint8_t*)&uncoreFIVRErrLog, |
| 509 | &cc) != PECI_CC_SUCCESS) |
| 510 | { |
| 511 | continue; |
| 512 | } |
| 513 | if (uncoreFIVRErrLog) |
| 514 | { |
| 515 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 516 | continue; |
| 517 | } |
| 518 | |
| 519 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 520 | // both zero, but MSEC bits 31:24 have either |
| 521 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 522 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 523 | // uncore FIVR fault |
| 524 | if (!coreFIVRErrLog && !uncoreFIVRErrLog && |
| 525 | ((mc4Status & (0x51 << 24)) || |
| 526 | (mc4Status & (0x52 << 24)))) |
| 527 | { |
| 528 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 529 | continue; |
| 530 | } |
| 531 | cpuIERRLog(cpu); |
| 532 | } |
| 533 | break; |
| 534 | } |
| 535 | case icx: |
| 536 | { |
| 537 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 538 | // that caused the IERR |
| 539 | uint32_t mcaErrSrcLog = 0; |
| 540 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 541 | &cc) != PECI_CC_SUCCESS) |
| 542 | { |
| 543 | continue; |
| 544 | } |
| 545 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 546 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 547 | { |
| 548 | // TODO: Light the CPU fault LED? |
| 549 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 550 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 551 | // Next check if it's a CPU/VR mismatch by reading the |
| 552 | // IA32_MC4_STATUS MSR (0x411) |
| 553 | uint64_t mc4Status = 0; |
| 554 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 555 | PECI_CC_SUCCESS) |
| 556 | { |
| 557 | continue; |
| 558 | } |
| 559 | // TODO: Update MSEC/MSCOD_31_24 check |
| 560 | // Check MSEC bits 31:24 for |
| 561 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 562 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 563 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 564 | if ((mc4Status & (0x40 << 24)) || |
| 565 | (mc4Status & (0x42 << 24)) || |
| 566 | (mc4Status & (0x43 << 24))) |
| 567 | { |
| 568 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 569 | continue; |
| 570 | } |
| 571 | |
| 572 | // Next check if it's a Core FIVR fault by looking for a |
| 573 | // non-zero value of CORE_FIVR_ERR_LOG (B(31) D30 F2 offsets |
| 574 | // C0h and C4h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 575 | uint32_t coreFIVRErrLog0 = 0; |
| 576 | uint32_t coreFIVRErrLog1 = 0; |
| 577 | if (peci_RdEndPointConfigPciLocal( |
| 578 | addr, 0, 14, 30, 2, 0xC0, sizeof(uint32_t), |
| 579 | (uint8_t*)&coreFIVRErrLog0, &cc) != PECI_CC_SUCCESS) |
| 580 | { |
| 581 | continue; |
| 582 | } |
| 583 | if (peci_RdEndPointConfigPciLocal( |
| 584 | addr, 0, 14, 30, 2, 0xC4, sizeof(uint32_t), |
| 585 | (uint8_t*)&coreFIVRErrLog1, &cc) != PECI_CC_SUCCESS) |
| 586 | { |
| 587 | continue; |
| 588 | } |
| 589 | if (coreFIVRErrLog0 || coreFIVRErrLog1) |
| 590 | { |
| 591 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 592 | continue; |
| 593 | } |
| 594 | |
| 595 | // Next check if it's an Uncore FIVR fault by looking for a |
| 596 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(31) D30 F2 |
| 597 | // offset 84h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 598 | uint32_t uncoreFIVRErrLog = 0; |
| 599 | if (peci_RdEndPointConfigPciLocal( |
| 600 | addr, 0, 14, 30, 2, 0x84, sizeof(uint32_t), |
| 601 | (uint8_t*)&uncoreFIVRErrLog, |
| 602 | &cc) != PECI_CC_SUCCESS) |
| 603 | { |
| 604 | continue; |
| 605 | } |
| 606 | if (uncoreFIVRErrLog) |
| 607 | { |
| 608 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 609 | continue; |
| 610 | } |
| 611 | |
| 612 | // TODO: Update MSEC/MSCOD_31_24 check |
| 613 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 614 | // both zero, but MSEC bits 31:24 have either |
| 615 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 616 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 617 | // uncore FIVR fault |
| 618 | if (!coreFIVRErrLog0 && !coreFIVRErrLog1 && |
| 619 | !uncoreFIVRErrLog && |
| 620 | ((mc4Status & (0x51 << 24)) || |
| 621 | (mc4Status & (0x52 << 24)))) |
| 622 | { |
| 623 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 624 | continue; |
| 625 | } |
| 626 | cpuIERRLog(cpu); |
| 627 | } |
| 628 | break; |
| 629 | } |
| 630 | } |
| 631 | } |
| 632 | return cpuIERRFound; |
| 633 | } |
| 634 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 635 | static void caterrAssertHandler() |
| 636 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 637 | caterrAssertTimer.expires_after(std::chrono::milliseconds(caterrTimeoutMs)); |
| 638 | caterrAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 639 | if (ec) |
| 640 | { |
| 641 | // operation_aborted is expected if timer is canceled |
| 642 | // before completion. |
| 643 | if (ec != boost::asio::error::operation_aborted) |
