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); |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 86 | //---------------------------------- |
| 87 | // CPU_MEM_THERM_EVENT function related definition |
| 88 | //---------------------------------- |
| 89 | static gpiod::line cpu1MemtripLine; |
| 90 | static boost::asio::posix::stream_descriptor cpu1MemtripEvent(io); |
| 91 | static gpiod::line cpu2MemtripLine; |
| 92 | static boost::asio::posix::stream_descriptor cpu2MemtripEvent(io); |
jayaprakash Mutyala | 53099c4 | 2020-03-15 00:16:26 +0000 | [diff] [blame] | 93 | //--------------------------------- |
| 94 | // CPU_MISMATCH function related definition |
| 95 | //--------------------------------- |
| 96 | static gpiod::line cpu1MismatchLine; |
| 97 | static gpiod::line cpu2MismatchLine; |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 98 | |
Yong Li | 061eb03 | 2020-02-26 15:06:18 +0800 | [diff] [blame] | 99 | // beep function for CPU error |
| 100 | const static constexpr uint8_t beepCPUErr2 = 5; |
| 101 | |
| 102 | static void beep(const uint8_t& beepPriority) |
| 103 | { |
| 104 | conn->async_method_call( |
| 105 | [](boost::system::error_code ec) { |
| 106 | if (ec) |
| 107 | { |
| 108 | std::cerr << "beep returned error with " |
| 109 | "async_method_call (ec = " |
| 110 | << ec << ")\n"; |
| 111 | return; |
| 112 | } |
| 113 | }, |
| 114 | "xyz.openbmc_project.BeepCode", "/xyz/openbmc_project/BeepCode", |
| 115 | "xyz.openbmc_project.BeepCode", "Beep", uint8_t(beepPriority)); |
| 116 | } |
| 117 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 118 | static void cpuIERRLog() |
| 119 | { |
| 120 | sd_journal_send("MESSAGE=HostError: IERR", "PRIORITY=%i", LOG_INFO, |
| 121 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 122 | "REDFISH_MESSAGE_ARGS=%s", "IERR", NULL); |
| 123 | } |
| 124 | |
| 125 | static void cpuIERRLog(const int cpuNum) |
| 126 | { |
| 127 | std::string msg = "IERR on CPU " + std::to_string(cpuNum + 1); |
| 128 | |
| 129 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 130 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 131 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 132 | } |
| 133 | |
| 134 | static void cpuIERRLog(const int cpuNum, const std::string& type) |
| 135 | { |
| 136 | std::string msg = type + " IERR on CPU " + std::to_string(cpuNum + 1); |
| 137 | |
| 138 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 139 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 140 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 141 | } |
| 142 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 143 | static void cpuERRXLog(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 144 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 145 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout"; |
| 146 | |
| 147 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 148 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 149 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 150 | } |
| 151 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 152 | static void cpuERRXLog(const int errPin, const int cpuNum) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 153 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 154 | std::string msg = "ERR" + std::to_string(errPin) + " Timeout on CPU " + |
| 155 | std::to_string(cpuNum + 1); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 156 | |
| 157 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 158 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 159 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 160 | } |
| 161 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 162 | static void smiTimeoutLog() |
| 163 | { |
| 164 | sd_journal_send("MESSAGE=HostError: SMI Timeout", "PRIORITY=%i", LOG_INFO, |
| 165 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 166 | "REDFISH_MESSAGE_ARGS=%s", "SMI Timeout", NULL); |
| 167 | } |
| 168 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 169 | static void cpuBootFIVRFaultLog(const int cpuNum) |
| 170 | { |
| 171 | std::string msg = "Boot FIVR Fault on CPU " + std::to_string(cpuNum); |
| 172 | |
| 173 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 174 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUError", |
| 175 | "REDFISH_MESSAGE_ARGS=%s", msg.c_str(), NULL); |
| 176 | } |
| 177 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 178 | static void cpuThermTripLog(const int cpuNum) |
| 179 | { |
| 180 | std::string msg = "CPU " + std::to_string(cpuNum) + " thermal trip"; |
| 181 | |
| 182 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 183 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 184 | "OpenBMC.0.1.CPUThermalTrip", "REDFISH_MESSAGE_ARGS=%d", |
| 185 | cpuNum, NULL); |
| 186 | } |
| 187 | |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 188 | static void memThermTripLog(const int cpuNum) |
| 189 | { |
| 190 | std::string cpuNumber = "CPU " + std::to_string(cpuNum); |
| 191 | std::string msg = cpuNumber + " Memory Thermal trip."; |
| 192 | |
| 193 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 194 | LOG_ERR, "REDFISH_MESSAGE_ID=%s", |
| 195 | "OpenBMC.0.1.MemoryThermTrip", "REDFISH_MESSAGE_ARGS=%s", |
| 196 | cpuNumber.c_str(), NULL); |
| 197 | } |
| 198 | |
jayaprakash Mutyala | 53099c4 | 2020-03-15 00:16:26 +0000 | [diff] [blame] | 199 | static void cpuMismatchLog(const int cpuNum) |
| 200 | { |
| 201 | std::string msg = "CPU " + std::to_string(cpuNum) + " mismatch"; |
| 202 | |
| 203 | sd_journal_send("MESSAGE= %s", msg.c_str(), "PRIORITY=%i", LOG_ERR, |
| 204 | "REDFISH_MESSAGE_ID=%s", "OpenBMC.0.1.CPUMismatch", |
| 205 | "REDFISH_MESSAGE_ARGS=%d", cpuNum, NULL); |
| 206 | } |
| 207 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 208 | static void cpuVRHotLog(const std::string& vr) |
| 209 | { |
| 210 | std::string msg = vr + " Voltage Regulator Overheated."; |
| 211 | |
| 212 | sd_journal_send("MESSAGE=HostError: %s", msg.c_str(), "PRIORITY=%i", |
| 213 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 214 | "OpenBMC.0.1.VoltageRegulatorOverheated", |
| 215 | "REDFISH_MESSAGE_ARGS=%s", vr.c_str(), NULL); |
| 216 | } |
| 217 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 218 | static void ssbThermTripLog() |
| 219 | { |
| 220 | sd_journal_send("MESSAGE=HostError: SSB thermal trip", "PRIORITY=%i", |
| 221 | LOG_INFO, "REDFISH_MESSAGE_ID=%s", |
| 222 | "OpenBMC.0.1.SsbThermalTrip", NULL); |
| 223 | } |
| 224 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 225 | static void initializeErrorState(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 226 | static void initializeHostState() |
| 227 | { |
| 228 | conn->async_method_call( |
| 229 | [](boost::system::error_code ec, |
| 230 | const std::variant<std::string>& property) { |
| 231 | if (ec) |
| 232 | { |
| 233 | return; |
| 234 | } |
| 235 | const std::string* state = std::get_if<std::string>(&property); |
| 236 | if (state == nullptr) |
| 237 | { |
| 238 | std::cerr << "Unable to read host state value\n"; |
| 239 | return; |
| 240 | } |
| 241 | hostOff = *state == "xyz.openbmc_project.State.Host.HostState.Off"; |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 242 | // If the system is on, initialize the error state |
| 243 | if (!hostOff) |
| 244 | { |
| 245 | initializeErrorState(); |
| 246 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 247 | }, |
| 248 | "xyz.openbmc_project.State.Host", "/xyz/openbmc_project/state/host0", |
| 249 | "org.freedesktop.DBus.Properties", "Get", |
| 250 | "xyz.openbmc_project.State.Host", "CurrentHostState"); |
| 251 | } |
| 252 | |
| 253 | static std::shared_ptr<sdbusplus::bus::match::match> startHostStateMonitor() |
| 254 | { |
| 255 | return std::make_shared<sdbusplus::bus::match::match>( |
| 256 | *conn, |
| 257 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 258 | "member='PropertiesChanged',arg0namespace='xyz.openbmc_project.State." |
| 259 | "Host'", |
| 260 | [](sdbusplus::message::message& msg) { |
