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