instrumentation/events/lttng-module/sched.h

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