| 261 | std::string interfaceName; |
| 262 | boost::container::flat_map<std::string, std::variant<std::string>> |
| 263 | propertiesChanged; |
| 264 | std::string state; |
| 265 | try |
| 266 | { |
| 267 | msg.read(interfaceName, propertiesChanged); |
| 268 | state = |
| 269 | std::get<std::string>(propertiesChanged.begin()->second); |
| 270 | } |
| 271 | catch (std::exception& e) |
| 272 | { |
| 273 | std::cerr << "Unable to read host state\n"; |
| 274 | return; |
| 275 | } |
| 276 | hostOff = state == "xyz.openbmc_project.State.Host.HostState.Off"; |
| 277 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 278 | if (hostOff) |
| 279 | { |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 280 | // No host events should fire while off, so cancel any pending |
| 281 | // timers |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 282 | caterrAssertTimer.cancel(); |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 283 | err0AssertTimer.cancel(); |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 284 | err1AssertTimer.cancel(); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 285 | err2AssertTimer.cancel(); |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 286 | smiAssertTimer.cancel(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 287 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 288 | else |
| 289 | { |
| 290 | // Handle any initial errors when the host turns on |
| 291 | initializeErrorState(); |
| 292 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 293 | }); |
| 294 | } |
| 295 | |
| 296 | static bool requestGPIOEvents( |
| 297 | const std::string& name, const std::function<void()>& handler, |
| 298 | gpiod::line& gpioLine, |
| 299 | boost::asio::posix::stream_descriptor& gpioEventDescriptor) |
| 300 | { |
| 301 | // Find the GPIO line |
| 302 | gpioLine = gpiod::find_line(name); |
| 303 | if (!gpioLine) |
| 304 | { |
| 305 | std::cerr << "Failed to find the " << name << " line\n"; |
| 306 | return false; |
| 307 | } |
| 308 | |
| 309 | try |
| 310 | { |
| 311 | gpioLine.request( |
| 312 | {"host-error-monitor", gpiod::line_request::EVENT_BOTH_EDGES}); |
| 313 | } |
| 314 | catch (std::exception&) |
| 315 | { |
| 316 | std::cerr << "Failed to request events for " << name << "\n"; |
| 317 | return false; |
| 318 | } |
| 319 | |
| 320 | int gpioLineFd = gpioLine.event_get_fd(); |
| 321 | if (gpioLineFd < 0) |
| 322 | { |
| 323 | std::cerr << "Failed to get " << name << " fd\n"; |
| 324 | return false; |
| 325 | } |
| 326 | |
| 327 | gpioEventDescriptor.assign(gpioLineFd); |
| 328 | |
| 329 | gpioEventDescriptor.async_wait( |
| 330 | boost::asio::posix::stream_descriptor::wait_read, |
| 331 | [&name, handler](const boost::system::error_code ec) { |
| 332 | if (ec) |
| 333 | { |
| 334 | std::cerr << name << " fd handler error: " << ec.message() |
| 335 | << "\n"; |
| 336 | return; |
| 337 | } |
| 338 | handler(); |
| 339 | }); |
| 340 | return true; |
| 341 | } |
| 342 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 343 | static bool requestGPIOInput(const std::string& name, gpiod::line& gpioLine) |
| 344 | { |
| 345 | // Find the GPIO line |
| 346 | gpioLine = gpiod::find_line(name); |
| 347 | if (!gpioLine) |
| 348 | { |
| 349 | std::cerr << "Failed to find the " << name << " line.\n"; |
| 350 | return false; |
| 351 | } |
| 352 | |
| 353 | // Request GPIO input |
| 354 | try |
| 355 | { |
| 356 | gpioLine.request({__FUNCTION__, gpiod::line_request::DIRECTION_INPUT}); |
| 357 | } |
| 358 | catch (std::exception&) |
| 359 | { |
| 360 | std::cerr << "Failed to request " << name << " input\n"; |
| 361 | return false; |
| 362 | } |
| 363 | |
| 364 | return true; |
| 365 | } |
| 366 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 367 | static void startPowerCycle() |
| 368 | { |
| 369 | conn->async_method_call( |
| 370 | [](boost::system::error_code ec) { |
| 371 | if (ec) |
| 372 | { |
| 373 | std::cerr << "failed to set Chassis State\n"; |
| 374 | } |
| 375 | }, |
| 376 | "xyz.openbmc_project.State.Chassis", |
| 377 | "/xyz/openbmc_project/state/chassis0", |
| 378 | "org.freedesktop.DBus.Properties", "Set", |
| 379 | "xyz.openbmc_project.State.Chassis", "RequestedPowerTransition", |
| 380 | std::variant<std::string>{ |
| 381 | "xyz.openbmc_project.State.Chassis.Transition.PowerCycle"}); |
| 382 | } |
| 383 | |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 384 | static void startCrashdumpAndRecovery(bool recoverSystem, |
| 385 | const std::string& triggerType) |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 386 | { |
| 387 | std::cout << "Starting crashdump\n"; |
| 388 | static std::shared_ptr<sdbusplus::bus::match::match> crashdumpCompleteMatch; |
| 389 | static boost::asio::steady_timer crashdumpTimer(io); |
| 390 | |
| 391 | crashdumpCompleteMatch = std::make_shared<sdbusplus::bus::match::match>( |
| 392 | *conn, |
| 393 | "type='signal',interface='org.freedesktop.DBus.Properties'," |
| 394 | "member='PropertiesChanged',arg0namespace='com.intel.crashdump'", |
| 395 | [recoverSystem](sdbusplus::message::message& msg) { |
| 396 | crashdumpTimer.cancel(); |
| 397 | std::cout << "Crashdump completed\n"; |
| 398 | if (recoverSystem) |
| 399 | { |
| 400 | std::cout << "Recovering the system\n"; |
| 401 | startPowerCycle(); |
| 402 | } |
| 403 | crashdumpCompleteMatch.reset(); |
| 404 | }); |
| 405 | |
| 406 | crashdumpTimer.expires_after(std::chrono::seconds(crashdumpTimeoutS)); |
| 407 | crashdumpTimer.async_wait([](const boost::system::error_code ec) { |
| 408 | if (ec) |
| 409 | { |
| 410 | // operation_aborted is expected if timer is canceled |
| 411 | if (ec != boost::asio::error::operation_aborted) |
| 412 | { |
| 413 | std::cerr << "Crashdump async_wait failed: " << ec.message() |
| 414 | << "\n"; |
| 415 | } |
| 416 | std::cout << "Crashdump timer canceled\n"; |
| 417 | return; |
| 418 | } |
| 419 | std::cerr << "Crashdump failed to complete before timeout\n"; |
| 420 | crashdumpCompleteMatch.reset(); |
| 421 | }); |
| 422 | |
| 423 | conn->async_method_call( |
| 424 | [](boost::system::error_code ec) { |
| 425 | if (ec) |
| 426 | { |
| 427 | std::cerr << "failed to start Crashdump\n"; |
| 428 | crashdumpTimer.cancel(); |
| 429 | crashdumpCompleteMatch.reset(); |
| 430 | } |
| 431 | }, |
| 432 | "com.intel.crashdump", "/com/intel/crashdump", |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 433 | "com.intel.crashdump.Stored", "GenerateStoredLog", triggerType); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 434 | } |
| 435 | |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 436 | static void incrementCPUErrorCount(int cpuNum) |
| 437 | { |
| 438 | std::string propertyName = "ErrorCountCPU" + std::to_string(cpuNum + 1); |
| 439 | |
| 440 | // Get the current count |
| 441 | conn->async_method_call( |
| 442 | [propertyName](boost::system::error_code ec, |
| 443 | const std::variant<uint8_t>& property) { |
| 444 | if (ec) |
| 445 | { |
| 446 | std::cerr << "Failed to read " << propertyName << ": " |
| 447 | << ec.message() << "\n"; |
| 448 | return; |
| 449 | } |
| 450 | const uint8_t* errorCountVariant = std::get_if<uint8_t>(&property); |
| 451 | if (errorCountVariant == nullptr) |
| 452 | { |
| 453 | std::cerr << propertyName << " invalid\n"; |
| 454 | return; |
| 455 | } |
| 456 | uint8_t errorCount = *errorCountVariant; |
| 457 | if (errorCount == std::numeric_limits<uint8_t>::max()) |
| 458 | { |
| 459 | std::cerr << "Maximum error count reached\n"; |
| 460 | return; |
| 461 | } |
| 462 | // Increment the count |
| 463 | errorCount++; |
| 464 | conn->async_method_call( |
| 465 | [propertyName](boost::system::error_code ec) { |
| 466 | if (ec) |
| 467 | { |
| 468 | std::cerr << "Failed to set " << propertyName << ": " |
| 469 | << ec.message() << "\n"; |
| 470 | } |
| 471 | }, |
| 472 | "xyz.openbmc_project.Settings", |
| 473 | "/xyz/openbmc_project/control/processor_error_config", |
| 474 | "org.freedesktop.DBus.Properties", "Set", |
