include/ras/ras_event.h

   1 #undef TRACE_SYSTEM
   2 #define TRACE_SYSTEM ras
   3 #define TRACE_INCLUDE_FILE ras_event
   4
   5 #if !defined(_TRACE_HW_EVENT_MC_H) || defined(TRACE_HEADER_MULTI_READ)
   6 #define _TRACE_HW_EVENT_MC_H
   7
   8 #include <linux/tracepoint.h>
   9 #include <linux/edac.h>
  10 #include <linux/ktime.h>
  11 #include <linux/pci.h>
  12 #include <linux/aer.h>
  13 #include <linux/cper.h>
  14 #include <linux/mm.h>
  15
  16 /*
  17  * MCE Extended Error Log trace event
  18  *
  19  * These events are generated when hardware detects a corrected or
  20  * uncorrected event.
  21  */
  22
  23 /* memory trace event */
  24
  25 #if defined(CONFIG_ACPI_EXTLOG) || defined(CONFIG_ACPI_EXTLOG_MODULE)
  26 TRACE_EVENT(extlog_mem_event,
  27         TP_PROTO(struct cper_sec_mem_err *mem,
  28                  u32 err_seq,
  29                  const uuid_le *fru_id,
  30                  const char *fru_text,
  31                  u8 sev),
  32
  33         TP_ARGS(mem, err_seq, fru_id, fru_text, sev),
  34
  35         TP_STRUCT__entry(
  36                 __field(u32, err_seq)
  37                 __field(u8, etype)
  38                 __field(u8, sev)
  39                 __field(u64, pa)
  40                 __field(u8, pa_mask_lsb)
  41                 __field_struct(uuid_le, fru_id)
  42                 __string(fru_text, fru_text)
  43                 __field_struct(struct cper_mem_err_compact, data)
  44         ),
  45
  46         TP_fast_assign(
  47                 __entry->err_seq = err_seq;
  48                 if (mem->validation_bits & CPER_MEM_VALID_ERROR_TYPE)
  49                         __entry->etype = mem->error_type;
  50                 else
  51                         __entry->etype = ~0;
  52                 __entry->sev = sev;
  53                 if (mem->validation_bits & CPER_MEM_VALID_PA)
  54                         __entry->pa = mem->physical_addr;
  55                 else
  56                         __entry->pa = ~0ull;
  57
  58                 if (mem->validation_bits & CPER_MEM_VALID_PA_MASK)
  59                         __entry->pa_mask_lsb = (u8)__ffs64(mem->physical_addr_mask);
  60                 else
  61                         __entry->pa_mask_lsb = ~0;
  62                 __entry->fru_id = *fru_id;
  63                 __assign_str(fru_text, fru_text);
  64                 cper_mem_err_pack(mem, &__entry->data);
  65         ),
  66
  67         TP_printk("{%d} %s error: %s physical addr: %016llx (mask lsb: %x) %sFRU: %pUl %.20s",
  68                   __entry->err_seq,
  69                   cper_severity_str(__entry->sev),
  70                   cper_mem_err_type_str(__entry->etype),
  71                   __entry->pa,
  72                   __entry->pa_mask_lsb,
  73                   cper_mem_err_unpack(p, &__entry->data),
  74                   &__entry->fru_id,
  75                   __get_str(fru_text))
  76 );
  77 #endif
  78
  79 /*
  80  * Hardware Events Report
  81  *
  82  * Those events are generated when hardware detected a corrected or
  83  * uncorrected event, and are meant to replace the current API to report
  84  * errors defined on both EDAC and MCE subsystems.
  85  *
  86  * FIXME: Add events for handling memory errors originated from the
  87  *        MCE subsystem.
