instrumentation/events/lttng-module/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #undef TRACE_SYSTEM
   3 #define TRACE_SYSTEM sched
   4
   5 #if !defined(LTTNG_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
   6 #define LTTNG_TRACE_SCHED_H
   7
   8 #include <probes/lttng-tracepoint-event.h>
   9 #include <linux/sched.h>
  10 #include <linux/pid_namespace.h>
  11 #include <linux/binfmts.h>
  12 #include <linux/version.h>
  13 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0))
  14 #include <linux/sched/rt.h>
  15 #endif
  16 #include <wrapper/namespace.h>
  17
  18 #define LTTNG_MAX_PID_NS_LEVEL 32
  19
  20 #ifndef _TRACE_SCHED_DEF_
  21 #define _TRACE_SCHED_DEF_
  22
  23 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,15,0))
  24
  25 static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
  26 {
  27         unsigned int state;
  28
  29 #ifdef CONFIG_SCHED_DEBUG
  30         BUG_ON(p != current);
  31 #endif /* CONFIG_SCHED_DEBUG */
  32
  33         /*
  34          * Preemption ignores task state, therefore preempted tasks are always
  35          * RUNNING (we will not have dequeued if state != RUNNING).
  36          */
  37         if (preempt)
  38                 return TASK_REPORT_MAX;
  39
  40         /*
  41          * task_state_index() uses fls() and returns a value from 0-8 range.
  42          * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
  43          * it for left shift operation to get the correct task->state
  44          * mapping.
  45          */
  46         state = task_state_index(p);
  47
  48         return state ? (1 << (state - 1)) : state;
  49 }
  50
  51 #elif (LINUX_VERSION_CODE >= KERNEL_VERSION(4,14,0))
  52
  53 static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
  54 {
  55         unsigned int state;
  56
  57 #ifdef CONFIG_SCHED_DEBUG
  58         BUG_ON(p != current);
  59 #endif /* CONFIG_SCHED_DEBUG */
  60
  61         /*
  62          * Preemption ignores task state, therefore preempted tasks are always
  63          * RUNNING (we will not have dequeued if state != RUNNING).
  64          */
  65         if (preempt)
  66                 return TASK_REPORT_MAX;
  67
  68         /*
  69          * __get_task_state() uses fls() and returns a value from 0-8 range.
  70          * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
  71          * it for left shift operation to get the correct task->state
  72          * mapping.
  73          */
  74         state = __get_task_state(p);
  75
  76         return state ? (1 << (state - 1)) : state;
  77 }
  78
  79 #elif (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
  80
  81 static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
  82 {
  83 #ifdef CONFIG_SCHED_DEBUG
  84         BUG_ON(p != current);
  85 #endif /* CONFIG_SCHED_DEBUG */
  86         /*
  87          * Preemption ignores task state, therefore preempted tasks are always RUNNING
  88          * (we will not have dequeued if state != RUNNING).
  89          */
  90         return preempt ? TASK_RUNNING | TASK_STATE_MAX : p->state;
  91 }
  92
  93 #elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,19,0))
  94
  95 static inline long __trace_sched_switch_state(struct task_struct *p)
  96 {
  97         long state = p->state;
  98
  99 #ifdef CONFIG_PREEMPT
 100 #ifdef CONFIG_SCHED_DEBUG
 101         BUG_ON(p != current);
 102 #endif /* CONFIG_SCHED_DEBUG */
 103         /*
 104          * For all intents and purposes a preempted task is a running task.
 105          */
 106         if (preempt_count() & PREEMPT_ACTIVE)
 107                 state = TASK_RUNNING | TASK_STATE_MAX;
 108 #endif  /* CONFIG_PREEMPT */
 109
 110         return state;
 111 }
 112
 113 #elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,13,0))
 114
 115 static inline long __trace_sched_switch_state(struct task_struct *p)
 116 {
 117         long state = p->state;
 118
 119 #ifdef CONFIG_PREEMPT
 120         /*
 121          * For all intents and purposes a preempted task is a running task.
