1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef __PERF_RECORD_H 3 #define __PERF_RECORD_H 4 /* 5 * The linux/stddef.h isn't need here, but is needed for __always_inline used 6 * in files included from uapi/linux/perf_event.h such as 7 * /usr/include/linux/swab.h and /usr/include/linux/byteorder/little_endian.h, 8 * detected in at least musl libc, used in Alpine Linux. -acme 9 */ 10 #include <stdio.h> 11 #include <linux/stddef.h> 12 #include <perf/event.h> 13 #include <linux/types.h> 14 15 #include "perf_regs.h" 16 17 struct dso; 18 struct machine; 19 struct perf_event_attr; 20 21 #ifdef __LP64__ 22 /* 23 * /usr/include/inttypes.h uses just 'lu' for PRIu64, but we end up defining 24 * __u64 as long long unsigned int, and then -Werror=format= kicks in and 25 * complains of the mismatched types, so use these two special extra PRI 26 * macros to overcome that. 27 */ 28 #define PRI_lu64 "l" PRIu64 29 #define PRI_lx64 "l" PRIx64 30 #define PRI_ld64 "l" PRId64 31 #else 32 #define PRI_lu64 PRIu64 33 #define PRI_lx64 PRIx64 34 #define PRI_ld64 PRId64 35 #endif 36 37 #define PERF_SAMPLE_MASK \ 38 (PERF_SAMPLE_IP | PERF_SAMPLE_TID | \ 39 PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR | \ 40 PERF_SAMPLE_ID | PERF_SAMPLE_STREAM_ID | \ 41 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD | \ 42 PERF_SAMPLE_IDENTIFIER) 43 44 /* perf sample has 16 bits size limit */ 45 #define PERF_SAMPLE_MAX_SIZE (1 << 16) 46 47 struct regs_dump { 48 u64 abi; 49 u64 mask; 50 u64 *regs; 51 52 /* Cached values/mask filled by first register access. */ 53 u64 cache_regs[PERF_REGS_MAX]; 54 u64 cache_mask; 55 }; 56 57 struct stack_dump { 58 u16 offset; 59 u64 size; 60 char *data; 61 }; 62 63 struct sample_read_value { 64 u64 value; 65 u64 id; 66 }; 67 68 struct sample_read { 69 u64 time_enabled; 70 u64 time_running; 71 union { 72 struct { 73 u64 nr; 74 struct sample_read_value *values; 75 } group; 76 struct sample_read_value one; 77 }; 78 }; 79 80 struct ip_callchain { 81 u64 nr; 82 u64 ips[]; 83 }; 84 85 struct branch_stack; 86 87 enum { 88 PERF_IP_FLAG_BRANCH = 1ULL << 0, 89 PERF_IP_FLAG_CALL = 1ULL << 1, 90 PERF_IP_FLAG_RETURN = 1ULL << 2, 91 PERF_IP_FLAG_CONDITIONAL = 1ULL << 3, 92 PERF_IP_FLAG_SYSCALLRET = 1ULL << 4, 93 PERF_IP_FLAG_ASYNC = 1ULL << 5, 94 PERF_IP_FLAG_INTERRUPT = 1ULL << 6, 95 PERF_IP_FLAG_TX_ABORT = 1ULL << 7, 96 PERF_IP_FLAG_TRACE_BEGIN = 1ULL << 8, 97 PERF_IP_FLAG_TRACE_END = 1ULL << 9, 98 PERF_IP_FLAG_IN_TX = 1ULL << 10, 99 PERF_IP_FLAG_VMENTRY = 1ULL << 11, 100 PERF_IP_FLAG_VMEXIT = 1ULL << 12, 101 }; 102 103 #define PERF_IP_FLAG_CHARS "bcrosyiABEx" 104 105 #define PERF_BRANCH_MASK (\ 106 PERF_IP_FLAG_BRANCH |\ 107 PERF_IP_FLAG_CALL |\ 108 PERF_IP_FLAG_RETURN |\ 109 PERF_IP_FLAG_CONDITIONAL |\ 110 PERF_IP_FLAG_SYSCALLRET |\ 111 PERF_IP_FLAG_ASYNC |\ 112 PERF_IP_FLAG_INTERRUPT |\ 113 PERF_IP_FLAG_TX_ABORT |\ 114 PERF_IP_FLAG_TRACE_BEGIN |\ 115 PERF_IP_FLAG_TRACE_END |\ 116 PERF_IP_FLAG_VMENTRY |\ 117 PERF_IP_FLAG_VMEXIT) 118 119 #define MAX_INSN 16 