include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13
  14 #include <linux/pid.h>
  15 #include <linux/sem.h>
  16 #include <linux/shm.h>
  17 #include <linux/mutex.h>
  18 #include <linux/plist.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/irqflags.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/nodemask.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/refcount.h>
  25 #include <linux/resource.h>
  26 #include <linux/latencytop.h>
  27 #include <linux/sched/prio.h>
  28 #include <linux/sched/types.h>
  29 #include <linux/signal_types.h>
  30 #include <linux/syscall_user_dispatch.h>
  31 #include <linux/mm_types_task.h>
  32 #include <linux/task_io_accounting.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/rseq.h>
  35 #include <linux/seqlock.h>
  36 #include <linux/kcsan.h>
  37 #include <asm/kmap_size.h>
  38
  39 /* task_struct member predeclarations (sorted alphabetically): */
  40 struct audit_context;
  41 struct backing_dev_info;
  42 struct bio_list;
  43 struct blk_plug;
  44 struct bpf_local_storage;
  45 struct capture_control;
  46 struct cfs_rq;
  47 struct fs_struct;
  48 struct futex_pi_state;
  49 struct io_context;
  50 struct io_uring_task;
  51 struct mempolicy;
  52 struct nameidata;
  53 struct nsproxy;
  54 struct perf_event_context;
  55 struct pid_namespace;
  56 struct pipe_inode_info;
  57 struct rcu_node;
  58 struct reclaim_state;
  59 struct robust_list_head;
  60 struct root_domain;
  61 struct rq;
  62 struct sched_attr;
  63 struct sched_param;
  64 struct seq_file;
  65 struct sighand_struct;
  66 struct signal_struct;
  67 struct task_delay_info;
  68 struct task_group;
  69
  70 /*
  71  * Task state bitmask. NOTE! These bits are also
  72  * encoded in fs/proc/array.c: get_task_state().
  73  *
  74  * We have two separate sets of flags: task->state
  75  * is about runnability, while task->exit_state are
  76  * about the task exiting. Confusing, but this way
  77  * modifying one set can't modify the other one by
  78  * mistake.
  79  */
  80
  81 /* Used in tsk->state: */
  82 #define TASK_RUNNING                    0x0000
  83 #define TASK_INTERRUPTIBLE              0x0001
  84 #define TASK_UNINTERRUPTIBLE            0x0002
  85 #define __TASK_STOPPED                  0x0004
  86 #define __TASK_TRACED                   0x0008
  87 /* Used in tsk->exit_state: */
  88 #define EXIT_DEAD                       0x0010
  89 #define EXIT_ZOMBIE                     0x0020
  90 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
  91 /* Used in tsk->state again: */
  92 #define TASK_PARKED                     0x0040
  93 #define TASK_DEAD                       0x0080
  94 #define TASK_WAKEKILL                   0x0100
  95 #define TASK_WAKING                     0x0200
  96 #define TASK_NOLOAD                     0x0400
  97 #define TASK_NEW                        0x0800
  98 /* RT specific auxilliary flag to mark RT lock waiters */
  99 #define TASK_RTLOCK_WAIT                0x1000
 100 #define TASK_STATE_MAX                  0x2000
 101
 102 /* Convenience macros for the sake of set_current_state: */
 103 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 104 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 105 #define TASK_TRACED                     (TASK_WAKEKILL | __TASK_TRACED)
 106
 107 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 108
 109 /* Convenience macros for the sake of wake_up(): */
 110 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 111
 112 /* get_task_state(): */
 113 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 114                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 115                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 116                                          TASK_PARKED)
 117
 118 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 119
 120 #define task_is_traced(task)            ((READ_ONCE(task->__state) & __TASK_TRACED) != 0)
 121
 122 #define task_is_stopped(task)           ((READ_ONCE(task->__state) & __TASK_STOPPED) != 0)
 123
 124 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->__state) & (__TASK_STOPPED | __TASK_TRACED)) != 0)
 125
 126 /*
 127  * Special states are those that do not use the normal wait-loop pattern. See
 128  * the comment with set_special_state().
 129  */
 130 #define is_special_task_state(state)                            \
 131         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
 132
 133 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 134 # define debug_normal_state_change(state_value)                         \
 135         do {                                                            \
 136                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 137                 current->task_state_change = _THIS_IP_;                 \
 138         } while (0)
 139
 140 # define debug_special_state_change(state_value)                        \
 141         do {                                                            \
 142                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 143                 current->task_state_change = _THIS_IP_;                 \
 144         } while (0)
 145
 146 # define debug_rtlock_wait_set_state()                                  \
 147         do {                                                             \
 148                 current->saved_state_change = current->task_state_change;\
 149                 current->task_state_change = _THIS_IP_;                  \
 150         } while (0)
 151
 152 # define debug_rtlock_wait_restore_state()                              \
 153         do {                                                             \
 154                 current->task_state_change = current->saved_state_change;\
 155         } while (0)
 156
 157 #else
 158 # define debug_normal_state_change(cond)        do { } while (0)
 159 # define debug_special_state_change(cond)       do { } while (0)
 160 # define debug_rtlock_wait_set_state()          do { } while (0)
 161 # define debug_rtlock_wait_restore_state()      do { } while (0)
 162 #endif
 163
 164 /*
 165  * set_current_state() includes a barrier so that the write of current->state
 166  * is correctly serialised wrt the caller's subsequent test of whether to
 167  * actually sleep:
 168  *
 169  *   for (;;) {
 170  *      set_current_state(TASK_UNINTERRUPTIBLE);
 171  *      if (CONDITION)
 172  *         break;
 173  *
 174  *      schedule();
 175  *   }
 176  *   __set_current_state(TASK_RUNNING);
 177  *
 178  * If the caller does not need such serialisation (because, for instance, the
 179  * CONDITION test and condition change and wakeup are under the same lock) then
 180  * use __set_current_state().
 181  *
 182  * The above is typically ordered against the wakeup, which does:
 183  *
 184  *   CONDITION = 1;
 185  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 186  *
 187  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 188  * accessing p->state.
 189  *
 190  * Wakeup will do: if (@state & p->state) p->state = TASK_RUNNING, that is,
 191  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 192  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 193  *
 194  * However, with slightly different timing the wakeup TASK_RUNNING store can
 195  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 196  * a problem either because that will result in one extra go around the loop
 197  * and our @cond test will save the day.
 198  *
 199  * Also see the comments of try_to_wake_up().
 200  */
 201 #define __set_current_state(state_value)                                \
 202         do {                                                            \
 203                 debug_normal_state_change((state_value));               \
 204                 WRITE_ONCE(current->__state, (state_value));            \
 205         } while (0)
 206
 207 #define set_current_state(state_value)                                  \
 208         do {                                                            \
 209                 debug_normal_state_change((state_value));               \
 210                 smp_store_mb(current->__state, (state_value));          \
 211         } while (0)
 212
 213 /*
 214  * set_special_state() should be used for those states when the blocking task
 215  * can not use the regular condition based wait-loop. In that case we must
 216  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 217  * stores will not collide with our state change.
 218  */
 219 #define set_special_state(state_value)                                  \
 220         do {                                                            \
 221                 unsigned long flags; /* may shadow */                   \
 222                                                                         \
 223                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 224                 debug_special_state_change((state_value));              \
 225                 WRITE_ONCE(current->__state, (state_value));            \
 226                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 227         } while (0)
 228
 229 /*
 230  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 231  *
 232  * RT's spin/rwlock substitutions are state preserving. The state of the
 233  * task when blocking on the lock is saved in task_struct::saved_state and
 234  * restored after the lock has been acquired.  These operations are
 235  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 236  * lock related wakeups while the task is blocked on the lock are
 237  * redirected to operate on task_struct::saved_state to ensure that these
 238  * are not dropped. On restore task_struct::saved_state is set to
 239  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 240  *
 241  * The lock operation looks like this:
 242  *
 243  *      current_save_and_set_rtlock_wait_state();
 244  *      for (;;) {
 245  *              if (try_lock())
 246  *                      break;
 247  *              raw_spin_unlock_irq(&lock->wait_lock);
 248  *              schedule_rtlock();
 249  *              raw_spin_lock_irq(&lock->wait_lock);
 250  *              set_current_state(TASK_RTLOCK_WAIT);
 251  *      }
 252  *      current_restore_rtlock_saved_state();
 253  */
 254 #define current_save_and_set_rtlock_wait_state()                        \
 255         do {                                                            \
 256                 lockdep_assert_irqs_disabled();                         \
 257                 raw_spin_lock(&current->pi_lock);                       \
 258                 current->saved_state = current->__state;                \
 259                 debug_rtlock_wait_set_state();                          \
 260                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 261                 raw_spin_unlock(&current->pi_lock);                     \
 262         } while (0);
 263
 264 #define current_restore_rtlock_saved_state()                            \
 265         do {                                                            \
 266                 lockdep_assert_irqs_disabled();                         \
 267                 raw_spin_lock(&current->pi_lock);                       \
 268                 debug_rtlock_wait_restore_state();                      \
 269                 WRITE_ONCE(current->__state, current->saved_state);     \
 270                 current->saved_state = TASK_RUNNING;                    \
 271                 raw_spin_unlock(&current->pi_lock);                     \
 272         } while (0);
 273
 274 #define get_current_state()     READ_ONCE(current->__state)
 275
 276 /* Task command name length: */
 277 #define TASK_COMM_LEN                   16
 278
 279 extern void scheduler_tick(void);
 280
 281 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 282
 283 extern long schedule_timeout(long timeout);
 284 extern long schedule_timeout_interruptible(long timeout);
 285 extern long schedule_timeout_killable(long timeout);
 286 extern long schedule_timeout_uninterruptible(long timeout);
 287 extern long schedule_timeout_idle(long timeout);
 288 asmlinkage void schedule(void);
 289 extern void schedule_preempt_disabled(void);
 290 asmlinkage void preempt_schedule_irq(void);
 291
 292 extern int __must_check io_schedule_prepare(void);
 293 extern void io_schedule_finish(int token);
 294 extern long io_schedule_timeout(long timeout);
 295 extern void io_schedule(void);
 296
 297 /**
 298  * struct prev_cputime - snapshot of system and user cputime
 299  * @utime: time spent in user mode
 300  * @stime: time spent in system mode
 301  * @lock: protects the above two fields
 302  *
 303  * Stores previous user/system time values such that we can guarantee
 304  * monotonicity.