| 475 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName, |
| 476 | std::variant<uint8_t>{errorCount}); |
| 477 | }, |
| 478 | "xyz.openbmc_project.Settings", |
| 479 | "/xyz/openbmc_project/control/processor_error_config", |
| 480 | "org.freedesktop.DBus.Properties", "Get", |
| 481 | "xyz.openbmc_project.Control.Processor.ErrConfig", propertyName); |
| 482 | } |
| 483 | |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 484 | static bool checkIERRCPUs() |
| 485 | { |
| 486 | bool cpuIERRFound = false; |
| 487 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 488 | cpu++, addr++) |
| 489 | { |
| 490 | uint8_t cc = 0; |
| 491 | CPUModel model{}; |
| 492 | uint8_t stepping = 0; |
| 493 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 494 | { |
| 495 | std::cerr << "Cannot get CPUID!\n"; |
| 496 | continue; |
| 497 | } |
| 498 | |
| 499 | switch (model) |
| 500 | { |
| 501 | case skx: |
| 502 | { |
| 503 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 504 | // that caused the IERR |
| 505 | uint32_t mcaErrSrcLog = 0; |
| 506 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 507 | &cc) != PECI_CC_SUCCESS) |
| 508 | { |
| 509 | continue; |
| 510 | } |
| 511 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 512 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 513 | { |
| 514 | // TODO: Light the CPU fault LED? |
| 515 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 516 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 517 | // Next check if it's a CPU/VR mismatch by reading the |
| 518 | // IA32_MC4_STATUS MSR (0x411) |
| 519 | uint64_t mc4Status = 0; |
| 520 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 521 | PECI_CC_SUCCESS) |
| 522 | { |
| 523 | continue; |
| 524 | } |
| 525 | // Check MSEC bits 31:24 for |
| 526 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 527 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 528 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 529 | if ((mc4Status & (0x40 << 24)) || |
| 530 | (mc4Status & (0x42 << 24)) || |
| 531 | (mc4Status & (0x43 << 24))) |
| 532 | { |
| 533 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 534 | continue; |
| 535 | } |
| 536 | |
| 537 | // Next check if it's a Core FIVR fault by looking for a |
| 538 | // non-zero value of CORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 539 | // 80h) |
| 540 | uint32_t coreFIVRErrLog = 0; |
| 541 | if (peci_RdPCIConfigLocal( |
| 542 | addr, 1, 30, 2, 0x80, sizeof(uint32_t), |
| 543 | (uint8_t*)&coreFIVRErrLog, &cc) != PECI_CC_SUCCESS) |
| 544 | { |
| 545 | continue; |
| 546 | } |
| 547 | if (coreFIVRErrLog) |
| 548 | { |
| 549 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 550 | continue; |
| 551 | } |
| 552 | |
| 553 | // Next check if it's an Uncore FIVR fault by looking for a |
| 554 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(1) D30 F2 offset |
| 555 | // 84h) |
| 556 | uint32_t uncoreFIVRErrLog = 0; |
| 557 | if (peci_RdPCIConfigLocal(addr, 1, 30, 2, 0x84, |
| 558 | sizeof(uint32_t), |
| 559 | (uint8_t*)&uncoreFIVRErrLog, |
| 560 | &cc) != PECI_CC_SUCCESS) |
| 561 | { |
| 562 | continue; |
| 563 | } |
| 564 | if (uncoreFIVRErrLog) |
| 565 | { |
| 566 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 567 | continue; |
| 568 | } |
| 569 | |
| 570 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 571 | // both zero, but MSEC bits 31:24 have either |
| 572 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 573 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 574 | // uncore FIVR fault |
| 575 | if (!coreFIVRErrLog && !uncoreFIVRErrLog && |
| 576 | ((mc4Status & (0x51 << 24)) || |
| 577 | (mc4Status & (0x52 << 24)))) |
| 578 | { |
| 579 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 580 | continue; |
| 581 | } |
| 582 | cpuIERRLog(cpu); |
| 583 | } |
| 584 | break; |
| 585 | } |
| 586 | case icx: |
| 587 | { |
| 588 | // First check the MCA_ERR_SRC_LOG to see if this is the CPU |
| 589 | // that caused the IERR |
| 590 | uint32_t mcaErrSrcLog = 0; |
| 591 | if (peci_RdPkgConfig(addr, 0, 5, 4, (uint8_t*)&mcaErrSrcLog, |
| 592 | &cc) != PECI_CC_SUCCESS) |
| 593 | { |
| 594 | continue; |
| 595 | } |
| 596 | // Check MSMI_INTERNAL (20) and IERR_INTERNAL (27) |
| 597 | if ((mcaErrSrcLog & (1 << 20)) || (mcaErrSrcLog & (1 << 27))) |
| 598 | { |
| 599 | // TODO: Light the CPU fault LED? |
| 600 | cpuIERRFound = true; |
Jason M. Bills | d1a19f6 | 2019-08-06 11:52:58 -0700 | [diff] [blame] | 601 | incrementCPUErrorCount(cpu); |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 602 | // Next check if it's a CPU/VR mismatch by reading the |
| 603 | // IA32_MC4_STATUS MSR (0x411) |
| 604 | uint64_t mc4Status = 0; |
| 605 | if (peci_RdIAMSR(addr, 0, 0x411, &mc4Status, &cc) != |
| 606 | PECI_CC_SUCCESS) |
| 607 | { |
| 608 | continue; |
| 609 | } |
| 610 | // TODO: Update MSEC/MSCOD_31_24 check |
| 611 | // Check MSEC bits 31:24 for |
| 612 | // MCA_SVID_VCCIN_VR_ICC_MAX_FAILURE (0x40), |
| 613 | // MCA_SVID_VCCIN_VR_VOUT_FAILURE (0x42), or |
| 614 | // MCA_SVID_CPU_VR_CAPABILITY_ERROR (0x43) |
| 615 | if ((mc4Status & (0x40 << 24)) || |
| 616 | (mc4Status & (0x42 << 24)) || |
| 617 | (mc4Status & (0x43 << 24))) |
| 618 | { |
| 619 | cpuIERRLog(cpu, "CPU/VR Mismatch"); |
| 620 | continue; |
| 621 | } |
| 622 | |
| 623 | // Next check if it's a Core FIVR fault by looking for a |
| 624 | // non-zero value of CORE_FIVR_ERR_LOG (B(31) D30 F2 offsets |
| 625 | // C0h and C4h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 626 | uint32_t coreFIVRErrLog0 = 0; |
| 627 | uint32_t coreFIVRErrLog1 = 0; |
| 628 | if (peci_RdEndPointConfigPciLocal( |
| 629 | addr, 0, 14, 30, 2, 0xC0, sizeof(uint32_t), |
| 630 | (uint8_t*)&coreFIVRErrLog0, &cc) != PECI_CC_SUCCESS) |
| 631 | { |
| 632 | continue; |
| 633 | } |
| 634 | if (peci_RdEndPointConfigPciLocal( |
| 635 | addr, 0, 14, 30, 2, 0xC4, sizeof(uint32_t), |
| 636 | (uint8_t*)&coreFIVRErrLog1, &cc) != PECI_CC_SUCCESS) |
| 637 | { |
| 638 | continue; |
| 639 | } |
| 640 | if (coreFIVRErrLog0 || coreFIVRErrLog1) |
| 641 | { |
| 642 | cpuIERRLog(cpu, "Core FIVR Fault"); |
| 643 | continue; |
| 644 | } |
| 645 | |
| 646 | // Next check if it's an Uncore FIVR fault by looking for a |
| 647 | // non-zero value of UNCORE_FIVR_ERR_LOG (B(31) D30 F2 |
| 648 | // offset 84h) (Note: Bus 31 is accessed on PECI as bus 14) |
| 649 | uint32_t uncoreFIVRErrLog = 0; |
| 650 | if (peci_RdEndPointConfigPciLocal( |
| 651 | addr, 0, 14, 30, 2, 0x84, sizeof(uint32_t), |
| 652 | (uint8_t*)&uncoreFIVRErrLog, |
| 653 | &cc) != PECI_CC_SUCCESS) |
| 654 | { |
| 655 | continue; |
| 656 | } |
| 657 | if (uncoreFIVRErrLog) |
| 658 | { |
| 659 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 660 | continue; |
| 661 | } |
| 662 | |
| 663 | // TODO: Update MSEC/MSCOD_31_24 check |
| 664 | // Last if CORE_FIVR_ERR_LOG and UNCORE_FIVR_ERR_LOG are |
| 665 | // both zero, but MSEC bits 31:24 have either |
| 666 | // MCA_FIVR_CATAS_OVERVOL_FAULT (0x51) or |
| 667 | // MCA_FIVR_CATAS_OVERCUR_FAULT (0x52), then log it as an |
| 668 | // uncore FIVR fault |
| 669 | if (!coreFIVRErrLog0 && !coreFIVRErrLog1 && |
| 670 | !uncoreFIVRErrLog && |
| 671 | ((mc4Status & (0x51 << 24)) || |
| 672 | (mc4Status & (0x52 << 24)))) |
| 673 | { |
| 674 | cpuIERRLog(cpu, "Uncore FIVR Fault"); |
| 675 | continue; |
| 676 | } |
| 677 | cpuIERRLog(cpu); |
| 678 | } |
| 679 | break; |
| 680 | } |
| 681 | } |
| 682 | } |
| 683 | return cpuIERRFound; |
| 684 | } |
| 685 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 686 | static void caterrAssertHandler() |
| 687 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 688 | caterrAssertTimer.expires_after(std::chrono::milliseconds(caterrTimeoutMs)); |
| 689 | caterrAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 690 | if (ec) |
| 691 | { |
| 692 | // operation_aborted is expected if timer is canceled |
| 693 | // before completion. |
| 694 | if (ec != boost::asio::error::operation_aborted) |
| 695 | { |
| 696 | std::cerr << "caterr timeout async_wait failed: " |