 122          */
 123         if (task_preempt_count(p) & PREEMPT_ACTIVE)
 124                 state = TASK_RUNNING | TASK_STATE_MAX;
 125 #endif
 126
 127         return state;
 128 }
 129
 130 #elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0))
 131
 132 static inline long __trace_sched_switch_state(struct task_struct *p)
 133 {
 134         long state = p->state;
 135
 136 #ifdef CONFIG_PREEMPT
 137         /*
 138          * For all intents and purposes a preempted task is a running task.
 139          */
 140         if (task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)
 141                 state = TASK_RUNNING | TASK_STATE_MAX;
 142 #endif
 143
 144         return state;
 145 }
 146
 147 #else
 148
 149 static inline long __trace_sched_switch_state(struct task_struct *p)
 150 {
 151         long state = p->state;
 152
 153 #ifdef CONFIG_PREEMPT
 154         /*
 155          * For all intents and purposes a preempted task is a running task.
 156          */
 157         if (task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)
 158                 state = TASK_RUNNING;
 159 #endif
 160
 161         return state;
 162 }
 163
 164 #endif
 165
 166 #endif /* _TRACE_SCHED_DEF_ */
 167
 168 /*
 169  * Tracepoint for calling kthread_stop, performed to end a kthread:
 170  */
 171 LTTNG_TRACEPOINT_EVENT(sched_kthread_stop,
 172
 173         TP_PROTO(struct task_struct *t),
 174
 175         TP_ARGS(t),
 176
 177         TP_FIELDS(
 178                 ctf_array_text(char, comm, t->comm, TASK_COMM_LEN)
 179                 ctf_integer(pid_t, tid, t->pid)
 180         )
 181 )
 182
 183 /*
 184  * Tracepoint for the return value of the kthread stopping:
 185  */
 186 LTTNG_TRACEPOINT_EVENT(sched_kthread_stop_ret,
 187
 188         TP_PROTO(int ret),
 189
 190         TP_ARGS(ret),
 191
 192         TP_FIELDS(
 193                 ctf_integer(int, ret, ret)
 194         )
 195 )
 196
 197 /*
 198  * Tracepoint for waking up a task:
 199  */
 200 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0) || \
 201         LTTNG_RT_KERNEL_RANGE(4,1,10,11, 4,2,0,0) || \
 202         LTTNG_RT_KERNEL_RANGE(3,18,27,26, 3,19,0,0) || \
 203         LTTNG_RT_KERNEL_RANGE(3,14,61,63, 3,15,0,0) || \
 204         LTTNG_RT_KERNEL_RANGE(3,12,54,73, 3,13,0,0) || \
 205         LTTNG_RT_KERNEL_RANGE(3,10,97,106, 3,11,0,0) || \
 206         LTTNG_RT_KERNEL_RANGE(3,4,110,139, 3,5,0,0) || \
 207         LTTNG_RT_KERNEL_RANGE(3,2,77,111, 3,3,0,0))
 208 LTTNG_TRACEPOINT_EVENT_CLASS(sched_wakeup_template,
 209
 210         TP_PROTO(struct task_struct *p),
 211
 212         TP_ARGS(p),
 213
 214         TP_FIELDS(
 215                 ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
 216                 ctf_integer(pid_t, tid, p->pid)
 217                 ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
 218                 ctf_integer(int, target_cpu, task_cpu(p))
 219         )
 220 )
 221 #else /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
 222 LTTNG_TRACEPOINT_EVENT_CLASS(sched_wakeup_template,
 223
 224         TP_PROTO(struct task_struct *p, int success),
 225
 226         TP_ARGS(p, success),
 227
 228         TP_FIELDS(
 229                 ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
 230                 ctf_integer(pid_t, tid, p->pid)
 231                 ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
 232                 ctf_integer(int, success, success)
 233                 ctf_integer(int, target_cpu, task_cpu(p))
 234         )
 235 )
 236 #endif /* #else #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
 237
 238 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0) || \
 239         LTTNG_RT_KERNEL_RANGE(4,1,10,11, 4,2,0,0) || \
 240         LTTNG_RT_KERNEL_RANGE(3,18,27,26, 3,19,0,0) || \
 241         LTTNG_RT_KERNEL_RANGE(3,14,61,63, 3,15,0,0) || \
 242         LTTNG_RT_KERNEL_RANGE(3,12,54,73, 3,13,0,0) || \
 243         LTTNG_RT_KERNEL_RANGE(3,10,97,106, 3,11,0,0) || \
 244         LTTNG_RT_KERNEL_RANGE(3,4,110,139, 3,5,0,0) || \
 245         LTTNG_RT_KERNEL_RANGE(3,2,77,111, 3,3,0,0))
 246
 247 /*
 248  * Tracepoint called when waking a task; this tracepoint is guaranteed to be
 249  * called from the waking context.
