1 /* 2 * trace task wakeup timings 3 * 4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com> 5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com> 6 * 7 * Based on code from the latency_tracer, that is: 8 * 9 * Copyright (C) 2004-2006 Ingo Molnar 10 * Copyright (C) 2004 Nadia Yvette Chambers 11 */ 12 #include <linux/module.h> 13 #include <linux/fs.h> 14 #include <linux/debugfs.h> 15 #include <linux/kallsyms.h> 16 #include <linux/uaccess.h> 17 #include <linux/ftrace.h> 18 #include <trace/events/sched.h> 19 20 #include "trace.h" 21 22 static struct trace_array *wakeup_trace; 23 static int __read_mostly tracer_enabled; 24 25 static struct task_struct *wakeup_task; 26 static int wakeup_cpu; 27 static int wakeup_current_cpu; 28 static unsigned wakeup_prio = -1; 29 static int wakeup_rt; 30 31 static arch_spinlock_t wakeup_lock = 32 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; 33 34 static void wakeup_reset(struct trace_array *tr); 35 static void __wakeup_reset(struct trace_array *tr); 36 static int wakeup_graph_entry(struct ftrace_graph_ent *trace); 37 static void wakeup_graph_return(struct ftrace_graph_ret *trace); 38 39 static int save_lat_flag; 40 41 #define TRACE_DISPLAY_GRAPH 1 42 43 static struct tracer_opt trace_opts[] = { 44 #ifdef CONFIG_FUNCTION_GRAPH_TRACER 45 /* display latency trace as call graph */ 46 { TRACER_OPT(display-graph, TRACE_DISPLAY_GRAPH) }, 47 #endif 48 { } /* Empty entry */ 49 }; 50 51 static struct tracer_flags tracer_flags = { 52 .val = 0, 53 .opts = trace_opts, 54 }; 55 56 #define is_graph() (tracer_flags.val & TRACE_DISPLAY_GRAPH) 57 58 #ifdef CONFIG_FUNCTION_TRACER 59 60 /* 61 * Prologue for the wakeup function tracers. 62 * 63 * Returns 1 if it is OK to continue, and preemption 64 * is disabled and data->disabled is incremented. 65 * 0 if the trace is to be ignored, and preemption 66 * is not disabled and data->disabled is 67 * kept the same. 68 * 69 * Note, this function is also used outside this ifdef but 70 * inside the #ifdef of the function graph tracer below. 71 * This is OK, since the function graph tracer is 72 * dependent on the function tracer. 73 */ 74 static int 75 func_prolog_preempt_disable(struct trace_array *tr, 76 struct trace_array_cpu **data, 77 int *pc) 78 { 79 long disabled; 80 int cpu; 81 82 if (likely(!wakeup_task)) 83 return 0; 84 85 *pc = preempt_count(); 86 preempt_disable_notrace(); 87 88 cpu = raw_smp_processor_id(); 89 if (cpu != wakeup_current_cpu) 90 goto out_enable; 91 92 *data = tr->data[cpu]; 93 disabled = atomic_inc_return(&(*data)->disabled); 94 if (unlikely(disabled != 1)) 95 goto out; 96 97 return 1; 98 99 out: 100 atomic_dec(&(*data)->disabled); 101 102 out_enable: 103 preempt_enable_notrace(); 104 return 0; 105 } 106 107 /* 108 * wakeup uses its own tracer function to keep the overhead down: 109 */ 110 static void 111 wakeup_tracer_call(unsigned long ip, unsigned long parent_ip, 112 struct ftrace_ops *op, struct pt_regs *pt_regs) 113 { 114 struct trace_array *tr = wakeup_trace; 115 struct trace_array_cpu *data; 116 unsigned long flags; 117 int pc; 118 119 if (!func_prolog_preempt_disable(tr, &data, &pc)) 120 return; 121 122 local_irq_save(flags); 123 trace_function(tr, ip, parent_ip, flags, pc); 124 