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 #ifdef CONFIG_PREEMPT_RT
 292  extern void schedule_rtlock(void);
 293 #endif
 294
 295 extern int __must_check io_schedule_prepare(void);
 296 extern void io_schedule_finish(int token);
 297 extern long io_schedule_timeout(long timeout);
 298 extern void io_schedule(void);
 299
 300 /**
 301  * struct prev_cputime - snapshot of system and user cputime
 302  * @utime: time spent in user mode
 303  * @stime: time spent in system mode
 304  * @lock: protects the above two fields
 305  *
 306  * Stores previous user/system time values such that we can guarantee
 307  * monotonicity.
 308  */
 309 struct prev_cputime {
 310 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 311         u64                             utime;
 312         u64                             stime;
 313         raw_spinlock_t                  lock;
 314 #endif
 315 };
 316
 317 enum vtime_state {
 318         /* Task is sleeping or running in a CPU with VTIME inactive: */
 319         VTIME_INACTIVE = 0,
 320         /* Task is idle */
 321         VTIME_IDLE,
 322         /* Task runs in kernelspace in a CPU with VTIME active: */
 323         VTIME_SYS,
 324         /* Task runs in userspace in a CPU with VTIME active: */
 325         VTIME_USER,
 326         /* Task runs as guests in a CPU with VTIME active: */
 327         VTIME_GUEST,
 328 };
 329
 330 struct vtime {
 331         seqcount_t              seqcount;
 332         unsigned long long      starttime;
 333         enum vtime_state        state;
 334         unsigned int            cpu;
 335         u64                     utime;
 336         u64                     stime;
 337         u64                     gtime;
 338 };
 339
 340 /*
 341  * Utilization clamp constraints.
 342  * @UCLAMP_MIN: Minimum utilization
 343  * @UCLAMP_MAX: Maximum utilization
 344  * @UCLAMP_CNT: Utilization clamp constraints count
 345  */
 346 enum uclamp_id {
 347         UCLAMP_MIN = 0,
 348         UCLAMP_MAX,
 349         UCLAMP_CNT
 350 };
 351
 352 #ifdef CONFIG_SMP
 353 extern struct root_domain def_root_domain;
 354 extern struct mutex sched_domains_mutex;
 355 #endif
 356
 357 struct sched_info {
 358 #ifdef CONFIG_SCHED_INFO
 359         /* Cumulative counters: */
 360
 361         /* # of times we have run on this CPU: */
 362         unsigned long                   pcount;
 363
 364         /* Time spent waiting on a runqueue: */
 365         unsigned long long              run_delay;
 366
 367         /* Timestamps: */
 368
 369         /* When did we last run on a CPU? */
 370         unsigned long long              last_arrival;
 371
 372         /* When were we last queued to run? */
 373         unsigned long long              last_queued;
 374
 375 #endif /* CONFIG_SCHED_INFO */
 376 };
 377
 378 /*
 379  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 380  * has a few: load, load_avg, util_avg, freq, and capacity.
 381  *
 382  * We define a basic fixed point arithmetic range, and then formalize
 383  * all these metrics based on that basic range.
 384  */
 385 # define SCHED_FIXEDPOINT_SHIFT         10
 386 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 387
 388 /* Increase resolution of cpu_capacity calculations */
 389 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 390 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 391
 392 struct load_weight {
 393         unsigned long                   weight;
 394         u32                             inv_weight;
 395 };
 396
 397 /**
 398  * struct util_est - Estimation utilization of FAIR tasks
 399  * @enqueued: instantaneous estimated utilization of a task/cpu
 400  * @ewma:     the Exponential Weighted Moving Average (EWMA)
 401  *            utilization of a task
 402  *
 403  * Support data structure to track an Exponential Weighted Moving Average
 404  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
 405  * average each time a task completes an activation. Sample's weight is chosen
 406  * so that the EWMA will be relatively insensitive to transient changes to the
 407  * task's workload.
 408  *
 409  * The enqueued attribute has a slightly different meaning for tasks and cpus:
 410  * - task:   the task's util_avg at last task dequeue time
 411  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
 412  * Thus, the util_est.enqueued of a task represents the contribution on the
 413  * estimated utilization of the CPU where that task is currently enqueued.
 414  *
 415  * Only for tasks we track a moving average of the past instantaneous
 416  * estimated utilization. This allows to absorb sporadic drops in utilization
 417  * of an otherwise almost periodic task.
 418  *
 419  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 420  * updates. When a task is dequeued, its util_est should not be updated if its
 421  * util_avg has not been updated in the meantime.
 422  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
 423  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
 424  * for a task) it is safe to use MSB.
 425  */
 426 struct util_est {
 427         unsigned int                    enqueued;
 428         unsigned int                    ewma;
 429 #define UTIL_EST_WEIGHT_SHIFT           2
 430 #define UTIL_AVG_UNCHANGED              0x80000000
 431 } __attribute__((__aligned__(sizeof(u64))));
 432
 433 /*
 434  * The load/runnable/util_avg accumulates an infinite geometric series
 435  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 436  *
 437  * [load_avg definition]
 438  *
 439  *   load_avg = runnable% * scale_load_down(load)
 440  *
 441  * [runnable_avg definition]
 442  *
 443  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 444  *
 445  * [util_avg definition]
 446  *
 447  *   util_avg = running% * SCHED_CAPACITY_SCALE
 448  *
 449  * where runnable% is the time ratio that a sched_entity is runnable and
 450  * running% the time ratio that a sched_entity is running.
 451  *
 452  * For cfs_rq, they are the aggregated values of all runnable and blocked
 453  * sched_entities.
 454  *
 455  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 456  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 457  * for computing those signals (see update_rq_clock_pelt())
 458  *
 459  * N.B., the above ratios (runnable% and running%) themselves are in the
 460  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 461  * to as large a range as necessary. This is for example reflected by
 462  * util_avg's SCHED_CAPACITY_SCALE.
 463  *
 464  * [Overflow issue]
 465  *
 466  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 467  * with the highest load (=88761), always runnable on a single cfs_rq,
 468  * and should not overflow as the number already hits PID_MAX_LIMIT.
 469  *
 470  * For all other cases (including 32-bit kernels), struct load_weight's
 471  * weight will overflow first before we do, because:
 472  *
 473  *    Max(load_avg) <= Max(load.weight)
 474  *
 475  * Then it is the load_weight's responsibility to consider overflow
 476  * issues.
 477  */
 478 struct sched_avg {
 479         u64                             last_update_time;
 480         u64                             load_sum;
 481         u64                             runnable_sum;
 482         u32                             util_sum;
 483         u32                             period_contrib;
 484         unsigned long                   load_avg;
 485         unsigned long                   runnable_avg;
 486         unsigned long                   util_avg;
 487         struct util_est                 util_est;
 488 } ____cacheline_aligned;
 489
 490 struct sched_statistics {
 491 #ifdef CONFIG_SCHEDSTATS
 492         u64                             wait_start;
 493         u64                             wait_max;
 494         u64                             wait_count;
 495         u64                             wait_sum;
 496         u64                             iowait_count;
 497         u64                             iowait_sum;
 498
 499         u64                             sleep_start;
 500         u64                             sleep_max;
 501         s64                             sum_sleep_runtime;
 502
 503         u64                             block_start;
 504         u64                             block_max;
 505         u64                             exec_max;
 506         u64                             slice_max;
 507
 508         u64                             nr_migrations_cold;
 509         u64                             nr_failed_migrations_affine;
 510         u64                             nr_failed_migrations_running;
 511         u64                             nr_failed_migrations_hot;
 512         u64                             nr_forced_migrations;
 513
 514         u64                             nr_wakeups;
 515         u64                             nr_wakeups_sync;
 516         u64                             nr_wakeups_migrate;
 517         u64                             nr_wakeups_local;
 518         u64                             nr_wakeups_remote;
 519         u64                             nr_wakeups_affine;
 520         u64                             nr_wakeups_affine_attempts;
 521         u64                             nr_wakeups_passive;
 522         u64                             nr_wakeups_idle;
 523 #endif
 524 };
 525
 526 struct sched_entity {
 527         /* For load-balancing: */
 528         struct load_weight              load;
 529         struct rb_node                  run_node;
 530         struct list_head                group_node;
 531         unsigned int                    on_rq;
 532
 533         u64                             exec_start;
 534         u64                             sum_exec_runtime;
 535         u64                             vruntime;
 536         u64                             prev_sum_exec_runtime;
 537
 538         u64                             nr_migrations;
 539
 540         struct sched_statistics         statistics;
 541
 542 #ifdef CONFIG_FAIR_GROUP_SCHED
 543         int                             depth;
 544         struct sched_entity             *parent;
 545         /* rq on which this entity is (to be) queued: */
 546         struct cfs_rq                   *cfs_rq;
 547         /* rq "owned" by this entity/group: */
 548         struct cfs_rq                   *my_q;
 549         /* cached value of my_q->h_nr_running */
 550         unsigned long                   runnable_weight;
 551 #endif
 552
 553 #ifdef CONFIG_SMP
 554         /*
 555          * Per entity load average tracking.
