1 #include "occ_ffdc.hpp"
2
3 #include "utils.hpp"
4
5 #include <errno.h>
6 #include <fcntl.h>
7 #include <stdio.h>
8 #include <sys/ioctl.h>
9 #include <unistd.h>
10
11 #include <nlohmann/json.hpp>
12 #include <org/open_power/OCC/Device/error.hpp>
13 #include <phosphor-logging/elog-errors.hpp>
14 #include <phosphor-logging/elog.hpp>
15 #include <phosphor-logging/log.hpp>
16 #include <xyz/openbmc_project/Common/error.hpp>
17 #include <xyz/openbmc_project/Logging/Create/server.hpp>
18
19 #include <format>
20
21 namespace open_power
22 {
23 namespace occ
24 {
25
26 static constexpr size_t max_ffdc_size = 8192;
27 static constexpr size_t sbe_status_header_size = 8;
28
29 static constexpr auto loggingObjectPath = "/xyz/openbmc_project/logging";
30 static constexpr auto opLoggingInterface = "org.open_power.Logging.PEL";
31
32 using namespace phosphor::logging;
33 using namespace sdbusplus::org::open_power::OCC::Device::Error;
34 using InternalFailure =
35 sdbusplus::xyz::openbmc_project::Common::Error::InternalFailure;
36
createPEL(const char * path,uint32_t src6,const char * msg,int fd)37 uint32_t FFDC::createPEL(const char* path, uint32_t src6, const char* msg,
38 int fd)
39 {
40 uint32_t plid = 0;
41 std::vector<std::tuple<
42 sdbusplus::xyz::openbmc_project::Logging::server::Create::FFDCFormat,
43 uint8_t, uint8_t, sdbusplus::message::unix_fd>>
44 pelFFDCInfo;
45
46 log<level::INFO>(
47 std::format("Creating PEL for OCC{} with SBE FFDC: {} - SRC6: 0x{:08X}",
48 src6 >> 16, path, src6)
49 .c_str());
50
51 if (fd > 0)
52 {
53 pelFFDCInfo.push_back(std::make_tuple(
54 sdbusplus::xyz::openbmc_project::Logging::server::Create::
55 FFDCFormat::Custom,
56 static_cast<uint8_t>(0xCB), static_cast<uint8_t>(0x01), fd));
57 }
58
59 // Add journal traces to PEL FFDC
60 auto occJournalFile =
61 addJournalEntries(pelFFDCInfo, "openpower-occ-control", 25);
62
63 std::map<std::string, std::string> additionalData;
64 additionalData.emplace("SRC6", std::to_string(src6));
65 additionalData.emplace("_PID", std::to_string(getpid()));
66 additionalData.emplace("SBE_ERR_MSG", msg);
67
68 auto& bus = utils::getBus();
69
70 try
71 {
72 std::string service =
73 utils::getService(loggingObjectPath, opLoggingInterface);
74 auto method =
75 bus.new_method_call(service.c_str(), loggingObjectPath,
76 opLoggingInterface, "CreatePELWithFFDCFiles");
77
78 // Set level to Notice (Informational). Error should trigger an OCC
79 // reset and if it does not recover, HTMGT/HBRT will create an
80 // unrecoverable error.
81 auto level =
82 sdbusplus::xyz::openbmc_project::Logging::server::convertForMessage(
83 sdbusplus::xyz::openbmc_project::Logging::server::Entry::Level::
84 Notice);
85
86 method.append(path, level, additionalData, pelFFDCInfo);
87 auto response = bus.call(method);
88 std::tuple<uint32_t, uint32_t> reply = {0, 0};
89
90 response.read(reply);
91 plid = std::get<1>(reply);
92 }
93 catch (const sdbusplus::exception_t& e)
94 {
95 log<level::ERR>(
96 std::format("Failed to create PEL: {}", e.what()).c_str());
97 }
98
99 return plid;
100 }
101
createOCCResetPEL(unsigned int instance,const char * path,int err,const char * callout)102 void FFDC::createOCCResetPEL(unsigned int instance, const char* path, int err,
103 const char* callout)
104 {
105 std::map<std::string, std::string> additionalData;
106
107 additionalData.emplace("_PID", std::to_string(getpid()));
108
109 if (err)
110 {
111 additionalData.emplace("CALLOUT_ERRNO", std::to_string(-err));
112 }
113
114 if (callout)
115 {
116 additionalData.emplace("CALLOUT_DEVICE_PATH", std::string(callout));
117 }
118
119 additionalData.emplace("OCC", std::to_string(instance));
120
121 log<level::INFO>(
122 std::format("Creating OCC Reset PEL for OCC{}: {}", instance, path)
123 .c_str());
124
125 auto& bus = utils::getBus();
126
127 try
128 {
129 FFDCFiles ffdc;
130 // Add journal traces to PEL FFDC
131 auto occJournalFile =
132 addJournalEntries(ffdc, "openpower-occ-control", 25);
133
134 std::string service =
135 utils::getService(loggingObjectPath, opLoggingInterface);
136 auto method =
137 bus.new_method_call(service.c_str(), loggingObjectPath,
138 opLoggingInterface, "CreatePELWithFFDCFiles");
139
140 // Set level to Notice (Informational). Error should trigger an OCC
141 // reset and if it does not recover, HTMGT/HBRT will create an
142 // unrecoverable error.