| 644 | { |
| 645 | std::cerr << "caterr timeout async_wait failed: " |
| 646 | << ec.message() << "\n"; |
| 647 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 648 | return; |
| 649 | } |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 650 | std::cerr << "CATERR asserted for " << std::to_string(caterrTimeoutMs) |
| 651 | << " ms\n"; |
| 652 | if (!checkIERRCPUs()) |
| 653 | { |
| 654 | cpuIERRLog(); |
| 655 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 656 | conn->async_method_call( |
| 657 | [](boost::system::error_code ec, |
| 658 | const std::variant<bool>& property) { |
| 659 | if (ec) |
| 660 | { |
| 661 | return; |
| 662 | } |
| 663 | const bool* reset = std::get_if<bool>(&property); |
| 664 | if (reset == nullptr) |
| 665 | { |
| 666 | std::cerr << "Unable to read reset on CATERR value\n"; |
| 667 | return; |
| 668 | } |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 669 | startCrashdumpAndRecovery(*reset, "IERR"); |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 670 | }, |
| 671 | "xyz.openbmc_project.Settings", |
| 672 | "/xyz/openbmc_project/control/processor_error_config", |
| 673 | "org.freedesktop.DBus.Properties", "Get", |
| 674 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnCATERR"); |
| 675 | }); |
| 676 | } |
| 677 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 678 | static void caterrHandler() |
| 679 | { |
| 680 | if (!hostOff) |
| 681 | { |
| 682 | gpiod::line_event gpioLineEvent = caterrLine.event_read(); |
| 683 | |
| 684 | bool caterr = |
| 685 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 686 | if (caterr) |
| 687 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 688 | caterrAssertHandler(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 689 | } |
| 690 | else |
| 691 | { |
| 692 | caterrAssertTimer.cancel(); |
| 693 | } |
| 694 | } |
| 695 | caterrEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 696 | [](const boost::system::error_code ec) { |
| 697 | if (ec) |
| 698 | { |
| 699 | std::cerr << "caterr handler error: " |
| 700 | << ec.message() << "\n"; |
| 701 | return; |
| 702 | } |
| 703 | caterrHandler(); |
| 704 | }); |
| 705 | } |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 706 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 707 | static void cpu1ThermtripAssertHandler() |
| 708 | { |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 709 | if (cpu1FIVRFaultLine.get_value() == 0) |
| 710 | { |
| 711 | cpuBootFIVRFaultLog(1); |
| 712 | } |
| 713 | else |
| 714 | { |
| 715 | cpuThermTripLog(1); |
| 716 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 717 | } |
| 718 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 719 | static void cpu1ThermtripHandler() |
| 720 | { |
| 721 | if (!hostOff) |
| 722 | { |
| 723 | gpiod::line_event gpioLineEvent = cpu1ThermtripLine.event_read(); |
| 724 | |
| 725 | bool cpu1Thermtrip = |
| 726 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 727 | if (cpu1Thermtrip) |
| 728 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 729 | cpu1ThermtripAssertHandler(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 730 | } |
| 731 | } |
| 732 | cpu1ThermtripEvent.async_wait( |
| 733 | boost::asio::posix::stream_descriptor::wait_read, |
| 734 | [](const boost::system::error_code ec) { |
| 735 | if (ec) |
| 736 | { |
| 737 | std::cerr << "CPU 1 Thermtrip handler error: " << ec.message() |
| 738 | << "\n"; |
| 739 | return; |
| 740 | } |
| 741 | cpu1ThermtripHandler(); |
| 742 | }); |
| 743 | } |
| 744 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 745 | static void cpu2ThermtripAssertHandler() |
| 746 | { |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 747 | if (cpu2FIVRFaultLine.get_value() == 0) |
| 748 | { |
| 749 | cpuBootFIVRFaultLog(2); |
| 750 | } |
| 751 | else |
| 752 | { |
| 753 | cpuThermTripLog(2); |
| 754 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 755 | } |
| 756 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 757 | static void cpu2ThermtripHandler() |
| 758 | { |
| 759 | if (!hostOff) |
| 760 | { |
| 761 | gpiod::line_event gpioLineEvent = cpu2ThermtripLine.event_read(); |
| 762 | |
| 763 | bool cpu2Thermtrip = |
| 764 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 765 | if (cpu2Thermtrip) |
| 766 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 767 | cpu2ThermtripAssertHandler(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 768 | } |
| 769 | } |
| 770 | cpu2ThermtripEvent.async_wait( |
| 771 | boost::asio::posix::stream_descriptor::wait_read, |
| 772 | [](const boost::system::error_code ec) { |
| 773 | if (ec) |
| 774 | { |
| 775 | std::cerr << "CPU 2 Thermtrip handler error: " << ec.message() |
| 776 | << "\n"; |
| 777 | return; |
| 778 | } |
| 779 | cpu2ThermtripHandler(); |
| 780 | }); |
| 781 | } |