 305  */
 306 struct prev_cputime {
 307 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 308         u64                             utime;
 309         u64                             stime;
 310         raw_spinlock_t                  lock;
 311 #endif
 312 };
 313
 314 enum vtime_state {
 315         /* Task is sleeping or running in a CPU with VTIME inactive: */
 316         VTIME_INACTIVE = 0,
 317         /* Task is idle */
 318         VTIME_IDLE,
 319         /* Task runs in kernelspace in a CPU with VTIME active: */
 320         VTIME_SYS,
 321         /* Task runs in userspace in a CPU with VTIME active: */
 322         VTIME_USER,
 323         /* Task runs as guests in a CPU with VTIME active: */
 324         VTIME_GUEST,
 325 };
 326
 327 struct vtime {
 328         seqcount_t              seqcount;
 329         unsigned long long      starttime;
 330         enum vtime_state        state;
 331         unsigned int            cpu;
 332         u64                     utime;
 333         u64                     stime;
 334         u64                     gtime;
 335 };
 336
 337 /*
 338  * Utilization clamp constraints.
 339  * @UCLAMP_MIN: Minimum utilization
 340  * @UCLAMP_MAX: Maximum utilization
 341  * @UCLAMP_CNT: Utilization clamp constraints count
 342  */
 343 enum uclamp_id {
 344         UCLAMP_MIN = 0,
 345         UCLAMP_MAX,
 346         UCLAMP_CNT
 347 };
 348
 349 #ifdef CONFIG_SMP
 350 extern struct root_domain def_root_domain;
 351 extern struct mutex sched_domains_mutex;
 352 #endif
 353
 354 struct sched_info {
 355 #ifdef CONFIG_SCHED_INFO
 356         /* Cumulative counters: */
 357
 358         /* # of times we have run on this CPU: */
 359         unsigned long                   pcount;
 360
 361         /* Time spent waiting on a runqueue: */
 362         unsigned long long              run_delay;
 363
 364         /* Timestamps: */
 365
 366         /* When did we last run on a CPU? */
 367         unsigned long long              last_arrival;
 368
 369         /* When were we last queued to run? */
 370         unsigned long long              last_queued;
 371
 372 #endif /* CONFIG_SCHED_INFO */
 373 };
 374
 375 /*
 376  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 377  * has a few: load, load_avg, util_avg, freq, and capacity.
 378  *
 379  * We define a basic fixed point arithmetic range, and then formalize
 380  * all these metrics based on that basic range.
 381  */
 382 # define SCHED_FIXEDPOINT_SHIFT         10
 383 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 384
 385 /* Increase resolution of cpu_capacity calculations */
 386 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 387 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 388
 389 struct load_weight {
 390         unsigned long                   weight;
 391         u32                             inv_weight;
 392 };
 393
 394 /**
 395  * struct util_est - Estimation utilization of FAIR tasks
 396  * @enqueued: instantaneous estimated utilization of a task/cpu
 397  * @ewma:     the Exponential Weighted Moving Average (EWMA)
 398  *            utilization of a task
 399  *
 400  * Support data structure to track an Exponential Weighted Moving Average
 401  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
 402  * average each time a task completes an activation. Sample's weight is chosen
 403  * so that the EWMA will be relatively insensitive to transient changes to the
 404  * task's workload.
 405  *
 406  * The enqueued attribute has a slightly different meaning for tasks and cpus:
 407  * - task:   the task's util_avg at last task dequeue time
 408  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
 409  * Thus, the util_est.enqueued of a task represents the contribution on the
 410  * estimated utilization of the CPU where that task is currently enqueued.
 411  *
 412  * Only for tasks we track a moving average of the past instantaneous
 413  * estimated utilization. This allows to absorb sporadic drops in utilization
 414  * of an otherwise almost periodic task.
 415  *
 416  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 417  * updates. When a task is dequeued, its util_est should not be updated if its
 418  * util_avg has not been updated in the meantime.
 419  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
 420  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
 421  * for a task) it is safe to use MSB.
 422  */
 423 struct util_est {
 424         unsigned int                    enqueued;
 425         unsigned int                    ewma;
 426 #define UTIL_EST_WEIGHT_SHIFT           2
 427 #define UTIL_AVG_UNCHANGED              0x80000000
 428 } __attribute__((__aligned__(sizeof(u64))));
 429
 430 /*
 431  * The load/runnable/util_avg accumulates an infinite geometric series
 432  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 433  *
 434  * [load_avg definition]
 435  *
 436  *   load_avg = runnable% * scale_load_down(load)
 437  *
 438  * [runnable_avg definition]
 439  *
 440  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 441  *
 442  * [util_avg definition]
 443  *
 444  *   util_avg = running% * SCHED_CAPACITY_SCALE
 445  *
 446  * where runnable% is the time ratio that a sched_entity is runnable and
 447  * running% the time ratio that a sched_entity is running.
 448  *
 449  * For cfs_rq, they are the aggregated values of all runnable and blocked
 450  * sched_entities.
 451  *
 452  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 453  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 454  * for computing those signals (see update_rq_clock_pelt())
 455  *
 456  * N.B., the above ratios (runnable% and running%) themselves are in the
 457  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 458  * to as large a range as necessary. This is for example reflected by
 459  * util_avg's SCHED_CAPACITY_SCALE.
 460  *
 461  * [Overflow issue]
 462  *
 463  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 464  * with the highest load (=88761), always runnable on a single cfs_rq,
 465  * and should not overflow as the number already hits PID_MAX_LIMIT.
 466  *
 467  * For all other cases (including 32-bit kernels), struct load_weight's
 468  * weight will overflow first before we do, because:
 469  *
 470  *    Max(load_avg) <= Max(load.weight)
 471  *
 472  * Then it is the load_weight's responsibility to consider overflow
 473  * issues.
 474  */
 475 struct sched_avg {
 476         u64                             last_update_time;
 477         u64                             load_sum;
 478         u64                             runnable_sum;
 479         u32                             util_sum;
 480         u32                             period_contrib;
 481         unsigned long                   load_avg;
 482         unsigned long                   runnable_avg;
 483         unsigned long                   util_avg;
 484         struct util_est                 util_est;
 485 } ____cacheline_aligned;
 486
 487 struct sched_statistics {
 488 #ifdef CONFIG_SCHEDSTATS
 489         u64                             wait_start;
 490         u64                             wait_max;
 491         u64                             wait_count;
 492         u64                             wait_sum;
 493         u64                             iowait_count;
 494         u64                             iowait_sum;
 495
 496         u64                             sleep_start;
 497         u64                             sleep_max;
 498         s64                             sum_sleep_runtime;
 499
 500         u64                             block_start;
 501         u64                             block_max;
 502         u64                             exec_max;
 503         u64                             slice_max;
 504
 505         u64                             nr_migrations_cold;
 506         u64                             nr_failed_migrations_affine;
 507         u64                             nr_failed_migrations_running;
 508         u64                             nr_failed_migrations_hot;
 509         u64                             nr_forced_migrations;
 510
 511         u64                             nr_wakeups;
 512         u64                             nr_wakeups_sync;
 513         u64                             nr_wakeups_migrate;
 514         u64                             nr_wakeups_local;
 515         u64                             nr_wakeups_remote;
 516         u64                             nr_wakeups_affine;
 517         u64                             nr_wakeups_affine_attempts;
 518         u64                             nr_wakeups_passive;
 519         u64                             nr_wakeups_idle;
 520 #endif
 521 };
 522
 523 struct sched_entity {
 524         /* For load-balancing: */
 525         struct load_weight              load;
 526         struct rb_node                  run_node;
 527         struct list_head                group_node;
 528         unsigned int                    on_rq;
 529
 530         u64                             exec_start;
 531         u64                             sum_exec_runtime;
 532         u64                             vruntime;
 533         u64                             prev_sum_exec_runtime;
 534
 535         u64                             nr_migrations;
 536
 537         struct sched_statistics         statistics;
 538
 539 #ifdef CONFIG_FAIR_GROUP_SCHED
 540         int                             depth;
 541         struct sched_entity             *parent;
 542         /* rq on which this entity is (to be) queued: */
 543         struct cfs_rq                   *cfs_rq;
 544         /* rq "owned" by this entity/group: */
 545         struct cfs_rq                   *my_q;
 546         /* cached value of my_q->h_nr_running */
 547         unsigned long                   runnable_weight;
 548 #endif
 549
 550 #ifdef CONFIG_SMP
 551         /*
 552          * Per entity load average tracking.
