ras_event.h 6.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238
  1. #undef TRACE_SYSTEM
  2. #define TRACE_SYSTEM ras
  3. #define TRACE_INCLUDE_FILE ras_event
  4. #if !defined(_TRACE_HW_EVENT_MC_H) || defined(TRACE_HEADER_MULTI_READ)
  5. #define _TRACE_HW_EVENT_MC_H
  6. #include <linux/tracepoint.h>
  7. #include <linux/edac.h>
  8. #include <linux/ktime.h>
  9. #include <linux/pci.h>
  10. #include <linux/aer.h>
  11. #include <linux/cper.h>
  12. /*
  13. * MCE Extended Error Log trace event
  14. *
  15. * These events are generated when hardware detects a corrected or
  16. * uncorrected event.
  17. */
  18. /* memory trace event */
  19. #if defined(CONFIG_ACPI_EXTLOG) || defined(CONFIG_ACPI_EXTLOG_MODULE)
  20. TRACE_EVENT(extlog_mem_event,
  21. TP_PROTO(struct cper_sec_mem_err *mem,
  22. u32 err_seq,
  23. const uuid_le *fru_id,
  24. const char *fru_text,
  25. u8 sev),
  26. TP_ARGS(mem, err_seq, fru_id, fru_text, sev),
  27. TP_STRUCT__entry(
  28. __field(u32, err_seq)
  29. __field(u8, etype)
  30. __field(u8, sev)
  31. __field(u64, pa)
  32. __field(u8, pa_mask_lsb)
  33. __field_struct(uuid_le, fru_id)
  34. __string(fru_text, fru_text)
  35. __field_struct(struct cper_mem_err_compact, data)
  36. ),
  37. TP_fast_assign(
  38. __entry->err_seq = err_seq;
  39. if (mem->validation_bits & CPER_MEM_VALID_ERROR_TYPE)
  40. __entry->etype = mem->error_type;
  41. else
  42. __entry->etype = ~0;
  43. __entry->sev = sev;
  44. if (mem->validation_bits & CPER_MEM_VALID_PA)
  45. __entry->pa = mem->physical_addr;
  46. else
  47. __entry->pa = ~0ull;
  48. if (mem->validation_bits & CPER_MEM_VALID_PA_MASK)
  49. __entry->pa_mask_lsb = (u8)__ffs64(mem->physical_addr_mask);
  50. else
  51. __entry->pa_mask_lsb = ~0;
  52. __entry->fru_id = *fru_id;
  53. __assign_str(fru_text, fru_text);
  54. cper_mem_err_pack(mem, &__entry->data);
  55. ),
  56. TP_printk("{%d} %s error: %s physical addr: %016llx (mask lsb: %x) %sFRU: %pUl %.20s",
  57. __entry->err_seq,
  58. cper_severity_str(__entry->sev),
  59. cper_mem_err_type_str(__entry->etype),
  60. __entry->pa,
  61. __entry->pa_mask_lsb,
  62. cper_mem_err_unpack(p, &__entry->data),
  63. &__entry->fru_id,
  64. __get_str(fru_text))
  65. );
  66. #endif
  67. /*
  68. * Hardware Events Report
  69. *
  70. * Those events are generated when hardware detected a corrected or
  71. * uncorrected event, and are meant to replace the current API to report
  72. * errors defined on both EDAC and MCE subsystems.
  73. *
  74. * FIXME: Add events for handling memory errors originated from the
  75. * MCE subsystem.
  76. */
  77. /*
  78. * Hardware-independent Memory Controller specific events
  79. */
  80. /*
  81. * Default error mechanisms for Memory Controller errors (CE and UE)
  82. */
  83. TRACE_EVENT(mc_event,
  84. TP_PROTO(const unsigned int err_type,
  85. const char *error_msg,
  86. const char *label,
  87. const int error_count,
  88. const u8 mc_index,
  89. const s8 top_layer,
  90. const s8 mid_layer,
  91. const s8 low_layer,
  92. unsigned long address,
  93. const u8 grain_bits,
  94. unsigned long syndrome,
  95. const char *driver_detail),
  96. TP_ARGS(err_type, error_msg, label, error_count, mc_index,
  97. top_layer, mid_layer, low_layer, address, grain_bits,
  98. syndrome, driver_detail),
  99. TP_STRUCT__entry(
  100. __field( unsigned int, error_type )
  101. __string( msg, error_msg )
  102. __string( label, label )
  103. __field( u16, error_count )
  104. __field( u8, mc_index )
  105. __field( s8, top_layer )
  106. __field( s8, middle_layer )
  107. __field( s8, lower_layer )
  108. __field( long, address )
  109. __field( u8, grain_bits )
  110. __field( long, syndrome )
  111. __string( driver_detail, driver_detail )
  112. ),
  113. TP_fast_assign(
  114. __entry->error_type = err_type;
  115. __assign_str(msg, error_msg);
  116. __assign_str(label, label);
  117. __entry->error_count = error_count;
  118. __entry->mc_index = mc_index;
  119. __entry->top_layer = top_layer;
  120. __entry->middle_layer = mid_layer;
  121. __entry->lower_layer = low_layer;
  122. __entry->address = address;
  123. __entry->grain_bits = grain_bits;
  124. __entry->syndrome = syndrome;
  125. __assign_str(driver_detail, driver_detail);
  126. ),
  127. TP_printk("%d %s error%s:%s%s on %s (mc:%d location:%d:%d:%d address:0x%08lx grain:%d syndrome:0x%08lx%s%s)",
  128. __entry->error_count,
  129. mc_event_error_type(__entry->error_type),
  130. __entry->error_count > 1 ? "s" : "",
  131. ((char *)__get_str(msg))[0] ? " " : "",
  132. __get_str(msg),
  133. __get_str(label),
  134. __entry->mc_index,
  135. __entry->top_layer,
  136. __entry->middle_layer,
  137. __entry->lower_layer,
  138. __entry->address,
  139. 1 << __entry->grain_bits,
  140. __entry->syndrome,
  141. ((char *)__get_str(driver_detail))[0] ? " " : "",
  142. __get_str(driver_detail))
  143. );
  144. /*
  145. * PCIe AER Trace event
  146. *
  147. * These events are generated when hardware detects a corrected or
  148. * uncorrected event on a PCIe device. The event report has
  149. * the following structure:
  150. *
  151. * char * dev_name - The name of the slot where the device resides
  152. * ([domain:]bus:device.function).
