xref: /openbmc/openpower-hw-diags/attn/attn_logging.cpp (revision 3bd32526ba36e623c65ffb4e77229cdb48cb0b37)
1 #include <unistd.h>
2 
3 #include <analyzer/analyzer_main.hpp>
4 #include <attn/attn_common.hpp>
5 #include <attn/attn_dbus.hpp>
6 #include <attn/attn_dump.hpp>
7 #include <attn/attn_logging.hpp>
8 #include <attn/pel/pel_minimal.hpp>
9 #include <phosphor-logging/log.hpp>
10 #include <util/dbus.hpp>
11 #include <util/ffdc.hpp>
12 #include <util/trace.hpp>
13 
14 namespace attn
15 {
16 /** @brief Tuple containing information about ffdc files */
17 using FFDCTuple =
18     std::tuple<util::FFDCFormat, uint8_t, uint8_t, sdbusplus::message::unix_fd>;
19 
20 /**
21  * Create FFDCTuple objects corresponding to the specified FFDC files.
22  *
23  * The D-Bus method to create an error log requires a vector of tuples to
24  * pass in the FFDC file information.
25  *
26  * @param   files - FFDC files
27  * @return  vector of FFDCTuple objects
28  */
createFFDCTuples(const std::vector<util::FFDCFile> & files)29 std::vector<FFDCTuple> createFFDCTuples(
30     const std::vector<util::FFDCFile>& files)
31 {
32     std::vector<FFDCTuple> ffdcTuples{};
33     util::transformFFDC(files, ffdcTuples);
34 
35     return ffdcTuples;
36 }
37 
38 /**
39  * @brief Create an FFDCFile object containing raw data
40  *
41  * Throws an exception if an error occurs.
42  *
43  * @param   i_buffer - raw data to add to ffdc faw data file
44  * @param   i_size - size of the raw data
45  * @return  FFDCFile object
46  */
createFFDCRawFile(void * i_buffer,size_t i_size)47 util::FFDCFile createFFDCRawFile(void* i_buffer, size_t i_size)
48 {
49     util::FFDCFile file{util::FFDCFormat::Custom};
50 
51     // Write buffer to file and then reset file description file offset
52     int fd = file.getFileDescriptor();
53     size_t numBytes = write(fd, static_cast<char*>(i_buffer), i_size);
54     if (i_size != numBytes)
55     {
56         trace::err("%s only %u of %u bytes written", file.getPath().c_str(),
57                    numBytes, i_size);
58     }
59 
60     lseek(fd, 0, SEEK_SET);
61 
62     return file;
63 }
64 
65 /**
66  * Create FFDCFile objects containing debug data to store in the error log.
67  *
68  * If an error occurs, the error is written to the journal but an exception
69  * is not thrown.
70  *
71  * @param   i_buffer - raw data (if creating raw dump ffdc entry in log)
72  * @return  vector of FFDCFile objects
73  */
createFFDCFiles(char * i_buffer=nullptr,size_t i_size=0)74 std::vector<util::FFDCFile> createFFDCFiles(char* i_buffer = nullptr,
75                                             size_t i_size = 0)
76 {
77     std::vector<util::FFDCFile> files{};
78 
79     // Create raw dump file
80     if ((nullptr != i_buffer) && (0 != i_size))
81     {
82         files.emplace_back(createFFDCRawFile(i_buffer, i_size));
83     }
84 
85     // Create trace dump file
86     util::createFFDCTraceFiles(files);
87 
88     // Add PRD scratch registers
89     addPrdScratchRegs(files);
90 
91     return files;
92 }
93 
94 /**
95  * Create a PEL from an existing PEL
96  *
97  * Create a new PEL based on the specified raw PEL and submit the new PEL
98  * to the backend logging code as a raw PEL. Note that  additional data map
99  * here contains data to be committed to the PEL and it can also be used to
100  * create the PEL as it contains needed information.