 250  */
 251 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_waking,
 252              TP_PROTO(struct task_struct *p),
 253              TP_ARGS(p))
 254
 255 /*
 256  * Tracepoint called when the task is actually woken; p->state == TASK_RUNNNG.
 257  * It it not always called from the waking context.
 258  */
 259 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup,
 260              TP_PROTO(struct task_struct *p),
 261              TP_ARGS(p))
 262
 263 /*
 264  * Tracepoint for waking up a new task:
 265  */
 266 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup_new,
 267              TP_PROTO(struct task_struct *p),
 268              TP_ARGS(p))
 269
 270 #else
 271
 272 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup,
 273              TP_PROTO(struct task_struct *p, int success),
 274              TP_ARGS(p, success))
 275
 276 /*
 277  * Tracepoint for waking up a new task:
 278  */
 279 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup_new,
 280              TP_PROTO(struct task_struct *p, int success),
 281              TP_ARGS(p, success))
 282
 283 #endif /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
 284
 285 /*
 286  * Tracepoint for task switches, performed by the scheduler:
 287  */
 288 LTTNG_TRACEPOINT_EVENT(sched_switch,
 289
 290 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
 291         TP_PROTO(bool preempt,
 292                  struct task_struct *prev,
 293                  struct task_struct *next),
 294
 295         TP_ARGS(preempt, prev, next),
 296 #else
 297         TP_PROTO(struct task_struct *prev,
 298                  struct task_struct *next),
 299
 300         TP_ARGS(prev, next),
 301 #endif /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)) */
 302
 303         TP_FIELDS(
 304                 ctf_array_text(char, prev_comm, prev->comm, TASK_COMM_LEN)
 305                 ctf_integer(pid_t, prev_tid, prev->pid)
 306                 ctf_integer(int, prev_prio, prev->prio - MAX_RT_PRIO)
 307 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
 308                 ctf_integer(long, prev_state, __trace_sched_switch_state(preempt, prev))
 309 #else
 310                 ctf_integer(long, prev_state, __trace_sched_switch_state(prev))
 311 #endif
 312                 ctf_array_text(char, next_comm, next->comm, TASK_COMM_LEN)
 313                 ctf_integer(pid_t, next_tid, next->pid)
 314                 ctf_integer(int, next_prio, next->prio - MAX_RT_PRIO)
 315         )
 316 )
 317
 318 /*
 319  * Tracepoint for a task being migrated:
 320  */
 321 LTTNG_TRACEPOINT_EVENT(sched_migrate_task,
 322
 323         TP_PROTO(struct task_struct *p, int dest_cpu),
 324
 325         TP_ARGS(p, dest_cpu),
 326
 327         TP_FIELDS(
 328                 ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
 329                 ctf_integer(pid_t, tid, p->pid)
 330                 ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
 331                 ctf_integer(int, orig_cpu, task_cpu(p))
 332                 ctf_integer(int, dest_cpu, dest_cpu)
 333         )
 334 )
 335
 336 LTTNG_TRACEPOINT_EVENT_CLASS(sched_process_template,
 337
 338         TP_PROTO(struct task_struct *p),
 339
 340         TP_ARGS(p),
 341
 342         TP_FIELDS(
 343                 ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
 344                 ctf_integer(pid_t, tid, p->pid)
 345                 ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
 346         )
 347 )
 348
 349 /*
 350  * Tracepoint for freeing a task:
 351  */
 352 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_process_free,
 353              TP_PROTO(struct task_struct *p),
 354              TP_ARGS(p))
 355
 356
 357 /*
 358  * Tracepoint for a task exiting:
 359  */
 360 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_process_exit,
 361              TP_PROTO(struct task_struct *p),
 362              TP_ARGS(p))
 363
 364 /*
 365  * Tracepoint for waiting on task to unschedule:
 366  */
 367 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_wait_task,
 368         TP_PROTO(struct task_struct *p),
 369         TP_ARGS(p))
 370
 371 /*
 372  * Tracepoint for a waiting task:
 373  */
 374 LTTNG_TRACEPOINT_EVENT(sched_process_wait,
 375
 376         TP_PROTO(struct pid *pid),
 377
 378         TP_ARGS(pid),
 379
 380         TP_FIELDS(
 381                 ctf_array_text(char, comm, current->comm, TASK_COMM_LEN)
 382                 ctf_integer(pid_t, tid, pid_nr(pid))
 383                 ctf_integer(int, prio, current->prio - MAX_RT_PRIO)
 384         )
 385 )
 386
 387 /*
 388  * Tracepoint for do_fork.