 553          *
 554          * Put into separate cache line so it does not
 555          * collide with read-mostly values above.
 556          */
 557         struct sched_avg                avg;
 558 #endif
 559 };
 560
 561 struct sched_rt_entity {
 562         struct list_head                run_list;
 563         unsigned long                   timeout;
 564         unsigned long                   watchdog_stamp;
 565         unsigned int                    time_slice;
 566         unsigned short                  on_rq;
 567         unsigned short                  on_list;
 568
 569         struct sched_rt_entity          *back;
 570 #ifdef CONFIG_RT_GROUP_SCHED
 571         struct sched_rt_entity          *parent;
 572         /* rq on which this entity is (to be) queued: */
 573         struct rt_rq                    *rt_rq;
 574         /* rq "owned" by this entity/group: */
 575         struct rt_rq                    *my_q;
 576 #endif
 577 } __randomize_layout;
 578
 579 struct sched_dl_entity {
 580         struct rb_node                  rb_node;
 581
 582         /*
 583          * Original scheduling parameters. Copied here from sched_attr
 584          * during sched_setattr(), they will remain the same until
 585          * the next sched_setattr().
 586          */
 587         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 588         u64                             dl_deadline;    /* Relative deadline of each instance   */
 589         u64                             dl_period;      /* Separation of two instances (period) */
 590         u64                             dl_bw;          /* dl_runtime / dl_period               */
 591         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 592
 593         /*
 594          * Actual scheduling parameters. Initialized with the values above,
 595          * they are continuously updated during task execution. Note that
 596          * the remaining runtime could be < 0 in case we are in overrun.
 597          */
 598         s64                             runtime;        /* Remaining runtime for this instance  */
 599         u64                             deadline;       /* Absolute deadline for this instance  */
 600         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 601
 602         /*
 603          * Some bool flags:
 604          *
 605          * @dl_throttled tells if we exhausted the runtime. If so, the
 606          * task has to wait for a replenishment to be performed at the
 607          * next firing of dl_timer.
 608          *
 609          * @dl_boosted tells if we are boosted due to DI. If so we are
 610          * outside bandwidth enforcement mechanism (but only until we
 611          * exit the critical section);
 612          *
 613          * @dl_yielded tells if task gave up the CPU before consuming
 614          * all its available runtime during the last job.
 615          *
 616          * @dl_non_contending tells if the task is inactive while still
 617          * contributing to the active utilization. In other words, it
 618          * indicates if the inactive timer has been armed and its handler
 619          * has not been executed yet. This flag is useful to avoid race
 620          * conditions between the inactive timer handler and the wakeup
 621          * code.
 622          *
 623          * @dl_overrun tells if the task asked to be informed about runtime
 624          * overruns.
 625          */
 626         unsigned int                    dl_throttled      : 1;
 627         unsigned int                    dl_yielded        : 1;
 628         unsigned int                    dl_non_contending : 1;
 629         unsigned int                    dl_overrun        : 1;
 630
 631         /*
 632          * Bandwidth enforcement timer. Each -deadline task has its
 633          * own bandwidth to be enforced, thus we need one timer per task.
 634          */
 635         struct hrtimer                  dl_timer;
 636
 637         /*
 638          * Inactive timer, responsible for decreasing the active utilization
 639          * at the "0-lag time". When a -deadline task blocks, it contributes
 640          * to GRUB's active utilization until the "0-lag time", hence a
 641          * timer is needed to decrease the active utilization at the correct
 642          * time.
 643          */
 644         struct hrtimer inactive_timer;
 645
 646 #ifdef CONFIG_RT_MUTEXES
 647         /*
 648          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 649          * pi_se points to the donor, otherwise points to the dl_se it belongs
 650          * to (the original one/itself).
 651          */
 652         struct sched_dl_entity *pi_se;
 653 #endif
 654 };
 655
 656 #ifdef CONFIG_UCLAMP_TASK
 657 /* Number of utilization clamp buckets (shorter alias) */
 658 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 659
 660 /*
 661  * Utilization clamp for a scheduling entity
 662  * @value:              clamp value "assigned" to a se
 663  * @bucket_id:          bucket index corresponding to the "assigned" value
 664  * @active:             the se is currently refcounted in a rq's bucket
 665  * @user_defined:       the requested clamp value comes from user-space
 666  *
 667  * The bucket_id is the index of the clamp bucket matching the clamp value
 668  * which is pre-computed and stored to avoid expensive integer divisions from
 669  * the fast path.
 670  *
 671  * The active bit is set whenever a task has got an "effective" value assigned,
 672  * which can be different from the clamp value "requested" from user-space.
 673  * This allows to know a task is refcounted in the rq's bucket corresponding
 674  * to the "effective" bucket_id.
 675  *
 676  * The user_defined bit is set whenever a task has got a task-specific clamp
 677  * value requested from userspace, i.e. the system defaults apply to this task
 678  * just as a restriction. This allows to relax default clamps when a less
 679  * restrictive task-specific value has been requested, thus allowing to
 680  * implement a "nice" semantic. For example, a task running with a 20%
 681  * default boost can still drop its own boosting to 0%.
 682  */
 683 struct uclamp_se {
 684         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 685         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 686         unsigned int active             : 1;
 687         unsigned int user_defined       : 1;
 688 };
 689 #endif /* CONFIG_UCLAMP_TASK */
 690
 691 union rcu_special {
 692         struct {
 693                 u8                      blocked;
 694                 u8                      need_qs;
 695                 u8                      exp_hint; /* Hint for performance. */
 696                 u8                      need_mb; /* Readers need smp_mb(). */
 697         } b; /* Bits. */
 698         u32 s; /* Set of bits. */
 699 };
 700
 701 enum perf_event_task_context {
 702         perf_invalid_context = -1,
 703         perf_hw_context = 0,
 704         perf_sw_context,
 705         perf_nr_task_contexts,
 706 };
 707
 708 struct wake_q_node {
 709         struct wake_q_node *next;
 710 };
 711
 712 struct kmap_ctrl {
 713 #ifdef CONFIG_KMAP_LOCAL
 714         int                             idx;
 715         pte_t                           pteval[KM_MAX_IDX];
 716 #endif
 717 };
 718
 719 struct task_struct {
 720 #ifdef CONFIG_THREAD_INFO_IN_TASK
 721         /*
 722          * For reasons of header soup (see current_thread_info()), this
 723          * must be the first element of task_struct.
 724          */
 725         struct thread_info              thread_info;
 726 #endif
 727         unsigned int                    __state;
 728
 729 #ifdef CONFIG_PREEMPT_RT
 730         /* saved state for "spinlock sleepers" */
 731         unsigned int                    saved_state;
 732 #endif
 733
 734         /*
 735          * This begins the randomizable portion of task_struct. Only
 736          * scheduling-critical items should be added above here.
 737          */
 738         randomized_struct_fields_start
 739
 740         void                            *stack;
 741         refcount_t                      usage;
 742         /* Per task flags (PF_*), defined further below: */
 743         unsigned int                    flags;
 744         unsigned int                    ptrace;
 745
 746 #ifdef CONFIG_SMP
 747         int                             on_cpu;
 748         struct __call_single_node       wake_entry;
 749 #ifdef CONFIG_THREAD_INFO_IN_TASK
 750         /* Current CPU: */
 751         unsigned int                    cpu;
 752 #endif
 753         unsigned int                    wakee_flips;
 754         unsigned long                   wakee_flip_decay_ts;
 755         struct task_struct              *last_wakee;
 756
 757         /*
 758          * recent_used_cpu is initially set as the last CPU used by a task
 759          * that wakes affine another task. Waker/wakee relationships can
 760          * push tasks around a CPU where each wakeup moves to the next one.
 761          * Tracking a recently used CPU allows a quick search for a recently
 762          * used CPU that may be idle.
 763          */
 764         int                             recent_used_cpu;
 765         int                             wake_cpu;
 766 #endif
 767         int                             on_rq;
 768
 769         int                             prio;
 770         int                             static_prio;
 771         int                             normal_prio;
 772         unsigned int                    rt_priority;
 773
 774         const struct sched_class        *sched_class;
 775         struct sched_entity             se;
 776         struct sched_rt_entity          rt;
 777         struct sched_dl_entity          dl;
 778
 779 #ifdef CONFIG_SCHED_CORE
 780         struct rb_node                  core_node;
 781         unsigned long                   core_cookie;
 782         unsigned int                    core_occupation;
 783 #endif
 784
 785 #ifdef CONFIG_CGROUP_SCHED
 786         struct task_group               *sched_task_group;
 787 #endif
 788
 789 #ifdef CONFIG_UCLAMP_TASK
 790         /*
 791          * Clamp values requested for a scheduling entity.