101  *
102  * @param   i_rawPel - buffer containing a raw PEL
103  * @param   i_additional - additional data to be added to the new PEL
104  */
createPelCustom(std::vector<uint8_t> & i_rawPel,std::map<std::string,std::string> i_additional)105 void createPelCustom(std::vector<uint8_t>& i_rawPel,
106                      std::map<std::string, std::string> i_additional)
107 {
108     // create PEL object from buffer
109     auto tiPel = std::make_unique<pel::PelMinimal>(i_rawPel);
110 
111     // The additional data contains the TI info as well as the value for the
112     // subystem that provided the TI info. Get the subystem from additional
113     // data and then populate the primary SRC and SRC words for the custom PEL
114     // based on the subsystem's TI info.
115     std::map<std::string, std::string>::iterator it;
116     uint8_t subsystem;
117 
118     it = i_additional.find("Subsystem");
119     if (it != i_additional.end())
120     {
121         subsystem = std::stoi(it->second);
122         tiPel->setSubsystem(subsystem);
123     }
124     else
125     {
126         // The entry with key "Subsystem" does not exist in the additional map.
127         // Log the error, create failure event, and return.
128         trace::err("Error the key Subsystem does not exist in the map.");
129         eventAttentionFail((int)AttnSection::attnLogging | ATTN_INVALID_KEY);
130         return;
131     }
132 
133     // If recoverable attentions are active we will call the analyzer and
134     // then link the custom pel to analyzer pel.
135     it = i_additional.find("recoverables");
136     if (it != i_additional.end() && "true" == it->second)
137     {
138         DumpParameters dumpParameters;
139         auto plid = analyzer::analyzeHardware(
140             analyzer::AnalysisType::TERMINATE_IMMEDIATE, dumpParameters);
141         if (0 != plid)
142         {
143             // Link the PLID if an attention was found and a PEL was generated.
144             tiPel->setPlid(plid);
145         }
146     }
147 
148     if (static_cast<uint8_t>(pel::SubsystemID::hypervisor) == subsystem)
149     {
150         // populate hypervisor SRC words
151         tiPel->setSrcWords(std::array<uint32_t, pel::numSrcWords>{
152             (uint32_t)std::stoul(i_additional["0x10 SRC Word 12"], nullptr, 16),
153             (uint32_t)std::stoul(i_additional["0x14 SRC Word 13"], nullptr, 16),
154             (uint32_t)std::stoul(i_additional["0x18 SRC Word 14"], nullptr, 16),
155             (uint32_t)std::stoul(i_additional["0x1c SRC Word 15"], nullptr, 16),
156             (uint32_t)std::stoul(i_additional["0x20 SRC Word 16"], nullptr, 16),
157             (uint32_t)std::stoul(i_additional["0x24 SRC Word 17"], nullptr, 16),
158             (uint32_t)std::stoul(i_additional["0x28 SRC Word 18"], nullptr, 16),
159             (uint32_t)std::stoul(i_additional["0x2c SRC Word 19"], nullptr,
160                                  16)});
161 
162         // Populate phyp primary SRC
163 
164         // char array for raw pel src
165         std::array<char, pel::asciiStringSize> srcChars{'0'};
166         std::string srcString;
167 
168         // src from TI info
169         it = i_additional.find("SrcAscii");
170         if (it != i_additional.end())
171         {
172             srcString = it->second;
173         }
174         else
175         {
176             // The entry with key "Subsystem" does not exist in the additional
177             // map. Log the error, create failure event, and return.
178             trace::err("Error the key SrcAscii does not exist in the map.");
179             eventAttentionFail(
180                 (int)AttnSection::attnLogging | ATTN_INVALID_KEY);
181             return;
182         }
183 
184         // copy from string to char array
185         srcString.copy(srcChars.data(),
186                        std::min(srcString.size(), pel::asciiStringSize), 0);
187 
188         tiPel->setAsciiString(srcChars); // pel object src is char array
189 
190         // set symptom-id
191         auto symptomId = (srcString.substr(0, 8) + '_');
192 
193         symptomId += (i_additional["0x10 SRC Word 12"]);
194         symptomId += (i_additional["0x14 SRC Word 13"] + '_');
195         symptomId += (i_additional["0x18 SRC Word 14"]);
196         symptomId += (i_additional["0x1c SRC Word 15"] + '_');
197         symptomId += (i_additional["0x20 SRC Word 16"]);
198         symptomId += (i_additional["0x24 SRC Word 17"] + '_');
199         symptomId += (i_additional["0x28 SRC Word 18"]);
200         symptomId += (i_additional["0x2c SRC Word 19"]);
201 
202         // setSymptomId will take care of required null-terminate and padding
203         tiPel->setSymptomId(symptomId);
204     }
205     else
206     {
207         // Populate hostboot SRC words - note HB word 0 from the shared info
208         // data (additional data "0x10 HB Word") is reflected in the PEL as
209         // "reason code" so we zero it here. Also note that the first word
210         // in this group of words starts at word 0 and word 1 does not exits.