143 auto level =
144 sdbusplus::xyz::openbmc_project::Logging::server::convertForMessage(
145 sdbusplus::xyz::openbmc_project::Logging::server::Entry::Level::
146 Notice);
147
148 method.append(path, level, additionalData, ffdc);
149 bus.call(method);
150 }
151 catch (const sdbusplus::exception_t& e)
152 {
153 log<level::ERR>(
154 std::format("Failed to create OCC Reset PEL: {}", e.what())
155 .c_str());
156 }
157 }
158
159 // Reads the SBE FFDC file and create an error log
analyzeEvent()160 void FFDC::analyzeEvent()
161 {
162 int tfd = -1;
163 size_t total = 0;
164 auto data = std::make_unique<unsigned char[]>(max_ffdc_size);
165 while (total < max_ffdc_size)
166 {
167 auto r = read(fd, data.get() + total, max_ffdc_size - total);
168 if (r < 0)
169 {
170 elog<ReadFailure>(
171 phosphor::logging::org::open_power::OCC::Device::ReadFailure::
172 CALLOUT_ERRNO(errno),
173 phosphor::logging::org::open_power::OCC::Device::ReadFailure::
174 CALLOUT_DEVICE_PATH(file.c_str()));
175 return;
176 }
177 if (!r)
178 {
179 break;
180 }
181 total += r;
182 }
183
184 lseek(fd, 0, SEEK_SET);
185
186 if (!total)
187 {
188 // no error
189 return;
190 }
191
192 uint32_t src6 = instance << 16;
193 src6 |= *(data.get() + 2) << 8;
194 src6 |= *(data.get() + 3);
195
196 if (total > sbe_status_header_size)
197 {
198 std::string templateString =
199 fs::temp_directory_path() / "OCC_FFDC_XXXXXX";
200 tfd = mkostemp(templateString.data(), O_RDWR);
201 if (tfd < 0)
202 {
203 log<level::ERR>("Couldn't create temporary FFDC file");
204 }
205 else
206 {
207 temporaryFiles.emplace_back(templateString, tfd);
208 size_t written = sbe_status_header_size;
209 while (written < total)
210 {
211 auto r = write(tfd, data.get() + written, total - written);
212 if (r < 0)
213 {
214 close(temporaryFiles.back().second);
215 fs::remove(temporaryFiles.back().first);
216 temporaryFiles.pop_back();
217 tfd = -1;
218 log<level::ERR>("Couldn't write temporary FFDC file");
219 break;
220 }
221 if (!r)
222 {
223 break;
224 }
225 written += r;
226 }
227 }
228 }
229
230 createPEL("org.open_power.Processor.Error.SbeChipOpFailure", src6,
231 "SBE command reported error", tfd);
232 }
233
234 // Create file with the latest journal entries for specified executable
addJournalEntries(FFDCFiles & fileList,const std::string & executable,unsigned int lines)235 std::unique_ptr<FFDCFile> FFDC::addJournalEntries(
236 FFDCFiles& fileList, const std::string& executable, unsigned int lines)
237 {
238 auto journalFile = makeJsonFFDCFile(getJournalEntries(lines, executable));
239 if (journalFile && journalFile->fd() != -1)
240 {
241 log<level::DEBUG>(
242 std::format(
243 "addJournalEntries: Added up to {} journal entries for {}",
244 lines, executable)
245 .c_str());
246 fileList.emplace_back(FFDCFormat::JSON, 0x01, 0x01, journalFile->fd());
247 }
248 else
249 {
250 log<level::ERR>(
251 std::format(
252 "addJournalEntries: Failed to add journal entries for {}",
253 executable)
254 .c_str());
255 }
256 return journalFile;
257 }
258
259 // Write JSON data into FFDC file and return the file
makeJsonFFDCFile(const nlohmann::json & ffdcData)260 std::unique_ptr<FFDCFile> FFDC::makeJsonFFDCFile(const nlohmann::json& ffdcData)
261 {
262 std::string tmpFile = fs::temp_directory_path() / "OCC_JOURNAL_XXXXXX";
263 auto fd = mkostemp(tmpFile.data(), O_RDWR);
264 if (fd != -1)
265 {
266 auto jsonString = ffdcData.dump();
267 auto rc = write(fd, jsonString.data(), jsonString.size());
268 close(fd);
269 if (rc != -1)
270 {
271 fs::path jsonFile{tmpFile};
272 return std::make_unique<FFDCFile>(jsonFile);
273 }
274 else
275 {
276 auto e = errno;
277 log<level::ERR>(
278 std::format(
279 "makeJsonFFDCFile: Failed call to write JSON FFDC file, errno={}",
280 e)
281 .c_str());
282 }
283 }
284 else
285 {
286 auto e = errno;
287 log<level::ERR>(
288 std::format("makeJsonFFDCFile: Failed called to mkostemp, errno={}",
289 e)
290 .c_str());
291 }
292 return nullptr;
293 }
294
295 // Collect the latest journal entries for a specified executable
getJournalEntries(int numLines,std::string executable)296 nlohmann::json FFDC::getJournalEntries(int numLines, std::string executable)
297 {
298 // Sleep 100ms; otherwise recent journal entries sometimes not available