 556          *
 557          * Put into separate cache line so it does not
 558          * collide with read-mostly values above.
 559          */
 560         struct sched_avg                avg;
 561 #endif
 562 };
 563
 564 struct sched_rt_entity {
 565         struct list_head                run_list;
 566         unsigned long                   timeout;
 567         unsigned long                   watchdog_stamp;
 568         unsigned int                    time_slice;
 569         unsigned short                  on_rq;
 570         unsigned short                  on_list;
 571
 572         struct sched_rt_entity          *back;
 573 #ifdef CONFIG_RT_GROUP_SCHED
 574         struct sched_rt_entity          *parent;
 575         /* rq on which this entity is (to be) queued: */
 576         struct rt_rq                    *rt_rq;
 577         /* rq "owned" by this entity/group: */
 578         struct rt_rq                    *my_q;
 579 #endif
 580 } __randomize_layout;
 581
 582 struct sched_dl_entity {
 583         struct rb_node                  rb_node;
 584
 585         /*
 586          * Original scheduling parameters. Copied here from sched_attr
 587          * during sched_setattr(), they will remain the same until
 588          * the next sched_setattr().
 589          */
 590         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 591         u64                             dl_deadline;    /* Relative deadline of each instance   */
 592         u64                             dl_period;      /* Separation of two instances (period) */
 593         u64                             dl_bw;          /* dl_runtime / dl_period               */
 594         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 595
 596         /*
 597          * Actual scheduling parameters. Initialized with the values above,
 598          * they are continuously updated during task execution. Note that
 599          * the remaining runtime could be < 0 in case we are in overrun.
 600          */
 601         s64                             runtime;        /* Remaining runtime for this instance  */
 602         u64                             deadline;       /* Absolute deadline for this instance  */
 603         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 604
 605         /*
 606          * Some bool flags:
 607          *
 608          * @dl_throttled tells if we exhausted the runtime. If so, the
 609          * task has to wait for a replenishment to be performed at the
 610          * next firing of dl_timer.
 611          *
 612          * @dl_boosted tells if we are boosted due to DI. If so we are
 613          * outside bandwidth enforcement mechanism (but only until we
 614          * exit the critical section);
 615          *
 616          * @dl_yielded tells if task gave up the CPU before consuming
 617          * all its available runtime during the last job.
 618          *
 619          * @dl_non_contending tells if the task is inactive while still
 620          * contributing to the active utilization. In other words, it
 621          * indicates if the inactive timer has been armed and its handler
 622          * has not been executed yet. This flag is useful to avoid race
 623          * conditions between the inactive timer handler and the wakeup
 624          * code.
 625          *
 626          * @dl_overrun tells if the task asked to be informed about runtime
 627          * overruns.
 628          */
 629         unsigned int                    dl_throttled      : 1;
 630         unsigned int                    dl_yielded        : 1;
 631         unsigned int                    dl_non_contending : 1;
 632         unsigned int                    dl_overrun        : 1;
 633
 634         /*
 635          * Bandwidth enforcement timer. Each -deadline task has its
 636          * own bandwidth to be enforced, thus we need one timer per task.
 637          */
 638         struct hrtimer                  dl_timer;
 639
 640         /*
 641          * Inactive timer, responsible for decreasing the active utilization
 642          * at the "0-lag time". When a -deadline task blocks, it contributes
 643          * to GRUB's active utilization until the "0-lag time", hence a
 644          * timer is needed to decrease the active utilization at the correct
 645          * time.
 646          */
 647         struct hrtimer inactive_timer;
 648
 649 #ifdef CONFIG_RT_MUTEXES
 650         /*
 651          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 652          * pi_se points to the donor, otherwise points to the dl_se it belongs
 653          * to (the original one/itself).
 654          */
 655         struct sched_dl_entity *pi_se;
 656 #endif
 657 };
 658
 659 #ifdef CONFIG_UCLAMP_TASK
 660 /* Number of utilization clamp buckets (shorter alias) */
 661 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 662
 663 /*
 664  * Utilization clamp for a scheduling entity
 665  * @value:              clamp value "assigned" to a se
 666  * @bucket_id:          bucket index corresponding to the "assigned" value
 667  * @active:             the se is currently refcounted in a rq's bucket
 668  * @user_defined:       the requested clamp value comes from user-space
 669  *
 670  * The bucket_id is the index of the clamp bucket matching the clamp value
 671  * which is pre-computed and stored to avoid expensive integer divisions from
 672  * the fast path.
 673  *
 674  * The active bit is set whenever a task has got an "effective" value assigned,
 675  * which can be different from the clamp value "requested" from user-space.
 676  * This allows to know a task is refcounted in the rq's bucket corresponding
 677  * to the "effective" bucket_id.
 678  *
 679  * The user_defined bit is set whenever a task has got a task-specific clamp
 680  * value requested from userspace, i.e. the system defaults apply to this task
 681  * just as a restriction. This allows to relax default clamps when a less
 682  * restrictive task-specific value has been requested, thus allowing to
 683  * implement a "nice" semantic. For example, a task running with a 20%
 684  * default boost can still drop its own boosting to 0%.
 685  */
 686 struct uclamp_se {
 687         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 688         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 689         unsigned int active             : 1;
 690         unsigned int user_defined       : 1;
 691 };
 692 #endif /* CONFIG_UCLAMP_TASK */
 693
 694 union rcu_special {
 695         struct {
 696                 u8                      blocked;
 697                 u8                      need_qs;
 698                 u8                      exp_hint; /* Hint for performance. */
 699                 u8                      need_mb; /* Readers need smp_mb(). */
 700         } b; /* Bits. */
 701         u32 s; /* Set of bits. */
 702 };
 703
 704 enum perf_event_task_context {
 705         perf_invalid_context = -1,
 706         perf_hw_context = 0,
 707         perf_sw_context,
 708         perf_nr_task_contexts,
 709 };
 710
 711 struct wake_q_node {
 712         struct wake_q_node *next;
 713 };
 714
 715 struct kmap_ctrl {
 716 #ifdef CONFIG_KMAP_LOCAL
 717         int                             idx;
 718         pte_t                           pteval[KM_MAX_IDX];
 719 #endif
 720 };
 721
 722 struct task_struct {
 723 #ifdef CONFIG_THREAD_INFO_IN_TASK
 724         /*
 725          * For reasons of header soup (see current_thread_info()), this
 726          * must be the first element of task_struct.
 727          */
 728         struct thread_info              thread_info;
 729 #endif
 730         unsigned int                    __state;
 731
 732 #ifdef CONFIG_PREEMPT_RT
 733         /* saved state for "spinlock sleepers" */
 734         unsigned int                    saved_state;
 735 #endif
 736
 737         /*
 738          * This begins the randomizable portion of task_struct. Only
 739          * scheduling-critical items should be added above here.
