429 lines · c
1/* SPDX-License-Identifier: GPL-2.0 */2#undef TRACE_SYSTEM3#define TRACE_SYSTEM ras4#define TRACE_INCLUDE_FILE ras_event5 6#if !defined(_TRACE_HW_EVENT_MC_H) || defined(TRACE_HEADER_MULTI_READ)7#define _TRACE_HW_EVENT_MC_H8 9#include <linux/tracepoint.h>10#include <linux/edac.h>11#include <linux/ktime.h>12#include <linux/pci.h>13#include <linux/aer.h>14#include <linux/cper.h>15#include <linux/mm.h>16 17/*18 * MCE Extended Error Log trace event19 *20 * These events are generated when hardware detects a corrected or21 * uncorrected event.22 */23 24/* memory trace event */25 26#if defined(CONFIG_ACPI_EXTLOG) || defined(CONFIG_ACPI_EXTLOG_MODULE)27TRACE_EVENT(extlog_mem_event,28 TP_PROTO(struct cper_sec_mem_err *mem,29 u32 err_seq,30 const guid_t *fru_id,31 const char *fru_text,32 u8 sev),33 34 TP_ARGS(mem, err_seq, fru_id, fru_text, sev),35 36 TP_STRUCT__entry(37 __field(u32, err_seq)38 __field(u8, etype)39 __field(u8, sev)40 __field(u64, pa)41 __field(u8, pa_mask_lsb)42 __field_struct(guid_t, fru_id)43 __string(fru_text, fru_text)44 __field_struct(struct cper_mem_err_compact, data)45 ),46 47 TP_fast_assign(48 __entry->err_seq = err_seq;49 if (mem->validation_bits & CPER_MEM_VALID_ERROR_TYPE)50 __entry->etype = mem->error_type;51 else52 __entry->etype = ~0;53 __entry->sev = sev;54 if (mem->validation_bits & CPER_MEM_VALID_PA)55 __entry->pa = mem->physical_addr;56 else57 __entry->pa = ~0ull;58 59 if (mem->validation_bits & CPER_MEM_VALID_PA_MASK)60 __entry->pa_mask_lsb = (u8)__ffs64(mem->physical_addr_mask);61 else62 __entry->pa_mask_lsb = ~0;63 __entry->fru_id = *fru_id;64 __assign_str(fru_text);65 cper_mem_err_pack(mem, &__entry->data);66 ),67 68 TP_printk("{%d} %s error: %s physical addr: %016llx (mask lsb: %x) %sFRU: %pUl %.20s",69 __entry->err_seq,70 cper_severity_str(__entry->sev),71 cper_mem_err_type_str(__entry->etype),72 __entry->pa,73 __entry->pa_mask_lsb,74 cper_mem_err_unpack(p, &__entry->data),75 &__entry->fru_id,76 __get_str(fru_text))77);78#endif79 80/*81 * Hardware Events Report82 *83 * Those events are generated when hardware detected a corrected or84 * uncorrected event, and are meant to replace the current API to report85 * errors defined on both EDAC and MCE subsystems.86 *87 * FIXME: Add events for handling memory errors originated from the88 * MCE subsystem.89 */90 91/*92 * Hardware-independent Memory Controller specific events93 */94 95/*96 * Default error mechanisms for Memory Controller errors (CE and UE)97 */98TRACE_EVENT(mc_event,99 100 TP_PROTO(const unsigned int err_type,101 const char *error_msg,102 const char *label,103 const int error_count,104 const u8 mc_index,105 const s8 top_layer,106 const s8 mid_layer,107 const s8 low_layer,108 unsigned long address,109 const u8 grain_bits,110 unsigned long syndrome,111 const char *driver_detail),112 113 TP_ARGS(err_type, error_msg, label, error_count, mc_index,114 top_layer, mid_layer, low_layer, address, grain_bits,115 syndrome, driver_detail),116 117 TP_STRUCT__entry(118 __field( unsigned int, error_type )119 __string( msg, error_msg )120 __string( label, label )121 __field( u16, error_count )122 __field( u8, mc_index )123 __field( s8, top_layer )124 __field( s8, middle_layer )125 __field( s8, lower_layer )126 __field( long, address )127 __field( u8, grain_bits )128 __field( long, syndrome )129 __string( driver_detail, driver_detail )130 ),131 132 TP_fast_assign(133 __entry->error_type = err_type;134 __assign_str(msg);135 __assign_str(label);136 __entry->error_count = error_count;137 __entry->mc_index = mc_index;138 __entry->top_layer = top_layer;139 __entry->middle_layer = mid_layer;140 __entry->lower_layer = low_layer;141 __entry->address = address;142 __entry->grain_bits = grain_bits;143 __entry->syndrome = syndrome;144 __assign_str(driver_detail);145 ),146 147 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)",148 __entry->error_count,149 mc_event_error_type(__entry->error_type),150 __entry->error_count > 1 ? "s" : "",151 __get_str(msg)[0] ? " " : "",152 __get_str(msg),153 __get_str(label),154 __entry->mc_index,155 __entry->top_layer,156 __entry->middle_layer,157 __entry->lower_layer,158 __entry->address,159 1 << __entry->grain_bits,160 __entry->syndrome,161 __get_str(driver_detail)[0] ? " " : "",162 __get_str(driver_detail))163);164 165/*166 * ARM Processor Events Report167 *168 * This event is generated when hardware detects an ARM processor error169 * has occurred. UEFI 2.6 spec section N.2.4.4.170 */171TRACE_EVENT(arm_event,172 173 TP_PROTO(const struct cper_sec_proc_arm *proc),174 175 TP_ARGS(proc),176 177 TP_STRUCT__entry(178 __field(u64, mpidr)179 __field(u64, midr)180 __field(u32, running_state)181 __field(u32, psci_state)182 __field(u8, affinity)183 ),184 185 TP_fast_assign(186 if (proc->validation_bits & CPER_ARM_VALID_AFFINITY_LEVEL)187 __entry->affinity = proc->affinity_level;188 else189 __entry->affinity = ~0;190 if (proc->validation_bits & CPER_ARM_VALID_MPIDR)191 __entry->mpidr = proc->mpidr;192 else193 __entry->mpidr = 0ULL;194 __entry->midr = proc->midr;195 if (proc->validation_bits & CPER_ARM_VALID_RUNNING_STATE) {196 __entry->running_state = proc->running_state;197 __entry->psci_state = proc->psci_state;198 } else {199 __entry->running_state = ~0;200 __entry->psci_state = ~0;201 }202 ),203 204 TP_printk("affinity level: %d; MPIDR: %016llx; MIDR: %016llx; "205 "running state: %d; PSCI state: %d",206 __entry->affinity, __entry->mpidr, __entry->midr,207 __entry->running_state, __entry->psci_state)208);209 210/*211 * Non-Standard Section Report212 *213 * This event is generated when hardware detected a hardware214 * error event, which may be of non-standard section as defined215 * in UEFI spec appendix "Common Platform Error Record", or may216 * be of sections for which TRACE_EVENT is not defined.217 *218 */219TRACE_EVENT(non_standard_event,220 221 TP_PROTO(const guid_t *sec_type,222 const guid_t *fru_id,223 const char *fru_text,224 const u8 sev,225 const u8 *err,226 const u32 len),227 228 TP_ARGS(sec_type, fru_id, fru_text, sev, err, len),229 230 TP_STRUCT__entry(231 __array(char, sec_type, UUID_SIZE)232 __array(char, fru_id, UUID_SIZE)233 __string(fru_text, fru_text)234 __field(u8, sev)235 __field(u32, len)236 __dynamic_array(u8, buf, len)237 ),238 