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  */
29 std::vector<FFDCTuple>
createFFDCTuples(const std::vector<util::FFDCFile> & files)30     createFFDCTuples(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"], 0, 16),
153             (uint32_t)std::stoul(i_additional["0x14 SRC Word 13"], 0, 16),
154             (uint32_t)std::stoul(i_additional["0x18 SRC Word 14"], 0, 16),
155             (uint32_t)std::stoul(i_additional["0x1c SRC Word 15"], 0, 16),
156             (uint32_t)std::stoul(i_additional["0x20 SRC Word 16"], 0, 16),
157             (uint32_t)std::stoul(i_additional["0x24 SRC Word 17"], 0, 16),
158             (uint32_t)std::stoul(i_additional["0x28 SRC Word 18"], 0, 16),
159             (uint32_t)std::stoul(i_additional["0x2c SRC Word 19"], 0, 16)});
160 
161         // Populate phyp primary SRC
162 
163         // char array for raw pel src
164         std::array<char, pel::asciiStringSize> srcChars{'0'};
165         std::string srcString;
166 
167         // src from TI info
168         it = i_additional.find("SrcAscii");
169         if (it != i_additional.end())
170         {
171             srcString = it->second;
172         }
173         else
174         {
175             // The entry with key "Subsystem" does not exist in the additional
176             // map. Log the error, create failure event, and return.
177             trace::err("Error the key SrcAscii does not exist in the map.");
178             eventAttentionFail((int)AttnSection::attnLogging |
179                                ATTN_INVALID_KEY);
180             return;
181         }
182 
183         // copy from string to char array
184         srcString.copy(srcChars.data(),
185                        std::min(srcString.size(), pel::asciiStringSize), 0);
186 
187         tiPel->setAsciiString(srcChars); // pel object src is char array
188 
189         // set symptom-id
190         auto symptomId = (srcString.substr(0, 8) + '_');
191 
192         symptomId += (i_additional["0x10 SRC Word 12"]);
193         symptomId += (i_additional["0x14 SRC Word 13"] + '_');
194         symptomId += (i_additional["0x18 SRC Word 14"]);
195         symptomId += (i_additional["0x1c SRC Word 15"] + '_');
196         symptomId += (i_additional["0x20 SRC Word 16"]);
197         symptomId += (i_additional["0x24 SRC Word 17"] + '_');
198         symptomId += (i_additional["0x28 SRC Word 18"]);
199         symptomId += (i_additional["0x2c SRC Word 19"]);
200 
201         // setSymptomId will take care of required null-terminate and padding
202         tiPel->setSymptomId(symptomId);
203     }
204     else
205     {
206         // Populate hostboot SRC words - note HB word 0 from the shared info
207         // data (additional data "0x10 HB Word") is reflected in the PEL as
208         // "reason code" so we zero it here. Also note that the first word
209         // in this group of words starts at word 0 and word 1 does not exits.
210         tiPel->setSrcWords(std::array<uint32_t, pel::numSrcWords>{
211             (uint32_t)0x00000000,
212             (uint32_t)std::stoul(i_additional["0x14 HB Word 2"], 0, 16),
213             (uint32_t)std::stoul(i_additional["0x18 HB Word 3"], 0, 16),
214             (uint32_t)std::stoul(i_additional["0x1c HB Word 4"], 0, 16),
215             (uint32_t)std::stoul(i_additional["0x20 HB Word 5"], 0, 16),
216             (uint32_t)std::stoul(i_additional["0x24 HB Word 6"], 0, 16),
217             (uint32_t)std::stoul(i_additional["0x28 HB Word 7"], 0, 16),
218             (uint32_t)std::stoul(i_additional["0x2c HB Word 8"], 0, 16)});
219 
220         // Populate hostboot primary SRC
221 
222         // char array for raw pel src
223         std::array<char, pel::asciiStringSize> srcChars{'0'};
224         std::string srcString;
225 
226         // src from TI info
227         it = i_additional.find("SrcAscii");
228         if (it != i_additional.end())
229         {
230             srcString = it->second;
231         }
232         else
233         {
234             // The entry with key "Subsystem" does not exist in the additional
235             // map. Log the error, create failure event, and return.