local_irq_restore(flags); 125 126 atomic_dec(&data->disabled); 127 preempt_enable_notrace(); 128 } 129 130 static struct ftrace_ops trace_ops __read_mostly = 131 { 132 .func = wakeup_tracer_call, 133 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE, 134 }; 135 #endif /* CONFIG_FUNCTION_TRACER */ 136 137 static int start_func_tracer(int graph) 138 { 139 int ret; 140 141 if (!graph) 142 ret = register_ftrace_function(&trace_ops); 143 else 144 ret = register_ftrace_graph(&wakeup_graph_return, 145 &wakeup_graph_entry); 146 147 if (!ret && tracing_is_enabled()) 148 tracer_enabled = 1; 149 else 150 tracer_enabled = 0; 151 152 return ret; 153 } 154 155 static void stop_func_tracer(int graph) 156 { 157 tracer_enabled = 0; 158 159 if (!graph) 160 unregister_ftrace_function(&trace_ops); 161 else 162 unregister_ftrace_graph(); 163 } 164 165 #ifdef CONFIG_FUNCTION_GRAPH_TRACER 166 static int wakeup_set_flag(u32 old_flags, u32 bit, int set) 167 { 168 169 if (!(bit & TRACE_DISPLAY_GRAPH)) 170 return -EINVAL; 171 172 if (!(is_graph() ^ set)) 173 return 0; 174 175 stop_func_tracer(!set); 176 177 wakeup_reset(wakeup_trace); 178 tracing_max_latency = 0; 179 180 return start_func_tracer(set); 181 } 182 183 static int wakeup_graph_entry(struct ftrace_graph_ent *trace) 184 { 185 struct trace_array *tr = wakeup_trace; 186 struct trace_array_cpu *data; 187 unsigned long flags; 188 int pc, ret = 0; 189 190 if (!func_prolog_preempt_disable(tr, &data, &pc)) 191 return 0; 192 193 local_save_flags(flags); 194 ret = __trace_graph_entry(tr, trace, flags, pc); 195 atomic_dec(&data->disabled); 196 preempt_enable_notrace(); 197 198 return ret; 199 } 200 201 static void wakeup_graph_return(struct ftrace_graph_ret *trace) 202 { 203 struct trace_array *tr = wakeup_trace; 204 struct trace_array_cpu *data; 205 unsigned long flags; 206 int pc; 207 208 if (!func_prolog_preempt_disable(tr, &data, &pc)) 209 return; 210 211 local_save_flags(flags); 212 __trace_graph_return(tr, trace, flags, pc); 213 atomic_dec(&data->disabled); 214 215 preempt_enable_notrace(); 216 return; 217 } 218 219 static void wakeup_trace_open(struct trace_iterator *iter) 220 { 221 if (is_graph()) 222 graph_trace_open(iter); 223 } 224 225 static void wakeup_trace_close(struct trace_iterator *iter) 226 { 227 if (iter->private) 228 graph_trace_close(iter); 229 } 230 231 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_PROC | \ 232 TRACE_GRAPH_PRINT_ABS_TIME | \ 233 TRACE_GRAPH_PRINT_DURATION) 234 235 static enum print_line_t wakeup_print_line(struct trace_iterator *iter) 236 { 237 /* 238 * In graph mode call the graph tracer output function, 239 * otherwise go with the TRACE_FN event handler 240 */ 241 if (is_graph()) 242 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS); 243 244 return TRACE_TYPE_UNHANDLED; 245 } 246 247 static void wakeup_print_header(struct seq_file *s) 248 { 249 if (is_graph()) 250 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS); 251 else 252 trace_default_header(s); 253 } 254 255 static void 256 __trace_function(struct trace_array *tr, 257 unsigned long ip, unsigned long parent_ip, 258 unsigned long flags, int pc) 259 { 260 if (is_graph()) 261 trace_graph_function(tr, ip, parent_ip, flags, pc); 262 else 263 