| 782 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 783 | static void cpu1VRHotAssertHandler() |
| 784 | { |
| 785 | cpuVRHotLog("CPU 1"); |
| 786 | } |
| 787 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 788 | static void cpu1VRHotHandler() |
| 789 | { |
| 790 | if (!hostOff) |
| 791 | { |
| 792 | gpiod::line_event gpioLineEvent = cpu1VRHotLine.event_read(); |
| 793 | |
| 794 | bool cpu1VRHot = |
| 795 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 796 | if (cpu1VRHot) |
| 797 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 798 | cpu1VRHotAssertHandler(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 799 | } |
| 800 | } |
| 801 | cpu1VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 802 | [](const boost::system::error_code ec) { |
| 803 | if (ec) |
| 804 | { |
| 805 | std::cerr << "CPU 1 VRHot handler error: " |
| 806 | << ec.message() << "\n"; |
| 807 | return; |
| 808 | } |
| 809 | cpu1VRHotHandler(); |
| 810 | }); |
| 811 | } |
| 812 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 813 | static void cpu1MemABCDVRHotAssertHandler() |
| 814 | { |
| 815 | cpuVRHotLog("CPU 1 Memory ABCD"); |
| 816 | } |
| 817 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 818 | static void cpu1MemABCDVRHotHandler() |
| 819 | { |
| 820 | if (!hostOff) |
| 821 | { |
| 822 | gpiod::line_event gpioLineEvent = cpu1MemABCDVRHotLine.event_read(); |
| 823 | |
| 824 | bool cpu1MemABCDVRHot = |
| 825 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 826 | if (cpu1MemABCDVRHot) |
| 827 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 828 | cpu1MemABCDVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 829 | } |
| 830 | } |
| 831 | cpu1MemABCDVRHotEvent.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 1 Memory ABCD VRHot handler error: " |
| 837 | << ec.message() << "\n"; |
| 838 | return; |
| 839 | } |
| 840 | cpu1MemABCDVRHotHandler(); |
| 841 | }); |
| 842 | } |
| 843 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 844 | static void cpu1MemEFGHVRHotAssertHandler() |
| 845 | { |
| 846 | cpuVRHotLog("CPU 1 Memory EFGH"); |
| 847 | } |
| 848 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 849 | static void cpu1MemEFGHVRHotHandler() |
| 850 | { |
| 851 | if (!hostOff) |
| 852 | { |
| 853 | gpiod::line_event gpioLineEvent = cpu1MemEFGHVRHotLine.event_read(); |
| 854 | |
| 855 | bool cpu1MemEFGHVRHot = |
| 856 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 857 | if (cpu1MemEFGHVRHot) |
| 858 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 859 | cpu1MemEFGHVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 860 | } |
| 861 | } |
| 862 | cpu1MemEFGHVRHotEvent.async_wait( |
| 863 | boost::asio::posix::stream_descriptor::wait_read, |
| 864 | [](const boost::system::error_code ec) { |
| 865 | if (ec) |
| 866 | { |
| 867 | std::cerr << "CPU 1 Memory EFGH VRHot handler error: " |
| 868 | << ec.message() << "\n"; |
| 869 | return; |
| 870 | } |
| 871 | cpu1MemEFGHVRHotHandler(); |
| 872 | }); |
| 873 | } |
| 874 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 875 | static void cpu2VRHotAssertHandler() |
| 876 | { |
| 877 | cpuVRHotLog("CPU 2"); |
| 878 | } |
| 879 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 880 | static void cpu2VRHotHandler() |
| 881 | { |
| 882 | if (!hostOff) |
| 883 | { |
| 884 | gpiod::line_event gpioLineEvent = cpu2VRHotLine.event_read(); |
| 885 | |
| 886 | bool cpu2VRHot = |
| 887 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 888 | if (cpu2VRHot) |
| 889 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 890 | cpu2VRHotAssertHandler(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 891 | } |
| 892 | } |
| 893 | cpu2VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 894 | [](const boost::system::error_code ec) { |
| 895 | if (ec) |
| 896 | { |
| 897 | std::cerr << "CPU 2 VRHot handler error: " |
| 898 | << ec.message() << "\n"; |
| 899 | return; |
| 900 | } |
| 901 | cpu2VRHotHandler(); |
| 902 | }); |
| 903 | } |
| 904 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 905 | static void cpu2MemABCDVRHotAssertHandler() |
| 906 | { |
| 907 | cpuVRHotLog("CPU 2 Memory ABCD"); |
| 908 | } |
| 909 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 910 | static void cpu2MemABCDVRHotHandler() |
| 911 | { |
| 912 | if (!hostOff) |
| 913 | { |
| 914 | gpiod::line_event gpioLineEvent = cpu2MemABCDVRHotLine.event_read(); |
| 915 | |
| 916 | bool cpu2MemABCDVRHot = |
| 917 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 918 | if (cpu2MemABCDVRHot) |
| 919 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 920 | cpu2MemABCDVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 921 | } |
| 922 | } |
| 923 | cpu2MemABCDVRHotEvent.async_wait( |
| 924 | boost::asio::posix::stream_descriptor::wait_read, |
| 925 | [](const boost::system::error_code ec) { |