 389  * Saving both TID and PID information, especially for the child, allows
 390  * trace analyzers to distinguish between creation of a new process and
 391  * creation of a new thread. Newly created processes will have child_tid
 392  * == child_pid, while creation of a thread yields to child_tid !=
 393  * child_pid.
 394  */
 395 LTTNG_TRACEPOINT_EVENT_CODE(sched_process_fork,
 396
 397         TP_PROTO(struct task_struct *parent, struct task_struct *child),
 398
 399         TP_ARGS(parent, child),
 400
 401         TP_locvar(
 402                 pid_t vtids[LTTNG_MAX_PID_NS_LEVEL];
 403                 unsigned int ns_level;
 404         ),
 405
 406         TP_code_pre(
 407                 if (child) {
 408                         struct pid *child_pid;
 409                         unsigned int i;
 410
 411                         child_pid = task_pid(child);
 412                         tp_locvar->ns_level =
 413                                 min_t(unsigned int, child_pid->level + 1,
 414                                         LTTNG_MAX_PID_NS_LEVEL);
 415                         for (i = 0; i < tp_locvar->ns_level; i++)
 416                                 tp_locvar->vtids[i] = child_pid->numbers[i].nr;
 417                 }
 418         ),
 419
 420         TP_FIELDS(
 421                 ctf_array_text(char, parent_comm, parent->comm, TASK_COMM_LEN)
 422                 ctf_integer(pid_t, parent_tid, parent->pid)
 423                 ctf_integer(pid_t, parent_pid, parent->tgid)
 424 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,8,0))
 425                 ctf_integer(unsigned int, parent_ns_inum,
 426                         ({
 427                                 unsigned int parent_ns_inum = 0;
 428
 429                                 if (parent) {
 430                                         struct pid_namespace *pid_ns;
 431
 432                                         pid_ns = task_active_pid_ns(parent);
 433                                         if (pid_ns)
 434                                                 parent_ns_inum =
 435                                                         pid_ns->lttng_ns_inum;
 436                                 }
 437                                 parent_ns_inum;
 438                         }))
 439 #endif
 440                 ctf_array_text(char, child_comm, child->comm, TASK_COMM_LEN)
 441                 ctf_integer(pid_t, child_tid, child->pid)
 442                 ctf_sequence(pid_t, vtids, tp_locvar->vtids, u8, tp_locvar->ns_level)
 443                 ctf_integer(pid_t, child_pid, child->tgid)
 444 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,8,0))
 445                 ctf_integer(unsigned int, child_ns_inum,
 446                         ({
 447                                 unsigned int child_ns_inum = 0;
 448
 449                                 if (child) {
 450                                         struct pid_namespace *pid_ns;
 451
 452                                         pid_ns = task_active_pid_ns(child);
 453                                         if (pid_ns)
 454                                                 child_ns_inum =
 455                                                         pid_ns->lttng_ns_inum;
 456                                 }
 457                                 child_ns_inum;
 458                         }))
 459 #endif
 460         ),
 461
 462         TP_code_post()
 463 )
 464
 465 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,4,0))
 466 /*
 467  * Tracepoint for exec:
 468  */
 469 LTTNG_TRACEPOINT_EVENT(sched_process_exec,
 470
 471         TP_PROTO(struct task_struct *p, pid_t old_pid,
 472                  struct linux_binprm *bprm),
 473
 474         TP_ARGS(p, old_pid, bprm),
 475
 476         TP_FIELDS(
 477                 ctf_string(filename, bprm->filename)
 478                 ctf_integer(pid_t, tid, p->pid)
 479                 ctf_integer(pid_t, old_tid, old_pid)
 480         )
 481 )
 482 #endif
 483
 484 /*
 485  * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
 486  *     adding sched_stat support to SCHED_FIFO/RR would be welcome.
