trace_boot.c (16166B)
1// SPDX-License-Identifier: GPL-2.0 2/* 3 * trace_boot.c 4 * Tracing kernel boot-time 5 */ 6 7#define pr_fmt(fmt) "trace_boot: " fmt 8 9#include <linux/bootconfig.h> 10#include <linux/cpumask.h> 11#include <linux/ftrace.h> 12#include <linux/init.h> 13#include <linux/kernel.h> 14#include <linux/mutex.h> 15#include <linux/string.h> 16#include <linux/slab.h> 17#include <linux/trace.h> 18#include <linux/trace_events.h> 19 20#include "trace.h" 21 22#define MAX_BUF_LEN 256 23 24static void __init 25trace_boot_set_instance_options(struct trace_array *tr, struct xbc_node *node) 26{ 27 struct xbc_node *anode; 28 const char *p; 29 char buf[MAX_BUF_LEN]; 30 unsigned long v = 0; 31 32 /* Common ftrace options */ 33 xbc_node_for_each_array_value(node, "options", anode, p) { 34 if (strlcpy(buf, p, ARRAY_SIZE(buf)) >= ARRAY_SIZE(buf)) { 35 pr_err("String is too long: %s\n", p); 36 continue; 37 } 38 39 if (trace_set_options(tr, buf) < 0) 40 pr_err("Failed to set option: %s\n", buf); 41 } 42 43 p = xbc_node_find_value(node, "tracing_on", NULL); 44 if (p && *p != '\0') { 45 if (kstrtoul(p, 10, &v)) 46 pr_err("Failed to set tracing on: %s\n", p); 47 if (v) 48 tracer_tracing_on(tr); 49 else 50 tracer_tracing_off(tr); 51 } 52 53 p = xbc_node_find_value(node, "trace_clock", NULL); 54 if (p && *p != '\0') { 55 if (tracing_set_clock(tr, p) < 0) 56 pr_err("Failed to set trace clock: %s\n", p); 57 } 58 59 p = xbc_node_find_value(node, "buffer_size", NULL); 60 if (p && *p != '\0') { 61 v = memparse(p, NULL); 62 if (v < PAGE_SIZE) 63 pr_err("Buffer size is too small: %s\n", p); 64 if (tracing_resize_ring_buffer(tr, v, RING_BUFFER_ALL_CPUS) < 0) 65 pr_err("Failed to resize trace buffer to %s\n", p); 66 } 67 68 p = xbc_node_find_value(node, "cpumask", NULL); 69 if (p && *p != '\0') { 70 cpumask_var_t new_mask; 71 72 if (alloc_cpumask_var(&new_mask, GFP_KERNEL)) { 73 if (cpumask_parse(p, new_mask) < 0 || 74 tracing_set_cpumask(tr, new_mask) < 0) 75 pr_err("Failed to set new CPU mask %s\n", p); 76 free_cpumask_var(new_mask); 77 } 78 } 79} 80 81#ifdef CONFIG_EVENT_TRACING 82static void __init 83trace_boot_enable_events(struct trace_array *tr, struct xbc_node *node) 84{ 85 struct xbc_node *anode; 86 char buf[MAX_BUF_LEN]; 87 const char *p; 88 89 xbc_node_for_each_array_value(node, "events", anode, p) { 90 if (strlcpy(buf, p, ARRAY_SIZE(buf)) >= ARRAY_SIZE(buf)) { 91 pr_err("String is too long: %s\n", p); 92 continue; 93 } 94 95 if (ftrace_set_clr_event(tr, buf, 1) < 0) 96 pr_err("Failed to enable event: %s\n", p); 97 } 98} 99 100#ifdef CONFIG_KPROBE_EVENTS 101static int __init 102trace_boot_add_kprobe_event(struct xbc_node *node, const char *event) 103{ 104 struct dynevent_cmd cmd; 105 struct xbc_node *anode; 106 char buf[MAX_BUF_LEN]; 107 const char *val; 108 int ret = 0; 109 110 xbc_node_for_each_array_value(node, "probes", anode, val) { 111 kprobe_event_cmd_init(&cmd, buf, MAX_BUF_LEN); 112 113 ret = kprobe_event_gen_cmd_start(&cmd, event, val); 114 if (ret) { 115 pr_err("Failed to generate probe: %s\n", buf); 116 break; 117 } 118 119 ret = kprobe_event_gen_cmd_end(&cmd); 120 if (ret) { 121 pr_err("Failed to add probe: %s\n", buf); 122 break; 123 } 124 } 125 126 return ret; 127} 128#else 129static inline int __init 130trace_boot_add_kprobe_event(struct xbc_node *node, const char *event) 131{ 132 pr_err("Kprobe event is not supported.