239 TP_fast_assign(240 memcpy(__entry->sec_type, sec_type, UUID_SIZE);241 memcpy(__entry->fru_id, fru_id, UUID_SIZE);242 __assign_str(fru_text);243 __entry->sev = sev;244 __entry->len = len;245 memcpy(__get_dynamic_array(buf), err, len);246 ),247 248 TP_printk("severity: %d; sec type:%pU; FRU: %pU %s; data len:%d; raw data:%s",249 __entry->sev, __entry->sec_type,250 __entry->fru_id, __get_str(fru_text),251 __entry->len,252 __print_hex(__get_dynamic_array(buf), __entry->len))253);254 255/*256 * PCIe AER Trace event257 *258 * These events are generated when hardware detects a corrected or259 * uncorrected event on a PCIe device. The event report has260 * the following structure:261 *262 * char * dev_name - The name of the slot where the device resides263 * ([domain:]bus:device.function).264 * u32 status - Either the correctable or uncorrectable register265 * indicating what error or errors have been seen266 * u8 severity - error severity 0:NONFATAL 1:FATAL 2:CORRECTED267 */268 269#define aer_correctable_errors \270 {PCI_ERR_COR_RCVR, "Receiver Error"}, \271 {PCI_ERR_COR_BAD_TLP, "Bad TLP"}, \272 {PCI_ERR_COR_BAD_DLLP, "Bad DLLP"}, \273 {PCI_ERR_COR_REP_ROLL, "RELAY_NUM Rollover"}, \274 {PCI_ERR_COR_REP_TIMER, "Replay Timer Timeout"}, \275 {PCI_ERR_COR_ADV_NFAT, "Advisory Non-Fatal Error"}, \276 {PCI_ERR_COR_INTERNAL, "Corrected Internal Error"}, \277 {PCI_ERR_COR_LOG_OVER, "Header Log Overflow"}278 279#define aer_uncorrectable_errors \280 {PCI_ERR_UNC_UND, "Undefined"}, \281 {PCI_ERR_UNC_DLP, "Data Link Protocol Error"}, \282 {PCI_ERR_UNC_SURPDN, "Surprise Down Error"}, \283 {PCI_ERR_UNC_POISON_TLP,"Poisoned TLP"}, \284 {PCI_ERR_UNC_FCP, "Flow Control Protocol Error"}, \285 {PCI_ERR_UNC_COMP_TIME, "Completion Timeout"}, \286 {PCI_ERR_UNC_COMP_ABORT,"Completer Abort"}, \287 {PCI_ERR_UNC_UNX_COMP, "Unexpected Completion"}, \288 {PCI_ERR_UNC_RX_OVER, "Receiver Overflow"}, \289 {PCI_ERR_UNC_MALF_TLP, "Malformed TLP"}, \290 {PCI_ERR_UNC_ECRC, "ECRC Error"}, \291 {PCI_ERR_UNC_UNSUP, "Unsupported Request Error"}, \292 {PCI_ERR_UNC_ACSV, "ACS Violation"}, \293 {PCI_ERR_UNC_INTN, "Uncorrectable Internal Error"},\294 {PCI_ERR_UNC_MCBTLP, "MC Blocked TLP"}, \295 {PCI_ERR_UNC_ATOMEG, "AtomicOp Egress Blocked"}, \296 {PCI_ERR_UNC_TLPPRE, "TLP Prefix Blocked Error"}297 298TRACE_EVENT(aer_event,299 TP_PROTO(const char *dev_name,300 const u32 status,301 const u8 severity,302 const u8 tlp_header_valid,303 struct pcie_tlp_log *tlp),304 305 TP_ARGS(dev_name, status, severity, tlp_header_valid, tlp),306 307 TP_STRUCT__entry(308 __string( dev_name, dev_name )309 __field( u32, status )310 __field( u8, severity )311 __field( u8, tlp_header_valid)312 __array( u32, tlp_header, 4 )313 ),314 315 TP_fast_assign(316 __assign_str(dev_name);317 __entry->status = status;318 __entry->severity = severity;319 __entry->tlp_header_valid = tlp_header_valid;320 if (tlp_header_valid) {321 __entry->tlp_header[0] = tlp->dw[0];322 __entry->tlp_header[1] = tlp->dw[1];323 __entry->tlp_header[2] = tlp->dw[2];324 __entry->tlp_header[3] = tlp->dw[3];325 }326 ),327 328 TP_printk("%s PCIe Bus Error: severity=%s, %s, TLP Header=%s\n",329 __get_str(dev_name),330 __entry->severity == AER_CORRECTABLE ? "Corrected" :331 __entry->severity == AER_FATAL ?332 "Fatal" : "Uncorrected, non-fatal",333 __entry->severity == AER_CORRECTABLE ?334 __print_flags(__entry->status, "|", aer_correctable_errors) :335 __print_flags(__entry->status, "|", aer_uncorrectable_errors),336 __entry->tlp_header_valid ?337 __print_array(__entry->tlp_header, 4, 4) :338 "Not available")339);340 341/*342 * memory-failure recovery action result event343 *344 * unsigned long pfn - Page Frame Number of the corrupted page345 * int type - Page types of the corrupted page346 * int result - Result of recovery action347 */348 349#ifdef CONFIG_MEMORY_FAILURE350#define MF_ACTION_RESULT \351 EM ( MF_IGNORED, "Ignored" ) \352 EM ( MF_FAILED, "Failed" ) \353 EM ( MF_DELAYED, "Delayed" ) \354 EMe ( MF_RECOVERED, "Recovered" )355 356#define MF_PAGE_TYPE \357 EM ( MF_MSG_KERNEL, "reserved kernel page" ) \358 EM ( MF_MSG_KERNEL_HIGH_ORDER, "high-order kernel page" ) \359 EM ( MF_MSG_HUGE, "huge page" ) \360 EM ( MF_MSG_FREE_HUGE, "free huge page" ) \361 EM ( MF_MSG_GET_HWPOISON, "get hwpoison page" ) \362 EM ( MF_MSG_UNMAP_FAILED, "unmapping failed page" ) \363 EM ( MF_MSG_DIRTY_SWAPCACHE, "dirty swapcache page" ) \364 EM ( MF_MSG_CLEAN_SWAPCACHE, "clean swapcache page" ) \365 EM ( MF_MSG_DIRTY_MLOCKED_LRU, "dirty mlocked LRU page" ) \366 EM ( MF_MSG_CLEAN_MLOCKED_LRU, "clean mlocked LRU page" ) \367 EM ( MF_MSG_DIRTY_UNEVICTABLE_LRU, "dirty unevictable LRU page" ) \368 EM ( MF_MSG_CLEAN_UNEVICTABLE_LRU, "clean unevictable LRU page" ) \369 EM ( MF_MSG_DIRTY_LRU, "dirty LRU page" ) \370 EM ( MF_MSG_CLEAN_LRU, "clean LRU page" ) \371 EM ( MF_MSG_TRUNCATED_LRU, "already truncated LRU page" ) \372 EM ( MF_MSG_BUDDY, "free buddy page" ) \373 EM ( MF_MSG_DAX, "dax page" ) \374 EM ( MF_MSG_UNSPLIT_THP, "unsplit thp" ) \375 EM ( MF_MSG_ALREADY_POISONED, "already poisoned" ) \376 EMe ( MF_MSG_UNKNOWN, "unknown page" )377 378/*379 * First define the enums in MM_ACTION_RESULT to be exported to userspace380 * via TRACE_DEFINE_ENUM().381 */382#undef EM383#undef EMe384#define EM(a, b) TRACE_DEFINE_ENUM(a);385#define EMe(a, b) TRACE_DEFINE_ENUM(a);386 387MF_ACTION_RESULT388MF_PAGE_TYPE389 390/*391 * Now redefine the EM() and EMe() macros to map the enums to the strings392 * that will be printed in the output.393 */394#undef EM395#undef EMe396#define EM(a, b) { a, b },397#define EMe(a, b) { a, b }398 399TRACE_EVENT(memory_failure_event,400 TP_PROTO(unsigned long pfn,401 int type,402 int result),403 404 TP_ARGS(pfn, type, result),405 406 TP_STRUCT__entry(407 __field(unsigned long, pfn)408 __field(int, type)409 __field(int, result)410 ),411 412 TP_fast_assign(413 __entry->pfn = pfn;414 __entry->type = type;415 __entry->result = result;416 ),417 418 TP_printk("pfn %#lx: recovery action for %s: %s",419 __entry->pfn,420 __print_symbolic(__entry->type, MF_PAGE_TYPE),421 __print_symbolic(__entry->result, MF_ACTION_RESULT)422 )423);424#endif /* CONFIG_MEMORY_FAILURE */425#endif /* _TRACE_HW_EVENT_MC_H */426 427/* This part must be outside protection */428#include <trace/define_trace.h>429