 792          * Must be updated with task_rq_lock() held.
 793          */
 794         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 795         /*
 796          * Effective clamp values used for a scheduling entity.
 797          * Must be updated with task_rq_lock() held.
 798          */
 799         struct uclamp_se                uclamp[UCLAMP_CNT];
 800 #endif
 801
 802 #ifdef CONFIG_PREEMPT_NOTIFIERS
 803         /* List of struct preempt_notifier: */
 804         struct hlist_head               preempt_notifiers;
 805 #endif
 806
 807 #ifdef CONFIG_BLK_DEV_IO_TRACE
 808         unsigned int                    btrace_seq;
 809 #endif
 810
 811         unsigned int                    policy;
 812         int                             nr_cpus_allowed;
 813         const cpumask_t                 *cpus_ptr;
 814         cpumask_t                       cpus_mask;
 815         void                            *migration_pending;
 816 #ifdef CONFIG_SMP
 817         unsigned short                  migration_disabled;
 818 #endif
 819         unsigned short                  migration_flags;
 820
 821 #ifdef CONFIG_PREEMPT_RCU
 822         int                             rcu_read_lock_nesting;
 823         union rcu_special               rcu_read_unlock_special;
 824         struct list_head                rcu_node_entry;
 825         struct rcu_node                 *rcu_blocked_node;
 826 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 827
 828 #ifdef CONFIG_TASKS_RCU
 829         unsigned long                   rcu_tasks_nvcsw;
 830         u8                              rcu_tasks_holdout;
 831         u8                              rcu_tasks_idx;
 832         int                             rcu_tasks_idle_cpu;
 833         struct list_head                rcu_tasks_holdout_list;
 834 #endif /* #ifdef CONFIG_TASKS_RCU */
 835
 836 #ifdef CONFIG_TASKS_TRACE_RCU
 837         int                             trc_reader_nesting;
 838         int                             trc_ipi_to_cpu;
 839         union rcu_special               trc_reader_special;
 840         bool                            trc_reader_checked;
 841         struct list_head                trc_holdout_list;
 842 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 843
 844         struct sched_info               sched_info;
 845
 846         struct list_head                tasks;
 847 #ifdef CONFIG_SMP
 848         struct plist_node               pushable_tasks;
 849         struct rb_node                  pushable_dl_tasks;
 850 #endif
 851
 852         struct mm_struct                *mm;
 853         struct mm_struct                *active_mm;
 854
 855         /* Per-thread vma caching: */
 856         struct vmacache                 vmacache;
 857
 858 #ifdef SPLIT_RSS_COUNTING
 859         struct task_rss_stat            rss_stat;
 860 #endif
 861         int                             exit_state;
 862         int                             exit_code;
 863         int                             exit_signal;
 864         /* The signal sent when the parent dies: */
 865         int                             pdeath_signal;
 866         /* JOBCTL_*, siglock protected: */
 867         unsigned long                   jobctl;
 868
 869         /* Used for emulating ABI behavior of previous Linux versions: */
 870         unsigned int                    personality;
 871
 872         /* Scheduler bits, serialized by scheduler locks: */
 873         unsigned                        sched_reset_on_fork:1;
 874         unsigned                        sched_contributes_to_load:1;
 875         unsigned                        sched_migrated:1;
 876 #ifdef CONFIG_PSI
 877         unsigned                        sched_psi_wake_requeue:1;
 878 #endif
 879
 880         /* Force alignment to the next boundary: */
 881         unsigned                        :0;
 882
 883         /* Unserialized, strictly 'current' */
 884
 885         /*
 886          * This field must not be in the scheduler word above due to wakelist
 887          * queueing no longer being serialized by p->on_cpu. However:
 888          *
 889          * p->XXX = X;                  ttwu()
 890          * schedule()                     if (p->on_rq && ..) // false
 891          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 892          *   deactivate_task()                ttwu_queue_wakelist())
 893          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 894          *
 895          * guarantees all stores of 'current' are visible before
 896          * ->sched_remote_wakeup gets used, so it can be in this word.
 897          */
 898         unsigned                        sched_remote_wakeup:1;
 899
 900         /* Bit to tell LSMs we're in execve(): */
 901         unsigned                        in_execve:1;
 902         unsigned                        in_iowait:1;
 903 #ifndef TIF_RESTORE_SIGMASK
 904         unsigned                        restore_sigmask:1;
 905 #endif
 906 #ifdef CONFIG_MEMCG
 907         unsigned                        in_user_fault:1;
 908 #endif
 909 #ifdef CONFIG_COMPAT_BRK
 910         unsigned                        brk_randomized:1;
 911 #endif
 912 #ifdef CONFIG_CGROUPS
 913         /* disallow userland-initiated cgroup migration */
 914         unsigned                        no_cgroup_migration:1;
 915         /* task is frozen/stopped (used by the cgroup freezer) */
 916         unsigned                        frozen:1;
 917 #endif
 918 #ifdef CONFIG_BLK_CGROUP
 919         unsigned                        use_memdelay:1;
 920 #endif
 921 #ifdef CONFIG_PSI
 922         /* Stalled due to lack of memory */
 923         unsigned                        in_memstall:1;
 924 #endif
 925 #ifdef CONFIG_PAGE_OWNER
 926         /* Used by page_owner=on to detect recursion in page tracking. */
 927         unsigned                        in_page_owner:1;
 928 #endif
 929
 930         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 931
 932         struct restart_block            restart_block;
 933
 934         pid_t                           pid;
 935         pid_t                           tgid;
 936
 937 #ifdef CONFIG_STACKPROTECTOR
 938         /* Canary value for the -fstack-protector GCC feature: */
 939         unsigned long                   stack_canary;
 940 #endif
 941         /*
 942          * Pointers to the (original) parent process, youngest child, younger sibling,
 943          * older sibling, respectively.  (p->father can be replaced with
 944          * p->real_parent->pid)
 945          */
 946
 947         /* Real parent process: */
 948         struct task_struct __rcu        *real_parent;
 949
 950         /* Recipient of SIGCHLD, wait4() reports: */
 951         struct task_struct __rcu        *parent;
 952
 953         /*
 954          * Children/sibling form the list of natural children:
 955          */
 956         struct list_head                children;
 957         struct list_head                sibling;
 958         struct task_struct              *group_leader;
 959
 960         /*
 961          * 'ptraced' is the list of tasks this task is using ptrace() on.
 962          *
 963          * This includes both natural children and PTRACE_ATTACH targets.
 964          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 965          */
 966         struct list_head                ptraced;
 967         struct list_head                ptrace_entry;
 968
 969         /* PID/PID hash table linkage. */
 970         struct pid                      *thread_pid;
 971         struct hlist_node               pid_links[PIDTYPE_MAX];
 972         struct list_head                thread_group;
 973         struct list_head                thread_node;
 974
 975         struct completion               *vfork_done;
 976
 977         /* CLONE_CHILD_SETTID: */
 978         int __user                      *set_child_tid;
 979
 980         /* CLONE_CHILD_CLEARTID: */
 981         int __user                      *clear_child_tid;
 982
 983         /* PF_IO_WORKER */
 984         void                            *pf_io_worker;
 985
 986         u64                             utime;
 987         u64                             stime;
 988 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 989         u64                             utimescaled;
 990         u64                             stimescaled;
 991 #endif
 992         u64                             gtime;
 993         struct prev_cputime             prev_cputime;
 994 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
 995         struct vtime                    vtime;
 996 #endif
 997
 998 #ifdef CONFIG_NO_HZ_FULL
 999         atomic_t                        tick_dep_mask;
1000 #endif
1001         /* Context switch counts: */
1002         unsigned long                   nvcsw;
1003         unsigned long                   nivcsw;
1004
1005         /* Monotonic time in nsecs: */
1006         u64                             start_time;
1007
1008         /* Boot based time in nsecs: */
1009         u64                             start_boottime;
1010
1011         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1012         unsigned long                   min_flt;
1013         unsigned long                   maj_flt;
1014
1015         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1016         struct posix_cputimers          posix_cputimers;
1017
1018 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1019         struct posix_cputimers_work     posix_cputimers_work;
1020 #endif
1021
1022         /* Process credentials: */
1023
1024         /* Tracer's credentials at attach: */
1025         const struct cred __rcu         *ptracer_cred;
1026
1027         /* Objective and real subjective task credentials (COW): */
1028         const struct cred __rcu         *real_cred;
1029
1030         /* Effective (overridable) subjective task credentials (COW): */
1031         const struct cred __rcu         *cred;
1032
1033 #ifdef CONFIG_KEYS
1034         /* Cached requested key. */
1035         struct key                      *cached_requested_key;
1036 #endif
1037
1038         /*
1039          * executable name, excluding path.