\n"); 133 return -ENOTSUPP; 134} 135#endif 136 137#ifdef CONFIG_SYNTH_EVENTS 138static int __init 139trace_boot_add_synth_event(struct xbc_node *node, const char *event) 140{ 141 struct dynevent_cmd cmd; 142 struct xbc_node *anode; 143 char buf[MAX_BUF_LEN]; 144 const char *p; 145 int ret; 146 147 synth_event_cmd_init(&cmd, buf, MAX_BUF_LEN); 148 149 ret = synth_event_gen_cmd_start(&cmd, event, NULL); 150 if (ret) 151 return ret; 152 153 xbc_node_for_each_array_value(node, "fields", anode, p) { 154 ret = synth_event_add_field_str(&cmd, p); 155 if (ret) 156 return ret; 157 } 158 159 ret = synth_event_gen_cmd_end(&cmd); 160 if (ret < 0) 161 pr_err("Failed to add synthetic event: %s\n", buf); 162 163 return ret; 164} 165#else 166static inline int __init 167trace_boot_add_synth_event(struct xbc_node *node, const char *event) 168{ 169 pr_err("Synthetic event is not supported.\n"); 170 return -ENOTSUPP; 171} 172#endif 173 174#ifdef CONFIG_HIST_TRIGGERS 175static int __init __printf(3, 4) 176append_printf(char **bufp, char *end, const char *fmt, ...) 177{ 178 va_list args; 179 int ret; 180 181 if (*bufp == end) 182 return -ENOSPC; 183 184 va_start(args, fmt); 185 ret = vsnprintf(*bufp, end - *bufp, fmt, args); 186 if (ret < end - *bufp) { 187 *bufp += ret; 188 } else { 189 *bufp = end; 190 ret = -ERANGE; 191 } 192 va_end(args); 193 194 return ret; 195} 196 197static int __init 198append_str_nospace(char **bufp, char *end, const char *str) 199{ 200 char *p = *bufp; 201 int len; 202 203 while (p < end - 1 && *str != '\0') { 204 if (!isspace(*str)) 205 *(p++) = *str; 206 str++; 207 } 208 *p = '\0'; 209 if (p == end - 1) { 210 *bufp = end; 211 return -ENOSPC; 212 } 213 len = p - *bufp; 214 *bufp = p; 215 return (int)len; 216} 217 218static int __init 219trace_boot_hist_add_array(struct xbc_node *hnode, char **bufp, 220 char *end, const char *key) 221{ 222 struct xbc_node *anode; 223 const char *p; 224 char sep; 225 226 p = xbc_node_find_value(hnode, key, &anode); 227 if (p) { 228 if (!anode) { 229 pr_err("hist.%s requires value(s).\n", key); 230 return -EINVAL; 231 } 232 233 append_printf(bufp, end, ":%s", key); 234 sep = '='; 235 xbc_array_for_each_value(anode, p) { 236 append_printf(bufp, end, "%c%s", sep, p); 237 if (sep == '=') 238 sep = ','; 239 } 240 } else 241 return -ENOENT; 242 243 return 0; 244} 245 246static int __init 247trace_boot_hist_add_one_handler(struct xbc_node *hnode, char **bufp, 248 char *end, const char *handler, 249 const char *param) 250{ 251 struct xbc_node *knode, *anode; 252 const char *p; 253 char sep; 254 255 /* Compose 'handler' parameter */ 256 p = xbc_node_find_value(hnode, param, NULL); 257 if (!p) { 258 pr_err("hist.%s requires '%s' option.\n", 259 xbc_node_get_data(hnode), param); 260 return -EINVAL; 261 } 262 append_printf(bufp, end, ":%s(%s)", handler, p); 263 264 /* Compose 'action' parameter */ 265 knode = xbc_node_find_subkey(hnode, "trace"); 266 if (!knode) 267 knode = xbc_node_find_subkey(hnode, "save"); 268 269 if (knode) { 270 anode = xbc_node_get_child(knode); 271 if (!anode || !xbc_node_is_value(anode)) { 272 pr_err("hist.%s.%s requires value(s).\n", 273 xbc_node_get_data(hnode), 274 xbc_node_get_data(knode)); 275 return -EINVAL; 276 } 277 278 append_printf(bufp, end, ".%s", xbc_node_get_data(knode)); 279 sep = '('; 280 xbc_array_for_each_value(anode, p) { 281 append_printf(bufp, end, "%c%s", sep, p); 282 if (sep == '(') 283 sep = ','; 284 } 285 append_printf(bufp, end, ")"); 286 } else if (xbc_node_find_subkey(hnode, "snapshot")) { 287 append_printf(bufp, end, ".snapshot()"); 288 } else { 289 pr_err("hist.%s requires an action.