 740          */
 741         randomized_struct_fields_start
 742
 743         void                            *stack;
 744         refcount_t                      usage;
 745         /* Per task flags (PF_*), defined further below: */
 746         unsigned int                    flags;
 747         unsigned int                    ptrace;
 748
 749 #ifdef CONFIG_SMP
 750         int                             on_cpu;
 751         struct __call_single_node       wake_entry;
 752 #ifdef CONFIG_THREAD_INFO_IN_TASK
 753         /* Current CPU: */
 754         unsigned int                    cpu;
 755 #endif
 756         unsigned int                    wakee_flips;
 757         unsigned long                   wakee_flip_decay_ts;
 758         struct task_struct              *last_wakee;
 759
 760         /*
 761          * recent_used_cpu is initially set as the last CPU used by a task
 762          * that wakes affine another task. Waker/wakee relationships can
 763          * push tasks around a CPU where each wakeup moves to the next one.
 764          * Tracking a recently used CPU allows a quick search for a recently
 765          * used CPU that may be idle.
 766          */
 767         int                             recent_used_cpu;
 768         int                             wake_cpu;
 769 #endif
 770         int                             on_rq;
 771
 772         int                             prio;
 773         int                             static_prio;
 774         int                             normal_prio;
 775         unsigned int                    rt_priority;
 776
 777         const struct sched_class        *sched_class;
 778         struct sched_entity             se;
 779         struct sched_rt_entity          rt;
 780         struct sched_dl_entity          dl;
 781
 782 #ifdef CONFIG_SCHED_CORE
 783         struct rb_node                  core_node;
 784         unsigned long                   core_cookie;
 785         unsigned int                    core_occupation;
 786 #endif
 787
 788 #ifdef CONFIG_CGROUP_SCHED
 789         struct task_group               *sched_task_group;
 790 #endif
 791
 792 #ifdef CONFIG_UCLAMP_TASK
 793         /*
 794          * Clamp values requested for a scheduling entity.
 795          * Must be updated with task_rq_lock() held.
 796          */
 797         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 798         /*
 799          * Effective clamp values used for a scheduling entity.
 800          * Must be updated with task_rq_lock() held.
 801          */
 802         struct uclamp_se                uclamp[UCLAMP_CNT];
 803 #endif
 804
 805 #ifdef CONFIG_PREEMPT_NOTIFIERS
 806         /* List of struct preempt_notifier: */
 807         struct hlist_head               preempt_notifiers;
 808 #endif
 809
 810 #ifdef CONFIG_BLK_DEV_IO_TRACE
 811         unsigned int                    btrace_seq;
 812 #endif
 813
 814         unsigned int                    policy;
 815         int                             nr_cpus_allowed;
 816         const cpumask_t                 *cpus_ptr;
 817         cpumask_t                       *user_cpus_ptr;
 818         cpumask_t                       cpus_mask;
 819         void                            *migration_pending;
 820 #ifdef CONFIG_SMP
 821         unsigned short                  migration_disabled;
 822 #endif
 823         unsigned short                  migration_flags;
 824
 825 #ifdef CONFIG_PREEMPT_RCU
 826         int                             rcu_read_lock_nesting;
 827         union rcu_special               rcu_read_unlock_special;
 828         struct list_head                rcu_node_entry;
 829         struct rcu_node                 *rcu_blocked_node;
 830 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 831
 832 #ifdef CONFIG_TASKS_RCU
 833         unsigned long                   rcu_tasks_nvcsw;
 834         u8                              rcu_tasks_holdout;
 835         u8                              rcu_tasks_idx;
 836         int                             rcu_tasks_idle_cpu;
 837         struct list_head                rcu_tasks_holdout_list;
 838 #endif /* #ifdef CONFIG_TASKS_RCU */
 839
 840 #ifdef CONFIG_TASKS_TRACE_RCU
 841         int                             trc_reader_nesting;
 842         int                             trc_ipi_to_cpu;
 843         union rcu_special               trc_reader_special;
 844         bool                            trc_reader_checked;
 845         struct list_head                trc_holdout_list;
 846 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 847
 848         struct sched_info               sched_info;
 849
 850         struct list_head                tasks;
 851 #ifdef CONFIG_SMP
 852         struct plist_node               pushable_tasks;
 853         struct rb_node                  pushable_dl_tasks;
 854 #endif
 855
 856         struct mm_struct                *mm;
 857         struct mm_struct                *active_mm;
 858
 859         /* Per-thread vma caching: */
 860         struct vmacache                 vmacache;
 861
 862 #ifdef SPLIT_RSS_COUNTING
 863         struct task_rss_stat            rss_stat;
 864 #endif
 865         int                             exit_state;
 866         int                             exit_code;
 867         int                             exit_signal;
 868         /* The signal sent when the parent dies: */
 869         int                             pdeath_signal;
 870         /* JOBCTL_*, siglock protected: */
 871         unsigned long                   jobctl;
 872
 873         /* Used for emulating ABI behavior of previous Linux versions: */
 874         unsigned int                    personality;
 875
 876         /* Scheduler bits, serialized by scheduler locks: */
 877         unsigned                        sched_reset_on_fork:1;
 878         unsigned                        sched_contributes_to_load:1;
 879         unsigned                        sched_migrated:1;
 880 #ifdef CONFIG_PSI
 881         unsigned                        sched_psi_wake_requeue:1;
 882 #endif
 883
 884         /* Force alignment to the next boundary: */
 885         unsigned                        :0;
 886
 887         /* Unserialized, strictly 'current' */
 888
 889         /*
 890          * This field must not be in the scheduler word above due to wakelist
 891          * queueing no longer being serialized by p->on_cpu. However:
 892          *
 893          * p->XXX = X;                  ttwu()
 894          * schedule()                     if (p->on_rq && ..) // false
 895          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 896          *   deactivate_task()                ttwu_queue_wakelist())
 897          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 898          *
 899          * guarantees all stores of 'current' are visible before
 900          * ->sched_remote_wakeup gets used, so it can be in this word.
 901          */
 902         unsigned                        sched_remote_wakeup:1;
 903
 904         /* Bit to tell LSMs we're in execve(): */
 905         unsigned                        in_execve:1;
 906         unsigned                        in_iowait:1;
 907 #ifndef TIF_RESTORE_SIGMASK
 908         unsigned                        restore_sigmask:1;
 909 #endif
 910 #ifdef CONFIG_MEMCG
 911         unsigned                        in_user_fault:1;
 912 #endif
 913 #ifdef CONFIG_COMPAT_BRK
 914         unsigned                        brk_randomized:1;
 915 #endif
 916 #ifdef CONFIG_CGROUPS
 917         /* disallow userland-initiated cgroup migration */
 918         unsigned                        no_cgroup_migration:1;
 919         /* task is frozen/stopped (used by the cgroup freezer) */
 920         unsigned                        frozen:1;
 921 #endif
 922 #ifdef CONFIG_BLK_CGROUP
 923         unsigned                        use_memdelay:1;
 924 #endif
 925 #ifdef CONFIG_PSI
 926         /* Stalled due to lack of memory */
 927         unsigned                        in_memstall:1;
 928 #endif
 929 #ifdef CONFIG_PAGE_OWNER
 930         /* Used by page_owner=on to detect recursion in page tracking. */
 931         unsigned                        in_page_owner:1;
 932 #endif
 933 #ifdef CONFIG_EVENTFD
 934         /* Recursion prevention for eventfd_signal() */
 935         unsigned                        in_eventfd_signal:1;
 936 #endif
 937
 938         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 939
 940         struct restart_block            restart_block;
 941
 942         pid_t                           pid;
 943         pid_t                           tgid;
 944
 945 #ifdef CONFIG_STACKPROTECTOR
 946         /* Canary value for the -fstack-protector GCC feature: */
 947         unsigned long                   stack_canary;
 948 #endif
 949         /*
 950          * Pointers to the (original) parent process, youngest child, younger sibling,
 951          * older sibling, respectively.  (p->father can be replaced with
 952          * p->real_parent->pid)
 953          */
 954
 955         /* Real parent process: */
 956         struct task_struct __rcu        *real_parent;
 957
 958         /* Recipient of SIGCHLD, wait4() reports: */
 959         struct task_struct __rcu        *parent;
 960
 961         /*
 962          * Children/sibling form the list of natural children:
 963          */
 964         struct list_head                children;
 965         struct list_head                sibling;
 966         struct task_struct              *group_leader;
 967
 968         /*
 969          * 'ptraced' is the list of tasks this task is using ptrace() on.