| 697 | << ec.message() << "\n"; |
| 698 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 699 | return; |
| 700 | } |
Jason M. Bills | a339793 | 2019-08-06 11:07:21 -0700 | [diff] [blame] | 701 | std::cerr << "CATERR asserted for " << std::to_string(caterrTimeoutMs) |
| 702 | << " ms\n"; |
| 703 | if (!checkIERRCPUs()) |
| 704 | { |
| 705 | cpuIERRLog(); |
| 706 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 707 | conn->async_method_call( |
| 708 | [](boost::system::error_code ec, |
| 709 | const std::variant<bool>& property) { |
| 710 | if (ec) |
| 711 | { |
| 712 | return; |
| 713 | } |
| 714 | const bool* reset = std::get_if<bool>(&property); |
| 715 | if (reset == nullptr) |
| 716 | { |
| 717 | std::cerr << "Unable to read reset on CATERR value\n"; |
| 718 | return; |
| 719 | } |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 720 | startCrashdumpAndRecovery(*reset, "IERR"); |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 721 | }, |
| 722 | "xyz.openbmc_project.Settings", |
| 723 | "/xyz/openbmc_project/control/processor_error_config", |
| 724 | "org.freedesktop.DBus.Properties", "Get", |
| 725 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnCATERR"); |
| 726 | }); |
| 727 | } |
| 728 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 729 | static void caterrHandler() |
| 730 | { |
| 731 | if (!hostOff) |
| 732 | { |
| 733 | gpiod::line_event gpioLineEvent = caterrLine.event_read(); |
| 734 | |
| 735 | bool caterr = |
| 736 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 737 | if (caterr) |
| 738 | { |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 739 | caterrAssertHandler(); |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 740 | } |
| 741 | else |
| 742 | { |
| 743 | caterrAssertTimer.cancel(); |
| 744 | } |
| 745 | } |
| 746 | caterrEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 747 | [](const boost::system::error_code ec) { |
| 748 | if (ec) |
| 749 | { |
| 750 | std::cerr << "caterr handler error: " |
| 751 | << ec.message() << "\n"; |
| 752 | return; |
| 753 | } |
| 754 | caterrHandler(); |
| 755 | }); |
| 756 | } |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 757 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 758 | static void cpu1ThermtripAssertHandler() |
| 759 | { |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 760 | if (cpu1FIVRFaultLine.get_value() == 0) |
| 761 | { |
| 762 | cpuBootFIVRFaultLog(1); |
| 763 | } |
| 764 | else |
| 765 | { |
| 766 | cpuThermTripLog(1); |
| 767 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 768 | } |
| 769 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 770 | static void cpu1ThermtripHandler() |
| 771 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 772 | gpiod::line_event gpioLineEvent = cpu1ThermtripLine.event_read(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 773 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 774 | bool cpu1Thermtrip = |
| 775 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 776 | if (cpu1Thermtrip) |
| 777 | { |
| 778 | cpu1ThermtripAssertHandler(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 779 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 780 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 781 | cpu1ThermtripEvent.async_wait( |
| 782 | boost::asio::posix::stream_descriptor::wait_read, |
| 783 | [](const boost::system::error_code ec) { |
| 784 | if (ec) |
| 785 | { |
| 786 | std::cerr << "CPU 1 Thermtrip handler error: " << ec.message() |
| 787 | << "\n"; |
| 788 | return; |
| 789 | } |
| 790 | cpu1ThermtripHandler(); |
| 791 | }); |
| 792 | } |
| 793 | |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 794 | static void cpu1MemtripHandler() |
| 795 | { |
| 796 | if (!hostOff) |
| 797 | { |
| 798 | gpiod::line_event gpioLineEvent = cpu1MemtripLine.event_read(); |
| 799 | |
| 800 | bool cpu1Memtrip = |
| 801 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 802 | if (cpu1Memtrip) |
| 803 | { |
| 804 | memThermTripLog(1); |
| 805 | } |
| 806 | } |
| 807 | cpu1MemtripEvent.async_wait( |
| 808 | boost::asio::posix::stream_descriptor::wait_read, |
| 809 | [](const boost::system::error_code ec) { |
| 810 | if (ec) |
| 811 | { |
| 812 | std::cerr << "CPU 1 Memory Thermaltrip handler error: " |
| 813 | << ec.message() << "\n"; |
| 814 | return; |
| 815 | } |
| 816 | cpu1MemtripHandler(); |
| 817 | }); |
| 818 | } |
| 819 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 820 | static void cpu2ThermtripAssertHandler() |
| 821 | { |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 822 | if (cpu2FIVRFaultLine.get_value() == 0) |
| 823 | { |
| 824 | cpuBootFIVRFaultLog(2); |
| 825 | } |
| 826 | else |
| 827 | { |
| 828 | cpuThermTripLog(2); |
| 829 | } |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 830 | } |
| 831 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 832 | static void cpu2ThermtripHandler() |
| 833 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 834 | gpiod::line_event gpioLineEvent = cpu2ThermtripLine.event_read(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 835 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 836 | bool cpu2Thermtrip = |
| 837 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 838 | if (cpu2Thermtrip) |
| 839 | { |
| 840 | cpu2ThermtripAssertHandler(); |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 841 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 842 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 843 | cpu2ThermtripEvent.async_wait( |
| 844 | boost::asio::posix::stream_descriptor::wait_read, |
| 845 | [](const boost::system::error_code ec) { |
| 846 | if (ec) |
| 847 | { |
| 848 | std::cerr << "CPU 2 Thermtrip handler error: " << ec.message() |
| 849 | << "\n"; |
| 850 | return; |
| 851 | } |
| 852 | cpu2ThermtripHandler(); |
| 853 | }); |
| 854 | } |
| 855 | |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 856 | static void cpu2MemtripHandler() |
| 857 | { |
| 858 | if (!hostOff) |
| 859 | { |
| 860 | gpiod::line_event gpioLineEvent = cpu2MemtripLine.event_read(); |
| 861 | |
| 862 | bool cpu2Memtrip = |
| 863 | gpioLineEvent.event_type == gpiod::line_event::RISING_EDGE; |
| 864 | if (cpu2Memtrip) |
| 865 | { |
| 866 | memThermTripLog(2); |
| 867 | } |
| 868 | } |
| 869 | cpu2MemtripEvent.async_wait( |
| 870 | boost::asio::posix::stream_descriptor::wait_read, |
| 871 | [](const boost::system::error_code ec) { |
| 872 | if (ec) |
| 873 | { |
| 874 | std::cerr << "CPU 2 Memory Thermaltrip handler error: " |
| 875 | << ec.message() << "\n"; |
| 876 | return; |
| 877 | } |
| 878 | cpu2MemtripHandler(); |
| 879 | }); |
| 880 | } |
| 881 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 882 | static void cpu1VRHotAssertHandler() |
| 883 | { |
| 884 | cpuVRHotLog("CPU 1"); |
| 885 | } |
| 886 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 887 | static void cpu1VRHotHandler() |
| 888 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 889 | gpiod::line_event gpioLineEvent = cpu1VRHotLine.event_read(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 890 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 891 | bool cpu1VRHot = |
| 892 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 893 | if (cpu1VRHot) |
| 894 | { |
| 895 | cpu1VRHotAssertHandler(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 896 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 897 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 898 | cpu1VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 899 | [](const boost::system::error_code ec) { |
| 900 | if (ec) |
| 901 | { |
| 902 | std::cerr << "CPU 1 VRHot handler error: " |
| 903 | << ec.message() << "\n"; |
| 904 | return; |
| 905 | } |
| 906 | cpu1VRHotHandler(); |
| 907 | }); |
| 908 | } |
| 909 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 910 | static void cpu1MemABCDVRHotAssertHandler() |
| 911 | { |
| 912 | cpuVRHotLog("CPU 1 Memory ABCD"); |
| 913 | } |