120 121 struct aux_sample { 122 u64 size; 123 void *data; 124 }; 125 126 struct perf_sample { 127 u64 ip; 128 u32 pid, tid; 129 u64 time; 130 u64 addr; 131 u64 id; 132 u64 stream_id; 133 u64 period; 134 u64 weight; 135 u64 transaction; 136 u64 insn_cnt; 137 u64 cyc_cnt; 138 u32 cpu; 139 u32 raw_size; 140 u64 data_src; 141 u64 phys_addr; 142 u64 data_page_size; 143 u64 code_page_size; 144 u64 cgroup; 145 u32 flags; 146 u16 insn_len; 147 u8 cpumode; 148 u16 misc; 149 u16 ins_lat; 150 bool no_hw_idx; /* No hw_idx collected in branch_stack */ 151 char insn[MAX_INSN]; 152 void *raw_data; 153 struct ip_callchain *callchain; 154 struct branch_stack *branch_stack; 155 struct regs_dump user_regs; 156 struct regs_dump intr_regs; 157 struct stack_dump user_stack; 158 struct sample_read read; 159 struct aux_sample aux_sample; 160 }; 161 162 #define PERF_MEM_DATA_SRC_NONE \ 163 (PERF_MEM_S(OP, NA) |\ 164 PERF_MEM_S(LVL, NA) |\ 165 PERF_MEM_S(SNOOP, NA) |\ 166 PERF_MEM_S(LOCK, NA) |\ 167 PERF_MEM_S(TLB, NA)) 168 169 /* Attribute type for custom synthesized events */ 170 #define PERF_TYPE_SYNTH (INT_MAX + 1U) 171 172 /* Attribute config for custom synthesized events */ 173 enum perf_synth_id { 174 PERF_SYNTH_INTEL_PTWRITE, 175 PERF_SYNTH_INTEL_MWAIT, 176 PERF_SYNTH_INTEL_PWRE, 177 PERF_SYNTH_INTEL_EXSTOP, 178 PERF_SYNTH_INTEL_PWRX, 179 PERF_SYNTH_INTEL_CBR, 180 PERF_SYNTH_INTEL_PSB, 181 }; 182 183 /* 184 * Raw data formats for synthesized events. Note that 4 bytes of padding are 185 * present to match the 'size' member of PERF_SAMPLE_RAW data which is always 186 * 8-byte aligned. That means we must dereference raw_data with an offset of 4. 187 * Refer perf_sample__synth_ptr() and perf_synth__raw_data(). It also means the 188 * structure sizes are 4 bytes bigger than the raw_size, refer 189 * perf_synth__raw_size(). 190 */ 191 192 struct perf_synth_intel_ptwrite { 193 u32 padding; 194 union { 195 struct { 196 u32 ip : 1, 197 reserved : 31; 198 }; 199 u32 flags; 200 }; 201 u64 payload; 202 }; 203 204 struct perf_synth_intel_mwait { 205 u32 padding; 206 u32 reserved; 207 union { 208 struct { 209 u64 hints : 8, 210 reserved1 : 24, 211 extensions : 2, 212 reserved2 : 30; 213 }; 214 u64 payload; 215 }; 216 }; 217 218 struct perf_synth_intel_pwre { 219 u32 padding; 220 u32 reserved; 221 union { 222 struct { 223 u64 reserved1 : 7, 224 hw : 1, 225 subcstate : 4, 226 cstate : 4, 227 reserved2 : 48; 228 }; 229 u64 payload; 230 }; 231 }; 232 233 struct perf_synth_intel_exstop { 234 u32 padding; 235 union { 236 struct { 237 u32 ip : 1, 238 reserved : 31; 239 }; 240 u32 flags; 241 }; 242 }; 243 244 struct perf_synth_intel_pwrx { 245 u32 padding; 246 u32 reserved; 247 union { 248 struct { 249 u64 deepest_cstate : 4, 250 last_cstate : 4, 251 wake_reason : 4, 252 reserved1 : 52; 253 }; 254 u64 payload; 255 }; 256 }; 257 258 struct perf_synth_intel_cbr { 259 u32 padding; 260 union { 261 struct { 262 u32 cbr : 8, 263 reserved1 : 8, 264 max_nonturbo : 8, 265 reserved2 : 8; 266 }; 267 u32 flags; 268 }; 269 u32 freq; 270 u32 reserved3; 271 }; 272 273 struct perf_synth_intel_psb { 274 u32 padding; 275 u32 reserved; 276 u64 offset; 277 }; 278 279 /* 280 * raw_data is always 4 bytes from an 8-byte boundary, so subtract 4 to get 281 * 8-byte alignment. 