  88  */
  89
  90 /*
  91  * Hardware-independent Memory Controller specific events
  92  */
  93
  94 /*
  95  * Default error mechanisms for Memory Controller errors (CE and UE)
  96  */
  97 TRACE_EVENT(mc_event,
  98
  99         TP_PROTO(const unsigned int err_type,
 100                  const char *error_msg,
 101                  const char *label,
 102                  const int error_count,
 103                  const u8 mc_index,
 104                  const s8 top_layer,
 105                  const s8 mid_layer,
 106                  const s8 low_layer,
 107                  unsigned long address,
 108                  const u8 grain_bits,
 109                  unsigned long syndrome,
 110                  const char *driver_detail),
 111
 112         TP_ARGS(err_type, error_msg, label, error_count, mc_index,
 113                 top_layer, mid_layer, low_layer, address, grain_bits,
 114                 syndrome, driver_detail),
 115
 116         TP_STRUCT__entry(
 117                 __field(        unsigned int,   error_type              )
 118                 __string(       msg,            error_msg               )
 119                 __string(       label,          label                   )
 120                 __field(        u16,            error_count             )
 121                 __field(        u8,             mc_index                )
 122                 __field(        s8,             top_layer               )
 123                 __field(        s8,             middle_layer            )
 124                 __field(        s8,             lower_layer             )
 125                 __field(        long,           address                 )
 126                 __field(        u8,             grain_bits              )
 127                 __field(        long,           syndrome                )
 128                 __string(       driver_detail,  driver_detail           )
 129         ),
 130
 131         TP_fast_assign(
 132                 __entry->error_type             = err_type;
 133                 __assign_str(msg, error_msg);
 134                 __assign_str(label, label);
 135                 __entry->error_count            = error_count;
 136                 __entry->mc_index               = mc_index;
 137                 __entry->top_layer              = top_layer;
 138                 __entry->middle_layer           = mid_layer;
 139                 __entry->lower_layer            = low_layer;
 140                 __entry->address                = address;
 141                 __entry->grain_bits             = grain_bits;
 142                 __entry->syndrome               = syndrome;
 143                 __assign_str(driver_detail, driver_detail);
 144         ),
 145
 146         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)",
 147                   __entry->error_count,
 148                   mc_event_error_type(__entry->error_type),
 149                   __entry->error_count > 1 ? "s" : "",
 150                   __get_str(msg)[0] ? " " : "",
 151                   __get_str(msg),
 152                   __get_str(label),
 153                   __entry->mc_index,
 154                   __entry->top_layer,
 155                   __entry->middle_layer,
 156                   __entry->lower_layer,
 157                   __entry->address,
 158                   1 << __entry->grain_bits,
 159                   __entry->syndrome,
 160                   __get_str(driver_detail)[0] ? " " : "",
 161                   __get_str(driver_detail))
 162 );
 163
 164 /*
 165  * Non-Standard Section Report
 166  *
 167  * This event is generated when hardware detected a hardware
 168  * error event, which may be of non-standard section as defined
 169  * in UEFI spec appendix "Common Platform Error Record", or may
 170  * be of sections for which TRACE_EVENT is not defined.
 171  *
 172  */
 173 TRACE_EVENT(non_standard_event,
 174
 175         TP_PROTO(const uuid_le *sec_type,
 176                  const uuid_le *fru_id,
 177                  const char *fru_text,
 178                  const u8 sev,
 179                  const u8 *err,
 180                  const u32 len),
 181
 182         TP_ARGS(sec_type, fru_id, fru_text, sev, err, len),
 183
 184         TP_STRUCT__entry(
 185                 __array(char, sec_type, UUID_SIZE)
 186                 __array(char, fru_id, UUID_SIZE)
 187                 __string(fru_text, fru_text)
 188                 __field(u8, sev)
 189                 __field(u32, len)
 190                 __dynamic_array(u8, buf, len)
 191         ),
 192
 193         TP_fast_assign(
 194                 memcpy(__entry->sec_type, sec_type, UUID_SIZE);
 195                 memcpy(__entry->fru_id, fru_id, UUID_SIZE);
 196                 __assign_str(fru_text, fru_text);
 197                 __entry->sev = sev;
 198                 __entry->len = len;
 199                 memcpy(__get_dynamic_array(buf), err, len);
 200         ),
 201
 202         TP_printk("severity: %d; sec type:%pU; FRU: %pU %s; data len:%d; raw data:%s",
 203                   __entry->sev, __entry->sec_type,
 204                   __entry->fru_id, __get_str(fru_text),
 205                   __entry->len,
 206                   __print_hex(__get_dynamic_array(buf), __entry->len))
 207 );
 208
 209 /*
 210  * PCIe AER Trace event
 211  *
 212  * These events are generated when hardware detects a corrected or
 213  * uncorrected event on a PCIe device. The event report has
 214  * the following structure:
 215  *
 216  * char * dev_name -    The name of the slot where the device resides
 217  *                      ([domain:]bus:device.function).