| 914 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 915 | static void cpu1MemABCDVRHotHandler() |
| 916 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 917 | gpiod::line_event gpioLineEvent = cpu1MemABCDVRHotLine.event_read(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 918 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 919 | bool cpu1MemABCDVRHot = |
| 920 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 921 | if (cpu1MemABCDVRHot) |
| 922 | { |
| 923 | cpu1MemABCDVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 924 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 925 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 926 | cpu1MemABCDVRHotEvent.async_wait( |
| 927 | boost::asio::posix::stream_descriptor::wait_read, |
| 928 | [](const boost::system::error_code ec) { |
| 929 | if (ec) |
| 930 | { |
| 931 | std::cerr << "CPU 1 Memory ABCD VRHot handler error: " |
| 932 | << ec.message() << "\n"; |
| 933 | return; |
| 934 | } |
| 935 | cpu1MemABCDVRHotHandler(); |
| 936 | }); |
| 937 | } |
| 938 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 939 | static void cpu1MemEFGHVRHotAssertHandler() |
| 940 | { |
| 941 | cpuVRHotLog("CPU 1 Memory EFGH"); |
| 942 | } |
| 943 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 944 | static void cpu1MemEFGHVRHotHandler() |
| 945 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 946 | gpiod::line_event gpioLineEvent = cpu1MemEFGHVRHotLine.event_read(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 947 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 948 | bool cpu1MemEFGHVRHot = |
| 949 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 950 | if (cpu1MemEFGHVRHot) |
| 951 | { |
| 952 | cpu1MemEFGHVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 953 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 954 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 955 | cpu1MemEFGHVRHotEvent.async_wait( |
| 956 | boost::asio::posix::stream_descriptor::wait_read, |
| 957 | [](const boost::system::error_code ec) { |
| 958 | if (ec) |
| 959 | { |
| 960 | std::cerr << "CPU 1 Memory EFGH VRHot handler error: " |
| 961 | << ec.message() << "\n"; |
| 962 | return; |
| 963 | } |
| 964 | cpu1MemEFGHVRHotHandler(); |
| 965 | }); |
| 966 | } |
| 967 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 968 | static void cpu2VRHotAssertHandler() |
| 969 | { |
| 970 | cpuVRHotLog("CPU 2"); |
| 971 | } |
| 972 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 973 | static void cpu2VRHotHandler() |
| 974 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 975 | gpiod::line_event gpioLineEvent = cpu2VRHotLine.event_read(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 976 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 977 | bool cpu2VRHot = |
| 978 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 979 | if (cpu2VRHot) |
| 980 | { |
| 981 | cpu2VRHotAssertHandler(); |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 982 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 983 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 984 | cpu2VRHotEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 985 | [](const boost::system::error_code ec) { |
| 986 | if (ec) |
| 987 | { |
| 988 | std::cerr << "CPU 2 VRHot handler error: " |
| 989 | << ec.message() << "\n"; |
| 990 | return; |
| 991 | } |
| 992 | cpu2VRHotHandler(); |
| 993 | }); |
| 994 | } |
| 995 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 996 | static void cpu2MemABCDVRHotAssertHandler() |
| 997 | { |
| 998 | cpuVRHotLog("CPU 2 Memory ABCD"); |
| 999 | } |
| 1000 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1001 | static void cpu2MemABCDVRHotHandler() |
| 1002 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1003 | gpiod::line_event gpioLineEvent = cpu2MemABCDVRHotLine.event_read(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1004 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1005 | bool cpu2MemABCDVRHot = |
| 1006 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1007 | if (cpu2MemABCDVRHot) |
| 1008 | { |
| 1009 | cpu2MemABCDVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1010 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1011 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1012 | cpu2MemABCDVRHotEvent.async_wait( |
| 1013 | boost::asio::posix::stream_descriptor::wait_read, |
| 1014 | [](const boost::system::error_code ec) { |
| 1015 | if (ec) |
| 1016 | { |
| 1017 | std::cerr << "CPU 2 Memory ABCD VRHot handler error: " |
| 1018 | << ec.message() << "\n"; |
| 1019 | return; |
| 1020 | } |
| 1021 | cpu2MemABCDVRHotHandler(); |
| 1022 | }); |
| 1023 | } |
| 1024 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 1025 | static void cpu2MemEFGHVRHotAssertHandler() |
| 1026 | { |
| 1027 | cpuVRHotLog("CPU 2 Memory EFGH"); |
| 1028 | } |
| 1029 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1030 | static void cpu2MemEFGHVRHotHandler() |
| 1031 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1032 | gpiod::line_event gpioLineEvent = cpu2MemEFGHVRHotLine.event_read(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1033 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1034 | bool cpu2MemEFGHVRHot = |
| 1035 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1036 | if (cpu2MemEFGHVRHot) |
| 1037 | { |
| 1038 | cpu2MemEFGHVRHotAssertHandler(); |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1039 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1040 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1041 | cpu2MemEFGHVRHotEvent.async_wait( |
| 1042 | boost::asio::posix::stream_descriptor::wait_read, |
| 1043 | [](const boost::system::error_code ec) { |
| 1044 | if (ec) |
| 1045 | { |
| 1046 | std::cerr << "CPU 2 Memory EFGH VRHot handler error: " |
| 1047 | << ec.message() << "\n"; |
| 1048 | return; |
| 1049 | } |
| 1050 | cpu2MemEFGHVRHotHandler(); |
| 1051 | }); |
| 1052 | } |
| 1053 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1054 | static void pchThermtripHandler() |
| 1055 | { |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1056 | gpiod::line_event gpioLineEvent = pchThermtripLine.event_read(); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1057 | |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1058 | bool pchThermtrip = |
| 1059 | gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1060 | if (pchThermtrip) |
| 1061 | { |
| 1062 | ssbThermTripLog(); |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1063 | } |
Jason M. Bills | 8495114 | 2020-04-17 15:57:11 -0700 | [diff] [blame^] | 1064 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1065 | pchThermtripEvent.async_wait( |
| 1066 | boost::asio::posix::stream_descriptor::wait_read, |
| 1067 | [](const boost::system::error_code ec) { |
| 1068 | if (ec) |
| 1069 | { |
| 1070 | std::cerr << "PCH Thermal trip handler error: " << ec.message() |
| 1071 | << "\n"; |
| 1072 | return; |
| 1073 | } |
| 1074 | pchThermtripHandler(); |
| 1075 | }); |
| 1076 | } |
| 1077 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1078 | static std::bitset<MAX_CPUS> checkERRPinCPUs(const int errPin) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1079 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1080 | int errPinSts = (1 << errPin); |
| 1081 | std::bitset<MAX_CPUS> errPinCPUs = 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1082 | for (int cpu = 0, addr = MIN_CLIENT_ADDR; addr <= MAX_CLIENT_ADDR; |
| 1083 | cpu++, addr++) |
| 1084 | { |
| 1085 | if (peci_Ping(addr) == PECI_CC_SUCCESS) |
| 1086 | { |
| 1087 | uint8_t cc = 0; |
| 1088 | CPUModel model{}; |
| 1089 | uint8_t stepping = 0; |
| 1090 | if (peci_GetCPUID(addr, &model, &stepping, &cc) != PECI_CC_SUCCESS) |
| 1091 | { |
| 1092 | std::cerr << "Cannot get CPUID!\n"; |
| 1093 | continue; |
| 1094 | } |
| 1095 | |
| 1096 | switch (model) |
| 1097 | { |
| 1098 | case skx: |
| 1099 | { |
| 1100 | // 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] | 1101 | // the ERRx (B(0) D8 F0 offset 210h) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1102 | uint32_t errpinsts = 0; |
