1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * intel-bts.c: Intel Processor Trace support 4 * Copyright (c) 2013-2015, Intel Corporation. 5 */ 6 7 #include "../../../util/intel-bts.h" 8 9 #include <errno.h> 10 11 #include <linux/bitops.h> 12 #include <linux/kernel.h> 13 #include <linux/log2.h> 14 #include <linux/types.h> 15 #include <linux/zalloc.h> 16 17 #include <internal/lib.h> // page_size 18 19 #include "../../../util/auxtrace.h" 20 #include "../../../util/cpumap.h" 21 #include "../../../util/debug.h" 22 #include "../../../util/event.h" 23 #include "../../../util/evlist.h" 24 #include "../../../util/evsel.h" 25 #include "../../../util/mmap.h" 26 #include "../../../util/pmu.h" 27 #include "../../../util/pmus.h" 28 #include "../../../util/record.h" 29 #include "../../../util/session.h" 30 #include "../../../util/tsc.h" 31 32 #define KiB(x) ((x) * 1024) 33 #define MiB(x) ((x) * 1024 * 1024) 34 #define KiB_MASK(x) (KiB(x) - 1) 35 #define MiB_MASK(x) (MiB(x) - 1) 36 37 struct intel_bts_snapshot_ref { 38 void *ref_buf; 39 size_t ref_offset; 40 bool wrapped; 41 }; 42 43 struct intel_bts_recording { 44 struct auxtrace_record itr; 45 struct perf_pmu *intel_bts_pmu; 46 struct evlist *evlist; 47 bool snapshot_mode; 48 size_t snapshot_size; 49 int snapshot_ref_cnt; 50 struct intel_bts_snapshot_ref *snapshot_refs; 51 }; 52 53 struct branch { 54 u64 from; 55 u64 to; 56 u64 misc; 57 }; 58 59 static size_t 60 intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused, 61 struct evlist *evlist __maybe_unused) 62 { 63 return INTEL_BTS_AUXTRACE_PRIV_SIZE; 64 } 65 66 static int intel_bts_info_fill(struct auxtrace_record *itr, 67 struct perf_session *session, 68 struct perf_record_auxtrace_info *auxtrace_info, 69 size_t priv_size) 70 { 71 struct intel_bts_recording *btsr = 72 container_of(itr, struct intel_bts_recording, itr); 73 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu; 74 struct perf_event_mmap_page *pc; 75 struct perf_tsc_conversion tc = { .time_mult = 0, }; 76 bool cap_user_time_zero = false; 77 int err; 78 79 if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE) 80 return -EINVAL; 81 82 if (!session->evlist->core.nr_mmaps) 83 return -EINVAL; 84 85 pc = session->evlist->mmap[0].core.base; 86 if (pc) { 87 err = perf_read_tsc_conversion(pc, &tc); 88 if (err) { 89 if (err != -EOPNOTSUPP) 90 return err; 91 } else { 92 cap_user_time_zero = tc.time_mult != 0; 93 } 94 if (!cap_user_time_zero) 95 ui__warning("Intel BTS: TSC not available\n"); 96 } 97 98 auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS; 99 auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type; 100 auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift; 101 auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult; 102 auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero; 103 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero; 104 auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode; 105 106 return 0; 107 } 108 109 static int intel_bts_recording_options(struct auxtrace_record *itr, 110 struct evlist *evlist, 111 struct record_opts *opts) 112 { 113 struct intel_bts_recording *btsr = 114 container_of(itr, struct intel_bts_recording, itr); 115 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu; 116 struct evsel *evsel, *intel_bts_evsel = NULL; 117 const struct perf_cpu_map *cpus = evlist->core.user_requested_cpus; 118 bool privileged = perf_event_paranoid_check(-1); 119 120 if (opts->auxtrace_sample_mode) { 121 pr_err("Intel BTS does not support AUX area sampling\n"); 122 return -EINVAL; 123 } 124 125 btsr->evlist = evlist; 126 btsr->snapshot_mode = opts->auxtrace_snapshot_mode; 127 128 evlist__for_each_entry(evlist, evsel) { 129 if (evsel->core.attr.type == intel_bts_pmu->type) { 130 if (intel_bts_evsel) { 131 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n"); 132 return -EINVAL; 133 } 134 evsel->core.attr.freq = 0; 135 evsel->core.attr.sample_period = 1; 136 evsel->needs_auxtrace_mmap = true; 137 intel_bts_evsel = evsel; 138 opts->full_auxtrace = true; 139 } 140 } 141 142 if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) { 143 pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n"); 144 return -EINVAL; 145 } 146 147 if (!opts->full_auxtrace) 148 return 0; 149 150 if (opts->full_auxtrace && !perf_cpu_map__is_any_cpu_or_is_empty(cpus)) { 151 pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n"); 152 return -EINVAL; 153 } 154 155 /* Set default sizes for snapshot mode */ 156 if (opts->auxtrace_snapshot_mode) { 157 if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) { 158 if (privileged) { 159 opts->auxtrace_mmap_pages = MiB(4) / page_size; 160 } else { 161 opts->auxtrace_mmap_pages = KiB(128) / page_size; 162 if (opts->mmap_pages == UINT_MAX) 163 opts->mmap_pages = KiB(256) / page_size; 164 } 165 } else if (!opts->auxtrace_mmap_pages && !privileged && 166 opts->mmap_pages == UINT_MAX) { 167 opts->mmap_pages = KiB(256) / page_size; 168 } 169 if (!opts->auxtrace_snapshot_size) 170 opts->auxtrace_snapshot_size = 171 opts->auxtrace_mmap_pages * (size_t)page_size; 172 if (!opts->auxtrace_mmap_pages) { 173 size_t sz = opts->auxtrace_snapshot_size; 174 175 sz = round_up(sz, page_size) / page_size; 176 opts->auxtrace_mmap_pages = roundup_pow_of_two(sz); 177 } 178 if (opts->auxtrace_snapshot_size > 179 opts->auxtrace_mmap_pages * (size_t)page_size) { 180 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n", 181 opts->auxtrace_snapshot_size, 182 opts->auxtrace_mmap_pages * (size_t)page_size); 183 return -EINVAL; 184 } 185 if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) { 186 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n"); 187 return -EINVAL; 188 } 189 pr_debug2("Intel BTS snapshot size: %zu\n", 190 opts->auxtrace_snapshot_size); 191 } 192 193 /* Set default sizes for full trace mode */ 194 if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) { 195 if (privileged) { 196 opts->auxtrace_mmap_pages = MiB(4) / page_size; 197 } else { 198 opts->auxtrace_mmap_pages = KiB(128) / page_size; 199 if (opts->mmap_pages == UINT_MAX) 200 opts->mmap_pages = KiB(256) / page_size; 201 } 202 } 203 204 /* Validate auxtrace_mmap_pages */ 205 if (opts->auxtrace_mmap_pages) { 206 size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size; 207 size_t min_sz; 208 209 if (opts->auxtrace_snapshot_mode) 210 min_sz = KiB(4); 211 else 212 min_sz = KiB(8); 213 214 if (sz < min_sz || !is_power_of_2(sz)) { 215 pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n", 216 min_sz / 1024); 217 return -EINVAL; 218 } 219 } 220 221 if (intel_bts_evsel) { 222 /* 223 * To obtain the auxtrace buffer file descriptor, the auxtrace event 224 * must come first. 225 */ 226 evlist__to_front(evlist, intel_bts_evsel); 227 /* 228 * In the case of per-cpu mmaps, we need the CPU on the 229 * AUX event. 230 */ 231 if (!perf_cpu_map__is_any_cpu_or_is_empty(cpus)) 232 evsel__set_sample_bit(intel_bts_evsel, CPU); 233 } 234 235 /* Add dummy event to keep tracking */ 236 if (opts->full_auxtrace) { 237 struct evsel *tracking_evsel; 238 int err; 239 240 err = parse_event(evlist, "dummy:u"); 241 if (err) 242 return err; 243 244 tracking_evsel = evlist__last(evlist); 245 246 evlist__set_tracking_event(evlist, tracking_evsel); 247 248 tracking_evsel->core.attr.freq = 0; 249 tracking_evsel->core.attr.sample_period = 1; 250 } 251 252 return 0; 253 } 254 255 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr, 256 struct record_opts *opts, 257 const char *str) 258 { 259 struct intel_bts_recording *btsr = 260 container_of(itr, struct intel_bts_recording, itr); 261 unsigned long long snapshot_size = 0; 262 char *endptr; 263 264 if (str) { 265 snapshot_size = strtoull(str, &endptr, 0); 266 if (*endptr || snapshot_size > SIZE_MAX) 267 return -1; 268 } 269 270 opts->auxtrace_snapshot_mode = true; 271 opts->auxtrace_snapshot_size = snapshot_size; 272 273 btsr->snapshot_size = snapshot_size; 274 275 return 0; 276 } 277 278 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused) 279 { 280 return rdtsc(); 281 } 282 283 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr, 284 int idx) 285 { 286 const size_t sz = sizeof(struct intel_bts_snapshot_ref); 287 int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2; 288 struct intel_bts_snapshot_ref *refs; 289 290 if (!new_cnt) 291 new_cnt = 16; 292 293 while (new_cnt <= idx) 294 new_cnt *= 2; 295 296 refs = calloc(new_cnt, sz); 