 218  * u32 status -         Either the correctable or uncorrectable register
 219  *                      indicating what error or errors have been seen
 220  * u8 severity -        error severity 0:NONFATAL 1:FATAL 2:CORRECTED
 221  */
 222
 223 #define aer_correctable_errors                                  \
 224         {PCI_ERR_COR_RCVR,      "Receiver Error"},              \
 225         {PCI_ERR_COR_BAD_TLP,   "Bad TLP"},                     \
 226         {PCI_ERR_COR_BAD_DLLP,  "Bad DLLP"},                    \
 227         {PCI_ERR_COR_REP_ROLL,  "RELAY_NUM Rollover"},          \
 228         {PCI_ERR_COR_REP_TIMER, "Replay Timer Timeout"},        \
 229         {PCI_ERR_COR_ADV_NFAT,  "Advisory Non-Fatal Error"},    \
 230         {PCI_ERR_COR_INTERNAL,  "Corrected Internal Error"},    \
 231         {PCI_ERR_COR_LOG_OVER,  "Header Log Overflow"}
 232
 233 #define aer_uncorrectable_errors                                \
 234         {PCI_ERR_UNC_UND,       "Undefined"},                   \
 235         {PCI_ERR_UNC_DLP,       "Data Link Protocol Error"},    \
 236         {PCI_ERR_UNC_SURPDN,    "Surprise Down Error"},         \
 237         {PCI_ERR_UNC_POISON_TLP,"Poisoned TLP"},                \
 238         {PCI_ERR_UNC_FCP,       "Flow Control Protocol Error"}, \
 239         {PCI_ERR_UNC_COMP_TIME, "Completion Timeout"},          \
 240         {PCI_ERR_UNC_COMP_ABORT,"Completer Abort"},             \
 241         {PCI_ERR_UNC_UNX_COMP,  "Unexpected Completion"},       \
 242         {PCI_ERR_UNC_RX_OVER,   "Receiver Overflow"},           \
 243         {PCI_ERR_UNC_MALF_TLP,  "Malformed TLP"},               \
 244         {PCI_ERR_UNC_ECRC,      "ECRC Error"},                  \
 245         {PCI_ERR_UNC_UNSUP,     "Unsupported Request Error"},   \
 246         {PCI_ERR_UNC_ACSV,      "ACS Violation"},               \
 247         {PCI_ERR_UNC_INTN,      "Uncorrectable Internal Error"},\
 248         {PCI_ERR_UNC_MCBTLP,    "MC Blocked TLP"},              \
 249         {PCI_ERR_UNC_ATOMEG,    "AtomicOp Egress Blocked"},     \
 250         {PCI_ERR_UNC_TLPPRE,    "TLP Prefix Blocked Error"}
 251
 252 TRACE_EVENT(aer_event,
 253         TP_PROTO(const char *dev_name,
 254                  const u32 status,
 255                  const u8 severity),
 256
 257         TP_ARGS(dev_name, status, severity),
 258
 259         TP_STRUCT__entry(
 260                 __string(       dev_name,       dev_name        )
 261                 __field(        u32,            status          )
 262                 __field(        u8,             severity        )
 263         ),
 264
 265         TP_fast_assign(
 266                 __assign_str(dev_name, dev_name);
 267                 __entry->status         = status;
 268                 __entry->severity       = severity;
 269         ),
 270
 271         TP_printk("%s PCIe Bus Error: severity=%s, %s\n",
 272                 __get_str(dev_name),
 273                 __entry->severity == AER_CORRECTABLE ? "Corrected" :
 274                         __entry->severity == AER_FATAL ?
 275                         "Fatal" : "Uncorrected, non-fatal",
 276                 __entry->severity == AER_CORRECTABLE ?