| 926 | if (ec) |
| 927 | { |
| 928 | std::cerr << "CPU 2 Memory ABCD VRHot handler error: " |
| 929 | << ec.message() << "\n"; |
| 930 | return; |
| 931 | } |
| 932 | cpu2MemABCDVRHotHandler(); |
| 933 | }); |
| 934 | } |
| 935 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 936 | static void cpu2MemEFGHVRHotAssertHandler() |
| 937 | { |
| 938 | cpuVRHotLog("CPU 2 Memory EFGH"); |
| 939 | } |
| 940 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 941 | static void cpu2MemEFGHVRHotHandler() |
| 942 | { |
| 943 | if (!hostOff) |
| 944 | { |
| 945 | gpiod::line_event gpioLineEvent = cpu2MemEFGHVRHotLine.event_read(); |
| 946 | |
| 947 | bool cpu2MemEFGHVRHot = |
| 948 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 949 | if (cpu2MemEFGHVRHot) |
| 950 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 951 | cpu2MemEFGHVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 952 | } |
| 953 | } |
| 954 | cpu2MemEFGHVRHotEvent.async_wait( |
| 955 | boost::asio::posix::stream_descriptor::wait_read, |
| 956 | [](const boost::system::error_code ec) { |
| 957 | if (ec) |
| 958 | { |
| 959 | std::cerr << "CPU 2 Memory EFGH VRHot handler error: " |
| 960 | << ec.message() << "\n"; |
| 961 | return; |
| 962 | } |
| 963 | cpu2MemEFGHVRHotHandler(); |
| 964 | }); |
| 965 | } |
| 966 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 967 | static void pchThermtripHandler() |
| 968 | { |
| 969 | if (!hostOff) |
| 970 | { |
| 971 | gpiod::line_event gpioLineEvent = pchThermtripLine.event_read(); |
| 972 | |
| 973 | bool pchThermtrip = |
| 974 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 975 | if (pchThermtrip) |
| 976 | { |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 977 | ssbThermTripLog(); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 978 | } |
| 979 | } |
| 980 | pchThermtripEvent.async_wait( |
| 981 | boost::asio::posix::stream_descriptor::wait_read, |
| 982 | [](const boost::system::error_code ec) { |
| 983 | if (ec) |
| 984 | { |
| 985 | std::cerr << "PCH Thermal trip handler error: " << ec.message() |
| 986 | << "\n"; |
| 987 | return; |
| 988 | } |
| 989 | pchThermtripHandler(); |
| 990 | }); |
| 991 | } |
| 992 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 993 | static std::bitset<MAX_CPUS> checkERRPinCPUs(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 994 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 995 | int errPinSts = (1 << errPin); |
| 996 | std::bitset<MAX_CPUS> errPinCPUs = 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 997 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 998 | cpu++, addr++) |
| 999 | { |
| 1000 | if (peci_Ping(addr) == PECI_CC_SUCCESS) |
| 1001 | { |
| 1002 | uint8_t cc = 0; |
| 1003 | CPUModel model{}; |
| 1004 | uint8_t stepping = 0; |
| 1005 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 1006 | { |
| 1007 | std::cerr << "Cannot get CPUID!\n"; |
| 1008 | continue; |
| 1009 | } |
| 1010 | |
| 1011 | switch (model) |
| 1012 | { |
| 1013 | case skx: |
| 1014 | { |
| 1015 | // 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] | 1016 | // the ERRx (B(0) D8 F0 offset 210h) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1017 | uint32_t errpinsts = 0; |
| 1018 | if (peci_RdPCIConfigLocal( |
| 1019 | addr, 0, 8, 0, 0x210, sizeof(uint32_t), |
| 1020 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 1021 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1022 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1023 | } |
| 1024 | break; |
| 1025 | } |
| 1026 | case icx: |
| 1027 | { |
| 1028 | // 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] | 1029 | // 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] | 1030 | // accessed on PECI as bus 13) |
| 1031 | uint32_t errpinsts = 0; |
| 1032 | if (peci_RdEndPointConfigPciLocal( |
| 1033 | addr, 0, 13, 0, 3, 0x274, sizeof(uint32_t), |
| 1034 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 1035 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1036 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1037 | } |
| 1038 | break; |
| 1039 | } |
| 1040 | } |
| 1041 | } |
| 1042 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1043 | return errPinCPUs; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1044 | } |
| 1045 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1046 | static void errXAssertHandler(const int errPin, |
| 1047 | boost::asio::steady_timer& errXAssertTimer) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1048 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1049 | // ERRx status is not guaranteed through the timeout, so save which |
| 1050 | // CPUs have it asserted |
| 1051 | std::bitset<MAX_CPUS> errPinCPUs = checkERRPinCPUs(errPin); |
| 1052 | errXAssertTimer.expires_after(std::chrono::milliseconds(errTimeoutMs)); |
