1 /* 2 * intel-bts.c: Intel Processor Trace support 3 * Copyright (c) 2013-2015, Intel Corporation. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 * 14 */ 15 16 #include <linux/kernel.h> 17 #include <linux/types.h> 18 #include <linux/bitops.h> 19 #include <linux/log2.h> 20 21 #include "../../util/cpumap.h" 22 #include "../../util/evsel.h" 23 #include "../../util/evlist.h" 24 #include "../../util/session.h" 25 #include "../../util/util.h" 26 #include "../../util/pmu.h" 27 #include "../../util/debug.h" 28 #include "../../util/tsc.h" 29 #include "../../util/auxtrace.h" 30 #include "../../util/intel-bts.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 #define INTEL_BTS_DFLT_SAMPLE_SIZE KiB(4) 38 39 #define INTEL_BTS_MAX_SAMPLE_SIZE KiB(60) 40 41 struct intel_bts_snapshot_ref { 42 void *ref_buf; 43 size_t ref_offset; 44 bool wrapped; 45 }; 46 47 struct intel_bts_recording { 48 struct auxtrace_record itr; 49 struct perf_pmu *intel_bts_pmu; 50 struct perf_evlist *evlist; 51 bool snapshot_mode; 52 size_t snapshot_size; 53 int snapshot_ref_cnt; 54 struct intel_bts_snapshot_ref *snapshot_refs; 55 }; 56 57 struct branch { 58 u64 from; 59 u64 to; 60 u64 misc; 61 }; 62 63 static size_t intel_bts_info_priv_size(struct auxtrace_record *itr __maybe_unused) 64 { 65 return INTEL_BTS_AUXTRACE_PRIV_SIZE; 66 } 67 68 static int intel_bts_info_fill(struct auxtrace_record *itr, 69 struct perf_session *session, 70 struct auxtrace_info_event *auxtrace_info, 71 size_t priv_size) 72 { 73 struct intel_bts_recording *btsr = 74 container_of(itr, struct intel_bts_recording, itr); 75 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu; 76 struct perf_event_mmap_page *pc; 77 struct perf_tsc_conversion tc = { .time_mult = 0, }; 78 bool cap_user_time_zero = false; 79 int err; 80 81 if (priv_size != INTEL_BTS_AUXTRACE_PRIV_SIZE) 82 return -EINVAL; 83 84 if (!session->evlist->nr_mmaps) 85 return -EINVAL; 86 87 pc = session->evlist->mmap[0].base; 88 if (pc) { 89 err = perf_read_tsc_conversion(pc, &tc); 90 if (err) { 91 if (err != -EOPNOTSUPP) 92 return err; 93 } else { 94 cap_user_time_zero = tc.time_mult != 0; 95 } 96 if (!cap_user_time_zero) 97 ui__warning("Intel BTS: TSC not available\n"); 98 } 99 100 auxtrace_info->type = PERF_AUXTRACE_INTEL_BTS; 101 auxtrace_info->priv[INTEL_BTS_PMU_TYPE] = intel_bts_pmu->type; 102 auxtrace_info->priv[INTEL_BTS_TIME_SHIFT] = tc.time_shift; 103 auxtrace_info->priv[INTEL_BTS_TIME_MULT] = tc.time_mult; 104 auxtrace_info->priv[INTEL_BTS_TIME_ZERO] = tc.time_zero; 105 auxtrace_info->priv[INTEL_BTS_CAP_USER_TIME_ZERO] = cap_user_time_zero; 106 auxtrace_info->priv[INTEL_BTS_SNAPSHOT_MODE] = btsr->snapshot_mode; 107 108 return 0; 109 } 110 111 static int intel_bts_recording_options(struct auxtrace_record *itr, 112 struct perf_evlist *evlist, 113 struct record_opts *opts) 114 { 115 struct intel_bts_recording *btsr = 116 container_of(itr, struct intel_bts_recording, itr); 117 struct perf_pmu *intel_bts_pmu = btsr->intel_bts_pmu; 118 struct perf_evsel *evsel, *intel_bts_evsel = NULL; 119 const struct cpu_map *cpus = evlist->cpus; 120 bool privileged = geteuid() == 0 || perf_event_paranoid() < 0; 121 122 btsr->evlist = evlist; 123 