 970          *
 971          * This includes both natural children and PTRACE_ATTACH targets.
 972          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 973          */
 974         struct list_head                ptraced;
 975         struct list_head                ptrace_entry;
 976
 977         /* PID/PID hash table linkage. */
 978         struct pid                      *thread_pid;
 979         struct hlist_node               pid_links[PIDTYPE_MAX];
 980         struct list_head                thread_group;
 981         struct list_head                thread_node;
 982
 983         struct completion               *vfork_done;
 984
 985         /* CLONE_CHILD_SETTID: */
 986         int __user                      *set_child_tid;
 987
 988         /* CLONE_CHILD_CLEARTID: */
 989         int __user                      *clear_child_tid;
 990
 991         /* PF_IO_WORKER */
 992         void                            *pf_io_worker;
 993
 994         u64                             utime;
 995         u64                             stime;
 996 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 997         u64                             utimescaled;
 998         u64                             stimescaled;
 999 #endif
1000         u64                             gtime;
1001         struct prev_cputime             prev_cputime;
1002 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1003         struct vtime                    vtime;
1004 #endif
1005
1006 #ifdef CONFIG_NO_HZ_FULL
1007         atomic_t                        tick_dep_mask;
1008 #endif
1009         /* Context switch counts: */
1010         unsigned long                   nvcsw;
1011         unsigned long                   nivcsw;
1012
1013         /* Monotonic time in nsecs: */
1014         u64                             start_time;
1015
1016         /* Boot based time in nsecs: */
1017         u64                             start_boottime;
1018
1019         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1020         unsigned long                   min_flt;
1021         unsigned long                   maj_flt;
1022
1023         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1024         struct posix_cputimers          posix_cputimers;
1025
1026 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1027         struct posix_cputimers_work     posix_cputimers_work;
1028 #endif
1029
1030         /* Process credentials: */
1031
1032         /* Tracer's credentials at attach: */
1033         const struct cred __rcu         *ptracer_cred;
1034
1035         /* Objective and real subjective task credentials (COW): */
1036         const struct cred __rcu         *real_cred;
1037
1038         /* Effective (overridable) subjective task credentials (COW): */
1039         const struct cred __rcu         *cred;
1040
1041 #ifdef CONFIG_KEYS
1042         /* Cached requested key. */
1043         struct key                      *cached_requested_key;
1044 #endif
1045
1046         /*
1047          * executable name, excluding path.
1048          *
1049          * - normally initialized setup_new_exec()
1050          * - access it with [gs]et_task_comm()
1051          * - lock it with task_lock()
1052          */
1053         char                            comm[TASK_COMM_LEN];
1054
1055         struct nameidata                *nameidata;
1056
1057 #ifdef CONFIG_SYSVIPC
1058         struct sysv_sem                 sysvsem;
1059         struct sysv_shm                 sysvshm;
1060 #endif
1061 #ifdef CONFIG_DETECT_HUNG_TASK
1062         unsigned long                   last_switch_count;
1063         unsigned long                   last_switch_time;
1064 #endif
1065         /* Filesystem information: */
1066         struct fs_struct                *fs;
1067
1068         /* Open file information: */
1069         struct files_struct             *files;
1070
1071 #ifdef CONFIG_IO_URING
1072         struct io_uring_task            *io_uring;
1073 #endif
1074
1075         /* Namespaces: */
1076         struct nsproxy                  *nsproxy;
1077
1078         /* Signal handlers: */
1079         struct signal_struct            *signal;
1080         struct sighand_struct __rcu             *sighand;
1081         sigset_t                        blocked;
1082         sigset_t                        real_blocked;
1083         /* Restored if set_restore_sigmask() was used: */
1084         sigset_t                        saved_sigmask;
1085         struct sigpending               pending;
1086         unsigned long                   sas_ss_sp;
1087         size_t                          sas_ss_size;
1088         unsigned int                    sas_ss_flags;
1089
1090         struct callback_head            *task_works;
1091
1092 #ifdef CONFIG_AUDIT
1093 #ifdef CONFIG_AUDITSYSCALL
1094         struct audit_context            *audit_context;
1095 #endif
1096         kuid_t                          loginuid;
1097         unsigned int                    sessionid;
1098 #endif
1099         struct seccomp                  seccomp;
1100         struct syscall_user_dispatch    syscall_dispatch;
1101
1102         /* Thread group tracking: */
1103         u64                             parent_exec_id;
1104         u64                             self_exec_id;
1105
1106         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1107         spinlock_t                      alloc_lock;
1108
1109         /* Protection of the PI data structures: */
1110         raw_spinlock_t                  pi_lock;
1111
1112         struct wake_q_node              wake_q;
1113
1114 #ifdef CONFIG_RT_MUTEXES
1115         /* PI waiters blocked on a rt_mutex held by this task: */
1116         struct rb_root_cached           pi_waiters;
1117         /* Updated under owner's pi_lock and rq lock */
1118         struct task_struct              *pi_top_task;
1119         /* Deadlock detection and priority inheritance handling: */
1120         struct rt_mutex_waiter          *pi_blocked_on;
1121 #endif
1122
1123 #ifdef CONFIG_DEBUG_MUTEXES
1124         /* Mutex deadlock detection: */
1125         struct mutex_waiter             *blocked_on;
1126 #endif
1127
1128 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1129         int                             non_block_count;
1130 #endif
1131
1132 #ifdef CONFIG_TRACE_IRQFLAGS
1133         struct irqtrace_events          irqtrace;
1134         unsigned int                    hardirq_threaded;
1135         u64                             hardirq_chain_key;
1136         int                             softirqs_enabled;
1137         int                             softirq_context;
1138         int                             irq_config;
1139 #endif
1140 #ifdef CONFIG_PREEMPT_RT
1141         int                             softirq_disable_cnt;
1142 #endif
1143
1144 #ifdef CONFIG_LOCKDEP
1145 # define MAX_LOCK_DEPTH                 48UL
1146         u64                             curr_chain_key;
1147         int                             lockdep_depth;
1148         unsigned int                    lockdep_recursion;
1149         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1150 #endif
1151
1152 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1153         unsigned int                    in_ubsan;
1154 #endif
1155
1156         /* Journalling filesystem info: */
1157         void                            *journal_info;
1158
1159         /* Stacked block device info: */
1160         struct bio_list                 *bio_list;
1161
1162 #ifdef CONFIG_BLOCK
1163         /* Stack plugging: */
1164         struct blk_plug                 *plug;
1165 #endif
1166
1167         /* VM state: */
1168         struct reclaim_state            *reclaim_state;
1169
1170         struct backing_dev_info         *backing_dev_info;
1171
1172         struct io_context               *io_context;
1173
1174 #ifdef CONFIG_COMPACTION
1175         struct capture_control          *capture_control;
1176 #endif
1177         /* Ptrace state: */
1178         unsigned long                   ptrace_message;
1179         kernel_siginfo_t                *last_siginfo;
1180
1181         struct task_io_accounting       ioac;
1182 #ifdef CONFIG_PSI
1183         /* Pressure stall state */
1184         unsigned int                    psi_flags;
1185 #endif
1186 #ifdef CONFIG_TASK_XACCT
1187         /* Accumulated RSS usage: */
1188         u64                             acct_rss_mem1;
1189         /* Accumulated virtual memory usage: */
1190         u64                             acct_vm_mem1;
1191         /* stime + utime since last update: */
1192         u64                             acct_timexpd;
1193 #endif
1194 #ifdef CONFIG_CPUSETS
1195         /* Protected by ->alloc_lock: */
1196         nodemask_t                      mems_allowed;
1197         /* Sequence number to catch updates: */
1198         seqcount_spinlock_t             mems_allowed_seq;
1199         int                             cpuset_mem_spread_rotor;
1200         int                             cpuset_slab_spread_rotor;
1201 #endif
1202 #ifdef CONFIG_CGROUPS
1203         /* Control Group info protected by css_set_lock: */
1204         struct css_set __rcu            *cgroups;
1205         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1206         struct list_head                cg_list;
1207 #endif
1208 #ifdef CONFIG_X86_CPU_RESCTRL