211         tiPel->setSrcWords(std::array<uint32_t, pel::numSrcWords>{
212             (uint32_t)0x00000000,
213             (uint32_t)std::stoul(i_additional["0x14 HB Word 2"], nullptr, 16),
214             (uint32_t)std::stoul(i_additional["0x18 HB Word 3"], nullptr, 16),
215             (uint32_t)std::stoul(i_additional["0x1c HB Word 4"], nullptr, 16),
216             (uint32_t)std::stoul(i_additional["0x20 HB Word 5"], nullptr, 16),
217             (uint32_t)std::stoul(i_additional["0x24 HB Word 6"], nullptr, 16),
218             (uint32_t)std::stoul(i_additional["0x28 HB Word 7"], nullptr, 16),
219             (uint32_t)std::stoul(i_additional["0x2c HB Word 8"], nullptr, 16)});
220 
221         // Populate hostboot primary SRC
222 
223         // char array for raw pel src
224         std::array<char, pel::asciiStringSize> srcChars{'0'};
225         std::string srcString;
226 
227         // src from TI info
228         it = i_additional.find("SrcAscii");
229         if (it != i_additional.end())
230         {
231             srcString = it->second;
232         }
233         else
234         {
235             // The entry with key "Subsystem" does not exist in the additional
236             // map. Log the error, create failure event, and return.
237             trace::err("Error the key SrcAscii does not exist in the map.");
238             eventAttentionFail(
239                 (int)AttnSection::attnLogging | ATTN_INVALID_KEY);
240             return;
241         }
242 
243         // copy from string to char array
244         srcString.copy(srcChars.data(),
245                        std::min(srcString.size(), pel::asciiStringSize), 0);
246 
247         tiPel->setAsciiString(srcChars); // pel object src is char array
248 
249         // set symptom-id
250         auto symptomId = (srcString.substr(0, 8) + '_');
251 
252         symptomId += (i_additional["0x10 HB Word 0"]);       // note: word 1
253         symptomId += (i_additional["0x14 HB Word 2"] + '_'); // does not exist
254         symptomId += (i_additional["0x18 HB Word 3"]);
255         symptomId += (i_additional["0x1c HB Word 4"] + '_');
256         symptomId += (i_additional["0x20 HB Word 5"]);
257         symptomId += (i_additional["0x24 HB Word 6"] + '_');
258         symptomId += (i_additional["0x28 HB Word 7"]);
259         symptomId += (i_additional["0x2c HB Word 8"]);
260 
261         // setSymptomId will take care of required null-terminate and padding
262         tiPel->setSymptomId(symptomId);
263     }
264 
265     // set severity, event type and action flags
266     tiPel->setSeverity(static_cast<uint8_t>(pel::Severity::termination));
267     tiPel->setType(static_cast<uint8_t>(pel::EventType::na));
268 
269     auto actionFlags = pel::ActionFlags::service | pel::ActionFlags::report |
270                        pel::ActionFlags::call;
271 
272     it = i_additional.find("hidden");
273     if (it != i_additional.end() && "true" == it->second)
274     {
275         trace::inf("making HB TI PEL hidden");
276         actionFlags = actionFlags | pel::ActionFlags::hidden;
277     }
278 
279     tiPel->setAction(static_cast<uint16_t>(actionFlags));
280 
281     // The raw PEL that we used as the basis for this custom PEL contains some
282     // user data sections that do not need to be in this PEL. However we do
283     // want to include the raw TI information.
