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
  15 /*
  16  * MCE Extended Error Log trace event
  17  *
  18  * These events are generated when hardware detects a corrected or
  19  * uncorrected event.
  20  */
  21
  22 /* memory trace event */
  23
  24 #if defined(CONFIG_ACPI_EXTLOG) || defined(CONFIG_ACPI_EXTLOG_MODULE)
  25 TRACE_EVENT(extlog_mem_event,
  26         TP_PROTO(struct cper_sec_mem_err *mem,
  27                  u32 err_seq,
  28                  const uuid_le *fru_id,
  29                  const char *fru_text,
  30                  u8 sev),
  31
  32         TP_ARGS(mem, err_seq, fru_id, fru_text, sev),
  33
  34         TP_STRUCT__entry(
  35                 __field(u32, err_seq)
  36                 __field(u8, etype)
  37                 __field(u8, sev)
  38                 __field(u64, pa)
  39                 __field(u8, pa_mask_lsb)
  40                 __field_struct(uuid_le, fru_id)
  41                 __string(fru_text, fru_text)
  42                 __field_struct(struct cper_mem_err_compact, data)
  43         ),
  44
  45         TP_fast_assign(
  46                 __entry->err_seq = err_seq;
  47                 if (mem->validation_bits & CPER_MEM_VALID_ERROR_TYPE)
  48                         __entry->etype = mem->error_type;
  49                 else
  50                         __entry->etype = ~0;
  51                 __entry->sev = sev;
  52                 if (mem->validation_bits & CPER_MEM_VALID_PA)
  53                         __entry->pa = mem->physical_addr;
  54                 else
  55                         __entry->pa = ~0ull;
  56
  57                 if (mem->validation_bits & CPER_MEM_VALID_PA_MASK)
  58                         __entry->pa_mask_lsb = (u8)__ffs64(mem->physical_addr_mask);
  59                 else
  60                         __entry->pa_mask_lsb = ~0;
  61                 __entry->fru_id = *fru_id;
  62                 __assign_str(fru_text, fru_text);
  63                 cper_mem_err_pack(mem, &__entry->data);
  64         ),
  65
  66         TP_printk("{%d} %s error: %s physical addr: %016llx (mask lsb: %x) %sFRU: %pUl %.20s",
  67                   __entry->err_seq,
  68                   cper_severity_str(__entry->sev),
  69                   cper_mem_err_type_str(__entry->etype),
  70                   __entry->pa,
  71                   __entry->pa_mask_lsb,
  72                   cper_mem_err_unpack(p, &__entry->data),
  73                   &__entry->fru_id,
  74                   __get_str(fru_text))
  75 );
  76 #endif
  77
  78 /*
  79  * Hardware Events Report
  80  *
  81  * Those events are generated when hardware detected a corrected or
  82  * uncorrected event, and are meant to replace the current API to report
  83  * errors defined on both EDAC and MCE subsystems.
  84  *
  85  * FIXME: Add events for handling memory errors originated from the
  86  *        MCE subsystem.
  87  */
  88
  89 /*
  90  * Hardware-independent Memory Controller specific events
  91  */
  92
  93 /*
  94  * Default error mechanisms for Memory Controller errors (CE and UE)
  95  */
  96 TRACE_EVENT(mc_event,
  97
  98         TP_PROTO(const unsigned int err_type,
  99                  const char *error_msg,
 100                  const char *label,
 101                  const int error_count,
 102                  const u8 mc_index,
 103                  const s8 top_layer,
 104                  const s8 mid_layer,
 105                  const s8 low_layer,
 106                  unsigned long address,
 107                  const u8 grain_bits,
 108                  unsigned long syndrome,
 109                  const char *driver_detail),
 110
 111         TP_ARGS(err_type, error_msg, label, error_count, mc_index,
 112                 top_layer, mid_layer, low_layer, address, grain_bits,
 113                 syndrome, driver_detail),
 114
 115         TP_STRUCT__entry(
 116                 __field(        unsigned int,   error_type              )
 117                 __string(       msg,            error_msg               )
 118                 __string(       label,          label                   )
 119                 __field(        u16,            error_count             )
 120                 __field(        u8,             mc_index                )
 121                 __field(        s8,             top_layer               )
 122                 __field(        s8,             middle_layer            )
 123                 __field(        s8,             lower_layer             )
 124                 __field(        long,           address                 )
 125                 __field(        u8,             grain_bits              )
 126                 __field(        long,           syndrome                )
 127                 __string(       driver_detail,  driver_detail           )
 128         ),
 129
 130         TP_fast_assign(
 131                 __entry->error_type             = err_type;
 132                 __assign_str(msg, error_msg);
 133                 __assign_str(label, label);
 134                 __entry->error_count            = error_count;
 135                 __entry->mc_index               = mc_index;
 136                 __entry->top_layer              = top_layer;
 137                 __entry->middle_layer           = mid_layer;
 138                 __entry->lower_layer            = low_layer;
 139                 __entry->address                = address;
 140                 __entry->grain_bits             = grain_bits;
 141                 __entry->syndrome               = syndrome;
 142                 __assign_str(driver_detail, driver_detail);
 143         ),
 144
 145         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)",
 146                   __entry->error_count,
 147                   mc_event_error_type(__entry->error_type),
 148                   __entry->error_count > 1 ? "s" : "",
 149                   ((char *)__get_str(msg))[0] ? " " : "",
 150                   __get_str(msg),
 151                   __get_str(label),
 152                   __entry->mc_index,
 153                   __entry->top_layer,
 154                   __entry->middle_layer,
 155                   __entry->lower_layer,
 156                   __entry->address,
 157                   1 << __entry->grain_bits,
 158                   __entry->syndrome,
 159                   ((char *)__get_str(driver_detail))[0] ? " " : "",
 160                   __get_str(driver_detail))
 161 );
 162
 163 /*
 164  * PCIe AER Trace event
 165  *
 166  * These events are generated when hardware detects a corrected or
 167  * uncorrected event on a PCIe device. The event report has
 168  * the following structure:
 169  *
 170  * char * dev_name -    The name of the slot where the device resides
 171  *                      ([domain:]bus:device.function).