1209         u32                             closid;
1210         u32                             rmid;
1211 #endif
1212 #ifdef CONFIG_FUTEX
1213         struct robust_list_head __user  *robust_list;
1214 #ifdef CONFIG_COMPAT
1215         struct compat_robust_list_head __user *compat_robust_list;
1216 #endif
1217         struct list_head                pi_state_list;
1218         struct futex_pi_state           *pi_state_cache;
1219         struct mutex                    futex_exit_mutex;
1220         unsigned int                    futex_state;
1221 #endif
1222 #ifdef CONFIG_PERF_EVENTS
1223         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1224         struct mutex                    perf_event_mutex;
1225         struct list_head                perf_event_list;
1226 #endif
1227 #ifdef CONFIG_DEBUG_PREEMPT
1228         unsigned long                   preempt_disable_ip;
1229 #endif
1230 #ifdef CONFIG_NUMA
1231         /* Protected by alloc_lock: */
1232         struct mempolicy                *mempolicy;
1233         short                           il_prev;
1234         short                           pref_node_fork;
1235 #endif
1236 #ifdef CONFIG_NUMA_BALANCING
1237         int                             numa_scan_seq;
1238         unsigned int                    numa_scan_period;
1239         unsigned int                    numa_scan_period_max;
1240         int                             numa_preferred_nid;
1241         unsigned long                   numa_migrate_retry;
1242         /* Migration stamp: */
1243         u64                             node_stamp;
1244         u64                             last_task_numa_placement;
1245         u64                             last_sum_exec_runtime;
1246         struct callback_head            numa_work;
1247
1248         /*
1249          * This pointer is only modified for current in syscall and
1250          * pagefault context (and for tasks being destroyed), so it can be read
1251          * from any of the following contexts:
1252          *  - RCU read-side critical section
1253          *  - current->numa_group from everywhere
1254          *  - task's runqueue locked, task not running
1255          */
1256         struct numa_group __rcu         *numa_group;
1257
1258         /*
1259          * numa_faults is an array split into four regions:
1260          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1261          * in this precise order.
1262          *
1263          * faults_memory: Exponential decaying average of faults on a per-node
1264          * basis. Scheduling placement decisions are made based on these
1265          * counts. The values remain static for the duration of a PTE scan.
1266          * faults_cpu: Track the nodes the process was running on when a NUMA
1267          * hinting fault was incurred.
1268          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1269          * during the current scan window. When the scan completes, the counts
1270          * in faults_memory and faults_cpu decay and these values are copied.
1271          */
1272         unsigned long                   *numa_faults;
1273         unsigned long                   total_numa_faults;
1274
1275         /*
1276          * numa_faults_locality tracks if faults recorded during the last
1277          * scan window were remote/local or failed to migrate. The task scan
1278          * period is adapted based on the locality of the faults with different
1279          * weights depending on whether they were shared or private faults
1280          */
1281         unsigned long                   numa_faults_locality[3];
1282
1283         unsigned long                   numa_pages_migrated;
1284 #endif /* CONFIG_NUMA_BALANCING */
1285
1286 #ifdef CONFIG_RSEQ
1287         struct rseq __user *rseq;
1288         u32 rseq_sig;
1289         /*
1290          * RmW on rseq_event_mask must be performed atomically
1291          * with respect to preemption.
1292          */
1293         unsigned long rseq_event_mask;
1294 #endif
1295
1296         struct tlbflush_unmap_batch     tlb_ubc;
1297
1298         union {
1299                 refcount_t              rcu_users;
1300                 struct rcu_head         rcu;
1301         };
1302
1303         /* Cache last used pipe for splice(): */
1304         struct pipe_inode_info          *splice_pipe;
1305
1306         struct page_frag                task_frag;
1307
1308 #ifdef CONFIG_TASK_DELAY_ACCT
1309         struct task_delay_info          *delays;
1310 #endif
1311
1312 #ifdef CONFIG_FAULT_INJECTION
1313         int                             make_it_fail;
1314         unsigned int                    fail_nth;
1315 #endif
1316         /*
1317          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1318          * balance_dirty_pages() for a dirty throttling pause:
1319          */
1320         int                             nr_dirtied;
1321         int                             nr_dirtied_pause;
1322         /* Start of a write-and-pause period: */
1323         unsigned long                   dirty_paused_when;
1324
1325 #ifdef CONFIG_LATENCYTOP
1326         int                             latency_record_count;
1327         struct latency_record           latency_record[LT_SAVECOUNT];
1328 #endif
1329         /*
1330          * Time slack values; these are used to round up poll() and
1331          * select() etc timeout values. These are in nanoseconds.
1332          */
1333         u64                             timer_slack_ns;
1334         u64                             default_timer_slack_ns;
1335
1336 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1337         unsigned int                    kasan_depth;
1338 #endif
1339
1340 #ifdef CONFIG_KCSAN
1341         struct kcsan_ctx                kcsan_ctx;
1342 #ifdef CONFIG_TRACE_IRQFLAGS
1343         struct irqtrace_events          kcsan_save_irqtrace;
1344 #endif
1345 #endif
1346
1347 #if IS_ENABLED(CONFIG_KUNIT)
1348         struct kunit                    *kunit_test;
1349 #endif
1350
1351 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1352         /* Index of current stored address in ret_stack: */
1353         int                             curr_ret_stack;
1354         int                             curr_ret_depth;
1355
1356         /* Stack of return addresses for return function tracing: */
1357         struct ftrace_ret_stack         *ret_stack;
1358
1359         /* Timestamp for last schedule: */
1360         unsigned long long              ftrace_timestamp;
1361
1362         /*
1363          * Number of functions that haven't been traced
1364          * because of depth overrun:
1365          */
1366         atomic_t                        trace_overrun;
1367
1368         /* Pause tracing: */
1369         atomic_t                        tracing_graph_pause;
1370 #endif
1371
1372 #ifdef CONFIG_TRACING
1373         /* State flags for use by tracers: */
1374         unsigned long                   trace;
1375
1376         /* Bitmask and counter of trace recursion: */
1377         unsigned long                   trace_recursion;
1378 #endif /* CONFIG_TRACING */
1379
1380 #ifdef CONFIG_KCOV
1381         /* See kernel/kcov.c for more details. */
1382
1383         /* Coverage collection mode enabled for this task (0 if disabled): */
1384         unsigned int                    kcov_mode;
1385
1386         /* Size of the kcov_area: */
1387         unsigned int                    kcov_size;
1388
1389         /* Buffer for coverage collection: */
1390         void                            *kcov_area;
1391
1392         /* KCOV descriptor wired with this task or NULL: */
1393         struct kcov                     *kcov;
1394
1395         /* KCOV common handle for remote coverage collection: */
1396         u64                             kcov_handle;
1397
1398         /* KCOV sequence number: */
1399         int                             kcov_sequence;
1400
1401         /* Collect coverage from softirq context: */
1402         unsigned int                    kcov_softirq;
1403 #endif
1404
1405 #ifdef CONFIG_MEMCG
1406         struct mem_cgroup               *memcg_in_oom;
1407         gfp_t                           memcg_oom_gfp_mask;
1408         int                             memcg_oom_order;
1409
1410         /* Number of pages to reclaim on returning to userland: */
1411         unsigned int                    memcg_nr_pages_over_high;
1412
1413         /* Used by memcontrol for targeted memcg charge: */
1414         struct mem_cgroup               *active_memcg;
1415 #endif
1416
1417 #ifdef CONFIG_BLK_CGROUP
1418         struct request_queue            *throttle_queue;
1419 #endif
1420
1421 #ifdef CONFIG_UPROBES
1422         struct uprobe_task              *utask;
1423 #endif
1424 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1425         unsigned int                    sequential_io;
1426         unsigned int                    sequential_io_avg;
1427 #endif
1428         struct kmap_ctrl                kmap_ctrl;
1429 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1430         unsigned long                   task_state_change;
1431 # ifdef CONFIG_PREEMPT_RT
1432         unsigned long                   saved_state_change;
1433 # endif
1434 #endif
1435         int                             pagefault_disabled;
1436 #ifdef CONFIG_MMU
1437         struct task_struct              *oom_reaper_list;