\n", 290 xbc_node_get_data(hnode)); 291 return -EINVAL; 292 } 293 294 return 0; 295} 296 297static int __init 298trace_boot_hist_add_handlers(struct xbc_node *hnode, char **bufp, 299 char *end, const char *param) 300{ 301 struct xbc_node *node; 302 const char *p, *handler; 303 int ret = 0; 304 305 handler = xbc_node_get_data(hnode); 306 307 xbc_node_for_each_subkey(hnode, node) { 308 p = xbc_node_get_data(node); 309 if (!isdigit(p[0])) 310 continue; 311 /* All digit started node should be instances. */ 312 ret = trace_boot_hist_add_one_handler(node, bufp, end, handler, param); 313 if (ret < 0) 314 break; 315 } 316 317 if (xbc_node_find_subkey(hnode, param)) 318 ret = trace_boot_hist_add_one_handler(hnode, bufp, end, handler, param); 319 320 return ret; 321} 322 323/* 324 * Histogram boottime tracing syntax. 325 * 326 * ftrace.[instance.INSTANCE.]event.GROUP.EVENT.hist[.N] { 327 * keys = <KEY>[,...] 328 * values = <VAL>[,...] 329 * sort = <SORT-KEY>[,...] 330 * size = <ENTRIES> 331 * name = <HISTNAME> 332 * var { <VAR> = <EXPR> ... } 333 * pause|continue|clear 334 * onmax|onchange[.N] { var = <VAR>; <ACTION> [= <PARAM>] } 335 * onmatch[.N] { event = <EVENT>; <ACTION> [= <PARAM>] } 336 * filter = <FILTER> 337 * } 338 * 339 * Where <ACTION> are; 340 * 341 * trace = <EVENT>, <ARG1>[, ...] 342 * save = <ARG1>[, ...] 343 * snapshot 344 */ 345static int __init 346trace_boot_compose_hist_cmd(struct xbc_node *hnode, char *buf, size_t size) 347{ 348 struct xbc_node *node, *knode; 349 char *end = buf + size; 350 const char *p; 351 int ret = 0; 352 353 append_printf(&buf, end, "hist"); 354 355 ret = trace_boot_hist_add_array(hnode, &buf, end, "keys"); 356 if (ret < 0) { 357 if (ret == -ENOENT) 358 pr_err("hist requires keys.\n"); 359 return -EINVAL; 360 } 361 362 ret = trace_boot_hist_add_array(hnode, &buf, end, "values"); 363 if (ret == -EINVAL) 364 return ret; 365 ret = trace_boot_hist_add_array(hnode, &buf, end, "sort"); 366 if (ret == -EINVAL) 367 return ret; 368 369 p = xbc_node_find_value(hnode, "size", NULL); 370 if (p) 371 append_printf(&buf, end, ":size=%s", p); 372 373 p = xbc_node_find_value(hnode, "name", NULL); 374 if (p) 375 append_printf(&buf, end, ":name=%s", p); 376 377 node = xbc_node_find_subkey(hnode, "var"); 378 if (node) { 379 xbc_node_for_each_key_value(node, knode, p) { 380 /* Expression must not include spaces. */ 381 append_printf(&buf, end, ":%s=", 382 xbc_node_get_data(knode)); 383 append_str_nospace(&buf, end, p); 384 } 385 } 386 387 /* Histogram control attributes (mutual exclusive) */ 388 if (xbc_node_find_value(hnode, "pause", NULL)) 389 append_printf(&buf, end, ":pause"); 390 else if (xbc_node_find_value(hnode, "continue", NULL)) 391 append_printf(&buf, end, ":continue"); 392 else if (xbc_node_find_value(hnode, "clear", NULL)) 393 append_printf(&buf, end, ":clear"); 394 395 /* Histogram handler and actions */ 396 node = xbc_node_find_subkey(hnode, "onmax"); 397 if (node && trace_boot_hist_add_handlers(node, &buf, end, "var") < 0) 398 return -EINVAL; 399 node = xbc_node_find_subkey(hnode, "onchange"); 400 if (node && trace_boot_hist_add_handlers(node, &buf, end, "var") < 0) 401 return -EINVAL; 402 node = xbc_node_find_subkey(hnode, "onmatch"); 403 if (node && trace_boot_hist_add_handlers(node, &buf, end, "event") < 0) 404 return -EINVAL; 405 406 p = xbc_node_find_value(hnode, "filter", NULL); 407 if (p) 408 append_printf(&buf, end, " if %s", p); 409 410 if (buf == end) { 411 pr_err("hist exceeds the max command length.