btsr->snapshot_mode = opts->auxtrace_snapshot_mode; 124 125 evlist__for_each(evlist, evsel) { 126 if (evsel->attr.type == intel_bts_pmu->type) { 127 if (intel_bts_evsel) { 128 pr_err("There may be only one " INTEL_BTS_PMU_NAME " event\n"); 129 return -EINVAL; 130 } 131 evsel->attr.freq = 0; 132 evsel->attr.sample_period = 1; 133 intel_bts_evsel = evsel; 134 opts->full_auxtrace = true; 135 } 136 } 137 138 if (opts->auxtrace_snapshot_mode && !opts->full_auxtrace) { 139 pr_err("Snapshot mode (-S option) requires " INTEL_BTS_PMU_NAME " PMU event (-e " INTEL_BTS_PMU_NAME ")\n"); 140 return -EINVAL; 141 } 142 143 if (!opts->full_auxtrace) 144 return 0; 145 146 if (opts->full_auxtrace && !cpu_map__empty(cpus)) { 147 pr_err(INTEL_BTS_PMU_NAME " does not support per-cpu recording\n"); 148 return -EINVAL; 149 } 150 151 /* Set default sizes for snapshot mode */ 152 if (opts->auxtrace_snapshot_mode) { 153 if (!opts->auxtrace_snapshot_size && !opts->auxtrace_mmap_pages) { 154 if (privileged) { 155 opts->auxtrace_mmap_pages = MiB(4) / page_size; 156 } else { 157 opts->auxtrace_mmap_pages = KiB(128) / page_size; 158 if (opts->mmap_pages == UINT_MAX) 159 opts->mmap_pages = KiB(256) / page_size; 160 } 161 } else if (!opts->auxtrace_mmap_pages && !privileged && 162 opts->mmap_pages == UINT_MAX) { 163 opts->mmap_pages = KiB(256) / page_size; 164 } 165 if (!opts->auxtrace_snapshot_size) 166 opts->auxtrace_snapshot_size = 167 opts->auxtrace_mmap_pages * (size_t)page_size; 168 if (!opts->auxtrace_mmap_pages) { 169 size_t sz = opts->auxtrace_snapshot_size; 170 171 sz = round_up(sz, page_size) / page_size; 172 opts->auxtrace_mmap_pages = roundup_pow_of_two(sz); 173 } 174 if (opts->auxtrace_snapshot_size > 175 opts->auxtrace_mmap_pages * (size_t)page_size) { 176 pr_err("Snapshot size %zu must not be greater than AUX area tracing mmap size %zu\n", 177 opts->auxtrace_snapshot_size, 178 opts->auxtrace_mmap_pages * (size_t)page_size); 179 return -EINVAL; 180 } 181 if (!opts->auxtrace_snapshot_size || !opts->auxtrace_mmap_pages) { 182 pr_err("Failed to calculate default snapshot size and/or AUX area tracing mmap pages\n"); 183 return -EINVAL; 184 } 185 pr_debug2("Intel BTS snapshot size: %zu\n", 186 opts->auxtrace_snapshot_size); 187 } 188 189 /* Set default sizes for full trace mode */ 190 if (opts->full_auxtrace && !opts->auxtrace_mmap_pages) { 191 if (privileged) { 192 opts->auxtrace_mmap_pages = MiB(4) / page_size; 193 } else { 194 opts->auxtrace_mmap_pages = KiB(128) / page_size; 195 if (opts->mmap_pages == UINT_MAX) 196 opts->mmap_pages = KiB(256) / page_size; 197 } 198 } 199 200 /* Validate auxtrace_mmap_pages */ 201 if (opts->auxtrace_mmap_pages) { 202 size_t sz = opts->auxtrace_mmap_pages * (size_t)page_size; 203 size_t min_sz; 204 205 if (opts->auxtrace_snapshot_mode) 206 min_sz = KiB(4); 207 else 208 min_sz = KiB(8); 209 210 if (sz < min_sz || !is_power_of_2(sz)) { 211 pr_err("Invalid mmap size for Intel BTS: must be at least %zuKiB and a power of 2\n", 212 min_sz / 1024); 213 return -EINVAL; 214 } 215 } 216 217 if (intel_bts_evsel) { 218 /* 219 * To obtain the auxtrace buffer file descriptor, the auxtrace event 220 * must come first. 