1438 #endif
1439 #ifdef CONFIG_VMAP_STACK
1440         struct vm_struct                *stack_vm_area;
1441 #endif
1442 #ifdef CONFIG_THREAD_INFO_IN_TASK
1443         /* A live task holds one reference: */
1444         refcount_t                      stack_refcount;
1445 #endif
1446 #ifdef CONFIG_LIVEPATCH
1447         int patch_state;
1448 #endif
1449 #ifdef CONFIG_SECURITY
1450         /* Used by LSM modules for access restriction: */
1451         void                            *security;
1452 #endif
1453 #ifdef CONFIG_BPF_SYSCALL
1454         /* Used by BPF task local storage */
1455         struct bpf_local_storage __rcu  *bpf_storage;
1456 #endif
1457
1458 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1459         unsigned long                   lowest_stack;
1460         unsigned long                   prev_lowest_stack;
1461 #endif
1462
1463 #ifdef CONFIG_X86_MCE
1464         void __user                     *mce_vaddr;
1465         __u64                           mce_kflags;
1466         u64                             mce_addr;
1467         __u64                           mce_ripv : 1,
1468                                         mce_whole_page : 1,
1469                                         __mce_reserved : 62;
1470         struct callback_head            mce_kill_me;
1471 #endif
1472
1473 #ifdef CONFIG_KRETPROBES
1474         struct llist_head               kretprobe_instances;
1475 #endif
1476
1477 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1478         /*
1479          * If L1D flush is supported on mm context switch
1480          * then we use this callback head to queue kill work
1481          * to kill tasks that are not running on SMT disabled
1482          * cores
1483          */
1484         struct callback_head            l1d_flush_kill;
1485 #endif
1486
1487         /*
1488          * New fields for task_struct should be added above here, so that
1489          * they are included in the randomized portion of task_struct.
1490          */
1491         randomized_struct_fields_end
1492
1493         /* CPU-specific state of this task: */
1494         struct thread_struct            thread;
1495
1496         /*
1497          * WARNING: on x86, 'thread_struct' contains a variable-sized
1498          * structure.  It *MUST* be at the end of 'task_struct'.
1499          *
1500          * Do not put anything below here!
1501          */
1502 };
1503
1504 static inline struct pid *task_pid(struct task_struct *task)
1505 {
1506         return task->thread_pid;
1507 }
1508
1509 /*
1510  * the helpers to get the task's different pids as they are seen
1511  * from various namespaces
1512  *
1513  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1514  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1515  *                     current.
1516  * task_xid_nr_ns()  : id seen from the ns specified;
1517  *
1518  * see also pid_nr() etc in include/linux/pid.h
1519  */
1520 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1521
1522 static inline pid_t task_pid_nr(struct task_struct *tsk)
1523 {
1524         return tsk->pid;
1525 }
1526
1527 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1528 {
1529         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1530 }
1531
1532 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1533 {
1534         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1535 }
1536
1537
1538 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1539 {
1540         return tsk->tgid;
1541 }
1542
1543 /**
1544  * pid_alive - check that a task structure is not stale
1545  * @p: Task structure to be checked.
1546  *
1547  * Test if a process is not yet dead (at most zombie state)
1548  * If pid_alive fails, then pointers within the task structure
1549  * can be stale and must not be dereferenced.
1550  *
1551  * Return: 1 if the process is alive. 0 otherwise.
1552  */
1553 static inline int pid_alive(const struct task_struct *p)
1554 {
1555         return p->thread_pid != NULL;
1556 }
1557
1558 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1559 {
1560         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1561 }
1562
1563 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1564 {
1565         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1566 }
1567
1568
1569 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1570 {
1571         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1572 }
1573
1574 static inline pid_t task_session_vnr(struct task_struct *tsk)
1575 {
1576         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1577 }
1578
1579 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1580 {
1581         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1582 }
1583
1584 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1585 {
1586         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1587 }
1588
1589 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1590 {
1591         pid_t pid = 0;
1592
1593         rcu_read_lock();
1594         if (pid_alive(tsk))
1595                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1596         rcu_read_unlock();
1597
1598         return pid;
1599 }
1600
1601 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1602 {
1603         return task_ppid_nr_ns(tsk, &init_pid_ns);
1604 }
1605
1606 /* Obsolete, do not use: */
1607 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1608 {
1609         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1610 }
1611
1612 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1613 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1614
1615 static inline unsigned int task_state_index(struct task_struct *tsk)
1616 {
1617         unsigned int tsk_state = READ_ONCE(tsk->__state);
1618         unsigned int state = (tsk_state | tsk->exit_state) & TASK_REPORT;
1619
1620         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1621
1622         if (tsk_state == TASK_IDLE)
1623                 state = TASK_REPORT_IDLE;
1624
1625         return fls(state);
1626 }
1627
1628 static inline char task_index_to_char(unsigned int state)
1629 {
1630         static const char state_char[] = "RSDTtXZPI";
1631
1632         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1633
1634         return state_char[state];
1635 }
1636
1637 static inline char task_state_to_char(struct task_struct *tsk)
1638 {
1639         return task_index_to_char(task_state_index(tsk));
1640 }
1641
1642 /**
1643  * is_global_init - check if a task structure is init. Since init
1644  * is free to have sub-threads we need to check tgid.
1645  * @tsk: Task structure to be checked.
1646  *
1647  * Check if a task structure is the first user space task the kernel created.
1648  *
1649  * Return: 1 if the task structure is init. 0 otherwise.
1650  */
1651 static inline int is_global_init(struct task_struct *tsk)
1652 {
1653         return task_tgid_nr(tsk) == 1;
1654 }
1655
1656 extern struct pid *cad_pid;
1657
1658 /*
1659  * Per process flags
1660  */
1661 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1662 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1663 #define PF_EXITING              0x00000004      /* Getting shut down */
1664 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1665 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1666 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1667 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1668 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1669 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1670 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1671 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1672 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1673 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1674 #define PF_USED_ASYNC           0x00004000      /* Used async_schedule*(), used by module init */
1675 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1676 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1677 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1678 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1679 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1680 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1681                                                  * I am cleaning dirty pages from some other bdi. */
1682 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1683 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1684 #define PF_SWAPWRITE            0x00800000      /* Allowed to write to swap */
1685 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1686 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1687 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1688 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1689 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1690
1691 /*
1692  * Only the _current_ task can read/write to tsk->flags, but other
1693  * tasks can access tsk->flags in readonly mode for example
1694  * with tsk_used_math (like during threaded core dumping).
1695  * There is however an exception to this rule during ptrace
1696  * or during fork: the ptracer task is allowed to write to the
1697  * child->flags of its traced child (same goes for fork, the parent
1698  * can write to the child->flags), because we're guaranteed the
1699  * child is not running and in turn not changing child->flags
1700  * at the same time the parent does it.
