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