282 */ 283 static inline void *perf_sample__synth_ptr(struct perf_sample *sample) 284 { 285 return sample->raw_data - 4; 286 } 287 288 static inline void *perf_synth__raw_data(void *p) 289 { 290 return p + 4; 291 } 292 293 #define perf_synth__raw_size(d) (sizeof(d) - 4) 294 295 #define perf_sample__bad_synth_size(s, d) ((s)->raw_size < sizeof(d) - 4) 296 297 enum { 298 PERF_STAT_ROUND_TYPE__INTERVAL = 0, 299 PERF_STAT_ROUND_TYPE__FINAL = 1, 300 }; 301 302 void perf_event__print_totals(void); 303 304 struct perf_cpu_map; 305 struct perf_record_stat_config; 306 struct perf_stat_config; 307 struct perf_tool; 308 309 void perf_event__read_stat_config(struct perf_stat_config *config, 310 struct perf_record_stat_config *event); 311 312 int perf_event__process_comm(struct perf_tool *tool, 313 union perf_event *event, 314 struct perf_sample *sample, 315 struct machine *machine); 316 int perf_event__process_lost(struct perf_tool *tool, 317 union perf_event *event, 318 struct perf_sample *sample, 319 struct machine *machine); 320 int perf_event__process_lost_samples(struct perf_tool *tool, 321 union perf_event *event, 322 struct perf_sample *sample, 323 struct machine *machine); 324 int perf_event__process_aux(struct perf_tool *tool, 325 union perf_event *event, 326 struct perf_sample *sample, 327 struct machine *machine); 328 int perf_event__process_itrace_start(struct perf_tool *tool, 329 union perf_event *event, 330 struct perf_sample *sample, 331 struct machine *machine); 332 int perf_event__process_switch(struct perf_tool *tool, 333 union perf_event *event, 334 struct perf_sample *sample, 335 struct machine *machine); 336 int perf_event__process_namespaces(struct perf_tool *tool, 337 union perf_event *event, 338 struct perf_sample *sample, 339 struct machine *machine); 340 int perf_event__process_cgroup(struct perf_tool *tool, 341 union perf_event *event, 342 struct perf_sample *sample, 343 struct machine *machine); 344 int perf_event__process_mmap(struct perf_tool *tool, 345 union perf_event *event, 346 struct perf_sample *sample, 347 struct machine *machine); 348 int perf_event__process_mmap2(struct perf_tool *tool, 349 union perf_event *event, 350 struct perf_sample *sample, 351 struct machine *machine); 352 int perf_event__process_fork(struct perf_tool *tool, 353 union perf_event *event, 354 struct perf_sample *sample, 355 struct machine *machine); 356 int perf_event__process_exit(struct perf_tool *tool, 357 union perf_event *event, 358 struct perf_sample *sample, 359 struct machine *machine); 360 int perf_event__process_ksymbol(struct perf_tool *tool, 361 union perf_event *event, 362 struct perf_sample *sample, 363 struct machine *machine); 364 int perf_event__process_bpf(struct perf_tool *tool, 365 