1040          *
1041          * - normally initialized setup_new_exec()
1042          * - access it with [gs]et_task_comm()
1043          * - lock it with task_lock()
1044          */
1045         char                            comm[TASK_COMM_LEN];
1046
1047         struct nameidata                *nameidata;
1048
1049 #ifdef CONFIG_SYSVIPC
1050         struct sysv_sem                 sysvsem;
1051         struct sysv_shm                 sysvshm;
1052 #endif
1053 #ifdef CONFIG_DETECT_HUNG_TASK
1054         unsigned long                   last_switch_count;
1055         unsigned long                   last_switch_time;
1056 #endif
1057         /* Filesystem information: */
1058         struct fs_struct                *fs;
1059
1060         /* Open file information: */
1061         struct files_struct             *files;
1062
1063 #ifdef CONFIG_IO_URING
1064         struct io_uring_task            *io_uring;
1065 #endif
1066
1067         /* Namespaces: */
1068         struct nsproxy                  *nsproxy;
1069
1070         /* Signal handlers: */
1071         struct signal_struct            *signal;
1072         struct sighand_struct __rcu             *sighand;
1073         sigset_t                        blocked;
1074         sigset_t                        real_blocked;
1075         /* Restored if set_restore_sigmask() was used: */
1076         sigset_t                        saved_sigmask;
1077         struct sigpending               pending;
1078         unsigned long                   sas_ss_sp;
1079         size_t                          sas_ss_size;
1080         unsigned int                    sas_ss_flags;
1081
1082         struct callback_head            *task_works;
1083
1084 #ifdef CONFIG_AUDIT
1085 #ifdef CONFIG_AUDITSYSCALL
1086         struct audit_context            *audit_context;
1087 #endif
1088         kuid_t                          loginuid;
1089         unsigned int                    sessionid;
1090 #endif
1091         struct seccomp                  seccomp;
1092         struct syscall_user_dispatch    syscall_dispatch;
1093
1094         /* Thread group tracking: */
1095         u64                             parent_exec_id;
1096         u64                             self_exec_id;
1097
1098         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1099         spinlock_t                      alloc_lock;
1100
1101         /* Protection of the PI data structures: */
1102         raw_spinlock_t                  pi_lock;
1103
1104         struct wake_q_node              wake_q;
1105
1106 #ifdef CONFIG_RT_MUTEXES
1107         /* PI waiters blocked on a rt_mutex held by this task: */
1108         struct rb_root_cached           pi_waiters;
1109         /* Updated under owner's pi_lock and rq lock */
1110         struct task_struct              *pi_top_task;
1111         /* Deadlock detection and priority inheritance handling: */
1112         struct rt_mutex_waiter          *pi_blocked_on;
1113 #endif
1114
1115 #ifdef CONFIG_DEBUG_MUTEXES
1116         /* Mutex deadlock detection: */
1117         struct mutex_waiter             *blocked_on;
1118 #endif
1119
1120 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1121         int                             non_block_count;
1122 #endif
1123
1124 #ifdef CONFIG_TRACE_IRQFLAGS
1125         struct irqtrace_events          irqtrace;
1126         unsigned int                    hardirq_threaded;
1127         u64                             hardirq_chain_key;
1128         int                             softirqs_enabled;
1129         int                             softirq_context;
1130         int                             irq_config;
1131 #endif
1132 #ifdef CONFIG_PREEMPT_RT
1133         int                             softirq_disable_cnt;
1134 #endif
1135
1136 #ifdef CONFIG_LOCKDEP
1137 # define MAX_LOCK_DEPTH                 48UL
1138         u64                             curr_chain_key;
1139         int                             lockdep_depth;
1140         unsigned int                    lockdep_recursion;
1141         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1142 #endif
1143
1144 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1145         unsigned int                    in_ubsan;
1146 #endif
1147
1148         /* Journalling filesystem info: */
1149         void                            *journal_info;
1150
1151         /* Stacked block device info: */
1152         struct bio_list                 *bio_list;
1153
1154 #ifdef CONFIG_BLOCK
1155         /* Stack plugging: */
1156         struct blk_plug                 *plug;
1157 #endif
1158
1159         /* VM state: */
1160         struct reclaim_state            *reclaim_state;
1161
1162         struct backing_dev_info         *backing_dev_info;
1163
1164         struct io_context               *io_context;
1165
1166 #ifdef CONFIG_COMPACTION
1167         struct capture_control          *capture_control;
1168 #endif
1169         /* Ptrace state: */
1170         unsigned long                   ptrace_message;
1171         kernel_siginfo_t                *last_siginfo;
1172
1173         struct task_io_accounting       ioac;
1174 #ifdef CONFIG_PSI
1175         /* Pressure stall state */
1176         unsigned int                    psi_flags;
1177 #endif
1178 #ifdef CONFIG_TASK_XACCT
1179         /* Accumulated RSS usage: */
1180         u64                             acct_rss_mem1;
1181         /* Accumulated virtual memory usage: */
1182         u64                             acct_vm_mem1;
1183         /* stime + utime since last update: */
1184         u64                             acct_timexpd;
1185 #endif
1186 #ifdef CONFIG_CPUSETS
1187         /* Protected by ->alloc_lock: */
1188         nodemask_t                      mems_allowed;
1189         /* Sequence number to catch updates: */
1190         seqcount_spinlock_t             mems_allowed_seq;
1191         int                             cpuset_mem_spread_rotor;
1192         int                             cpuset_slab_spread_rotor;
1193 #endif
1194 #ifdef CONFIG_CGROUPS
1195         /* Control Group info protected by css_set_lock: */
1196         struct css_set __rcu            *cgroups;
1197         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1198         struct list_head                cg_list;
1199 #endif
1200 #ifdef CONFIG_X86_CPU_RESCTRL
1201         u32                             closid;
1202         u32                             rmid;
1203 #endif
1204 #ifdef CONFIG_FUTEX
1205         struct robust_list_head __user  *robust_list;
1206 #ifdef CONFIG_COMPAT
1207         struct compat_robust_list_head __user *compat_robust_list;
1208 #endif
1209         struct list_head                pi_state_list;
1210         struct futex_pi_state           *pi_state_cache;
1211         struct mutex                    futex_exit_mutex;
1212         unsigned int                    futex_state;
1213 #endif
1214 #ifdef CONFIG_PERF_EVENTS
1215         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1216         struct mutex                    perf_event_mutex;
1217         struct list_head                perf_event_list;
1218 #endif
1219 #ifdef CONFIG_DEBUG_PREEMPT
1220         unsigned long                   preempt_disable_ip;
1221 #endif
1222 #ifdef CONFIG_NUMA
1223         /* Protected by alloc_lock: */
1224         struct mempolicy                *mempolicy;
1225         short                           il_prev;
1226         short                           pref_node_fork;
1227 #endif
1228 #ifdef CONFIG_NUMA_BALANCING
1229         int                             numa_scan_seq;
1230         unsigned int                    numa_scan_period;
1231         unsigned int                    numa_scan_period_max;
1232         int                             numa_preferred_nid;
1233         unsigned long                   numa_migrate_retry;
1234         /* Migration stamp: */
1235         u64                             node_stamp;
1236         u64                             last_task_numa_placement;
1237         u64                             last_sum_exec_runtime;
1238         struct callback_head            numa_work;
1239
1240         /*
1241          * This pointer is only modified for current in syscall and
1242          * pagefault context (and for tasks being destroyed), so it can be read
1243          * from any of the following contexts:
1244          *  - RCU read-side critical section
1245          *  - current->numa_group from everywhere
1246          *  - task's runqueue locked, task not running
1247          */
1248         struct numa_group __rcu         *numa_group;
1249
1250         /*
1251          * numa_faults is an array split into four regions:
1252          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1253          * in this precise order.
1254          *
1255          * faults_memory: Exponential decaying average of faults on a per-node
1256          * basis. Scheduling placement decisions are made based on these
1257          * counts. The values remain static for the duration of a PTE scan.
1258          * faults_cpu: Track the nodes the process was running on when a NUMA
1259          * hinting fault was incurred.
1260          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1261          * during the current scan window. When the scan completes, the counts
1262          * in faults_memory and faults_cpu decay and these values are copied.
1263          */
1264         unsigned long                   *numa_faults;
1265         unsigned long                   total_numa_faults;
1266
1267         /*
1268          * numa_faults_locality tracks if faults recorded during the last
1269          * scan window were remote/local or failed to migrate. The task scan
1270          * period is adapted based on the locality of the faults with different
1271          * weights depending on whether they were shared or private faults
1272          */
1273         unsigned long                   numa_faults_locality[3];
1274
1275         unsigned long                   numa_pages_migrated;
1276 #endif /* CONFIG_NUMA_BALANCING */
1277
1278 #ifdef CONFIG_RSEQ
1279         struct rseq __user *rseq;
1280         u32 rseq_sig;
1281         /*
1282          * RmW on rseq_event_mask must be performed atomically
1283          * with respect to preemption.