| 1053 | errXAssertTimer.async_wait([errPin, errPinCPUs]( |
| 1054 | const boost::system::error_code ec) { |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1055 | if (ec) |
| 1056 | { |
| 1057 | // operation_aborted is expected if timer is canceled before |
| 1058 | // completion. |
| 1059 | if (ec != boost::asio::error::operation_aborted) |
| 1060 | { |
| 1061 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 1062 | << "\n"; |
| 1063 | } |
| 1064 | return; |
| 1065 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1066 | std::cerr << "ERR" << std::to_string(errPin) << " asserted for " |
| 1067 | << std::to_string(errTimeoutMs) << " ms\n"; |
| 1068 | if (errPinCPUs.count()) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1069 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1070 | for (int i = 0; i < errPinCPUs.size(); i++) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1071 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1072 | if (errPinCPUs[i]) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1073 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1074 | cpuERRXLog(errPin, i); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1075 | } |
| 1076 | } |
| 1077 | } |
| 1078 | else |
| 1079 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1080 | cpuERRXLog(errPin); |
| 1081 | } |
| 1082 | }); |
| 1083 | } |
| 1084 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1085 | static void err0AssertHandler() |
| 1086 | { |
| 1087 | // Handle the standard ERR0 detection and logging |
| 1088 | const static constexpr int err0 = 0; |
| 1089 | errXAssertHandler(err0, err0AssertTimer); |
| 1090 | } |
| 1091 | |
| 1092 | static void err0Handler() |
| 1093 | { |
| 1094 | if (!hostOff) |
| 1095 | { |
| 1096 | gpiod::line_event gpioLineEvent = err0Line.event_read(); |
| 1097 | |
| 1098 | bool err0 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1099 | if (err0) |
| 1100 | { |
| 1101 | err0AssertHandler(); |
| 1102 | } |
| 1103 | else |
| 1104 | { |
| 1105 | err0AssertTimer.cancel(); |
| 1106 | } |
| 1107 | } |
| 1108 | err0Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1109 | [](const boost::system::error_code ec) { |
| 1110 | if (ec) |
| 1111 | { |
| 1112 | std::cerr |
| 1113 | << "err0 handler error: " << ec.message() |
| 1114 | << "\n"; |
| 1115 | return; |
| 1116 | } |
| 1117 | err0Handler(); |
| 1118 | }); |
| 1119 | } |
| 1120 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1121 | static void err1AssertHandler() |
| 1122 | { |
| 1123 | // Handle the standard ERR1 detection and logging |
| 1124 | const static constexpr int err1 = 1; |
| 1125 | errXAssertHandler(err1, err1AssertTimer); |
| 1126 | } |
| 1127 | |
| 1128 | static void err1Handler() |
| 1129 | { |
| 1130 | if (!hostOff) |
| 1131 | { |
| 1132 | gpiod::line_event gpioLineEvent = err1Line.event_read(); |
| 1133 | |
| 1134 | bool err1 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1135 | if (err1) |
| 1136 | { |
| 1137 | err1AssertHandler(); |
| 1138 | } |
| 1139 | else |
| 1140 | { |
| 1141 | err1AssertTimer.cancel(); |
| 1142 | } |
| 1143 | } |
| 1144 | err1Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1145 | [](const boost::system::error_code ec) { |
| 1146 | if (ec) |
| 1147 | { |
| 1148 | std::cerr |
| 1149 | << "err1 handler error: " << ec.message() |
| 1150 | << "\n"; |
| 1151 | return; |
| 1152 | } |
| 1153 | err1Handler(); |
| 1154 | }); |
| 1155 | } |
| 1156 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1157 | static void err2AssertHandler() |
| 1158 | { |
| 1159 | // Handle the standard ERR2 detection and logging |
| 1160 | const static constexpr int err2 = 2; |
| 1161 | errXAssertHandler(err2, err2AssertTimer); |
| 1162 | // Also handle reset for ERR2 |
| 1163 | err2AssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1164 | if (ec) |
| 1165 | { |
| 1166 | // operation_aborted is expected if timer is canceled before |
| 1167 | // completion. |
| 1168 | if (ec != boost::asio::error::operation_aborted) |
| 1169 | { |
| 1170 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 1171 | << "\n"; |
| 1172 | } |
| 1173 | return; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1174 | } |
| 1175 | conn->async_method_call( |
| 1176 | [](boost::system::error_code ec, |
| 1177 | const std::variant<bool>& property) { |
| 1178 | if (ec) |
| 1179 | { |
| 1180 | return; |
| 1181 | } |
| 1182 | const bool* reset = std::get_if<bool>(&property); |
| 1183 | if (reset == nullptr) |
| 1184 | { |
| 1185 | std::cerr << "Unable to read reset on ERR2 value\n"; |
| 1186 | return; |
| 1187 | } |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 1188 | startCrashdumpAndRecovery(*reset, "ERR2 Timeout"); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1189 | }, |
| 1190 | "xyz.openbmc_project.Settings", |
| 1191 | "/xyz/openbmc_project/control/processor_error_config", |