trace_function(tr, ip, parent_ip, flags, pc); 264 } 265 #else 266 #define __trace_function trace_function 267 268 static int wakeup_set_flag(u32 old_flags, u32 bit, int set) 269 { 270 return -EINVAL; 271 } 272 273 static int wakeup_graph_entry(struct ftrace_graph_ent *trace) 274 { 275 return -1; 276 } 277 278 static enum print_line_t wakeup_print_line(struct trace_iterator *iter) 279 { 280 return TRACE_TYPE_UNHANDLED; 281 } 282 283 static void wakeup_graph_return(struct ftrace_graph_ret *trace) { } 284 static void wakeup_trace_open(struct trace_iterator *iter) { } 285 static void wakeup_trace_close(struct trace_iterator *iter) { } 286 287 #ifdef CONFIG_FUNCTION_TRACER 288 static void wakeup_print_header(struct seq_file *s) 289 { 290 trace_default_header(s); 291 } 292 #else 293 static void wakeup_print_header(struct seq_file *s) 294 { 295 trace_latency_header(s); 296 } 297 #endif /* CONFIG_FUNCTION_TRACER */ 298 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */ 299 300 /* 301 * Should this new latency be reported/recorded? 302 */ 303 static int report_latency(cycle_t delta) 304 { 305 if (tracing_thresh) { 306 if (delta < tracing_thresh) 307 return 0; 308 } else { 309 if (delta <= tracing_max_latency) 310 return 0; 311 } 312 return 1; 313 } 314 315 static void 316 probe_wakeup_migrate_task(void *ignore, struct task_struct *task, int cpu) 317 { 318 if (task != wakeup_task) 319 return; 320 321 wakeup_current_cpu = cpu; 322 } 323 324 static void notrace 325 probe_wakeup_sched_switch(void *ignore, 326 struct task_struct *prev, struct task_struct *next) 327 { 328 struct trace_array_cpu *data; 329 cycle_t T0, T1, delta; 330 unsigned long flags; 331 long disabled; 332 int cpu; 333 int pc; 334 335 tracing_record_cmdline(prev); 336 337 if (unlikely(!tracer_enabled)) 338 return; 339 340 /* 341 * When we start a new trace, we set wakeup_task to NULL 342 * and then set tracer_enabled = 1. We want to make sure 343 * that another CPU does not see the tracer_enabled = 1 344 * and the wakeup_task with an older task, that might 345 * actually be the same as next. 346 */ 347 smp_rmb(); 348 349 if (next != wakeup_task) 350 return; 351 352 pc = preempt_count(); 353 354 /* disable local data, not wakeup_cpu data */ 355 cpu = raw_smp_processor_id(); 356 disabled = atomic_inc_return(&wakeup_trace->data[cpu]->disabled); 357 if (likely(disabled != 1)) 358 goto out; 359 360 local_irq_save(flags); 361 arch_spin_lock(&wakeup_lock); 362 363 /* We could race with grabbing wakeup_lock */ 364 if (unlikely(!tracer_enabled || next != wakeup_task)) 365 goto out_unlock; 366 367 /* The task we are waiting for is waking up */ 368 data = wakeup_trace->data[wakeup_cpu]; 369 370 __trace_function(wakeup_trace, CALLER_ADDR0, CALLER_ADDR1, flags, pc); 371 tracing_sched_switch_trace(wakeup_trace, prev, next, flags, pc); 372 373 T0 = data->preempt_timestamp; 374 T1 = ftrace_now(cpu); 375 delta = T1-T0; 376 377 if (!report_latency(delta)) 378 goto out_unlock; 379 380 if (likely(!is_tracing_stopped())) { 381 tracing_max_latency = delta; 382 update_max_tr(wakeup_trace, wakeup_task, wakeup_cpu); 383 } 384 385 out_unlock: 386 __wakeup_reset(wakeup_trace); 387 arch_spin_unlock(&wakeup_lock); 388 local_irq_restore(flags); 389 out: 390 