284     int ffdcCount = 0;
285     it = i_additional.find("FFDC count");
286     if (it != i_additional.end())
287     {
288         // remove all sections except 1 (raw Ti info)
289         ffdcCount = std::stoi(it->second) - 1;
290     }
291     tiPel->setSectionCount(tiPel->getSectionCount() - ffdcCount);
292 
293     // Update the raw PEL with the new custom PEL data
294     tiPel->raw(i_rawPel);
295 
296     // create PEL from raw data
297     createPelRaw(i_rawPel);
298 }
299 
300 /**
301  * Log an event handled by the attention handler
302  *
303  * Basic (non TI) events will generate a standard message-registry based PEL
304  *
305  * TI events will create two PEL's. One PEL will be informational and will
306  * contain trace information relevent to attention handler. The second PEL
307  * will be specific to the TI type (including the primary SRC) and will be
308  * based off of the TI information provided to the attention handler through
309  * shared TI info data area.
310  *
311  * @param  i_event - The event type
312  * @param  i_additional - Additional PEL data
313  * @param  i_ffdc - FFDC PEL data
314  * @param  i_severity - Severity level
315  * @return Event log Id (0 if no event log generated)
316  */
event(EventType i_event,std::map<std::string,std::string> & i_additional,const std::vector<util::FFDCFile> & i_ffdc,std::string i_severity=levelPelError)317 uint32_t event(EventType i_event,
318                std::map<std::string, std::string>& i_additional,
319                const std::vector<util::FFDCFile>& i_ffdc,
320                std::string i_severity = levelPelError)
321 {
322     uint32_t pelId = 0;      // assume no event log generated
323 
324     bool eventValid = false; // assume no event created
325     bool tiEvent = false;    // assume not a terminate event
326 
327     // count user data sections so we can fixup custom PEL
328     i_additional["FFDC count"] = std::to_string(i_ffdc.size());
329 
330     std::string eventName;
331 
332     switch (i_event)
333     {
334         case EventType::Checkstop:
335             eventName = "org.open_power.HwDiags.Error.Checkstop";
336             eventValid = true;
337             break;
338         case EventType::Terminate:
339             eventName = "org.open_power.Attn.Error.Terminate";
340             eventValid = true;
341             tiEvent = true;
342             break;
343         case EventType::Vital:
344             eventName = "org.open_power.Attn.Error.Vital";
345             eventValid = true;
346             break;
347         case EventType::HwDiagsFail:
348         case EventType::AttentionFail:
349             eventName = "org.open_power.Attn.Error.Fail";
350             eventValid = true;
351             break;
352         default:
353             eventValid = false;
354             break;
355     }
356 
357     if (true == eventValid)
358     {
359         // Create PEL with additional data and FFDC data. The newly created
360         // PEL's platform log-id will be returned.
361         pelId = util::dbus::createPel(eventName, i_severity, i_additional,
362                                       createFFDCTuples(i_ffdc));
363 
364         // If this is a TI event we will create an additional PEL that is
365         // specific to the subsystem that generated the TI.
366         if ((0 != pelId) && (true == tiEvent))
367         {
368             // get file descriptor and size of information PEL
369             int pelFd = getPel(pelId);
370 
371             // if PEL found, read into buffer
372             if (-1 != pelFd)
373             {
374                 auto pelSize = lseek(pelFd, 0, SEEK_END);
375                 lseek(pelFd, 0, SEEK_SET);
376 
377                 // read information PEL into buffer
378                 std::vector<uint8_t> buffer(pelSize);
379                 size_t numBytes = read(pelFd, buffer.data(), buffer.size());
380                 if (buffer.size() != numBytes)
381                 {
382                     trace::err("Error reading event log: %u of %u bytes read",
383                                numBytes, buffer.size());
384                 }
385                 else
386                 {
387                     // create PEL from buffer
388                     createPelCustom(buffer, i_additional);
389                 }
390 
391                 close(pelFd);
392             }
393 
394             std::map<std::string, std::string>::iterator it;
395             uint8_t subsystem;
396 
397             it = i_additional.find("Subsystem");
398             if (it != i_additional.end())
399             {
400                 subsystem = std::stoi(it->second);
401             }
402             else
403             {
404                 // The entry with key "Subsystem" does not exist in the
405                 // additional map. Log the error, create failure event, and
406                 // return.