221 */ 222 perf_evlist__to_front(evlist, intel_bts_evsel); 223 /* 224 * In the case of per-cpu mmaps, we need the CPU on the 225 * AUX event. 226 */ 227 if (!cpu_map__empty(cpus)) 228 perf_evsel__set_sample_bit(intel_bts_evsel, CPU); 229 } 230 231 /* Add dummy event to keep tracking */ 232 if (opts->full_auxtrace) { 233 struct perf_evsel *tracking_evsel; 234 int err; 235 236 err = parse_events(evlist, "dummy:u", NULL); 237 if (err) 238 return err; 239 240 tracking_evsel = perf_evlist__last(evlist); 241 242 perf_evlist__set_tracking_event(evlist, tracking_evsel); 243 244 tracking_evsel->attr.freq = 0; 245 tracking_evsel->attr.sample_period = 1; 246 } 247 248 return 0; 249 } 250 251 static int intel_bts_parse_snapshot_options(struct auxtrace_record *itr, 252 struct record_opts *opts, 253 const char *str) 254 { 255 struct intel_bts_recording *btsr = 256 container_of(itr, struct intel_bts_recording, itr); 257 unsigned long long snapshot_size = 0; 258 char *endptr; 259 260 if (str) { 261 snapshot_size = strtoull(str, &endptr, 0); 262 if (*endptr || snapshot_size > SIZE_MAX) 263 return -1; 264 } 265 266 opts->auxtrace_snapshot_mode = true; 267 opts->auxtrace_snapshot_size = snapshot_size; 268 269 btsr->snapshot_size = snapshot_size; 270 271 return 0; 272 } 273 274 static u64 intel_bts_reference(struct auxtrace_record *itr __maybe_unused) 275 { 276 return rdtsc(); 277 } 278 279 static int intel_bts_alloc_snapshot_refs(struct intel_bts_recording *btsr, 280 int idx) 281 { 282 const size_t sz = sizeof(struct intel_bts_snapshot_ref); 283 int cnt = btsr->snapshot_ref_cnt, new_cnt = cnt * 2; 284 struct intel_bts_snapshot_ref *refs; 285 286 if (!new_cnt) 287 new_cnt = 16; 288 289 while (new_cnt <= idx) 290 new_cnt *= 2; 291 292 refs = calloc(new_cnt, sz); 293 if (!refs) 294 return -ENOMEM; 295 296 memcpy(refs, btsr->snapshot_refs, cnt * sz); 297 298 btsr->snapshot_refs = refs; 299 btsr->snapshot_ref_cnt = new_cnt; 300 301 return 0; 302 } 303 304 static void intel_bts_free_snapshot_refs(struct intel_bts_recording *btsr) 305 { 306 int i; 307 308 for (i = 0; i < btsr->snapshot_ref_cnt; i++) 309 zfree(&btsr->snapshot_refs[i].ref_buf); 310 zfree(&btsr->snapshot_refs); 311 } 312 313 static void intel_bts_recording_free(struct auxtrace_record *itr) 314 { 315 struct intel_bts_recording *btsr = 316 container_of(itr, struct intel_bts_recording, itr); 317 318 intel_bts_free_snapshot_refs(btsr); 319 free(btsr); 320 } 321 322 static int intel_bts_snapshot_start(struct auxtrace_record *itr) 323 { 324 struct intel_bts_recording *btsr = 325 container_of(itr, struct intel_bts_recording, itr); 326 struct perf_evsel *evsel; 327 328 evlist__for_each(btsr->evlist, evsel) { 329 if (evsel->attr.type == btsr->intel_bts_pmu->type) 330 return perf_evlist__disable_event(btsr->evlist, evsel); 331 } 332 return -EINVAL; 333 } 334 335 static int intel_bts_snapshot_finish(struct auxtrace_record *itr) 336 { 337 struct intel_bts_recording *btsr = 338 container_of(itr, struct intel_bts_recording, itr); 339 struct perf_evsel *evsel; 340 341 evlist__for_each(btsr->evlist, evsel) { 342 if (evsel->attr.type == btsr->intel_bts_pmu->type) 343 return perf_evlist__enable_event(btsr->evlist, evsel); 344 } 345 return -EINVAL; 346 } 347 348 static bool intel_bts_first_wrap(u64 *data, size_t buf_size) 349 { 350 int i, a, b; 351 352 b = buf_size >> 3; 353 a = b - 512; 354 if (a < 0) 355 a = 0; 356 357 for (i = a; i < b; i++) { 358 if (data[i]) 359 return true; 360 } 361 362 return false; 363 } 364 365 static int intel_bts_find_snapshot(struct auxtrace_record *itr, int idx, 366 struct auxtrace_mmap *mm, unsigned char *data, 367 u64 *head, u64 *old) 368 { 369 struct intel_bts_recording *btsr = 370 container_of(itr, struct intel_bts_recording, itr); 371 bool wrapped; 372 int err; 373 374 pr_debug3("%s: mmap index %d old head %zu new head %zu\n", 375 __func__, idx, (size_t)*old, (size_t)*head); 376 377 if (idx >= btsr->snapshot_ref_cnt) { 378 err = intel_bts_alloc_snapshot_refs(btsr, idx); 379 if (err) 380 goto out_err; 381 } 382 383 wrapped = btsr->snapshot_refs[idx].wrapped; 384 if (!wrapped && intel_bts_first_wrap((u64 *)data, mm->len)) { 385 btsr->snapshot_refs[idx].wrapped = true; 386 wrapped = true; 387 } 388 389 /* 390 * In full trace mode 'head' continually increases. However in snapshot 391 * mode 'head' is an offset within the buffer. Here 'old' and 'head' 392 * are adjusted to match the full trace case which expects that 'old' is 393 * always less than 'head'. 394 */ 395 if (wrapped) { 396 *old = *head; 397 *head += mm->len; 398 } else { 399 if (mm->mask) 400 *old &= mm->mask; 401 else 402 *old %= mm->len; 403 if (*old > *head) 404 *head += mm->len; 405 } 406 407 pr_debug3("%s: wrap-around %sdetected, adjusted old head %zu adjusted new head %zu\n", 408 __func__, wrapped ? "" : "not ", (size_t)*old, (size_t)*head); 409 410 return 0; 411 412 out_err: 413 pr_err("%s: failed, error %d\n", __func__, err); 414 return err; 415 } 416 417 static int intel_bts_read_finish(struct auxtrace_record *itr, int idx) 418 { 419 struct intel_bts_recording *btsr = 420 container_of(itr, struct intel_bts_recording, itr); 421 struct perf_evsel *evsel; 422 423 evlist__for_each(btsr->evlist, evsel) { 424 if (evsel->attr.type == btsr->intel_bts_pmu->type) 425 return perf_evlist__enable_event_idx(btsr->evlist, 426 evsel, idx); 427 } 428 return -EINVAL; 429 } 430 431 struct auxtrace_record *intel_bts_recording_init(int *err) 432 { 433 struct perf_pmu *intel_bts_pmu = perf_pmu__find(INTEL_BTS_PMU_NAME); 434 struct intel_bts_recording *btsr; 435 436 if (!intel_bts_pmu) 437 return NULL; 438 439 btsr = zalloc(sizeof(struct intel_bts_recording)); 440 if (!btsr) { 441 *err = -ENOMEM; 442 return NULL; 443 } 444 445 btsr->intel_bts_pmu = intel_bts_pmu; 446 btsr->itr.recording_options = intel_bts_recording_options; 447 btsr->itr.info_priv_size = intel_bts_info_priv_size; 448 btsr->itr.info_fill = intel_bts_info_fill; 449 btsr->itr.free = intel_bts_recording_free; 450 btsr->itr.snapshot_start = intel_bts_snapshot_start; 451 btsr->itr.snapshot_finish = intel_bts_snapshot_finish; 452 btsr->itr.find_snapshot = intel_bts_find_snapshot; 453 btsr->itr.parse_snapshot_options = intel_bts_parse_snapshot_options; 454 btsr->itr.reference = intel_bts_reference; 455 btsr->itr.read_finish = intel_bts_read_finish; 456 btsr->itr.alignment = sizeof(struct branch); 457 return &btsr->itr; 458 } 459