| 1192 | "org.freedesktop.DBus.Properties", "Get", |
| 1193 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnERR2"); |
| 1194 | }); |
| 1195 | } |
| 1196 | |
| 1197 | static void err2Handler() |
| 1198 | { |
| 1199 | if (!hostOff) |
| 1200 | { |
| 1201 | gpiod::line_event gpioLineEvent = err2Line.event_read(); |
| 1202 | |
| 1203 | bool err2 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1204 | if (err2) |
| 1205 | { |
| 1206 | err2AssertHandler(); |
| 1207 | } |
| 1208 | else |
| 1209 | { |
| 1210 | err2AssertTimer.cancel(); |
| 1211 | } |
| 1212 | } |
| 1213 | err2Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1214 | [](const boost::system::error_code ec) { |
| 1215 | if (ec) |
| 1216 | { |
| 1217 | std::cerr |
| 1218 | << "err2 handler error: " << ec.message() |
| 1219 | << "\n"; |
| 1220 | return; |
| 1221 | } |
| 1222 | err2Handler(); |
| 1223 | }); |
| 1224 | } |
| 1225 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1226 | static void smiAssertHandler() |
| 1227 | { |
| 1228 | smiAssertTimer.expires_after(std::chrono::milliseconds(smiTimeoutMs)); |
| 1229 | smiAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1230 | if (ec) |
| 1231 | { |
| 1232 | // operation_aborted is expected if timer is canceled before |
| 1233 | // completion. |
| 1234 | if (ec != boost::asio::error::operation_aborted) |
| 1235 | { |
| 1236 | std::cerr << "smi timeout async_wait failed: " << ec.message() |
| 1237 | << "\n"; |
| 1238 | } |
| 1239 | return; |
| 1240 | } |
| 1241 | std::cerr << "SMI asserted for " << std::to_string(smiTimeoutMs) |
| 1242 | << " ms\n"; |
| 1243 | smiTimeoutLog(); |
| 1244 | conn->async_method_call( |
| 1245 | [](boost::system::error_code ec, |
| 1246 | const std::variant<bool>& property) { |
| 1247 | if (ec) |
| 1248 | { |
| 1249 | return; |
| 1250 | } |
| 1251 | const bool* reset = std::get_if<bool>(&property); |
| 1252 | if (reset == nullptr) |
| 1253 | { |
| 1254 | std::cerr << "Unable to read reset on SMI value\n"; |
| 1255 | return; |
| 1256 | } |
Jason M. Bills | 9478544 | 2020-01-07 15:22:09 -0800 | [diff] [blame] | 1257 | #ifdef HOST_ERROR_CRASHDUMP_ON_SMI_TIMEOUT |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 1258 | startCrashdumpAndRecovery(*reset, "SMI Timeout"); |
Jason M. Bills | 9478544 | 2020-01-07 15:22:09 -0800 | [diff] [blame] | 1259 | #else |
| 1260 | if (*reset) |
| 1261 | { |
| 1262 | std::cout << "Recovering the system\n"; |
| 1263 | startPowerCycle(); |
| 1264 | } |
| 1265 | #endif |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1266 | }, |
| 1267 | "xyz.openbmc_project.Settings", |
| 1268 | "/xyz/openbmc_project/control/bmc_reset_disables", |
| 1269 | "org.freedesktop.DBus.Properties", "Get", |
| 1270 | "xyz.openbmc_project.Control.ResetDisables", "ResetOnSMI"); |
| 1271 | }); |
| 1272 | } |
| 1273 | |
| 1274 | static void smiHandler() |
| 1275 | { |
| 1276 | if (!hostOff) |
| 1277 | { |
| 1278 | gpiod::line_event gpioLineEvent = smiLine.event_read(); |
| 1279 | |
| 1280 | bool smi = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1281 | if (smi) |
| 1282 | { |
| 1283 | smiAssertHandler(); |
| 1284 | } |
| 1285 | else |
| 1286 | { |
| 1287 | smiAssertTimer.cancel(); |
| 1288 | } |
| 1289 | } |
| 1290 | smiEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1291 | [](const boost::system::error_code ec) { |
| 1292 | if (ec) |
| 1293 | { |
| 1294 | std::cerr |
| 1295 | << "smi handler error: " << ec.message() |
| 1296 | << "\n"; |
| 1297 | return; |
| 1298 | } |
| 1299 | smiHandler(); |
| 1300 | }); |
| 1301 | } |
| 1302 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1303 | static void initializeErrorState() |
| 1304 | { |
| 1305 | // Handle CPU_CATERR if it's asserted now |
| 1306 | if (caterrLine.get_value() == 0) |
| 1307 | { |
| 1308 | caterrAssertHandler(); |
| 1309 | } |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1310 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1311 | // Handle CPU_ERR0 if it's asserted now |
| 1312 | if (err0Line.get_value() == 0) |
| 1313 | { |
| 1314 | err0AssertHandler(); |
| 1315 | } |
| 1316 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1317 | // Handle CPU_ERR1 if it's asserted now |
| 1318 | if (err1Line.get_value() == 0) |
| 1319 | { |
| 1320 | err1AssertHandler(); |
| 1321 | } |
| 1322 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1323 | // Handle CPU_ERR2 if it's asserted now |
| 1324 | if (err2Line.get_value() == 0) |
| 1325 | { |
| 1326 | err2AssertHandler(); |
| 1327 | } |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1328 | |
| 1329 | // Handle SMI if it's asserted now |
| 1330 | if (smiLine.get_value() == 0) |
| 1331 | { |
| 1332 | smiAssertHandler(); |
| 1333 | } |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 1334 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 1335 | // Handle CPU1_THERMTRIP if it's asserted now |