236             trace::err("Error the key SrcAscii does not exist in the map.");
237             eventAttentionFail((int)AttnSection::attnLogging |
238                                ATTN_INVALID_KEY);
239             return;
240         }
241 
242         // copy from string to char array
243         srcString.copy(srcChars.data(),
244                        std::min(srcString.size(), pel::asciiStringSize), 0);
245 
246         tiPel->setAsciiString(srcChars); // pel object src is char array
247 
248         // set symptom-id
249         auto symptomId = (srcString.substr(0, 8) + '_');
250 
251         symptomId += (i_additional["0x10 HB Word 0"]);       // note: word 1
252         symptomId += (i_additional["0x14 HB Word 2"] + '_'); // does not exist
253         symptomId += (i_additional["0x18 HB Word 3"]);
254         symptomId += (i_additional["0x1c HB Word 4"] + '_');
255         symptomId += (i_additional["0x20 HB Word 5"]);
256         symptomId += (i_additional["0x24 HB Word 6"] + '_');
257         symptomId += (i_additional["0x28 HB Word 7"]);
258         symptomId += (i_additional["0x2c HB Word 8"]);
259 
260         // setSymptomId will take care of required null-terminate and padding
261         tiPel->setSymptomId(symptomId);
262     }
263 
264     // set severity, event type and action flags
265     tiPel->setSeverity(static_cast<uint8_t>(pel::Severity::termination));
266     tiPel->setType(static_cast<uint8_t>(pel::EventType::na));
267 
268     auto actionFlags = pel::ActionFlags::service | pel::ActionFlags::report |
269                        pel::ActionFlags::call;
270 
271     it = i_additional.find("hidden");
272     if (it != i_additional.end() && "true" == it->second)
273     {
274         trace::inf("making HB TI PEL hidden");
275         actionFlags = actionFlags | pel::ActionFlags::hidden;
276     }
277 
278     tiPel->setAction(static_cast<uint16_t>(actionFlags));
279 
280     // The raw PEL that we used as the basis for this custom PEL contains some
281     // user data sections that do not need to be in this PEL. However we do
282     // want to include the raw TI information.
283     int ffdcCount = 0;
284     it = i_additional.find("FFDC count");
285     if (it != i_additional.end())
286     {
287         // remove all sections except 1 (raw Ti info)
288         ffdcCount = std::stoi(it->second) - 1;
289     }
290     tiPel->setSectionCount(tiPel->getSectionCount() - ffdcCount);
291 
292     // Update the raw PEL with the new custom PEL data
293     tiPel->raw(i_rawPel);
294 
295     // create PEL from raw data
296     createPelRaw(i_rawPel);
297 }
298 
299 /**
300  * Log an event handled by the attention handler
301  *
302  * Basic (non TI) events will generate a standard message-registry based PEL
303  *
304  * TI events will create two PEL's. One PEL will be informational and will
305  * contain trace information relevent to attention handler. The second PEL
306  * will be specific to the TI type (including the primary SRC) and will be
307  * based off of the TI information provided to the attention handler through
308  * shared TI info data area.
309  *
310  * @param  i_event - The event type
311  * @param  i_additional - Additional PEL data
312  * @param  i_ffdc - FFDC PEL data
313  * @param  i_severity - Severity level
314  * @return Event log Id (0 if no event log generated)
315  */
event(EventType i_event,std::map<std::string,std::string> & i_additional,const std::vector<util::FFDCFile> & i_ffdc,std::string i_severity=levelPelError)316 uint32_t event(EventType i_event,
317                std::map<std::string, std::string>& i_additional,
318                const std::vector<util::FFDCFile>& i_ffdc,
319                std::string i_severity = levelPelError)
320 {
321     uint32_t pelId = 0;      // assume no event log generated
322 
323     bool eventValid = false; // assume no event created
324     bool tiEvent = false;    // assume not a terminate event
325 
326     // count user data sections so we can fixup custom PEL
327     i_additional["FFDC count"] = std::to_string(i_ffdc.size());
328 
329     std::string eventName;
330 
331     switch (i_event)
332     {
333         case EventType::Checkstop:
334             eventName = "org.open_power.HwDiags.Error.Checkstop";
335             eventValid = true;
336             break;
337         case EventType::Terminate:
338             eventName = "org.open_power.Attn.Error.Terminate";
339             eventValid = true;
340             tiEvent = true;
341             break;
342         case EventType::Vital:
343             eventName = "org.open_power.Attn.Error.Vital";
344             eventValid = true;
345             break;
346         case EventType::HwDiagsFail:
347         case EventType::AttentionFail:
348             eventName = "org.open_power.Attn.Error.Fail";
349             eventValid = true;
350             break;
351         default:
352             eventValid = false;
353             break;
354     }
355 
356     if (true == eventValid)
357     {
358         // Create PEL with additional data and FFDC data. The newly created
359         // PEL's platform log-id will be returned.
360         pelId = util::dbus::createPel(eventName, i_severity, i_additional,
361                                       createFFDCTuples(i_ffdc));
362 
363         // If this is a TI event we will create an additional PEL that is
364         // specific to the subsystem that generated the TI.