1284          */
1285         unsigned long rseq_event_mask;
1286 #endif
1287
1288         struct tlbflush_unmap_batch     tlb_ubc;
1289
1290         union {
1291                 refcount_t              rcu_users;
1292                 struct rcu_head         rcu;
1293         };
1294
1295         /* Cache last used pipe for splice(): */
1296         struct pipe_inode_info          *splice_pipe;
1297
1298         struct page_frag                task_frag;
1299
1300 #ifdef CONFIG_TASK_DELAY_ACCT
1301         struct task_delay_info          *delays;
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305         int                             make_it_fail;
1306         unsigned int                    fail_nth;
1307 #endif
1308         /*
1309          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1310          * balance_dirty_pages() for a dirty throttling pause:
1311          */
1312         int                             nr_dirtied;
1313         int                             nr_dirtied_pause;
1314         /* Start of a write-and-pause period: */
1315         unsigned long                   dirty_paused_when;
1316
1317 #ifdef CONFIG_LATENCYTOP
1318         int                             latency_record_count;
1319         struct latency_record           latency_record[LT_SAVECOUNT];
1320 #endif
1321         /*
1322          * Time slack values; these are used to round up poll() and
1323          * select() etc timeout values. These are in nanoseconds.
1324          */
1325         u64                             timer_slack_ns;
1326         u64                             default_timer_slack_ns;
1327
1328 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1329         unsigned int                    kasan_depth;
1330 #endif
1331
1332 #ifdef CONFIG_KCSAN
1333         struct kcsan_ctx                kcsan_ctx;
1334 #ifdef CONFIG_TRACE_IRQFLAGS
1335         struct irqtrace_events          kcsan_save_irqtrace;
1336 #endif
1337 #endif
1338
1339 #if IS_ENABLED(CONFIG_KUNIT)
1340         struct kunit                    *kunit_test;
1341 #endif
1342
1343 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1344         /* Index of current stored address in ret_stack: */
1345         int                             curr_ret_stack;
1346         int                             curr_ret_depth;
1347
1348         /* Stack of return addresses for return function tracing: */
1349         struct ftrace_ret_stack         *ret_stack;
1350
1351         /* Timestamp for last schedule: */
1352         unsigned long long              ftrace_timestamp;
1353
1354         /*
1355          * Number of functions that haven't been traced
1356          * because of depth overrun:
1357          */
1358         atomic_t                        trace_overrun;
1359
1360         /* Pause tracing: */
1361         atomic_t                        tracing_graph_pause;
1362 #endif
1363
1364 #ifdef CONFIG_TRACING
1365         /* State flags for use by tracers: */
1366         unsigned long                   trace;
1367
1368         /* Bitmask and counter of trace recursion: */
1369         unsigned long                   trace_recursion;
1370 #endif /* CONFIG_TRACING */
1371
1372 #ifdef CONFIG_KCOV
1373         /* See kernel/kcov.c for more details. */
1374
1375         /* Coverage collection mode enabled for this task (0 if disabled): */
1376         unsigned int                    kcov_mode;
1377
1378         /* Size of the kcov_area: */
1379         unsigned int                    kcov_size;
1380
1381         /* Buffer for coverage collection: */
1382         void                            *kcov_area;
1383
1384         /* KCOV descriptor wired with this task or NULL: */
1385         struct kcov                     *kcov;
1386
1387         /* KCOV common handle for remote coverage collection: */
1388         u64                             kcov_handle;
1389
1390         /* KCOV sequence number: */
1391         int                             kcov_sequence;
1392
1393         /* Collect coverage from softirq context: */
1394         unsigned int                    kcov_softirq;
1395 #endif
1396
1397 #ifdef CONFIG_MEMCG
1398         struct mem_cgroup               *memcg_in_oom;
1399         gfp_t                           memcg_oom_gfp_mask;
1400         int                             memcg_oom_order;
1401
1402         /* Number of pages to reclaim on returning to userland: */
1403         unsigned int                    memcg_nr_pages_over_high;
1404
1405         /* Used by memcontrol for targeted memcg charge: */
1406         struct mem_cgroup               *active_memcg;
1407 #endif
1408
1409 #ifdef CONFIG_BLK_CGROUP
1410         struct request_queue            *throttle_queue;
1411 #endif
1412
1413 #ifdef CONFIG_UPROBES
1414         struct uprobe_task              *utask;
1415 #endif
1416 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1417         unsigned int                    sequential_io;
1418         unsigned int                    sequential_io_avg;
1419 #endif
1420         struct kmap_ctrl                kmap_ctrl;
1421 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1422         unsigned long                   task_state_change;
1423 # ifdef CONFIG_PREEMPT_RT
1424         unsigned long                   saved_state_change;
1425 # endif
1426 #endif
1427         int                             pagefault_disabled;
1428 #ifdef CONFIG_MMU
1429         struct task_struct              *oom_reaper_list;
1430 #endif
1431 #ifdef CONFIG_VMAP_STACK
1432         struct vm_struct                *stack_vm_area;
1433 #endif
1434 #ifdef CONFIG_THREAD_INFO_IN_TASK
1435         /* A live task holds one reference: */
1436         refcount_t                      stack_refcount;
1437 #endif
1438 #ifdef CONFIG_LIVEPATCH
1439         int patch_state;
1440 #endif
1441 #ifdef CONFIG_SECURITY
1442         /* Used by LSM modules for access restriction: */
1443         void                            *security;
1444 #endif
1445 #ifdef CONFIG_BPF_SYSCALL
1446         /* Used by BPF task local storage */
1447         struct bpf_local_storage __rcu  *bpf_storage;
1448 #endif
1449
1450 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1451         unsigned long                   lowest_stack;
1452         unsigned long                   prev_lowest_stack;
1453 #endif
1454
1455 #ifdef CONFIG_X86_MCE
1456         void __user                     *mce_vaddr;
1457         __u64                           mce_kflags;
1458         u64                             mce_addr;
1459         __u64                           mce_ripv : 1,
1460                                         mce_whole_page : 1,
1461                                         __mce_reserved : 62;
1462         struct callback_head            mce_kill_me;
1463 #endif
1464
1465 #ifdef CONFIG_KRETPROBES
1466         struct llist_head               kretprobe_instances;
1467 #endif
1468
1469         /*
1470          * New fields for task_struct should be added above here, so that
1471          * they are included in the randomized portion of task_struct.
1472          */
1473         randomized_struct_fields_end
1474
1475         /* CPU-specific state of this task: */
1476         struct thread_struct            thread;
1477
1478         /*
1479          * WARNING: on x86, 'thread_struct' contains a variable-sized
1480          * structure.  It *MUST* be at the end of 'task_struct'.
1481          *
1482          * Do not put anything below here!
1483          */
1484 };
1485
1486 static inline struct pid *task_pid(struct task_struct *task)
1487 {
1488         return task->thread_pid;
1489 }
1490
1491 /*
1492  * the helpers to get the task's different pids as they are seen
1493  * from various namespaces
1494  *
1495  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1496  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1497  *                     current.
1498  * task_xid_nr_ns()  : id seen from the ns specified;
1499  *
1500  * see also pid_nr() etc in include/linux/pid.h
1501  */
1502 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1503
1504 static inline pid_t task_pid_nr(struct task_struct *tsk)
1505 {
1506         return tsk->pid;
1507 }
1508
1509 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1510 {
1511         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1512 }
1513
1514 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1515 {
1516         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1517 }
1518
1519
1520 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1521 {
1522         return tsk->tgid;
1523 }
1524
1525 /**
1526  * pid_alive - check that a task structure is not stale
1527  * @p: Task structure to be checked.
1528  *
1529  * Test if a process is not yet dead (at most zombie state)
1530  * If pid_alive fails, then pointers within the task structure
1531  * can be stale and must not be dereferenced.
1532  *
1533  * Return: 1 if the process is alive. 0 otherwise.
1534  */
1535 static inline int pid_alive(const struct task_struct *p)
1536 {
1537         return p->thread_pid != NULL;
1538 }
1539
1540 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1541 {
1542         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1543 }
1544
1545 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1546 {
1547         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1548 }
1549
1550
1551 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1552 {
1553         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1554 }
1555
1556 static inline pid_t task_session_vnr(struct task_struct *tsk)
1557 {
1558         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1559 }
1560
1561 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1562 {
1563         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1564 }
1565
1566 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1567 {
1568         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1569 }
1570
1571 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1572 {
1573         pid_t pid = 0;
1574
1575         rcu_read_lock();
1576         if (pid_alive(tsk))
1577                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1578         rcu_read_unlock();
1579
1580         return pid;
1581 }
1582
1583 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1584 {
1585         return task_ppid_nr_ns(tsk, &init_pid_ns);
1586 }
1587
1588 /* Obsolete, do not use: */
1589 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1590 {
1591         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1592 }
1593
1594 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1595 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1596
1597 static inline unsigned int task_state_index(struct task_struct *tsk)
1598 {
1599         unsigned int tsk_state = READ_ONCE(tsk->__state);
1600         unsigned int state = (tsk_state | tsk->exit_state) & TASK_REPORT;
1601
1602         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1603
1604         if (tsk_state == TASK_IDLE)
1605                 state = TASK_REPORT_IDLE;
1606
1607         return fls(state);
1608 }
1609
1610 static inline char task_index_to_char(unsigned int state)
1611 {
1612         static const char state_char[] = "RSDTtXZPI";
1613
1614         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1615
1616         return state_char[state];
1617 }
1618
1619 static inline char task_state_to_char(struct task_struct *tsk)
1620 {
1621         return task_index_to_char(task_state_index(tsk));
1622 }
1623
1624 /**
1625  * is_global_init - check if a task structure is init. Since init
1626  * is free to have sub-threads we need to check tgid.