\n"); 412 return -E2BIG; 413 } 414 415 return 0; 416} 417 418static void __init 419trace_boot_init_histograms(struct trace_event_file *file, 420 struct xbc_node *hnode, char *buf, size_t size) 421{ 422 struct xbc_node *node; 423 const char *p; 424 char *tmp; 425 426 xbc_node_for_each_subkey(hnode, node) { 427 p = xbc_node_get_data(node); 428 if (!isdigit(p[0])) 429 continue; 430 /* All digit started node should be instances. */ 431 if (trace_boot_compose_hist_cmd(node, buf, size) == 0) { 432 tmp = kstrdup(buf, GFP_KERNEL); 433 if (!tmp) 434 return; 435 if (trigger_process_regex(file, buf) < 0) 436 pr_err("Failed to apply hist trigger: %s\n", tmp); 437 kfree(tmp); 438 } 439 } 440 441 if (xbc_node_find_subkey(hnode, "keys")) { 442 if (trace_boot_compose_hist_cmd(hnode, buf, size) == 0) { 443 tmp = kstrdup(buf, GFP_KERNEL); 444 if (!tmp) 445 return; 446 if (trigger_process_regex(file, buf) < 0) 447 pr_err("Failed to apply hist trigger: %s\n", tmp); 448 kfree(tmp); 449 } 450 } 451} 452#else 453static void __init 454trace_boot_init_histograms(struct trace_event_file *file, 455 struct xbc_node *hnode, char *buf, size_t size) 456{ 457 /* do nothing */ 458} 459#endif 460 461static void __init 462trace_boot_init_one_event(struct trace_array *tr, struct xbc_node *gnode, 463 struct xbc_node *enode) 464{ 465 struct trace_event_file *file; 466 struct xbc_node *anode; 467 char buf[MAX_BUF_LEN]; 468 const char *p, *group, *event; 469 470 group = xbc_node_get_data(gnode); 471 event = xbc_node_get_data(enode); 472 473 if (!strcmp(group, "kprobes")) 474 if (trace_boot_add_kprobe_event(enode, event) < 0) 475 return; 476 if (!strcmp(group, "synthetic")) 477 if (trace_boot_add_synth_event(enode, event) < 0) 478 return; 479 480 mutex_lock(&event_mutex); 481 file = find_event_file(tr, group, event); 482 if (!file) { 483 pr_err("Failed to find event: %s:%s\n", group, event); 484 goto out; 485 } 486 487 p = xbc_node_find_value(enode, "filter", NULL); 488 if (p && *p != '\0') { 489 if (strlcpy(buf, p, ARRAY_SIZE(buf)) >= ARRAY_SIZE(buf)) 490 pr_err("filter string is too long: %s\n", p); 491 else if (apply_event_filter(file, buf) < 0) 492 pr_err("Failed to apply filter: %s\n", buf); 493 } 494 495 if (IS_ENABLED(CONFIG_HIST_TRIGGERS)) { 496 xbc_node_for_each_array_value(enode, "actions", anode, p) { 497 if (strlcpy(buf, p, ARRAY_SIZE(buf)) >= ARRAY_SIZE(buf)) 498 pr_err("action string is too long: %s\n", p); 499 else if (trigger_process_regex(file, buf) < 0) 500 pr_err("Failed to apply an action: %s\n", p); 501 } 502 anode = xbc_node_find_subkey(enode, "hist"); 503 if (anode) 504 trace_boot_init_histograms(file, anode, buf, ARRAY_SIZE(buf)); 505 } else if (xbc_node_find_value(enode, "actions", NULL)) 506 pr_err("Failed to apply event actions because CONFIG_HIST_TRIGGERS is not set.\n"); 507 508 if (xbc_node_find_value(enode, "enable", NULL)) { 509 if (trace_event_enable_disable(file, 1, 0) < 0) 510 pr_err("Failed to enable event node: %s:%s\n", 511 group, event); 512 } 513out: 514 mutex_unlock(&event_mutex); 515} 516 517static void __init 518trace_boot_init_events(struct trace_array *tr, struct xbc_node *node) 519{ 520 struct xbc_node *gnode, *enode; 521 bool enable, enable_all = false; 522 const char *data; 523 524 node = xbc_node_find_subkey(node, "event"); 525 if (!node) 526 return; 527 /* per-event key starts with "event.GROUP.EVENT" */ 528 xbc_node_for_each_subkey(node, gnode) { 529 data = xbc_node_get_data(gnode); 530 if (!strcmp(data, "enable")) { 531 enable_all = true; 532 continue; 533 } 534 enable = false; 535 xbc_node_for_each_subkey(gnode, enode) { 536 data = xbc_node_get_data(enode); 537 if (!strcmp(data, "enable")) { 538 enable = true; 539 continue; 540 } 541 trace_boot_init_one_event(tr, gnode, enode); 542 } 543 /* Event enablement must be done after event settings */ 544 if (enable) { 545 data = xbc_node_get_data(gnode); 546 trace_array_set_clr_event(tr, data, NULL, true); 547 } 548 } 549 /* Ditto */ 550 if (enable_all) 551 trace_array_set_clr_event(tr, NULL, NULL, true); 552} 553#else 554#define trace_boot_enable_events(tr, node) do {} while (0) 555#define trace_boot_init_events(tr, node) do {} while (0) 556#endif 557 558#ifdef CONFIG_DYNAMIC_FTRACE 559static void __init 560trace_boot_set_ftrace_filter(struct trace_array *tr, struct xbc_node *node) 561{ 562 struct xbc_node *anode; 563 const char *p; 564 char *q; 565 566 xbc_node_for_each_array_value(node, "ftrace.filters", anode, p) { 567 q = kstrdup(p, GFP_KERNEL); 568 if (!q) 569 return; 570 if (ftrace_set_filter(tr->ops, q, strlen(q), 0) < 0) 571 pr_err("Failed to add %s to ftrace filter\n", p); 572 else 573 ftrace_filter_param = true; 574 kfree(q); 575 } 576 xbc_node_for_each_array_value(node, "ftrace.notraces", anode, p) { 577 q = kstrdup(p, GFP_KERNEL); 578 if (!q) 579 return; 580 if (ftrace_set_notrace(tr->ops, q, strlen(q), 0) < 0) 581 pr_err("Failed to add %s to ftrace filter\n", p); 582 else 583 ftrace_filter_param = true; 584 kfree(q); 585 } 586} 587#else 588#define trace_boot_set_ftrace_filter(tr, node) do {} while (0) 589#endif 590 591static void __init 592trace_boot_enable_tracer(struct trace_array *tr, struct xbc_node *node) 593{ 594 const char *p; 595 596 trace_boot_set_ftrace_filter(tr, node); 597 598 p = xbc_node_find_value(node, "tracer", NULL); 599 if (p && *p != '\0') { 600 if (tracing_set_tracer(tr, p) < 0) 601 pr_err("Failed to set given tracer: %s\n", p); 602 } 603 604 /* Since tracer can free snapshot buffer, allocate snapshot here.*/ 605 if (xbc_node_find_value(node, "alloc_snapshot", NULL)) { 606 if (tracing_alloc_snapshot_instance(tr) < 0) 607 pr_err("Failed to allocate snapshot buffer\n"); 608 } 609} 610 611static void __init 612trace_boot_init_one_instance(struct trace_array *tr, struct xbc_node *node) 613{ 614 trace_boot_set_instance_options(tr, node); 615 trace_boot_init_events(tr, node); 616 trace_boot_enable_events(tr, node); 617 trace_boot_enable_tracer(tr, node); 618} 619 620static void __init 621trace_boot_init_instances(struct xbc_node *node) 622{ 623 struct xbc_node *inode; 624 struct trace_array *tr; 625 const char *p; 626 627 node = xbc_node_find_subkey(node, "instance"); 628 if (!node) 629 return; 630 631 xbc_node_for_each_subkey(node, inode) { 632 p = xbc_node_get_data(inode); 633 if (!p || *p == '\0') 634 continue; 635 636 tr = trace_array_get_by_name(p); 637 if (!tr) { 638 pr_err("Failed to get trace instance %s\n", p); 639 continue; 640 } 641 trace_boot_init_one_instance(tr, inode); 642 trace_array_put(tr); 643 } 644} 645 646static int __init trace_boot_init(void) 647{ 648 struct xbc_node *trace_node; 649 struct trace_array *tr; 650 651 trace_node = xbc_find_node("ftrace"); 652 if (!trace_node) 653 return 0; 654 655 tr = top_trace_array(); 656 if (!tr) 657 return 0; 658 659 /* Global trace array is also one instance */ 660 trace_boot_init_one_instance(tr, trace_node); 661 trace_boot_init_instances(trace_node); 662 663 disable_tracing_selftest("running boot-time tracing"); 664 665 return 0; 666} 667/* 668 * Start tracing at the end of core-initcall, so that it starts tracing 669 * from the beginning of postcore_initcall. 670 */ 671core_initcall_sync(trace_boot_init);