include/instrumentation/events/sched.h

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