1627  * @tsk: Task structure to be checked.
1628  *
1629  * Check if a task structure is the first user space task the kernel created.
1630  *
1631  * Return: 1 if the task structure is init. 0 otherwise.
1632  */
1633 static inline int is_global_init(struct task_struct *tsk)
1634 {
1635         return task_tgid_nr(tsk) == 1;
1636 }
1637
1638 extern struct pid *cad_pid;
1639
1640 /*
1641  * Per process flags
1642  */
1643 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1644 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1645 #define PF_EXITING              0x00000004      /* Getting shut down */
1646 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1647 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1648 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1649 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1650 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1651 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1652 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1653 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1654 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1655 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1656 #define PF_USED_ASYNC           0x00004000      /* Used async_schedule*(), used by module init */
1657 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1658 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1659 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1660 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1661 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1662 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1663                                                  * I am cleaning dirty pages from some other bdi. */
1664 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1665 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1666 #define PF_SWAPWRITE            0x00800000      /* Allowed to write to swap */
1667 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1668 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1669 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1670 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1671 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1672
1673 /*
1674  * Only the _current_ task can read/write to tsk->flags, but other
1675  * tasks can access tsk->flags in readonly mode for example
1676  * with tsk_used_math (like during threaded core dumping).
1677  * There is however an exception to this rule during ptrace
1678  * or during fork: the ptracer task is allowed to write to the
1679  * child->flags of its traced child (same goes for fork, the parent
1680  * can write to the child->flags), because we're guaranteed the
1681  * child is not running and in turn not changing child->flags
1682  * at the same time the parent does it.
1683  */
1684 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1685 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1686 #define clear_used_math()                       clear_stopped_child_used_math(current)
1687 #define set_used_math()                         set_stopped_child_used_math(current)
1688
1689 #define conditional_stopped_child_used_math(condition, child) \
1690         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1691
1692 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1693
1694 #define copy_to_stopped_child_used_math(child) \
1695         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1696
1697 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1698 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1699 #define used_math()                             tsk_used_math(current)
1700
1701 static inline bool is_percpu_thread(void)
1702 {
1703 #ifdef CONFIG_SMP
1704         return (current->flags & PF_NO_SETAFFINITY) &&
1705                 (current->nr_cpus_allowed  == 1);
1706 #else
1707         return true;
1708 #endif
1709 }
1710
1711 /* Per-process atomic flags. */
1712 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1713 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1714 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1715 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1716 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1717 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1718 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1719 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1720
1721 #define TASK_PFA_TEST(name, func)                                       \
1722         static inline bool task_##func(struct task_struct *p)           \
1723         { return test_bit(PFA_##name, &p->atomic_flags); }
1724
1725 #define TASK_PFA_SET(name, func)                                        \
1726         static inline void task_set_##func(struct task_struct *p)       \
1727         { set_bit(PFA_##name, &p->atomic_flags); }
1728
1729 #define TASK_PFA_CLEAR(name, func)                                      \
1730         static inline void task_clear_##func(struct task_struct *p)     \
1731         { clear_bit(PFA_##name, &p->atomic_flags); }
1732
1733 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1734 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1735
1736 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1737 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1738 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1739
1740 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1741 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1742 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1743
1744 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1745 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1746 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1747
1748 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1749 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1750 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1751
1752 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1753 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1754
1755 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1756 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1757 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1758
1759 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1760 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1761
1762 static inline void
1763 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1764 {
1765         current->flags &= ~flags;
1766         current->flags |= orig_flags & flags;
1767 }
1768
1769 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1770 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1771 #ifdef CONFIG_SMP
1772 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1773 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1774 #else
1775 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1776 {
1777 }
1778 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1779 {
1780         if (!cpumask_test_cpu(0, new_mask))
1781                 return -EINVAL;
1782         return 0;
1783 }
1784 #endif
1785
1786 extern int yield_to(struct task_struct *p, bool preempt);
1787 extern void set_user_nice(struct task_struct *p, long nice);
1788 extern int task_prio(const struct task_struct *p);
1789
1790 /**
1791  * task_nice - return the nice value of a given task.
1792  * @p: the task in question.
1793  *
1794  * Return: The nice value [ -20 ... 0 ... 19 ].
1795  */
1796 static inline int task_nice(const struct task_struct *p)
1797 {
1798         return PRIO_TO_NICE((p)->static_prio);
1799 }
1800
1801 extern int can_nice(const struct task_struct *p, const int nice);
1802 extern int task_curr(const struct task_struct *p);
1803 extern int idle_cpu(int cpu);
1804 extern int available_idle_cpu(int cpu);
1805 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1806 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1807 extern void sched_set_fifo(struct task_struct *p);
1808 extern void sched_set_fifo_low(struct task_struct *p);
1809 extern void sched_set_normal(struct task_struct *p, int nice);
1810 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1811 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1812 extern struct task_struct *idle_task(int cpu);
1813
1814 /**
1815  * is_idle_task - is the specified task an idle task?
1816  * @p: the task in question.
1817  *
1818  * Return: 1 if @p is an idle task. 0 otherwise.
1819  */
1820 static __always_inline bool is_idle_task(const struct task_struct *p)
1821 {
1822         return !!(p->flags & PF_IDLE);
1823 }
1824
1825 extern struct task_struct *curr_task(int cpu);
1826 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1827
1828 void yield(void);
1829
1830 union thread_union {
1831 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1832         struct task_struct task;
1833 #endif
1834 #ifndef CONFIG_THREAD_INFO_IN_TASK
1835         struct thread_info thread_info;
1836 #endif
1837         unsigned long stack[THREAD_SIZE/sizeof(long)];
1838 };
1839
1840 #ifndef CONFIG_THREAD_INFO_IN_TASK
1841 extern struct thread_info init_thread_info;
1842 #endif
1843
1844 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1845
1846 #ifdef CONFIG_THREAD_INFO_IN_TASK
1847 static inline struct thread_info *task_thread_info(struct task_struct *task)
1848 {
1849         return &task->thread_info;
1850 }
1851 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1852 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1853 #endif
1854
1855 /*
1856  * find a task by one of its numerical ids
1857  *
1858  * find_task_by_pid_ns():
1859  *      finds a task by its pid in the specified namespace
1860  * find_task_by_vpid():
1861  *      finds a task by its virtual pid
1862  *
1863  * see also find_vpid() etc in include/linux/pid.h
1864  */
1865
1866 extern struct task_struct *find_task_by_vpid(pid_t nr);
1867 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1868
1869 /*
1870  * find a task by its virtual pid and get the task struct
1871  */
1872 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1873
1874 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1875 extern int wake_up_process(struct task_struct *tsk);
1876 extern void wake_up_new_task(struct task_struct *tsk);
1877
1878 #ifdef CONFIG_SMP
1879 extern void kick_process(struct task_struct *tsk);
1880 #else
1881 static inline void kick_process(struct task_struct *tsk) { }
1882 #endif
1883
1884 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1885
1886 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1887 {
1888         __set_task_comm(tsk, from, false);
1889 }
1890
1891 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1892 #define get_task_comm(buf, tsk) ({                      \
1893         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1894         __get_task_comm(buf, sizeof(buf), tsk);         \
1895 })
1896
1897 #ifdef CONFIG_SMP
1898 static __always_inline void scheduler_ipi(void)
1899 {
1900         /*
1901          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1902          * TIF_NEED_RESCHED remotely (for the first time) will also send
1903          * this IPI.
1904          */
1905         preempt_fold_need_resched();
1906 }
1907 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1908 #else
1909 static inline void scheduler_ipi(void) { }
1910 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1911 {
1912         return 1;
1913 }
1914 #endif
1915
1916 /*
1917  * Set thread flags in other task's structures.
1918  * See asm/thread_info.h for TIF_xxxx flags available:
1919  */
1920 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1921 {
1922         set_ti_thread_flag(task_thread_info(tsk), flag);
1923 }
1924
1925 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1926 {
1927         clear_ti_thread_flag(task_thread_info(tsk), flag);
1928 }
1929
1930 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1931                                           bool value)
1932 {
1933         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1934 }
1935
1936 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1937 {
1938         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1939 }
1940
1941 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1942 {
1943         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1944 }
1945
1946 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1947 {
1948         return test_ti_thread_flag(task_thread_info(tsk), flag);
1949 }
1950
1951 static inline void set_tsk_need_resched(struct task_struct *tsk)
1952 {
1953         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1954 }
1955
1956 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1957 {
1958         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1959 }
1960
1961 static inline int test_tsk_need_resched(struct task_struct *tsk)
1962 {
1963         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
1964 }
1965
1966 /*
1967  * cond_resched() and cond_resched_lock(): latency reduction via
1968  * explicit rescheduling in places that are safe. The return
1969  * value indicates whether a reschedule was done in fact.