1701  */
1702 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1703 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1704 #define clear_used_math()                       clear_stopped_child_used_math(current)
1705 #define set_used_math()                         set_stopped_child_used_math(current)
1706
1707 #define conditional_stopped_child_used_math(condition, child) \
1708         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1709
1710 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1711
1712 #define copy_to_stopped_child_used_math(child) \
1713         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1714
1715 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1716 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1717 #define used_math()                             tsk_used_math(current)
1718
1719 static inline bool is_percpu_thread(void)
1720 {
1721 #ifdef CONFIG_SMP
1722         return (current->flags & PF_NO_SETAFFINITY) &&
1723                 (current->nr_cpus_allowed  == 1);
1724 #else
1725         return true;
1726 #endif
1727 }
1728
1729 /* Per-process atomic flags. */
1730 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1731 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1732 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1733 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1734 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1735 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1736 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1737 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1738
1739 #define TASK_PFA_TEST(name, func)                                       \
1740         static inline bool task_##func(struct task_struct *p)           \
1741         { return test_bit(PFA_##name, &p->atomic_flags); }
1742
1743 #define TASK_PFA_SET(name, func)                                        \
1744         static inline void task_set_##func(struct task_struct *p)       \
1745         { set_bit(PFA_##name, &p->atomic_flags); }
1746
1747 #define TASK_PFA_CLEAR(name, func)                                      \
1748         static inline void task_clear_##func(struct task_struct *p)     \
1749         { clear_bit(PFA_##name, &p->atomic_flags); }
1750
1751 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1752 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1753
1754 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1755 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1756 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1757
1758 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1759 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1760 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1761
1762 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1763 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1764 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1765
1766 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1767 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1768 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1769
1770 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1771 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1772
1773 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1774 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1775 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1776
1777 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1778 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1779
1780 static inline void
1781 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1782 {
1783         current->flags &= ~flags;
1784         current->flags |= orig_flags & flags;
1785 }
1786
1787 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1788 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1789 #ifdef CONFIG_SMP
1790 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1791 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1792 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1793 extern void release_user_cpus_ptr(struct task_struct *p);
1794 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1795 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1796 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1797 #else
1798 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1799 {
1800 }
1801 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1802 {
1803         if (!cpumask_test_cpu(0, new_mask))
1804                 return -EINVAL;
1805         return 0;
1806 }
1807 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1808 {
1809         if (src->user_cpus_ptr)
1810                 return -EINVAL;
1811         return 0;
1812 }
1813 static inline void release_user_cpus_ptr(struct task_struct *p)
1814 {
1815         WARN_ON(p->user_cpus_ptr);
1816 }
1817
1818 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1819 {
1820         return 0;
1821 }
1822 #endif
1823
1824 extern int yield_to(struct task_struct *p, bool preempt);
1825 extern void set_user_nice(struct task_struct *p, long nice);
1826 extern int task_prio(const struct task_struct *p);
1827
1828 /**
1829  * task_nice - return the nice value of a given task.
1830  * @p: the task in question.
1831  *
1832  * Return: The nice value [ -20 ... 0 ... 19 ].
1833  */
1834 static inline int task_nice(const struct task_struct *p)
1835 {
1836         return PRIO_TO_NICE((p)->static_prio);
1837 }
1838
1839 extern int can_nice(const struct task_struct *p, const int nice);
1840 extern int task_curr(const struct task_struct *p);
1841 extern int idle_cpu(int cpu);
1842 extern int available_idle_cpu(int cpu);
1843 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1844 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1845 extern void sched_set_fifo(struct task_struct *p);
1846 extern void sched_set_fifo_low(struct task_struct *p);
1847 extern void sched_set_normal(struct task_struct *p, int nice);
1848 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1849 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1850 extern struct task_struct *idle_task(int cpu);
1851
1852 /**
1853  * is_idle_task - is the specified task an idle task?
1854  * @p: the task in question.
1855  *
1856  * Return: 1 if @p is an idle task. 0 otherwise.
1857  */
1858 static __always_inline bool is_idle_task(const struct task_struct *p)
1859 {
1860         return !!(p->flags & PF_IDLE);
1861 }
1862
1863 extern struct task_struct *curr_task(int cpu);
1864 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1865
1866 void yield(void);
1867
1868 union thread_union {
1869 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1870         struct task_struct task;
1871 #endif
1872 #ifndef CONFIG_THREAD_INFO_IN_TASK
1873         struct thread_info thread_info;
1874 #endif
1875         unsigned long stack[THREAD_SIZE/sizeof(long)];
1876 };
1877
1878 #ifndef CONFIG_THREAD_INFO_IN_TASK
1879 extern struct thread_info init_thread_info;
1880 #endif
1881
1882 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1883
1884 #ifdef CONFIG_THREAD_INFO_IN_TASK
1885 static inline struct thread_info *task_thread_info(struct task_struct *task)
1886 {
1887         return &task->thread_info;
1888 }
1889 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1890 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1891 #endif
1892
1893 /*
1894  * find a task by one of its numerical ids
1895  *
1896  * find_task_by_pid_ns():
1897  *      finds a task by its pid in the specified namespace
1898  * find_task_by_vpid():
1899  *      finds a task by its virtual pid
1900  *
1901  * see also find_vpid() etc in include/linux/pid.h
1902  */
1903
1904 extern struct task_struct *find_task_by_vpid(pid_t nr);
1905 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1906
1907 /*
1908  * find a task by its virtual pid and get the task struct
1909  */
1910 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1911
1912 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1913 extern int wake_up_process(struct task_struct *tsk);
1914 extern void wake_up_new_task(struct task_struct *tsk);
1915
1916 #ifdef CONFIG_SMP
1917 extern void kick_process(struct task_struct *tsk);
1918 #else
1919 static inline void kick_process(struct task_struct *tsk) { }
1920 #endif
1921
1922 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1923
1924 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1925 {
1926         __set_task_comm(tsk, from, false);
1927 }
1928
1929 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1930 #define get_task_comm(buf, tsk) ({                      \
1931         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1932         __get_task_comm(buf, sizeof(buf), tsk);         \
1933 })
1934
1935 #ifdef CONFIG_SMP
1936 static __always_inline void scheduler_ipi(void)
1937 {
1938         /*
1939          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1940          * TIF_NEED_RESCHED remotely (for the first time) will also send
1941          * this IPI.
1942          */
1943         preempt_fold_need_resched();
1944 }
1945 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1946 #else
1947 static inline void scheduler_ipi(void) { }
1948 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1949 {
1950         return 1;
1951 }
1952 #endif
1953
1954 /*
1955  * Set thread flags in other task's structures.
1956  * See asm/thread_info.h for TIF_xxxx flags available:
1957  */
1958 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1959 {
1960         set_ti_thread_flag(task_thread_info(tsk), flag);
1961 }
1962
1963 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1964 {
1965         clear_ti_thread_flag(task_thread_info(tsk), flag);
1966 }
1967
1968 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1969                                           bool value)
1970 {
1971         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1972 }
1973
1974 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1975 {
1976         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1977 }
1978
1979 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1980 {
1981         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1982 }
1983
1984 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1985 {
1986         return test_ti_thread_flag(task_thread_info(tsk), flag);
1987 }
1988
1989 static inline void set_tsk_need_resched(struct task_struct *tsk)
1990 {
1991         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1992 }
1993
1994 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1995 {
1996         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1997 }
1998
1999 static inline int test_tsk_need_resched(struct task_struct *tsk)
2000 {
2001         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2002 }
2003
2004 /*
2005  * cond_resched() and cond_resched_lock(): latency reduction via
2006  * explicit rescheduling in places that are safe. The return
2007  * value indicates whether a reschedule was done in fact.