297 if (!refs) 298 return -ENOMEM; 299 300 memcpy(refs, btsr->snapshot_refs, cnt * sz); 301 302 btsr->snapshot_refs = refs; 303 btsr->snapshot_ref_cnt = new_cnt; 304 305 return 0; 306 } 307 308 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr) 309 { 310 int i; 311 312 for (i = 0; i < btsr->snapshot_ref_cnt; i++) 313 zfree(&btsr->snapshot_refs[i].ref_buf); 314 zfree(&btsr->snapshot_refs); 315 } 316 317 static void intel_bts_recording_free(struct auxtrace_record *itr) 318 { 319 struct intel_bts_recording *btsr = 320 container_of(itr, struct intel_bts_recording, itr); 321 322 intel_bts_free_snapshot_refs(btsr); 323 free(btsr); 324 } 325 326 static int intel_bts_snapshot_start(struct auxtrace_record *itr) 327 { 328 struct intel_bts_recording *btsr = 329 container_of(itr, struct intel_bts_recording, itr); 330 struct evsel *evsel; 331 332 evlist__for_each_entry(btsr->evlist, evsel) { 333 if (evsel->core.attr.type == btsr->intel_bts_pmu->type) 334 return evsel__disable(evsel); 335 } 336 return -EINVAL; 337 } 338 339 static int intel_bts_snapshot_finish(struct auxtrace_record *itr) 340 { 341 struct intel_bts_recording *btsr = 342 container_of(itr, struct intel_bts_recording, itr); 343 struct evsel *evsel; 344 345 evlist__for_each_entry(btsr->evlist, evsel) { 346 if (evsel->core.attr.type == btsr->intel_bts_pmu->type) 347 return evsel__enable(evsel); 348 } 349 return -EINVAL; 350 } 351 352 static bool intel_bts_first_wrap(u64 *data, size_t buf_size) 353 { 354 int i, a, b; 355 356 b = buf_size >> 3; 357 a = b - 512; 358 if (a < 0) 359 a = 0; 360 361 for (i = a; i < b; i++) { 362 if (data[i]) 363 return true; 364 } 365 366 return false; 367 } 368 369 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx, 370 struct auxtrace_mmap *mm, unsigned char *data, 371 u64 *head, u64 *old) 372 { 373 struct intel_bts_recording *btsr = 374 container_of(itr, struct intel_bts_recording, itr); 375 bool wrapped; 376 int err; 377 378 pr_debug3("%s: mmap index %d old head %zu new head %zu\n", 379 __func__, idx, (size_t)*old, (size_t)*head); 380 381 if (idx >= btsr->snapshot_ref_cnt) { 382 err = intel_bts_alloc_snapshot_refs(btsr, idx); 383 if (err) 384 goto out_err; 385 } 386 387 wrapped = btsr->snapshot_refs[idx].wrapped; 388 if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) { 389 btsr->snapshot_refs[idx].wrapped = true; 390 wrapped = true; 391 } 392 393 /* 394 * In full trace mode 'head' continually increases. However in snapshot 395 * mode 'head' is an offset within the buffer. Here 'old' and 'head' 396 * are adjusted to match the full trace case which expects that 'old' is 397 * always less than 'head'. 398 */ 399 if (wrapped) { 400 *old = *head; 401 *head += mm->len; 402 } else { 403 if (mm->mask) 404 *old &= mm->mask; 405 else 406 *old %= mm->len; 407 if (*old > *head) 408 *head += mm->len; 409 } 410 411 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n", 412 __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head); 413 414 return 0; 415 416 out_err: 417 pr_err("%s: failed, error %d\n", __func__, err); 418 return err; 419 } 420 421 struct auxtrace_record *intel_bts_recording_init(int *err) 422 { 423 struct perf_pmu *intel_bts_pmu = perf_pmus__find(INTEL_BTS_PMU_NAME); 424 struct intel_bts_recording *btsr; 425 426 if (!intel_bts_pmu) 427 return NULL; 428 429 if (setenv("JITDUMP_USE_ARCH_TIMESTAMP", "1", 1)) { 430 *err = -errno; 431 return NULL; 432 } 433 434 btsr = zalloc(sizeof(struct intel_bts_recording)); 435 if (!btsr) { 436 *err = -ENOMEM; 437 return NULL; 438 } 439 440 btsr->intel_bts_pmu = intel_bts_pmu; 441 btsr->itr.recording_options = intel_bts_recording_options; 442 btsr->itr.info_priv_size = intel_bts_info_priv_size; 443 btsr->itr.info_fill = intel_bts_info_fill; 444 btsr->itr.free = intel_bts_recording_free; 445 btsr->itr.snapshot_start = intel_bts_snapshot_start; 446 btsr->itr.snapshot_finish = intel_bts_snapshot_finish; 447 btsr->itr.find_snapshot = intel_bts_find_snapshot; 448 btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options; 449 btsr->itr.reference = intel_bts_reference; 450 btsr->itr.read_finish = auxtrace_record__read_finish; 451 btsr->itr.alignment = sizeof(struct branch); 452 return &btsr->itr; 453 } 454