  153. * u32 status - Either the correctable or uncorrectable register
  154. * indicating what error or errors have been seen
  155. * u8 severity - error severity 0:NONFATAL 1:FATAL 2:CORRECTED
  156. */
  157. #define aer_correctable_errors \
  158. {PCI_ERR_COR_RCVR, "Receiver Error"}, \
  159. {PCI_ERR_COR_BAD_TLP, "Bad TLP"}, \
  160. {PCI_ERR_COR_BAD_DLLP, "Bad DLLP"}, \
  161. {PCI_ERR_COR_REP_ROLL, "RELAY_NUM Rollover"}, \
  162. {PCI_ERR_COR_REP_TIMER, "Replay Timer Timeout"}, \
  163. {PCI_ERR_COR_ADV_NFAT, "Advisory Non-Fatal Error"}, \
  164. {PCI_ERR_COR_INTERNAL, "Corrected Internal Error"}, \
  165. {PCI_ERR_COR_LOG_OVER, "Header Log Overflow"}
  166. #define aer_uncorrectable_errors \
  167. {PCI_ERR_UNC_UND, "Undefined"}, \
  168. {PCI_ERR_UNC_DLP, "Data Link Protocol Error"}, \
  169. {PCI_ERR_UNC_SURPDN, "Surprise Down Error"}, \
  170. {PCI_ERR_UNC_POISON_TLP,"Poisoned TLP"}, \
  171. {PCI_ERR_UNC_FCP, "Flow Control Protocol Error"}, \
  172. {PCI_ERR_UNC_COMP_TIME, "Completion Timeout"}, \
  173. {PCI_ERR_UNC_COMP_ABORT,"Completer Abort"}, \
  174. {PCI_ERR_UNC_UNX_COMP, "Unexpected Completion"}, \
  175. {PCI_ERR_UNC_RX_OVER, "Receiver Overflow"}, \
  176. {PCI_ERR_UNC_MALF_TLP, "Malformed TLP"}, \
  177. {PCI_ERR_UNC_ECRC, "ECRC Error"}, \
  178. {PCI_ERR_UNC_UNSUP, "Unsupported Request Error"}, \
  179. {PCI_ERR_UNC_ACSV, "ACS Violation"}, \
  180. {PCI_ERR_UNC_INTN, "Uncorrectable Internal Error"},\
  181. {PCI_ERR_UNC_MCBTLP, "MC Blocked TLP"}, \
  182. {PCI_ERR_UNC_ATOMEG, "AtomicOp Egress Blocked"}, \
  183. {PCI_ERR_UNC_TLPPRE, "TLP Prefix Blocked Error"}
  184. TRACE_EVENT(aer_event,
  185. TP_PROTO(const char *dev_name,
  186. const u32 status,
  187. const u8 severity),
  188. TP_ARGS(dev_name, status, severity),
  189. TP_STRUCT__entry(
  190. __string( dev_name, dev_name )
  191. __field( u32, status )
  192. __field( u8, severity )
  193. ),
  194. TP_fast_assign(
  195. __assign_str(dev_name, dev_name);
  196. __entry->status = status;
  197. __entry->severity = severity;
  198. ),
  199. TP_printk("%s PCIe Bus Error: severity=%s, %s\n",
  200. __get_str(dev_name),
  201. __entry->severity == AER_CORRECTABLE ? "Corrected" :
  202. __entry->severity == AER_FATAL ?
  203. "Fatal" : "Uncorrected, non-fatal",
  204. __entry->severity == AER_CORRECTABLE ?
  205. __print_flags(__entry->status, "|", aer_correctable_errors) :
  206. __print_flags(__entry->status, "|", aer_uncorrectable_errors))
  207. );
  208. #endif /* _TRACE_HW_EVENT_MC_H */
  209. /* This part must be outside protection */
  210. #include <trace/define_trace.h>