union perf_event *event, 366 struct perf_sample *sample, 367 struct machine *machine); 368 int perf_event__process_text_poke(struct perf_tool *tool, 369 union perf_event *event, 370 struct perf_sample *sample, 371 struct machine *machine); 372 int perf_event__process(struct perf_tool *tool, 373 union perf_event *event, 374 struct perf_sample *sample, 375 struct machine *machine); 376 377 struct addr_location; 378 379 int machine__resolve(struct machine *machine, struct addr_location *al, 380 struct perf_sample *sample); 381 382 void addr_location__put(struct addr_location *al); 383 384 struct thread; 385 386 bool is_bts_event(struct perf_event_attr *attr); 387 bool sample_addr_correlates_sym(struct perf_event_attr *attr); 388 void thread__resolve(struct thread *thread, struct addr_location *al, 389 struct perf_sample *sample); 390 391 const char *perf_event__name(unsigned int id); 392 393 size_t perf_event__fprintf_comm(union perf_event *event, FILE *fp); 394 size_t perf_event__fprintf_mmap(union perf_event *event, FILE *fp); 395 size_t perf_event__fprintf_mmap2(union perf_event *event, FILE *fp); 396 size_t perf_event__fprintf_task(union perf_event *event, FILE *fp); 397 size_t perf_event__fprintf_aux(union perf_event *event, FILE *fp); 398 size_t perf_event__fprintf_itrace_start(union perf_event *event, FILE *fp); 399 size_t perf_event__fprintf_switch(union perf_event *event, FILE *fp); 400 size_t perf_event__fprintf_thread_map(union perf_event *event, FILE *fp); 401 size_t perf_event__fprintf_cpu_map(union perf_event *event, FILE *fp); 402 size_t perf_event__fprintf_namespaces(union perf_event *event, FILE *fp); 403 size_t perf_event__fprintf_cgroup(union perf_event *event, FILE *fp); 404 size_t perf_event__fprintf_ksymbol(union perf_event *event, FILE *fp); 405 size_t perf_event__fprintf_bpf(union perf_event *event, FILE *fp); 406 size_t perf_event__fprintf_text_poke(union perf_event *event, struct machine *machine,FILE *fp); 407 size_t perf_event__fprintf(union perf_event *event, struct machine *machine, FILE *fp); 408 409 int kallsyms__get_function_start(const char *kallsyms_filename, 410 const char *symbol_name, u64 *addr); 411 412 void *cpu_map_data__alloc(struct perf_cpu_map *map, size_t *size, u16 *type, int *max); 413 void cpu_map_data__synthesize(struct perf_record_cpu_map_data *data, struct perf_cpu_map *map, 414 u16 type, int max); 415 416 void event_attr_init(struct perf_event_attr *attr); 417 418 int perf_event_paranoid(void); 419 bool perf_event_paranoid_check(int max_level); 420 421 extern int sysctl_perf_event_max_stack; 422 extern int sysctl_perf_event_max_contexts_per_stack; 423 extern unsigned int proc_map_timeout; 424 425 #define PAGE_SIZE_NAME_LEN 32 426 char *get_page_size_name(u64 size, char *str); 427 428 void arch_perf_parse_sample_weight(struct perf_sample *data, const __u64 *array, u64 type); 429 void arch_perf_synthesize_sample_weight(const struct perf_sample *data, __u64 *array, u64 type); 430 431 #endif /* __PERF_RECORD_H */ 432