| 1103 | if (peci_RdPCIConfigLocal( |
| 1104 | addr, 0, 8, 0, 0x210, sizeof(uint32_t), |
| 1105 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 1106 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1107 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1108 | } |
| 1109 | break; |
| 1110 | } |
| 1111 | case icx: |
| 1112 | { |
| 1113 | // 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] | 1114 | // 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] | 1115 | // accessed on PECI as bus 13) |
| 1116 | uint32_t errpinsts = 0; |
| 1117 | if (peci_RdEndPointConfigPciLocal( |
| 1118 | addr, 0, 13, 0, 3, 0x274, sizeof(uint32_t), |
| 1119 | (uint8_t*)&errpinsts, &cc) == PECI_CC_SUCCESS) |
| 1120 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1121 | errPinCPUs[cpu] = (errpinsts & errPinSts) != 0; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1122 | } |
| 1123 | break; |
| 1124 | } |
| 1125 | } |
| 1126 | } |
| 1127 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1128 | return errPinCPUs; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1129 | } |
| 1130 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1131 | static void errXAssertHandler(const int errPin, |
| 1132 | boost::asio::steady_timer& errXAssertTimer) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1133 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1134 | // ERRx status is not guaranteed through the timeout, so save which |
| 1135 | // CPUs have it asserted |
| 1136 | std::bitset<MAX_CPUS> errPinCPUs = checkERRPinCPUs(errPin); |
| 1137 | errXAssertTimer.expires_after(std::chrono::milliseconds(errTimeoutMs)); |
| 1138 | errXAssertTimer.async_wait([errPin, errPinCPUs]( |
| 1139 | const boost::system::error_code ec) { |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1140 | if (ec) |
| 1141 | { |
| 1142 | // operation_aborted is expected if timer is canceled before |
| 1143 | // completion. |
| 1144 | if (ec != boost::asio::error::operation_aborted) |
| 1145 | { |
| 1146 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 1147 | << "\n"; |
| 1148 | } |
| 1149 | return; |
| 1150 | } |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1151 | std::cerr << "ERR" << std::to_string(errPin) << " asserted for " |
| 1152 | << std::to_string(errTimeoutMs) << " ms\n"; |
| 1153 | if (errPinCPUs.count()) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1154 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1155 | for (int i = 0; i < errPinCPUs.size(); i++) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1156 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1157 | if (errPinCPUs[i]) |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1158 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1159 | cpuERRXLog(errPin, i); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1160 | } |
| 1161 | } |
| 1162 | } |
| 1163 | else |
| 1164 | { |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1165 | cpuERRXLog(errPin); |
| 1166 | } |
| 1167 | }); |
| 1168 | } |
| 1169 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1170 | static void err0AssertHandler() |
| 1171 | { |
| 1172 | // Handle the standard ERR0 detection and logging |
| 1173 | const static constexpr int err0 = 0; |
| 1174 | errXAssertHandler(err0, err0AssertTimer); |
| 1175 | } |
| 1176 | |
| 1177 | static void err0Handler() |
| 1178 | { |
| 1179 | if (!hostOff) |
| 1180 | { |
| 1181 | gpiod::line_event gpioLineEvent = err0Line.event_read(); |
| 1182 | |
| 1183 | bool err0 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1184 | if (err0) |
| 1185 | { |
| 1186 | err0AssertHandler(); |
| 1187 | } |
| 1188 | else |
| 1189 | { |
| 1190 | err0AssertTimer.cancel(); |
| 1191 | } |
| 1192 | } |
| 1193 | err0Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1194 | [](const boost::system::error_code ec) { |
| 1195 | if (ec) |
| 1196 | { |
| 1197 | std::cerr |
| 1198 | << "err0 handler error: " << ec.message() |
| 1199 | << "\n"; |
| 1200 | return; |
| 1201 | } |
| 1202 | err0Handler(); |
| 1203 | }); |
| 1204 | } |
| 1205 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1206 | static void err1AssertHandler() |
| 1207 | { |
| 1208 | // Handle the standard ERR1 detection and logging |
| 1209 | const static constexpr int err1 = 1; |
| 1210 | errXAssertHandler(err1, err1AssertTimer); |
| 1211 | } |
| 1212 | |
| 1213 | static void err1Handler() |
| 1214 | { |
| 1215 | if (!hostOff) |
| 1216 | { |
| 1217 | gpiod::line_event gpioLineEvent = err1Line.event_read(); |
| 1218 | |
| 1219 | bool err1 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1220 | if (err1) |
| 1221 | { |
| 1222 | err1AssertHandler(); |
| 1223 | } |
| 1224 | else |
| 1225 | { |
| 1226 | err1AssertTimer.cancel(); |
| 1227 | } |
| 1228 | } |
| 1229 | err1Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1230 | [](const boost::system::error_code ec) { |
| 1231 | if (ec) |
| 1232 | { |
| 1233 | std::cerr |
| 1234 | << "err1 handler error: " << ec.message() |
| 1235 | << "\n"; |
| 1236 | return; |
| 1237 | } |
| 1238 | err1Handler(); |
| 1239 | }); |
| 1240 | } |
| 1241 | |
Jason M. Bills | cbf7853 | 2019-08-16 15:32:11 -0700 | [diff] [blame] | 1242 | static void err2AssertHandler() |
| 1243 | { |
| 1244 | // Handle the standard ERR2 detection and logging |
| 1245 | const static constexpr int err2 = 2; |
| 1246 | errXAssertHandler(err2, err2AssertTimer); |
| 1247 | // Also handle reset for ERR2 |
| 1248 | err2AssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1249 | if (ec) |
| 1250 | { |
| 1251 | // operation_aborted is expected if timer is canceled before |
| 1252 | // completion. |
| 1253 | if (ec != boost::asio::error::operation_aborted) |
| 1254 | { |
| 1255 | std::cerr << "err2 timeout async_wait failed: " << ec.message() |
| 1256 | << "\n"; |
| 1257 | } |
| 1258 | return; |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1259 | } |
| 1260 | conn->async_method_call( |
| 1261 | [](boost::system::error_code ec, |
| 1262 | const std::variant<bool>& property) { |
| 1263 | if (ec) |
| 1264 | { |
| 1265 | return; |
| 1266 | } |
| 1267 | const bool* reset = std::get_if<bool>(&property); |
| 1268 | if (reset == nullptr) |
| 1269 | { |
| 1270 | std::cerr << "Unable to read reset on ERR2 value\n"; |
| 1271 | return; |
| 1272 | } |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 1273 | startCrashdumpAndRecovery(*reset, "ERR2 Timeout"); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1274 | }, |
| 1275 | "xyz.openbmc_project.Settings", |
| 1276 | "/xyz/openbmc_project/control/processor_error_config", |
| 1277 | "org.freedesktop.DBus.Properties", "Get", |
| 1278 | "xyz.openbmc_project.Control.Processor.ErrConfig", "ResetOnERR2"); |
Yong Li | 061eb03 | 2020-02-26 15:06:18 +0800 | [diff] [blame] | 1279 | |
| 1280 | beep(beepCPUErr2); |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1281 | }); |
| 1282 | } |
| 1283 | |
| 1284 | static void err2Handler() |
| 1285 | { |
| 1286 | if (!hostOff) |
| 1287 | { |
| 1288 | gpiod::line_event gpioLineEvent = err2Line.event_read(); |
| 1289 | |
| 1290 | bool err2 = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1291 | if (err2) |
| 1292 | { |
| 1293 | err2AssertHandler(); |
| 1294 | } |
| 1295 | else |
| 1296 | { |
| 1297 | err2AssertTimer.cancel(); |
| 1298 | } |
| 1299 | } |
| 1300 | err2Event.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1301 | [](const boost::system::error_code ec) { |
| 1302 | if (ec) |
| 1303 | { |
| 1304 | std::cerr |
| 1305 | << "err2 handler error: " << ec.message() |
| 1306 | << "\n"; |
| 1307 | return; |
| 1308 | } |
| 1309 | err2Handler(); |
| 1310 | }); |
| 1311 | } |
| 1312 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1313 | static void smiAssertHandler() |
| 1314 | { |
| 1315 | smiAssertTimer.expires_after(std::chrono::milliseconds(smiTimeoutMs)); |
| 1316 | smiAssertTimer.async_wait([](const boost::system::error_code ec) { |