299 using namespace std::chrono_literals;
300 std::this_thread::sleep_for(100ms);
301
302 std::vector<std::string> entries;
303
304 // Open the journal
305 sd_journal* journal;
306 int rc = sd_journal_open(&journal, SD_JOURNAL_LOCAL_ONLY);
307 if (rc < 0)
308 {
309 // Build one line string containing field values
310 entries.push_back("[Internal error: sd_journal_open(), rc=" +
311 std::string(strerror(rc)) + "]");
312 return nlohmann::json(entries);
313 }
314
315 // Create object to automatically close journal
316 JournalCloser closer{journal};
317
318 // Add match so we only loop over entries with specified field value
319 std::string field{"SYSLOG_IDENTIFIER"};
320 std::string match{field + '=' + executable};
321 rc = sd_journal_add_match(journal, match.c_str(), 0);
322 if (rc < 0)
323 {
324 // Build one line string containing field values
325 entries.push_back("[Internal error: sd_journal_add_match(), rc=" +
326 std::string(strerror(rc)) + "]");
327 }
328 else
329 {
330 int count{1};
331 entries.reserve(numLines);
332 std::string syslogID, pid, message, timeStamp;
333
334 // Loop through journal entries from newest to oldest
335 SD_JOURNAL_FOREACH_BACKWARDS(journal)
336 {
337 // Get relevant journal entry fields
338 timeStamp = getTimeStamp(journal);
339 syslogID = getFieldValue(journal, "SYSLOG_IDENTIFIER");
340 pid = getFieldValue(journal, "_PID");
341 message = getFieldValue(journal, "MESSAGE");
342
343 // Build one line string containing field values
344 entries.push_back(
345 timeStamp + " " + syslogID + "[" + pid + "]: " + message);
346
347 // Stop after number of lines was read
348 if (count++ >= numLines)
349 {
350 break;
351 }
352 }
353 }
354
355 // put the journal entries in chronological order
356 std::reverse(entries.begin(), entries.end());
357
358 return nlohmann::json(entries);
359 }
360
getTimeStamp(sd_journal * journal)361 std::string FFDC::getTimeStamp(sd_journal* journal)
362 {
363 // Get realtime (wallclock) timestamp of current journal entry. The
364 // timestamp is in microseconds since the epoch.
365 uint64_t usec{0};
366 int rc = sd_journal_get_realtime_usec(journal, &usec);
367 if (rc < 0)
368 {
369 return "[Internal error: sd_journal_get_realtime_usec(), rc=" +
370 std::string(strerror(rc)) + "]";
371 }
372
373 // Convert to number of seconds since the epoch
374 time_t secs = usec / 1000000;
375
376 // Convert seconds to tm struct required by strftime()
377 struct tm* timeStruct = localtime(&secs);
378 if (timeStruct == nullptr)
379 {
380 return "[Internal error: localtime() returned nullptr]";
381 }
382
383 // Convert tm struct into a date/time string
384 char timeStamp[80];
385 strftime(timeStamp, sizeof(timeStamp), "%b %d %H:%M:%S", timeStruct);
386
387 return timeStamp;
388 }
389
getFieldValue(sd_journal * journal,const std::string & field)390 std::string FFDC::getFieldValue(sd_journal* journal, const std::string& field)
391 {
392 std::string value{};
393
394 // Get field data from current journal entry
395 const void* data{nullptr};
396 size_t length{0};
397 int rc = sd_journal_get_data(journal, field.c_str(), &data, &length);
398 if (rc < 0)
399 {
400 if (-rc == ENOENT)
401 {
402 // Current entry does not include this field; return empty value
403 return value;
404 }
405 else
406 {
407 return "[Internal error: sd_journal_get_data() rc=" +
408 std::string(strerror(rc)) + "]";
409 }
410 }
411
412 // Get value from field data. Field data in format "FIELD=value".
413 std::string dataString{static_cast<const char*>(data), length};
414 std::string::size_type pos = dataString.find('=');
415 if ((pos != std::string::npos) && ((pos + 1) < dataString.size()))
416 {
417 // Value is substring after the '='
418 value = dataString.substr(pos + 1);
419 }
420
421 return value;
422 }
423
424 // Create temporary file that will automatically get removed when destructed
FFDCFile(const fs::path & name)425 FFDCFile::FFDCFile(const fs::path& name) :
426 _fd(open(name.c_str(), O_RDONLY)), _name(name)
427 {
428 if (_fd() == -1)
429 {
430 auto e = errno;
431 log<level::ERR>(
432 std::format("FFDCFile: Could not open FFDC file {}. errno {}",
433 _name.string(), e)
434 .c_str());
435 }
436 }
437
438 } // namespace occ
439 } // namespace open_power
440