365         if ((0 != pelId) && (true == tiEvent))
366         {
367             // get file descriptor and size of information PEL
368             int pelFd = getPel(pelId);
369 
370             // if PEL found, read into buffer
371             if (-1 != pelFd)
372             {
373                 auto pelSize = lseek(pelFd, 0, SEEK_END);
374                 lseek(pelFd, 0, SEEK_SET);
375 
376                 // read information PEL into buffer
377                 std::vector<uint8_t> buffer(pelSize);
378                 size_t numBytes = read(pelFd, buffer.data(), buffer.size());
379                 if (buffer.size() != numBytes)
380                 {
381                     trace::err("Error reading event log: %u of %u bytes read",
382                                numBytes, buffer.size());
383                 }
384                 else
385                 {
386                     // create PEL from buffer
387                     createPelCustom(buffer, i_additional);
388                 }
389 
390                 close(pelFd);
391             }
392 
393             std::map<std::string, std::string>::iterator it;
394             uint8_t subsystem;
395 
396             it = i_additional.find("Subsystem");
397             if (it != i_additional.end())
398             {
399                 subsystem = std::stoi(it->second);
400             }
401             else
402             {
403                 // The entry with key "Subsystem" does not exist in the
404                 // additional map. Log the error, create failure event, and
405                 // return.
406                 trace::err(
407                     "Error the key Subsystem does not exist in the map.");
408                 eventAttentionFail((int)AttnSection::attnLogging |
409                                    ATTN_INVALID_KEY);
410                 return 0;
411             }
412 
413             // If not hypervisor TI
414             if (static_cast<uint8_t>(pel::SubsystemID::hypervisor) != subsystem)
415             {
416                 // Request a dump and transition the host
417                 if ("true" == i_additional["Dump"])
418                 {
419                     // will not return until dump is complete
420                     requestDump(pelId, DumpParameters{0, DumpType::Hostboot});
421                 }
422             }
423         }
424     }
425     return pelId;
426 }
427 
428 /**
429  * Commit special attention TI event to log
430  *
431  * Create a event log with provided additional information and standard
432  * FFDC data plus TI FFDC data
433  *
434  * @param i_additional - Additional log data
435  * @param i_ti_InfoData - TI FFDC data
436  */
eventTerminate(std::map<std::string,std::string> i_additionalData,char * i_tiInfoData)437 void eventTerminate(std::map<std::string, std::string> i_additionalData,
438                     char* i_tiInfoData)
439 {
440     uint32_t tiInfoSize = 0; // assume TI info was not available
441 
442     if (nullptr != i_tiInfoData)
443     {
444         tiInfoSize = 56; // assume not hypervisor TI
445 
446         std::map<std::string, std::string>::iterator it;
447         uint8_t subsystem;
448 
449         it = i_additionalData.find("Subsystem");
450         if (it != i_additionalData.end())
451         {
452             subsystem = std::stoi(it->second);
453         }
454         else
455         {
456             // The entry with key "Subsystem" does not exist in the additional
457             // map. Log the error, create failure event, and return.
458             trace::err("Error the key Subsystem does not exist in the map.");
459             eventAttentionFail((int)AttnSection::attnLogging |
460                                ATTN_INVALID_KEY);
461             return;
462         }
463 
464         // If hypervisor
465         if (static_cast<uint8_t>(pel::SubsystemID::hypervisor) == subsystem)
466         {
467             tiInfoSize = 1024; // assume hypervisor max
468 
469             // hypervisor may just want some of the data
470             if (0 == (*(i_tiInfoData + 0x09) & 0x01))
471             {
472                 uint32_t* additionalLength = (uint32_t*)(i_tiInfoData + 0x50);
473                 uint32_t tiAdditional = be32toh(*additionalLength);
474                 tiInfoSize = std::min(tiInfoSize, (84 + tiAdditional));
475             }
476         }
477     }
478 
479     trace::inf("TI info size = %u", tiInfoSize);
480 
481     event(EventType::Terminate, i_additionalData,
482           createFFDCFiles(i_tiInfoData, tiInfoSize));
483 }
484 
485 /** @brief Commit SBE vital event to log, returns event log ID */
eventVital(std::string severity)486 uint32_t eventVital(std::string severity)
487 {
488     // Additional data for log
489     std::map<std::string, std::string> additionalData;
490 
491     // Create log event with additional data and FFDC data
492     return event(EventType::Vital, additionalData, createFFDCFiles(nullptr, 0),
493                  severity);
494 }
495 
496 /**
497  * Commit attention handler failure event to log
498  *
499  * Create an event log containing the specified error code.
500  *
501  * @param i_error - Error code
502  */
eventAttentionFail(int i_error)503 void eventAttentionFail(int i_error)
504 {
505     // Additional data for log
506     std::map<std::string, std::string> additionalData;
507     additionalData["ERROR_CODE"] = std::to_string(i_error);
508 
509     // Create log event with additional data and FFDC data
510     event(EventType::AttentionFail, additionalData,
511           createFFDCFiles(nullptr, 0));
512 }
513 
514 } // namespace attn
515