 277                 __print_flags(__entry->status, "|", aer_correctable_errors) :
 278                 __print_flags(__entry->status, "|", aer_uncorrectable_errors))
 279 );
 280
 281 /*
 282  * memory-failure recovery action result event
 283  *
 284  * unsigned long pfn -  Page Frame Number of the corrupted page
 285  * int type     -       Page types of the corrupted page
 286  * int result   -       Result of recovery action
 287  */
 288
 289 #ifdef CONFIG_MEMORY_FAILURE
 290 #define MF_ACTION_RESULT        \
 291         EM ( MF_IGNORED, "Ignored" )    \
 292         EM ( MF_FAILED,  "Failed" )     \
 293         EM ( MF_DELAYED, "Delayed" )    \
 294         EMe ( MF_RECOVERED, "Recovered" )
 295
 296 #define MF_PAGE_TYPE            \
 297         EM ( MF_MSG_KERNEL, "reserved kernel page" )                    \
 298         EM ( MF_MSG_KERNEL_HIGH_ORDER, "high-order kernel page" )       \
 299         EM ( MF_MSG_SLAB, "kernel slab page" )                          \
 300         EM ( MF_MSG_DIFFERENT_COMPOUND, "different compound page after locking" ) \
 301         EM ( MF_MSG_POISONED_HUGE, "huge page already hardware poisoned" )      \
 302         EM ( MF_MSG_HUGE, "huge page" )                                 \
 303         EM ( MF_MSG_FREE_HUGE, "free huge page" )                       \
 304         EM ( MF_MSG_UNMAP_FAILED, "unmapping failed page" )             \
 305         EM ( MF_MSG_DIRTY_SWAPCACHE, "dirty swapcache page" )           \
 306         EM ( MF_MSG_CLEAN_SWAPCACHE, "clean swapcache page" )           \
 307         EM ( MF_MSG_DIRTY_MLOCKED_LRU, "dirty mlocked LRU page" )       \
 308         EM ( MF_MSG_CLEAN_MLOCKED_LRU, "clean mlocked LRU page" )       \
 309         EM ( MF_MSG_DIRTY_UNEVICTABLE_LRU, "dirty unevictable LRU page" )       \
 310         EM ( MF_MSG_CLEAN_UNEVICTABLE_LRU, "clean unevictable LRU page" )       \
 311         EM ( MF_MSG_DIRTY_LRU, "dirty LRU page" )                       \
 312         EM ( MF_MSG_CLEAN_LRU, "clean LRU page" )                       \
 313         EM ( MF_MSG_TRUNCATED_LRU, "already truncated LRU page" )       \
 314         EM ( MF_MSG_BUDDY, "free buddy page" )                          \
 315         EM ( MF_MSG_BUDDY_2ND, "free buddy page (2nd try)" )            \
 316         EMe ( MF_MSG_UNKNOWN, "unknown page" )
 317
 318 /*
 319  * First define the enums in MM_ACTION_RESULT to be exported to userspace
 320  * via TRACE_DEFINE_ENUM().
 321  */
 322 #undef EM
 323 #undef EMe
 324 #define EM(a, b) TRACE_DEFINE_ENUM(a);
 325 #define EMe(a, b)       TRACE_DEFINE_ENUM(a);
 326
 327 MF_ACTION_RESULT
 328 MF_PAGE_TYPE
 329
 330 /*
 331  * Now redefine the EM() and EMe() macros to map the enums to the strings
 332  * that will be printed in the output.
 333  */
 334 #undef EM
 335 #undef EMe
 336 #define EM(a, b)                { a, b },
 337 #define EMe(a, b)       { a, b }
 338
 339 TRACE_EVENT(memory_failure_event,
 340         TP_PROTO(unsigned long pfn,
 341                  int type,
 342                  int result),
 343
 344         TP_ARGS(pfn, type, result),
 345
 346         TP_STRUCT__entry(
 347                 __field(unsigned long, pfn)
 348                 __field(int, type)
 349                 __field(int, result)
 350         ),
 351
 352         TP_fast_assign(
 353                 __entry->pfn    = pfn;
 354                 __entry->type   = type;
 355                 __entry->result = result;
 356         ),
 357
 358         TP_printk("pfn %#lx: recovery action for %s: %s",
 359                 __entry->pfn,
 360                 __print_symbolic(__entry->type, MF_PAGE_TYPE),
 361                 __print_symbolic(__entry->result, MF_ACTION_RESULT)
 362         )
 363 );
 364 #endif /* CONFIG_MEMORY_FAILURE */
 365 #endif /* _TRACE_HW_EVENT_MC_H */
 366
 367 /* This part must be outside protection */
 368 #include <trace/define_trace.h>