2008  * cond_resched_lock() will drop the spinlock before scheduling,
2009  */
2010 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2011 extern int __cond_resched(void);
2012
2013 #ifdef CONFIG_PREEMPT_DYNAMIC
2014
2015 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2016
2017 static __always_inline int _cond_resched(void)
2018 {
2019         return static_call_mod(cond_resched)();
2020 }
2021
2022 #else
2023
2024 static inline int _cond_resched(void)
2025 {
2026         return __cond_resched();
2027 }
2028
2029 #endif /* CONFIG_PREEMPT_DYNAMIC */
2030
2031 #else
2032
2033 static inline int _cond_resched(void) { return 0; }
2034
2035 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2036
2037 #define cond_resched() ({                       \
2038         ___might_sleep(__FILE__, __LINE__, 0);  \
2039         _cond_resched();                        \
2040 })
2041
2042 extern int __cond_resched_lock(spinlock_t *lock);
2043 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2044 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2045
2046 #define cond_resched_lock(lock) ({                              \
2047         ___might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET);\
2048         __cond_resched_lock(lock);                              \
2049 })
2050
2051 #define cond_resched_rwlock_read(lock) ({                       \
2052         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2053         __cond_resched_rwlock_read(lock);                       \
2054 })
2055
2056 #define cond_resched_rwlock_write(lock) ({                      \
2057         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2058         __cond_resched_rwlock_write(lock);                      \
2059 })
2060
2061 static inline void cond_resched_rcu(void)
2062 {
2063 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2064         rcu_read_unlock();
2065         cond_resched();
2066         rcu_read_lock();
2067 #endif
2068 }
2069
2070 /*
2071  * Does a critical section need to be broken due to another
2072  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2073  * but a general need for low latency)
2074  */
2075 static inline int spin_needbreak(spinlock_t *lock)
2076 {
2077 #ifdef CONFIG_PREEMPTION
2078         return spin_is_contended(lock);
2079 #else
2080         return 0;
2081 #endif
2082 }
2083
2084 /*
2085  * Check if a rwlock is contended.
2086  * Returns non-zero if there is another task waiting on the rwlock.
2087  * Returns zero if the lock is not contended or the system / underlying
2088  * rwlock implementation does not support contention detection.
2089  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2090  * for low latency.
2091  */
2092 static inline int rwlock_needbreak(rwlock_t *lock)
2093 {
2094 #ifdef CONFIG_PREEMPTION
2095         return rwlock_is_contended(lock);
2096 #else
2097         return 0;
2098 #endif
2099 }
2100
2101 static __always_inline bool need_resched(void)
2102 {
2103         return unlikely(tif_need_resched());
2104 }
2105
2106 /*
2107  * Wrappers for p->thread_info->cpu access. No-op on UP.
2108  */
2109 #ifdef CONFIG_SMP
2110
2111 static inline unsigned int task_cpu(const struct task_struct *p)
2112 {
2113 #ifdef CONFIG_THREAD_INFO_IN_TASK
2114         return READ_ONCE(p->cpu);
2115 #else
2116         return READ_ONCE(task_thread_info(p)->cpu);
2117 #endif
2118 }
2119
2120 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2121
2122 #else
2123
2124 static inline unsigned int task_cpu(const struct task_struct *p)
2125 {
2126         return 0;
2127 }
2128
2129 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2130 {
2131 }
2132
2133 #endif /* CONFIG_SMP */
2134
2135 extern bool sched_task_on_rq(struct task_struct *p);
2136
2137 /*
2138  * In order to reduce various lock holder preemption latencies provide an
2139  * interface to see if a vCPU is currently running or not.
2140  *
2141  * This allows us to terminate optimistic spin loops and block, analogous to
2142  * the native optimistic spin heuristic of testing if the lock owner task is
2143  * running or not.
2144  */
2145 #ifndef vcpu_is_preempted
2146 static inline bool vcpu_is_preempted(int cpu)
2147 {
2148         return false;
2149 }
2150 #endif
2151
2152 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2153 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2154
2155 #ifndef TASK_SIZE_OF
2156 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2157 #endif
2158
2159 #ifdef CONFIG_SMP
2160 /* Returns effective CPU energy utilization, as seen by the scheduler */
2161 unsigned long sched_cpu_util(int cpu, unsigned long max);
2162 #endif /* CONFIG_SMP */
2163
2164 #ifdef CONFIG_RSEQ
2165
2166 /*
2167  * Map the event mask on the user-space ABI enum rseq_cs_flags
2168  * for direct mask checks.
2169  */
2170 enum rseq_event_mask_bits {
2171         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2172         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2173         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2174 };
2175
2176 enum rseq_event_mask {
2177         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2178         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2179         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2180 };
2181
2182 static inline void rseq_set_notify_resume(struct task_struct *t)
2183 {
2184         if (t->rseq)
2185                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2186 }
2187
2188 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2189
2190 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2191                                              struct pt_regs *regs)
2192 {
2193         if (current->rseq)
2194                 __rseq_handle_notify_resume(ksig, regs);
2195 }
2196
2197 static inline void rseq_signal_deliver(struct ksignal *ksig,
2198                                        struct pt_regs *regs)
2199 {
2200         preempt_disable();
2201         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2202         preempt_enable();
2203         rseq_handle_notify_resume(ksig, regs);
2204 }
2205
2206 /* rseq_preempt() requires preemption to be disabled. */
2207 static inline void rseq_preempt(struct task_struct *t)
2208 {
2209         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2210         rseq_set_notify_resume(t);
2211 }
2212
2213 /* rseq_migrate() requires preemption to be disabled. */
2214 static inline void rseq_migrate(struct task_struct *t)
2215 {
2216         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2217         rseq_set_notify_resume(t);
2218 }
2219
2220 /*
2221  * If parent process has a registered restartable sequences area, the
2222  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2223  */
2224 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2225 {
2226         if (clone_flags & CLONE_VM) {
2227                 t->rseq = NULL;
2228                 t->rseq_sig = 0;
2229                 t->rseq_event_mask = 0;
2230         } else {
2231                 t->rseq = current->rseq;
2232                 t->rseq_sig = current->rseq_sig;
2233                 t->rseq_event_mask = current->rseq_event_mask;
2234         }
2235 }
2236
2237 static inline void rseq_execve(struct task_struct *t)
2238 {
2239         t->rseq = NULL;
2240         t->rseq_sig = 0;
2241         t->rseq_event_mask = 0;
2242 }
2243
2244 #else
2245
2246 static inline void rseq_set_notify_resume(struct task_struct *t)
2247 {
2248 }
2249 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2250                                              struct pt_regs *regs)
2251 {
2252 }
2253 static inline void rseq_signal_deliver(struct ksignal *ksig,
2254                                        struct pt_regs *regs)
2255 {
2256 }
2257 static inline void rseq_preempt(struct task_struct *t)
2258 {
2259 }
2260 static inline void rseq_migrate(struct task_struct *t)
2261 {
2262 }
2263 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2264 {
2265 }
2266 static inline void rseq_execve(struct task_struct *t)
2267 {
2268 }
2269
2270 #endif
2271
2272 #ifdef CONFIG_DEBUG_RSEQ
2273
2274 void rseq_syscall(struct pt_regs *regs);
2275
2276 #else
2277
2278 static inline void rseq_syscall(struct pt_regs *regs)
2279 {
2280 }
2281
2282 #endif
2283
2284 const struct sched_avg *sched_trace_cfs_rq_avg(struct cfs_rq *cfs_rq);
2285 char *sched_trace_cfs_rq_path(struct cfs_rq *cfs_rq, char *str, int len);
2286 int sched_trace_cfs_rq_cpu(struct cfs_rq *cfs_rq);
2287
2288 const struct sched_avg *sched_trace_rq_avg_rt(struct rq *rq);
2289 const struct sched_avg *sched_trace_rq_avg_dl(struct rq *rq);
2290 const struct sched_avg *sched_trace_rq_avg_irq(struct rq *rq);
2291
2292 int sched_trace_rq_cpu(struct rq *rq);
2293 int sched_trace_rq_cpu_capacity(struct rq *rq);
2294 int sched_trace_rq_nr_running(struct rq *rq);
2295
2296 const struct cpumask *sched_trace_rd_span(struct root_domain *rd);
2297
2298 #ifdef CONFIG_SCHED_CORE
2299 extern void sched_core_free(struct task_struct *tsk);
2300 extern void sched_core_fork(struct task_struct *p);
2301 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2302                                 unsigned long uaddr);
2303 #else
2304 static inline void sched_core_free(struct task_struct *tsk) { }
2305 static inline void sched_core_fork(struct task_struct *p) { }
2306 #endif
2307
2308 #endif