| 1317 | if (ec) |
| 1318 | { |
| 1319 | // operation_aborted is expected if timer is canceled before |
| 1320 | // completion. |
| 1321 | if (ec != boost::asio::error::operation_aborted) |
| 1322 | { |
| 1323 | std::cerr << "smi timeout async_wait failed: " << ec.message() |
| 1324 | << "\n"; |
| 1325 | } |
| 1326 | return; |
| 1327 | } |
| 1328 | std::cerr << "SMI asserted for " << std::to_string(smiTimeoutMs) |
| 1329 | << " ms\n"; |
| 1330 | smiTimeoutLog(); |
| 1331 | conn->async_method_call( |
| 1332 | [](boost::system::error_code ec, |
| 1333 | const std::variant<bool>& property) { |
| 1334 | if (ec) |
| 1335 | { |
| 1336 | return; |
| 1337 | } |
| 1338 | const bool* reset = std::get_if<bool>(&property); |
| 1339 | if (reset == nullptr) |
| 1340 | { |
| 1341 | std::cerr << "Unable to read reset on SMI value\n"; |
| 1342 | return; |
| 1343 | } |
Jason M. Bills | 9478544 | 2020-01-07 15:22:09 -0800 | [diff] [blame] | 1344 | #ifdef HOST_ERROR_CRASHDUMP_ON_SMI_TIMEOUT |
Jason M. Bills | b61766b | 2019-11-26 17:02:44 -0800 | [diff] [blame] | 1345 | startCrashdumpAndRecovery(*reset, "SMI Timeout"); |
Jason M. Bills | 9478544 | 2020-01-07 15:22:09 -0800 | [diff] [blame] | 1346 | #else |
| 1347 | if (*reset) |
| 1348 | { |
| 1349 | std::cout << "Recovering the system\n"; |
| 1350 | startPowerCycle(); |
| 1351 | } |
| 1352 | #endif |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1353 | }, |
| 1354 | "xyz.openbmc_project.Settings", |
| 1355 | "/xyz/openbmc_project/control/bmc_reset_disables", |
| 1356 | "org.freedesktop.DBus.Properties", "Get", |
| 1357 | "xyz.openbmc_project.Control.ResetDisables", "ResetOnSMI"); |
| 1358 | }); |
| 1359 | } |
| 1360 | |
| 1361 | static void smiHandler() |
| 1362 | { |
| 1363 | if (!hostOff) |
| 1364 | { |
| 1365 | gpiod::line_event gpioLineEvent = smiLine.event_read(); |
| 1366 | |
| 1367 | bool smi = gpioLineEvent.event_type == gpiod::line_event::FALLING_EDGE; |
| 1368 | if (smi) |
| 1369 | { |
| 1370 | smiAssertHandler(); |
| 1371 | } |
| 1372 | else |
| 1373 | { |
| 1374 | smiAssertTimer.cancel(); |
| 1375 | } |
| 1376 | } |
| 1377 | smiEvent.async_wait(boost::asio::posix::stream_descriptor::wait_read, |
| 1378 | [](const boost::system::error_code ec) { |
| 1379 | if (ec) |
| 1380 | { |
| 1381 | std::cerr |
| 1382 | << "smi handler error: " << ec.message() |
| 1383 | << "\n"; |
| 1384 | return; |
| 1385 | } |
| 1386 | smiHandler(); |
| 1387 | }); |
| 1388 | } |
| 1389 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1390 | static void initializeErrorState() |
| 1391 | { |
jayaprakash Mutyala | 53099c4 | 2020-03-15 00:16:26 +0000 | [diff] [blame] | 1392 | // Handle CPU1_MISMATCH if it's asserted now |
| 1393 | if (cpu1MismatchLine.get_value() == 1) |
| 1394 | { |
| 1395 | cpuMismatchLog(1); |
| 1396 | } |
| 1397 | |
| 1398 | // Handle CPU2_MISMATCH if it's asserted now |
| 1399 | if (cpu2MismatchLine.get_value() == 1) |
| 1400 | { |
| 1401 | cpuMismatchLog(2); |
| 1402 | } |
| 1403 | |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1404 | // Handle CPU_CATERR if it's asserted now |
| 1405 | if (caterrLine.get_value() == 0) |
| 1406 | { |
| 1407 | caterrAssertHandler(); |
| 1408 | } |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1409 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1410 | // Handle CPU_ERR0 if it's asserted now |
| 1411 | if (err0Line.get_value() == 0) |
| 1412 | { |
| 1413 | err0AssertHandler(); |
| 1414 | } |
| 1415 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1416 | // Handle CPU_ERR1 if it's asserted now |
| 1417 | if (err1Line.get_value() == 0) |
| 1418 | { |
| 1419 | err1AssertHandler(); |
| 1420 | } |
| 1421 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1422 | // Handle CPU_ERR2 if it's asserted now |
| 1423 | if (err2Line.get_value() == 0) |
| 1424 | { |
| 1425 | err2AssertHandler(); |
| 1426 | } |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1427 | |
| 1428 | // Handle SMI if it's asserted now |
| 1429 | if (smiLine.get_value() == 0) |
| 1430 | { |
| 1431 | smiAssertHandler(); |
| 1432 | } |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 1433 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 1434 | // Handle CPU1_THERMTRIP if it's asserted now |
| 1435 | if (cpu1ThermtripLine.get_value() == 0) |
| 1436 | { |
| 1437 | cpu1ThermtripAssertHandler(); |
| 1438 | } |
| 1439 | |
| 1440 | // Handle CPU2_THERMTRIP if it's asserted now |
| 1441 | if (cpu2ThermtripLine.get_value() == 0) |
| 1442 | { |
| 1443 | cpu2ThermtripAssertHandler(); |
| 1444 | } |
| 1445 | |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 1446 | // Handle CPU1_MEM_THERM_EVENT (CPU1 DIMM Thermal trip) if it's asserted now |
| 1447 | if (cpu1MemtripLine.get_value() == 0) |
| 1448 | { |
| 1449 | memThermTripLog(1); |
| 1450 | } |
| 1451 | |
| 1452 | // Handle CPU2_MEM_THERM_EVENT (CPU2 DIMM Thermal trip) if it's asserted now |
| 1453 | if (cpu2MemtripLine.get_value() == 0) |
| 1454 | { |
| 1455 | memThermTripLog(2); |
| 1456 | } |
| 1457 | |
Jason M. Bills | e94f5e1 | 2019-09-13 11:11:34 -0700 | [diff] [blame] | 1458 | // Handle CPU1_VRHOT if it's asserted now |
| 1459 | if (cpu1VRHotLine.get_value() == 0) |
| 1460 | { |
| 1461 | cpu1VRHotAssertHandler(); |
| 1462 | } |
| 1463 | |
| 1464 | // Handle CPU1_MEM_ABCD_VRHOT if it's asserted now |
| 1465 | if (cpu1MemABCDVRHotLine.get_value() == 0) |
| 1466 | { |
| 1467 | cpu1MemABCDVRHotAssertHandler(); |
| 1468 | } |
| 1469 | |
| 1470 | // Handle CPU1_MEM_EFGH_VRHOT if it's asserted now |
| 1471 | if (cpu1MemEFGHVRHotLine.get_value() == 0) |
| 1472 | { |
| 1473 | cpu1MemEFGHVRHotAssertHandler(); |
| 1474 | } |
| 1475 | |
| 1476 | // Handle CPU2_VRHOT if it's asserted now |
| 1477 | if (cpu2VRHotLine.get_value() == 0) |
| 1478 | { |
| 1479 | cpu2VRHotAssertHandler(); |
| 1480 | } |
| 1481 | |
| 1482 | // Handle CPU2_MEM_ABCD_VRHOT if it's asserted now |
| 1483 | if (cpu2MemABCDVRHotLine.get_value() == 0) |
| 1484 | { |
| 1485 | cpu2MemABCDVRHotAssertHandler(); |
| 1486 | } |
| 1487 | |
| 1488 | // Handle CPU2_MEM_EFGH_VRHOT if it's asserted now |
| 1489 | if (cpu2MemEFGHVRHotLine.get_value() == 0) |
| 1490 | { |
| 1491 | cpu2MemEFGHVRHotAssertHandler(); |
| 1492 | } |
| 1493 | |
Jason M. Bills | 0886654 | 2019-08-16 12:04:19 -0700 | [diff] [blame] | 1494 | // Handle PCH_BMC_THERMTRIP if it's asserted now |
| 1495 | if (pchThermtripLine.get_value() == 0) |
| 1496 | { |
| 1497 | ssbThermTripLog(); |
| 1498 | } |
Jason M. Bills | a15c252 | 2019-08-16 10:01:44 -0700 | [diff] [blame] | 1499 | } |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1500 | } // namespace host_error_monitor |
| 1501 | |
| 1502 | int main(int argc, char* argv[]) |
| 1503 | { |
| 1504 | // setup connection to dbus |
| 1505 | host_error_monitor::conn = |
| 1506 | std::make_shared<sdbusplus::asio::connection>(host_error_monitor::io); |
| 1507 | |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame] | 1508 | // Host Error Monitor Service |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1509 | host_error_monitor::conn->request_name( |
| 1510 | "xyz.openbmc_project.HostErrorMonitor"); |
| 1511 | sdbusplus::asio::object_server server = |
| 1512 | sdbusplus::asio::object_server(host_error_monitor::conn); |
| 1513 | |
Jason M. Bills | c4b91f2 | 2019-11-26 17:04:50 -0800 | [diff] [blame] | 1514 | // Restart Cause Interface |
| 1515 | host_error_monitor::hostErrorTimeoutIface = |
| 1516 | server.add_interface("/xyz/openbmc_project/host_error_monitor", |
| 1517 | "xyz.openbmc_project.HostErrorMonitor.Timeout"); |
| 1518 | |
| 1519 | host_error_monitor::hostErrorTimeoutIface->register_property( |
| 1520 | "IERRTimeoutMs", host_error_monitor::caterrTimeoutMs, |
| 1521 | [](const std::size_t& requested, std::size_t& resp) { |
| 1522 | if (requested > host_error_monitor::caterrTimeoutMsMax) |
| 1523 | { |
| 1524 | std::cerr << "IERRTimeoutMs update to " << requested |
| 1525 | << "ms rejected. Cannot be greater than " |
| 1526 | << host_error_monitor::caterrTimeoutMsMax << "ms.\n"; |