| 1336 | if (cpu1ThermtripLine.get_value() == 0) |
| 1337 | { |
| 1338 | cpu1ThermtripAssertHandler(); |
| 1339 | } |
| 1340 | |
| 1341 | // Handle CPU2_THERMTRIP if it's asserted now |
| 1342 | if (cpu2ThermtripLine.get_value() == 0) |
| 1343 | { |
| 1344 | cpu2ThermtripAssertHandler(); |
| 1345 | } |
| 1346 | |
| 1347 | // Handle CPU1_VRHOT if it's asserted now |
| 1348 | if (cpu1VRHotLine.get_value() == 0) |
| 1349 | { |
| 1350 | cpu1VRHotAssertHandler(); |
| 1351 | } |
| 1352 | |
| 1353 | // Handle CPU1_MEM_ABCD_VRHOT if it's asserted now |
| 1354 | if (cpu1MemABCDVRHotLine.get_value() == 0) |
| 1355 | { |
| 1356 | cpu1MemABCDVRHotAssertHandler(); |
| 1357 | } |
| 1358 | |
| 1359 | // Handle CPU1_MEM_EFGH_VRHOT if it's asserted now |
| 1360 | if (cpu1MemEFGHVRHotLine.get_value() == 0) |
| 1361 | { |
| 1362 | cpu1MemEFGHVRHotAssertHandler(); |
| 1363 | } |
| 1364 | |
| 1365 | // Handle CPU2_VRHOT if it's asserted now |
| 1366 | if (cpu2VRHotLine.get_value() == 0) |
| 1367 | { |
| 1368 | cpu2VRHotAssertHandler(); |
| 1369 | } |
| 1370 | |
| 1371 | // Handle CPU2_MEM_ABCD_VRHOT if it's asserted now |
| 1372 | if (cpu2MemABCDVRHotLine.get_value() == 0) |
| 1373 | { |
| 1374 | cpu2MemABCDVRHotAssertHandler(); |
| 1375 | } |
| 1376 | |
| 1377 | // Handle CPU2_MEM_EFGH_VRHOT if it's asserted now |
| 1378 | if (cpu2MemEFGHVRHotLine.get_value() == 0) |
| 1379 | { |
| 1380 | cpu2MemEFGHVRHotAssertHandler(); |
| 1381 | } |
| 1382 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 1383 | // Handle PCH_BMC_THERMTRIP if it's asserted now |
| 1384 | if (pchThermtripLine.get_value() == 0) |
| 1385 | { |
| 1386 | ssbThermTripLog(); |
| 1387 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1388 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1389 | } // namespace host_error_monitor |
| 1390 | |
| 1391 | int main(int argc, char* argv[]) |
| 1392 | { |
| 1393 | // setup connection to dbus |
| 1394 | host_error_monitor::conn = |
| 1395 | std::make_shared<sdbusplus::asio::connection>(host_error_monitor::io); |
| 1396 | |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame^] | 1397 | // Host Error Monitor Service |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1398 | host_error_monitor::conn->request_name( |
| 1399 | "xyz.openbmc_project.HostErrorMonitor"); |
| 1400 | sdbusplus::asio::object_server server = |
| 1401 | sdbusplus::asio::object_server(host_error_monitor::conn); |
| 1402 | |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame^] | 1403 | // Restart Cause Interface |
| 1404 | host_error_monitor::hostErrorTimeoutIface = |
| 1405 | server.add_interface("/xyz/openbmc_project/host_error_monitor", |
| 1406 | "xyz.openbmc_project.HostErrorMonitor.Timeout"); |
| 1407 | |
| 1408 | host_error_monitor::hostErrorTimeoutIface->register_property( |
| 1409 | "IERRTimeoutMs", host_error_monitor::caterrTimeoutMs, |
| 1410 | [](const std::size_t& requested, std::size_t& resp) { |
| 1411 | if (requested > host_error_monitor::caterrTimeoutMsMax) |
| 1412 | { |
| 1413 | std::cerr << "IERRTimeoutMs update to " << requested |
| 1414 | << "ms rejected. Cannot be greater than " |
| 1415 | << host_error_monitor::caterrTimeoutMsMax << "ms.\n"; |
| 1416 | return 0; |
| 1417 | } |
| 1418 | std::cerr << "IERRTimeoutMs updated to " << requested << "ms\n"; |
| 1419 | host_error_monitor::caterrTimeoutMs = requested; |
| 1420 | resp = requested; |
| 1421 | return 1; |
| 1422 | }, |
| 1423 | [](std::size_t& resp) { return host_error_monitor::caterrTimeoutMs; }); |
| 1424 | host_error_monitor::hostErrorTimeoutIface->initialize(); |
| 1425 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1426 | // Start tracking host state |
| 1427 | std::shared_ptr<sdbusplus::bus::match::match> hostStateMonitor = |
| 1428 | host_error_monitor::startHostStateMonitor(); |
| 1429 | |
| 1430 | // Initialize the host state |
| 1431 | host_error_monitor::initializeHostState(); |
| 1432 | |
| 1433 | // Request CPU_CATERR GPIO events |
| 1434 | if (!host_error_monitor::requestGPIOEvents( |
| 1435 | "CPU_CATERR", host_error_monitor::caterrHandler, |
| 1436 | host_error_monitor::caterrLine, host_error_monitor::caterrEvent)) |
| 1437 | { |
| 1438 | return -1; |
| 1439 | } |
| 1440 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1441 | // Request CPU_ERR0 GPIO events |
| 1442 | if (!host_error_monitor::requestGPIOEvents( |
| 1443 | "CPU_ERR0", host_error_monitor::err0Handler, |
| 1444 | host_error_monitor::err0Line, host_error_monitor::err0Event)) |
| 1445 | { |
| 1446 | return -1; |
| 1447 | } |
| 1448 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1449 | // Request CPU_ERR1 GPIO events |
| 1450 | if (!host_error_monitor::requestGPIOEvents( |
| 1451 | "CPU_ERR1", host_error_monitor::err1Handler, |