atomic_dec(&wakeup_trace->data[cpu]->disabled); 391 } 392 393 static void __wakeup_reset(struct trace_array *tr) 394 { 395 wakeup_cpu = -1; 396 wakeup_prio = -1; 397 398 if (wakeup_task) 399 put_task_struct(wakeup_task); 400 401 wakeup_task = NULL; 402 } 403 404 static void wakeup_reset(struct trace_array *tr) 405 { 406 unsigned long flags; 407 408 tracing_reset_online_cpus(tr); 409 410 local_irq_save(flags); 411 arch_spin_lock(&wakeup_lock); 412 __wakeup_reset(tr); 413 arch_spin_unlock(&wakeup_lock); 414 local_irq_restore(flags); 415 } 416 417 static void 418 probe_wakeup(void *ignore, struct task_struct *p, int success) 419 { 420 struct trace_array_cpu *data; 421 int cpu = smp_processor_id(); 422 unsigned long flags; 423 long disabled; 424 int pc; 425 426 if (likely(!tracer_enabled)) 427 return; 428 429 tracing_record_cmdline(p); 430 tracing_record_cmdline(current); 431 432 if ((wakeup_rt && !rt_task(p)) || 433 p->prio >= wakeup_prio || 434 p->prio >= current->prio) 435 return; 436 437 pc = preempt_count(); 438 disabled = atomic_inc_return(&wakeup_trace->data[cpu]->disabled); 439 if (unlikely(disabled != 1)) 440 goto out; 441 442 /* interrupts should be off from try_to_wake_up */ 443 arch_spin_lock(&wakeup_lock); 444 445 /* check for races. */ 446 if (!tracer_enabled || p->prio >= wakeup_prio) 447 goto out_locked; 448 449 /* reset the trace */ 450 __wakeup_reset(wakeup_trace); 451 452 wakeup_cpu = task_cpu(p); 453 wakeup_current_cpu = wakeup_cpu; 454 wakeup_prio = p->prio; 455 456 wakeup_task = p; 457 get_task_struct(wakeup_task); 458 459 local_save_flags(flags); 460 461 data = wakeup_trace->data[wakeup_cpu]; 462 data->preempt_timestamp = ftrace_now(cpu); 463 tracing_sched_wakeup_trace(wakeup_trace, p, current, flags, pc); 464 465 /* 466 * We must be careful in using CALLER_ADDR2. But since wake_up 467 * is not called by an assembly function (where as schedule is) 468 * it should be safe to use it here. 469 */ 470 __trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc); 471 472 out_locked: 473 arch_spin_unlock(&wakeup_lock); 474 out: 475 atomic_dec(&wakeup_trace->data[cpu]->disabled); 476 } 477 478 static void start_wakeup_tracer(struct trace_array *tr) 479 { 480 int ret; 481 482 ret = register_trace_sched_wakeup(probe_wakeup, NULL); 483 if (ret) { 484 pr_info("wakeup trace: Couldn't activate tracepoint" 485 " probe to kernel_sched_wakeup\n"); 486 return; 487 } 488 489 ret = register_trace_sched_wakeup_new(probe_wakeup, NULL); 490 if (ret) { 491 pr_info("wakeup trace: Couldn't activate tracepoint" 492 " probe to kernel_sched_wakeup_new\n"); 493 goto fail_deprobe; 494 } 495 496 ret = register_trace_sched_switch(probe_wakeup_sched_switch, NULL); 497 if (ret) { 498 pr_info("sched trace: Couldn't activate tracepoint" 499 " probe to kernel_sched_switch\n"); 500 goto fail_deprobe_wake_new; 501 } 502 503 ret = register_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL); 504 if (ret) { 505 pr_info("wakeup trace: Couldn't activate tracepoint" 506 " probe to kernel_sched_migrate_task\n"); 507 return; 508 } 509 510 wakeup_reset(tr); 511 512 /* 513 * Don't let the tracer_enabled = 1 show up before 514 * the wakeup_task is reset. This may be overkill since 515 * wakeup_reset does a spin_unlock after setting the 516 * wakeup_task to NULL, but I want to be safe. 517 * This is a slow path anyway. 