1970  * cond_resched_lock() will drop the spinlock before scheduling,
1971  */
1972 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
1973 extern int __cond_resched(void);
1974
1975 #ifdef CONFIG_PREEMPT_DYNAMIC
1976
1977 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
1978
1979 static __always_inline int _cond_resched(void)
1980 {
1981         return static_call_mod(cond_resched)();
1982 }
1983
1984 #else
1985
1986 static inline int _cond_resched(void)
1987 {
1988         return __cond_resched();
1989 }
1990
1991 #endif /* CONFIG_PREEMPT_DYNAMIC */
1992
1993 #else
1994
1995 static inline int _cond_resched(void) { return 0; }
1996
1997 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
1998
1999 #define cond_resched() ({                       \
2000         ___might_sleep(__FILE__, __LINE__, 0);  \
2001         _cond_resched();                        \
2002 })
2003
2004 extern int __cond_resched_lock(spinlock_t *lock);
2005 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2006 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2007
2008 #define cond_resched_lock(lock) ({                              \
2009         ___might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET);\
2010         __cond_resched_lock(lock);                              \
2011 })
2012
2013 #define cond_resched_rwlock_read(lock) ({                       \
2014         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2015         __cond_resched_rwlock_read(lock);                       \
2016 })
2017
2018 #define cond_resched_rwlock_write(lock) ({                      \
2019         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2020         __cond_resched_rwlock_write(lock);                      \
2021 })
2022
2023 static inline void cond_resched_rcu(void)
2024 {
2025 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2026         rcu_read_unlock();
2027         cond_resched();
2028         rcu_read_lock();
2029 #endif
2030 }
2031
2032 /*
2033  * Does a critical section need to be broken due to another
2034  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2035  * but a general need for low latency)
2036  */
2037 static inline int spin_needbreak(spinlock_t *lock)
2038 {
2039 #ifdef CONFIG_PREEMPTION
2040         return spin_is_contended(lock);
2041 #else
2042         return 0;
2043 #endif
2044 }
2045
2046 /*
2047  * Check if a rwlock is contended.
2048  * Returns non-zero if there is another task waiting on the rwlock.
2049  * Returns zero if the lock is not contended or the system / underlying
2050  * rwlock implementation does not support contention detection.
2051  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2052  * for low latency.
2053  */
2054 static inline int rwlock_needbreak(rwlock_t *lock)
2055 {
2056 #ifdef CONFIG_PREEMPTION
2057         return rwlock_is_contended(lock);
2058 #else
2059         return 0;
2060 #endif
2061 }
2062
2063 static __always_inline bool need_resched(void)
2064 {
2065         return unlikely(tif_need_resched());
2066 }
2067
2068 /*
2069  * Wrappers for p->thread_info->cpu access. No-op on UP.
2070  */
2071 #ifdef CONFIG_SMP
2072
2073 static inline unsigned int task_cpu(const struct task_struct *p)
2074 {
2075 #ifdef CONFIG_THREAD_INFO_IN_TASK
2076         return READ_ONCE(p->cpu);
2077 #else
2078         return READ_ONCE(task_thread_info(p)->cpu);
2079 #endif
2080 }
2081
2082 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2083
2084 #else
2085
2086 static inline unsigned int task_cpu(const struct task_struct *p)
2087 {
2088         return 0;
2089 }
2090
2091 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2092 {
2093 }
2094
2095 #endif /* CONFIG_SMP */
2096
2097 extern bool sched_task_on_rq(struct task_struct *p);
2098
2099 /*
2100  * In order to reduce various lock holder preemption latencies provide an
2101  * interface to see if a vCPU is currently running or not.
2102  *
2103  * This allows us to terminate optimistic spin loops and block, analogous to
2104  * the native optimistic spin heuristic of testing if the lock owner task is
2105  * running or not.
2106  */
2107 #ifndef vcpu_is_preempted
2108 static inline bool vcpu_is_preempted(int cpu)
2109 {
2110         return false;
2111 }
2112 #endif
2113
2114 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2115 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2116
2117 #ifndef TASK_SIZE_OF
2118 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2119 #endif
2120
2121 #ifdef CONFIG_SMP
2122 /* Returns effective CPU energy utilization, as seen by the scheduler */
2123 unsigned long sched_cpu_util(int cpu, unsigned long max);
2124 #endif /* CONFIG_SMP */
2125
2126 #ifdef CONFIG_RSEQ
2127
2128 /*
2129  * Map the event mask on the user-space ABI enum rseq_cs_flags
2130  * for direct mask checks.
2131  */
2132 enum rseq_event_mask_bits {
2133         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2134         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2135         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2136 };
2137
2138 enum rseq_event_mask {
2139         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2140         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2141         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2142 };
2143
2144 static inline void rseq_set_notify_resume(struct task_struct *t)
2145 {
2146         if (t->rseq)
2147                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2148 }
2149
2150 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2151
2152 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2153                                              struct pt_regs *regs)
2154 {
2155         if (current->rseq)
2156                 __rseq_handle_notify_resume(ksig, regs);
2157 }
2158
2159 static inline void rseq_signal_deliver(struct ksignal *ksig,
2160                                        struct pt_regs *regs)
2161 {
2162         preempt_disable();
2163         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2164         preempt_enable();
2165         rseq_handle_notify_resume(ksig, regs);
2166 }
2167
2168 /* rseq_preempt() requires preemption to be disabled. */
2169 static inline void rseq_preempt(struct task_struct *t)
2170 {
2171         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2172         rseq_set_notify_resume(t);
2173 }
2174
2175 /* rseq_migrate() requires preemption to be disabled. */
2176 static inline void rseq_migrate(struct task_struct *t)
2177 {
2178         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2179         rseq_set_notify_resume(t);
2180 }
2181
2182 /*
2183  * If parent process has a registered restartable sequences area, the
2184  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2185  */
2186 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2187 {
2188         if (clone_flags & CLONE_VM) {
2189                 t->rseq = NULL;
2190                 t->rseq_sig = 0;
2191                 t->rseq_event_mask = 0;
2192         } else {
2193                 t->rseq = current->rseq;
2194                 t->rseq_sig = current->rseq_sig;
2195                 t->rseq_event_mask = current->rseq_event_mask;
2196         }
2197 }
2198
2199 static inline void rseq_execve(struct task_struct *t)
2200 {
2201         t->rseq = NULL;
2202         t->rseq_sig = 0;
2203         t->rseq_event_mask = 0;
2204 }
2205
2206 #else
2207
2208 static inline void rseq_set_notify_resume(struct task_struct *t)
2209 {
2210 }
2211 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2212                                              struct pt_regs *regs)
2213 {
2214 }
2215 static inline void rseq_signal_deliver(struct ksignal *ksig,
2216                                        struct pt_regs *regs)
2217 {
2218 }
2219 static inline void rseq_preempt(struct task_struct *t)
2220 {
2221 }
2222 static inline void rseq_migrate(struct task_struct *t)
2223 {
2224 }
2225 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2226 {
2227 }
2228 static inline void rseq_execve(struct task_struct *t)
2229 {
2230 }
2231
2232 #endif
2233
2234 #ifdef CONFIG_DEBUG_RSEQ
2235
2236 void rseq_syscall(struct pt_regs *regs);
2237
2238 #else
2239
2240 static inline void rseq_syscall(struct pt_regs *regs)
2241 {
2242 }
2243
2244 #endif
2245
2246 const struct sched_avg *sched_trace_cfs_rq_avg(struct cfs_rq *cfs_rq);
2247 char *sched_trace_cfs_rq_path(struct cfs_rq *cfs_rq, char *str, int len);
2248 int sched_trace_cfs_rq_cpu(struct cfs_rq *cfs_rq);
2249
2250 const struct sched_avg *sched_trace_rq_avg_rt(struct rq *rq);
2251 const struct sched_avg *sched_trace_rq_avg_dl(struct rq *rq);
2252 const struct sched_avg *sched_trace_rq_avg_irq(struct rq *rq);
2253
2254 int sched_trace_rq_cpu(struct rq *rq);
2255 int sched_trace_rq_cpu_capacity(struct rq *rq);
2256 int sched_trace_rq_nr_running(struct rq *rq);
2257
2258 const struct cpumask *sched_trace_rd_span(struct root_domain *rd);
2259
2260 #ifdef CONFIG_SCHED_CORE
2261 extern void sched_core_free(struct task_struct *tsk);
2262 extern void sched_core_fork(struct task_struct *p);
2263 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2264                                 unsigned long uaddr);
2265 #else
2266 static inline void sched_core_free(struct task_struct *tsk) { }
2267 static inline void sched_core_fork(struct task_struct *p) { }
2268 #endif
2269
2270 #endif