407                 trace::err(
408                     "Error the key Subsystem does not exist in the map.");
409                 eventAttentionFail(
410                     (int)AttnSection::attnLogging | ATTN_INVALID_KEY);
411                 return 0;
412             }
413 
414             // If not hypervisor TI
415             if (static_cast<uint8_t>(pel::SubsystemID::hypervisor) != subsystem)
416             {
417                 // Request a dump and transition the host
418                 if ("true" == i_additional["Dump"])
419                 {
420                     // will not return until dump is complete
421                     requestDump(pelId, DumpParameters{0, DumpType::Hostboot});
422                 }
423             }
424         }
425     }
426     return pelId;
427 }
428 
429 /**
430  * Commit special attention TI event to log
431  *
432  * Create a event log with provided additional information and standard
433  * FFDC data plus TI FFDC data
434  *
435  * @param i_additional - Additional log data
436  * @param i_ti_InfoData - TI FFDC data
437  */
eventTerminate(std::map<std::string,std::string> i_additionalData,char * i_tiInfoData)438 void eventTerminate(std::map<std::string, std::string> i_additionalData,
439                     char* i_tiInfoData)
440 {
441     uint32_t tiInfoSize = 0; // assume TI info was not available
442 
443     if (nullptr != i_tiInfoData)
444     {
445         tiInfoSize = 56; // assume not hypervisor TI
446 
447         std::map<std::string, std::string>::iterator it;
448         uint8_t subsystem;
449 
450         it = i_additionalData.find("Subsystem");
451         if (it != i_additionalData.end())
452         {
453             subsystem = std::stoi(it->second);
454         }
455         else
456         {
457             // The entry with key "Subsystem" does not exist in the additional
458             // map. Log the error, create failure event, and return.
459             trace::err("Error the key Subsystem does not exist in the map.");
460             eventAttentionFail(
461                 (int)AttnSection::attnLogging | ATTN_INVALID_KEY);
462             return;
463         }
464 
465         // If hypervisor
466         if (static_cast<uint8_t>(pel::SubsystemID::hypervisor) == subsystem)
467         {
468             tiInfoSize = 1024; // assume hypervisor max
469 
470             // hypervisor may just want some of the data
471             if (0 == (*(i_tiInfoData + 0x09) & 0x01))
472             {
473                 uint32_t* additionalLength = (uint32_t*)(i_tiInfoData + 0x50);
474                 uint32_t tiAdditional = be32toh(*additionalLength);
475                 tiInfoSize = std::min(tiInfoSize, (84 + tiAdditional));
476             }
477         }
478     }
479 
480     trace::inf("TI info size = %u", tiInfoSize);
481 
482     event(EventType::Terminate, i_additionalData,
483           createFFDCFiles(i_tiInfoData, tiInfoSize));
484 }
485 
486 /** @brief Commit SBE vital event to log, returns event log ID */
eventVital(std::string severity)487 uint32_t eventVital(std::string severity)
488 {
489     // Additional data for log
490     std::map<std::string, std::string> additionalData;
491 
492     // Create log event with additional data and FFDC data
493     return event(EventType::Vital, additionalData, createFFDCFiles(nullptr, 0),
494                  severity);
495 }
496 
497 /**
498  * Commit attention handler failure event to log
499  *
500  * Create an event log containing the specified error code.
501  *
502  * @param i_error - Error code
503  */
eventAttentionFail(int i_error)504 void eventAttentionFail(int i_error)
505 {
506     // Additional data for log
507     std::map<std::string, std::string> additionalData;
508     additionalData["ERROR_CODE"] = std::to_string(i_error);
509 
510     // Create log event with additional data and FFDC data
511     event(EventType::AttentionFail, additionalData,
512           createFFDCFiles(nullptr, 0));
513 }
514 
515 } // namespace attn
516