518 */ 519 smp_wmb(); 520 521 if (start_func_tracer(is_graph())) 522 printk(KERN_ERR "failed to start wakeup tracer\n"); 523 524 return; 525 fail_deprobe_wake_new: 526 unregister_trace_sched_wakeup_new(probe_wakeup, NULL); 527 fail_deprobe: 528 unregister_trace_sched_wakeup(probe_wakeup, NULL); 529 } 530 531 static void stop_wakeup_tracer(struct trace_array *tr) 532 { 533 tracer_enabled = 0; 534 stop_func_tracer(is_graph()); 535 unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL); 536 unregister_trace_sched_wakeup_new(probe_wakeup, NULL); 537 unregister_trace_sched_wakeup(probe_wakeup, NULL); 538 unregister_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL); 539 } 540 541 static int __wakeup_tracer_init(struct trace_array *tr) 542 { 543 save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT; 544 trace_flags |= TRACE_ITER_LATENCY_FMT; 545 546 tracing_max_latency = 0; 547 wakeup_trace = tr; 548 start_wakeup_tracer(tr); 549 return 0; 550 } 551 552 static int wakeup_tracer_init(struct trace_array *tr) 553 { 554 wakeup_rt = 0; 555 return __wakeup_tracer_init(tr); 556 } 557 558 static int wakeup_rt_tracer_init(struct trace_array *tr) 559 { 560 wakeup_rt = 1; 561 return __wakeup_tracer_init(tr); 562 } 563 564 static void wakeup_tracer_reset(struct trace_array *tr) 565 { 566 stop_wakeup_tracer(tr); 567 /* make sure we put back any tasks we are tracing */ 568 wakeup_reset(tr); 569 570 if (!save_lat_flag) 571 trace_flags &= ~TRACE_ITER_LATENCY_FMT; 572 } 573 574 static void wakeup_tracer_start(struct trace_array *tr) 575 { 576 wakeup_reset(tr); 577 tracer_enabled = 1; 578 } 579 580 static void wakeup_tracer_stop(struct trace_array *tr) 581 { 582 tracer_enabled = 0; 583 } 584 585 static struct tracer wakeup_tracer __read_mostly = 586 { 587 .name = "wakeup", 588 .init = wakeup_tracer_init, 589 .reset = wakeup_tracer_reset, 590 .start = wakeup_tracer_start, 591 .stop = wakeup_tracer_stop, 592 .print_max = true, 593 .print_header = wakeup_print_header, 594 .print_line = wakeup_print_line, 595 .flags = &tracer_flags, 596 .set_flag = wakeup_set_flag, 597 #ifdef CONFIG_FTRACE_SELFTEST 598 .selftest = trace_selftest_startup_wakeup, 599 #endif 600 .open = wakeup_trace_open, 601 .close = wakeup_trace_close, 602 .use_max_tr = true, 603 }; 604 605 static struct tracer wakeup_rt_tracer __read_mostly = 606 { 607 .name = "wakeup_rt", 608 .init = wakeup_rt_tracer_init, 609 .reset = wakeup_tracer_reset, 610 .start = wakeup_tracer_start, 611 .stop = wakeup_tracer_stop, 612 .wait_pipe = poll_wait_pipe, 613 .print_max = true, 614 .print_header = wakeup_print_header, 615 .print_line = wakeup_print_line, 616 .flags = &tracer_flags, 617 .set_flag = wakeup_set_flag, 618 #ifdef CONFIG_FTRACE_SELFTEST 619 .selftest = trace_selftest_startup_wakeup, 620 #endif 621 .open = wakeup_trace_open, 622 .close = wakeup_trace_close, 623 .use_max_tr = true, 624 }; 625 626 __init static int init_wakeup_tracer(void) 627 { 628 int ret; 629 630 ret = register_tracer(&wakeup_tracer); 631 if (ret) 632 return ret; 633 634 ret = register_tracer(&wakeup_rt_tracer); 635 if (ret) 636 return ret; 637 638 return 0; 639 } 640 core_initcall(init_wakeup_tracer); 641