| 1527 | return 0; |
| 1528 | } |
| 1529 | std::cerr << "IERRTimeoutMs updated to " << requested << "ms\n"; |
| 1530 | host_error_monitor::caterrTimeoutMs = requested; |
| 1531 | resp = requested; |
| 1532 | return 1; |
| 1533 | }, |
| 1534 | [](std::size_t& resp) { return host_error_monitor::caterrTimeoutMs; }); |
| 1535 | host_error_monitor::hostErrorTimeoutIface->initialize(); |
| 1536 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1537 | // Start tracking host state |
| 1538 | std::shared_ptr<sdbusplus::bus::match::match> hostStateMonitor = |
| 1539 | host_error_monitor::startHostStateMonitor(); |
| 1540 | |
jayaprakash Mutyala | 53099c4 | 2020-03-15 00:16:26 +0000 | [diff] [blame] | 1541 | // Request CPU1_MISMATCH GPIO events |
| 1542 | if (!host_error_monitor::requestGPIOInput( |
| 1543 | "CPU1_MISMATCH", host_error_monitor::cpu1MismatchLine)) |
| 1544 | { |
| 1545 | return -1; |
| 1546 | } |
| 1547 | |
| 1548 | // Request CPU2_MISMATCH GPIO events |
| 1549 | if (!host_error_monitor::requestGPIOInput( |
| 1550 | "CPU2_MISMATCH", host_error_monitor::cpu2MismatchLine)) |
| 1551 | { |
| 1552 | return -1; |
| 1553 | } |
| 1554 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1555 | // Initialize the host state |
| 1556 | host_error_monitor::initializeHostState(); |
| 1557 | |
| 1558 | // Request CPU_CATERR GPIO events |
| 1559 | if (!host_error_monitor::requestGPIOEvents( |
| 1560 | "CPU_CATERR", host_error_monitor::caterrHandler, |
| 1561 | host_error_monitor::caterrLine, host_error_monitor::caterrEvent)) |
| 1562 | { |
| 1563 | return -1; |
| 1564 | } |
| 1565 | |
Jason M. Bills | 8c58439 | 2019-08-19 11:05:51 -0700 | [diff] [blame] | 1566 | // Request CPU_ERR0 GPIO events |
| 1567 | if (!host_error_monitor::requestGPIOEvents( |
| 1568 | "CPU_ERR0", host_error_monitor::err0Handler, |
| 1569 | host_error_monitor::err0Line, host_error_monitor::err0Event)) |
| 1570 | { |
| 1571 | return -1; |
| 1572 | } |
| 1573 | |
Jason M. Bills | 75af396 | 2019-08-19 11:07:17 -0700 | [diff] [blame] | 1574 | // Request CPU_ERR1 GPIO events |
| 1575 | if (!host_error_monitor::requestGPIOEvents( |
| 1576 | "CPU_ERR1", host_error_monitor::err1Handler, |
| 1577 | host_error_monitor::err1Line, host_error_monitor::err1Event)) |
| 1578 | { |
| 1579 | return -1; |
| 1580 | } |
| 1581 | |
Jason M. Bills | 6a2cb69 | 2019-08-06 11:03:49 -0700 | [diff] [blame] | 1582 | // Request CPU_ERR2 GPIO events |
| 1583 | if (!host_error_monitor::requestGPIOEvents( |
| 1584 | "CPU_ERR2", host_error_monitor::err2Handler, |
| 1585 | host_error_monitor::err2Line, host_error_monitor::err2Event)) |
| 1586 | { |
| 1587 | return -1; |
| 1588 | } |
| 1589 | |
Jason M. Bills | 89922f8 | 2019-08-06 11:10:02 -0700 | [diff] [blame] | 1590 | // Request SMI GPIO events |
| 1591 | if (!host_error_monitor::requestGPIOEvents( |
| 1592 | "SMI", host_error_monitor::smiHandler, host_error_monitor::smiLine, |
| 1593 | host_error_monitor::smiEvent)) |
| 1594 | { |
| 1595 | return -1; |
| 1596 | } |
| 1597 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 1598 | // Request CPU1_FIVR_FAULT GPIO input |
| 1599 | if (!host_error_monitor::requestGPIOInput( |
| 1600 | "CPU1_FIVR_FAULT", host_error_monitor::cpu1FIVRFaultLine)) |
| 1601 | { |
| 1602 | return -1; |
| 1603 | } |
| 1604 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 1605 | // Request CPU1_THERMTRIP GPIO events |
| 1606 | if (!host_error_monitor::requestGPIOEvents( |
| 1607 | "CPU1_THERMTRIP", host_error_monitor::cpu1ThermtripHandler, |
| 1608 | host_error_monitor::cpu1ThermtripLine, |
| 1609 | host_error_monitor::cpu1ThermtripEvent)) |
| 1610 | { |
| 1611 | return -1; |
| 1612 | } |
| 1613 | |
Jason M. Bills | 45e87e0 | 2019-09-09 14:45:38 -0700 | [diff] [blame] | 1614 | // Request CPU2_FIVR_FAULT GPIO input |
| 1615 | if (!host_error_monitor::requestGPIOInput( |
| 1616 | "CPU2_FIVR_FAULT", host_error_monitor::cpu2FIVRFaultLine)) |
| 1617 | { |
| 1618 | return -1; |
| 1619 | } |
| 1620 | |
Jason M. Bills | 78c5eed | 2019-08-28 14:00:40 -0700 | [diff] [blame] | 1621 | // Request CPU2_THERMTRIP GPIO events |
| 1622 | if (!host_error_monitor::requestGPIOEvents( |
| 1623 | "CPU2_THERMTRIP", host_error_monitor::cpu2ThermtripHandler, |
| 1624 | host_error_monitor::cpu2ThermtripLine, |
| 1625 | host_error_monitor::cpu2ThermtripEvent)) |
| 1626 | { |
| 1627 | return -1; |
| 1628 | } |
| 1629 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1630 | // Request CPU1_VRHOT GPIO events |
| 1631 | if (!host_error_monitor::requestGPIOEvents( |
| 1632 | "CPU1_VRHOT", host_error_monitor::cpu1VRHotHandler, |
| 1633 | host_error_monitor::cpu1VRHotLine, |
| 1634 | host_error_monitor::cpu1VRHotEvent)) |
| 1635 | { |
| 1636 | return -1; |
| 1637 | } |
| 1638 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1639 | // Request CPU1_MEM_ABCD_VRHOT GPIO events |
| 1640 | if (!host_error_monitor::requestGPIOEvents( |
| 1641 | "CPU1_MEM_ABCD_VRHOT", host_error_monitor::cpu1MemABCDVRHotHandler, |
| 1642 | host_error_monitor::cpu1MemABCDVRHotLine, |
| 1643 | host_error_monitor::cpu1MemABCDVRHotEvent)) |
| 1644 | { |
| 1645 | return -1; |
| 1646 | } |
| 1647 | |
| 1648 | // Request CPU1_MEM_EFGH_VRHOT GPIO events |
| 1649 | if (!host_error_monitor::requestGPIOEvents( |
| 1650 | "CPU1_MEM_EFGH_VRHOT", host_error_monitor::cpu1MemEFGHVRHotHandler, |
| 1651 | host_error_monitor::cpu1MemEFGHVRHotLine, |
| 1652 | host_error_monitor::cpu1MemEFGHVRHotEvent)) |
| 1653 | { |
| 1654 | return -1; |
| 1655 | } |
| 1656 | |
Jason M. Bills | 250fa63 | 2019-08-28 15:58:25 -0700 | [diff] [blame] | 1657 | // Request CPU2_VRHOT GPIO events |
| 1658 | if (!host_error_monitor::requestGPIOEvents( |
| 1659 | "CPU2_VRHOT", host_error_monitor::cpu2VRHotHandler, |
| 1660 | host_error_monitor::cpu2VRHotLine, |
| 1661 | host_error_monitor::cpu2VRHotEvent)) |
| 1662 | { |
| 1663 | return -1; |
| 1664 | } |
| 1665 | |
Jason M. Bills | 9647ba7 | 2019-08-29 14:19:19 -0700 | [diff] [blame] | 1666 | // Request CPU2_MEM_ABCD_VRHOT GPIO events |
| 1667 | if (!host_error_monitor::requestGPIOEvents( |
| 1668 | "CPU2_MEM_ABCD_VRHOT", host_error_monitor::cpu2MemABCDVRHotHandler, |
| 1669 | host_error_monitor::cpu2MemABCDVRHotLine, |
| 1670 | host_error_monitor::cpu2MemABCDVRHotEvent)) |
| 1671 | { |
| 1672 | return -1; |
| 1673 | } |
| 1674 | |
| 1675 | // Request CPU2_MEM_EFGH_VRHOT GPIO events |
| 1676 | if (!host_error_monitor::requestGPIOEvents( |
| 1677 | "CPU2_MEM_EFGH_VRHOT", host_error_monitor::cpu2MemEFGHVRHotHandler, |
| 1678 | host_error_monitor::cpu2MemEFGHVRHotLine, |
| 1679 | host_error_monitor::cpu2MemEFGHVRHotEvent)) |
| 1680 | { |
| 1681 | return -1; |
| 1682 | } |
| 1683 | |
Chen Yugang | e6c0f1c | 2019-08-02 20:36:42 +0800 | [diff] [blame] | 1684 | // Request PCH_BMC_THERMTRIP GPIO events |
| 1685 | if (!host_error_monitor::requestGPIOEvents( |
| 1686 | "PCH_BMC_THERMTRIP", host_error_monitor::pchThermtripHandler, |
| 1687 | host_error_monitor::pchThermtripLine, |
| 1688 | host_error_monitor::pchThermtripEvent)) |
| 1689 | { |
| 1690 | return -1; |
| 1691 | } |
| 1692 | |
jayaprakash Mutyala | 009adbc | 2019-12-24 22:08:07 +0000 | [diff] [blame] | 1693 | // Request CPU1_MEM_THERM_EVENT GPIO events |
| 1694 | if (!host_error_monitor::requestGPIOEvents( |
| 1695 | "CPU1_MEM_THERM_EVENT", host_error_monitor::cpu1MemtripHandler, |
| 1696 | host_error_monitor::cpu1MemtripLine, |
| 1697 | host_error_monitor::cpu1MemtripEvent)) |
| 1698 | { |
| 1699 | return -1; |
| 1700 | } |
| 1701 | |
| 1702 | // Request CPU2_MEM_THERM_EVENT GPIO events |
| 1703 | if (!host_error_monitor::requestGPIOEvents( |
| 1704 | "CPU2_MEM_THERM_EVENT", host_error_monitor::cpu2MemtripHandler, |
| 1705 | host_error_monitor::cpu2MemtripLine, |
| 1706 | host_error_monitor::cpu2MemtripEvent)) |
| 1707 | { |
| 1708 | return -1; |
| 1709 | } |
| 1710 | |
Jason M. Bills | 1490b14 | 2019-07-01 15:48:43 -0700 | [diff] [blame] | 1711 | host_error_monitor::io.run(); |
| 1712 | |
| 1713 | return 0; |
| 1714 | } |