| 1452 | host_error_monitor::err1Line, host_error_monitor::err1Event)) |
| 1453 | { |
| 1454 | return -1; |
| 1455 | } |
| 1456 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1457 | // Request CPU_ERR2 GPIO events |
| 1458 | if (!host_error_monitor::requestGPIOEvents( |
| 1459 | "CPU_ERR2", host_error_monitor::err2Handler, |
| 1460 | host_error_monitor::err2Line, host_error_monitor::err2Event)) |
| 1461 | { |
| 1462 | return -1; |
| 1463 | } |
| 1464 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1465 | // Request SMI GPIO events |
| 1466 | if (!host_error_monitor::requestGPIOEvents( |
| 1467 | "SMI", host_error_monitor::smiHandler, host_error_monitor::smiLine, |
| 1468 | host_error_monitor::smiEvent)) |
| 1469 | { |
| 1470 | return -1; |
| 1471 | } |
| 1472 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 1473 | // Request CPU1_FIVR_FAULT GPIO input |
| 1474 | if (!host_error_monitor::requestGPIOInput( |
| 1475 | "CPU1_FIVR_FAULT", host_error_monitor::cpu1FIVRFaultLine)) |
| 1476 | { |
| 1477 | return -1; |
| 1478 | } |
| 1479 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 1480 | // Request CPU1_THERMTRIP GPIO events |
| 1481 | if (!host_error_monitor::requestGPIOEvents( |
| 1482 | "CPU1_THERMTRIP", host_error_monitor::cpu1ThermtripHandler, |
| 1483 | host_error_monitor::cpu1ThermtripLine, |
| 1484 | host_error_monitor::cpu1ThermtripEvent)) |
| 1485 | { |
| 1486 | return -1; |
| 1487 | } |
| 1488 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 1489 | // Request CPU2_FIVR_FAULT GPIO input |
| 1490 | if (!host_error_monitor::requestGPIOInput( |
| 1491 | "CPU2_FIVR_FAULT", host_error_monitor::cpu2FIVRFaultLine)) |
| 1492 | { |
| 1493 | return -1; |
| 1494 | } |
| 1495 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 1496 | // Request CPU2_THERMTRIP GPIO events |
| 1497 | if (!host_error_monitor::requestGPIOEvents( |
| 1498 | "CPU2_THERMTRIP", host_error_monitor::cpu2ThermtripHandler, |
| 1499 | host_error_monitor::cpu2ThermtripLine, |
| 1500 | host_error_monitor::cpu2ThermtripEvent)) |
| 1501 | { |
| 1502 | return -1; |
| 1503 | } |
| 1504 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1505 | // Request CPU1_VRHOT GPIO events |
| 1506 | if (!host_error_monitor::requestGPIOEvents( |
| 1507 | "CPU1_VRHOT", host_error_monitor::cpu1VRHotHandler, |
| 1508 | host_error_monitor::cpu1VRHotLine, |
| 1509 | host_error_monitor::cpu1VRHotEvent)) |
| 1510 | { |
| 1511 | return -1; |
| 1512 | } |
| 1513 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1514 | // Request CPU1_MEM_ABCD_VRHOT GPIO events |
| 1515 | if (!host_error_monitor::requestGPIOEvents( |
| 1516 | "CPU1_MEM_ABCD_VRHOT", host_error_monitor::cpu1MemABCDVRHotHandler, |
| 1517 | host_error_monitor::cpu1MemABCDVRHotLine, |
| 1518 | host_error_monitor::cpu1MemABCDVRHotEvent)) |
| 1519 | { |
| 1520 | return -1; |
| 1521 | } |
| 1522 | |
| 1523 | // Request CPU1_MEM_EFGH_VRHOT GPIO events |
| 1524 | if (!host_error_monitor::requestGPIOEvents( |
| 1525 | "CPU1_MEM_EFGH_VRHOT", host_error_monitor::cpu1MemEFGHVRHotHandler, |
| 1526 | host_error_monitor::cpu1MemEFGHVRHotLine, |
| 1527 | host_error_monitor::cpu1MemEFGHVRHotEvent)) |
| 1528 | { |
| 1529 | return -1; |
| 1530 | } |
| 1531 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1532 | // Request CPU2_VRHOT GPIO events |
| 1533 | if (!host_error_monitor::requestGPIOEvents( |
| 1534 | "CPU2_VRHOT", host_error_monitor::cpu2VRHotHandler, |
| 1535 | host_error_monitor::cpu2VRHotLine, |
| 1536 | host_error_monitor::cpu2VRHotEvent)) |
| 1537 | { |
| 1538 | return -1; |
| 1539 | } |
| 1540 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1541 | // Request CPU2_MEM_ABCD_VRHOT GPIO events |
| 1542 | if (!host_error_monitor::requestGPIOEvents( |
| 1543 | "CPU2_MEM_ABCD_VRHOT", host_error_monitor::cpu2MemABCDVRHotHandler, |
| 1544 | host_error_monitor::cpu2MemABCDVRHotLine, |
| 1545 | host_error_monitor::cpu2MemABCDVRHotEvent)) |
| 1546 | { |
| 1547 | return -1; |
| 1548 | } |
| 1549 | |
| 1550 | // Request CPU2_MEM_EFGH_VRHOT GPIO events |
| 1551 | if (!host_error_monitor::requestGPIOEvents( |
| 1552 | "CPU2_MEM_EFGH_VRHOT", host_error_monitor::cpu2MemEFGHVRHotHandler, |
| 1553 | host_error_monitor::cpu2MemEFGHVRHotLine, |
| 1554 | host_error_monitor::cpu2MemEFGHVRHotEvent)) |
| 1555 | { |
| 1556 | return -1; |
| 1557 | } |
| 1558 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1559 | // Request PCH_BMC_THERMTRIP GPIO events |
| 1560 | if (!host_error_monitor::requestGPIOEvents( |
| 1561 | "PCH_BMC_THERMTRIP", host_error_monitor::pchThermtripHandler, |
| 1562 | host_error_monitor::pchThermtripLine, |
| 1563 | host_error_monitor::pchThermtripEvent)) |
| 1564 | { |
| 1565 | return -1; |
| 1566 | } |
| 1567 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1568 | host_error_monitor::io.run(); |
| 1569 | |
| 1570 | return 0; |
| 1571 | } |