 172  * u32 status -         Either the correctable or uncorrectable register
 173  *                      indicating what error or errors have been seen
 174  * u8 severity -        error severity 0:NONFATAL 1:FATAL 2:CORRECTED
 175  */
 176
 177 #define aer_correctable_errors                                  \
 178         {PCI_ERR_COR_RCVR,      "Receiver Error"},              \
 179         {PCI_ERR_COR_BAD_TLP,   "Bad TLP"},                     \
 180         {PCI_ERR_COR_BAD_DLLP,  "Bad DLLP"},                    \
 181         {PCI_ERR_COR_REP_ROLL,  "RELAY_NUM Rollover"},          \
 182         {PCI_ERR_COR_REP_TIMER, "Replay Timer Timeout"},        \
 183         {PCI_ERR_COR_ADV_NFAT,  "Advisory Non-Fatal Error"},    \
 184         {PCI_ERR_COR_INTERNAL,  "Corrected Internal Error"},    \
 185         {PCI_ERR_COR_LOG_OVER,  "Header Log Overflow"}
 186
 187 #define aer_uncorrectable_errors                                \
 188         {PCI_ERR_UNC_UND,       "Undefined"},                   \
 189         {PCI_ERR_UNC_DLP,       "Data Link Protocol Error"},    \
 190         {PCI_ERR_UNC_SURPDN,    "Surprise Down Error"},         \
 191         {PCI_ERR_UNC_POISON_TLP,"Poisoned TLP"},                \
 192         {PCI_ERR_UNC_FCP,       "Flow Control Protocol Error"}, \
 193         {PCI_ERR_UNC_COMP_TIME, "Completion Timeout"},          \
 194         {PCI_ERR_UNC_COMP_ABORT,"Completer Abort"},             \
 195         {PCI_ERR_UNC_UNX_COMP,  "Unexpected Completion"},       \
 196         {PCI_ERR_UNC_RX_OVER,   "Receiver Overflow"},           \
 197         {PCI_ERR_UNC_MALF_TLP,  "Malformed TLP"},               \
 198         {PCI_ERR_UNC_ECRC,      "ECRC Error"},                  \
 199         {PCI_ERR_UNC_UNSUP,     "Unsupported Request Error"},   \
 200         {PCI_ERR_UNC_ACSV,      "ACS Violation"},               \
 201         {PCI_ERR_UNC_INTN,      "Uncorrectable Internal Error"},\
 202         {PCI_ERR_UNC_MCBTLP,    "MC Blocked TLP"},              \
 203         {PCI_ERR_UNC_ATOMEG,    "AtomicOp Egress Blocked"},     \
 204         {PCI_ERR_UNC_TLPPRE,    "TLP Prefix Blocked Error"}
 205
 206 TRACE_EVENT(aer_event,
 207         TP_PROTO(const char *dev_name,
 208                  const u32 status,
 209                  const u8 severity),
 210
 211         TP_ARGS(dev_name, status, severity),
 212
 213         TP_STRUCT__entry(
 214                 __string(       dev_name,       dev_name        )
 215                 __field(        u32,            status          )
 216                 __field(        u8,             severity        )
 217         ),
 218
 219         TP_fast_assign(
 220                 __assign_str(dev_name, dev_name);
 221                 __entry->status         = status;
 222                 __entry->severity       = severity;
 223         ),
 224
 225         TP_printk("%s PCIe Bus Error: severity=%s, %s\n",
 226                 __get_str(dev_name),
 227                 __entry->severity == AER_CORRECTABLE ? "Corrected" :
 228                         __entry->severity == AER_FATAL ?
 229                         "Fatal" : "Uncorrected, non-fatal",
 230                 __entry->severity == AER_CORRECTABLE ?
 231                 __print_flags(__entry->status, "|", aer_correctable_errors) :
 232                 __print_flags(__entry->status, "|", aer_uncorrectable_errors))
 233 );
 234
 235 #endif /* _TRACE_HW_EVENT_MC_H */
 236
 237 /* This part must be outside protection */
 238 #include <trace/define_trace.h>