 487  */
 488 LTTNG_TRACEPOINT_EVENT_CLASS(sched_stat_template,
 489
 490         TP_PROTO(struct task_struct *tsk, u64 delay),
 491
 492         TP_ARGS(tsk, delay),
 493
 494         TP_FIELDS(
 495                 ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
 496                 ctf_integer(pid_t, tid, tsk->pid)
 497                 ctf_integer(u64, delay, delay)
 498         )
 499 )
 500
 501
 502 /*
 503  * Tracepoint for accounting wait time (time the task is runnable
 504  * but not actually running due to scheduler contention).
 505  */
 506 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_wait,
 507              TP_PROTO(struct task_struct *tsk, u64 delay),
 508              TP_ARGS(tsk, delay))
 509
 510 /*
 511  * Tracepoint for accounting sleep time (time the task is not runnable,
 512  * including iowait, see below).
 513  */
 514 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_sleep,
 515              TP_PROTO(struct task_struct *tsk, u64 delay),
 516              TP_ARGS(tsk, delay))
 517
 518 /*
 519  * Tracepoint for accounting iowait time (time the task is not runnable
 520  * due to waiting on IO to complete).
 521  */
 522 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_iowait,
 523              TP_PROTO(struct task_struct *tsk, u64 delay),
 524              TP_ARGS(tsk, delay))
 525
 526 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,3,0))
 527 /*
 528  * Tracepoint for accounting blocked time (time the task is in uninterruptible).
 529  */
 530 LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_blocked,
 531              TP_PROTO(struct task_struct *tsk, u64 delay),
 532              TP_ARGS(tsk, delay))
 533 #endif
 534
 535 /*
 536  * Tracepoint for accounting runtime (time the task is executing
 537  * on a CPU).
 538  */
 539 LTTNG_TRACEPOINT_EVENT(sched_stat_runtime,
 540
 541         TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
 542
 543         TP_ARGS(tsk, runtime, vruntime),
 544
 545         TP_FIELDS(
 546                 ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
 547                 ctf_integer(pid_t, tid, tsk->pid)
 548                 ctf_integer(u64, runtime, runtime)
 549                 ctf_integer(u64, vruntime, vruntime)
 550         )
 551 )
 552
 553 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,12,0) || \
 554         LTTNG_RT_KERNEL_RANGE(4,9,27,18, 4,10,0,0) || \
 555         LTTNG_RT_KERNEL_RANGE(4,11,5,1, 4,12,0,0))
 556 /*
 557  * Tracepoint for showing priority inheritance modifying a tasks
 558  * priority.
 559  */
 560 LTTNG_TRACEPOINT_EVENT(sched_pi_setprio,
 561
 562         TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
 563
 564         TP_ARGS(tsk, pi_task),
 565
 566         TP_FIELDS(
 567                 ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
 568                 ctf_integer(pid_t, tid, tsk->pid)
 569                 ctf_integer(int, oldprio, tsk->prio - MAX_RT_PRIO)
 570                 ctf_integer(int, newprio, pi_task ? pi_task->prio - MAX_RT_PRIO : tsk->prio - MAX_RT_PRIO)
 571         )
 572 )
 573 #else
 574 /*
 575  * Tracepoint for showing priority inheritance modifying a tasks
 576  * priority.
 577  */
 578 LTTNG_TRACEPOINT_EVENT(sched_pi_setprio,
 579
 580         TP_PROTO(struct task_struct *tsk, int newprio),
 581
 582         TP_ARGS(tsk, newprio),
 583
 584         TP_FIELDS(
 585                 ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
 586                 ctf_integer(pid_t, tid, tsk->pid)
 587                 ctf_integer(int, oldprio, tsk->prio - MAX_RT_PRIO)
 588                 ctf_integer(int, newprio, newprio - MAX_RT_PRIO)
 589         )
 590 )
 591 #endif
 592
 593 #endif /* LTTNG_TRACE_SCHED_H */
 594
 595 /* This part must be outside protection */
 596 #include <probes/define_trace.h>