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 #include <asm/processor.h>
  14 #include <linux/thread_info.h>
  15 #include <linux/preempt.h>
  16 #include <linux/cpumask.h>
  17
  18 #include <linux/cache.h>
  19 #include <linux/irqflags_types.h>
  20 #include <linux/smp_types.h>
  21 #include <linux/pid_types.h>
  22 #include <linux/sem_types.h>
  23 #include <linux/shm.h>
  24 #include <linux/kmsan_types.h>
  25 #include <linux/mutex_types.h>
  26 #include <linux/plist_types.h>
  27 #include <linux/hrtimer_types.h>
  28 #include <linux/timer_types.h>
  29 #include <linux/seccomp_types.h>
  30 #include <linux/nodemask_types.h>
  31 #include <linux/refcount_types.h>
  32 #include <linux/resource.h>
  33 #include <linux/latencytop.h>
  34 #include <linux/sched/prio.h>
  35 #include <linux/sched/types.h>
  36 #include <linux/signal_types.h>
  37 #include <linux/syscall_user_dispatch_types.h>
  38 #include <linux/mm_types_task.h>
  39 #include <linux/task_io_accounting.h>
  40 #include <linux/posix-timers_types.h>
  41 #include <linux/restart_block.h>
  42 #include <uapi/linux/rseq.h>
  43 #include <linux/seqlock_types.h>
  44 #include <linux/kcsan.h>
  45 #include <linux/rv.h>
  46 #include <linux/livepatch_sched.h>
  47 #include <linux/uidgid_types.h>
  48 #include <asm/kmap_size.h>
  49
  50 /* task_struct member predeclarations (sorted alphabetically): */
  51 struct audit_context;
  52 struct bio_list;
  53 struct blk_plug;
  54 struct bpf_local_storage;
  55 struct bpf_run_ctx;
  56 struct capture_control;
  57 struct cfs_rq;
  58 struct fs_struct;
  59 struct futex_pi_state;
  60 struct io_context;
  61 struct io_uring_task;
  62 struct mempolicy;
  63 struct nameidata;
  64 struct nsproxy;
  65 struct perf_event_context;
  66 struct pid_namespace;
  67 struct pipe_inode_info;
  68 struct rcu_node;
  69 struct reclaim_state;
  70 struct robust_list_head;
  71 struct root_domain;
  72 struct rq;
  73 struct sched_attr;
  74 struct sched_dl_entity;
  75 struct seq_file;
  76 struct sighand_struct;
  77 struct signal_struct;
  78 struct task_delay_info;
  79 struct task_group;
  80 struct task_struct;
  81 struct user_event_mm;
  82
  83 /*
  84  * Task state bitmask. NOTE! These bits are also
  85  * encoded in fs/proc/array.c: get_task_state().
  86  *
  87  * We have two separate sets of flags: task->__state
  88  * is about runnability, while task->exit_state are
  89  * about the task exiting. Confusing, but this way
  90  * modifying one set can't modify the other one by
  91  * mistake.
  92  */
  93
  94 /* Used in tsk->__state: */
  95 #define TASK_RUNNING                    0x00000000
  96 #define TASK_INTERRUPTIBLE              0x00000001
  97 #define TASK_UNINTERRUPTIBLE            0x00000002
  98 #define __TASK_STOPPED                  0x00000004
  99 #define __TASK_TRACED                   0x00000008
 100 /* Used in tsk->exit_state: */
 101 #define EXIT_DEAD                       0x00000010
 102 #define EXIT_ZOMBIE                     0x00000020
 103 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
 104 /* Used in tsk->__state again: */
 105 #define TASK_PARKED                     0x00000040
 106 #define TASK_DEAD                       0x00000080
 107 #define TASK_WAKEKILL                   0x00000100
 108 #define TASK_WAKING                     0x00000200
 109 #define TASK_NOLOAD                     0x00000400
 110 #define TASK_NEW                        0x00000800
 111 #define TASK_RTLOCK_WAIT                0x00001000
 112 #define TASK_FREEZABLE                  0x00002000
 113 #define __TASK_FREEZABLE_UNSAFE        (0x00004000 * IS_ENABLED(CONFIG_LOCKDEP))
 114 #define TASK_FROZEN                     0x00008000
 115 #define TASK_STATE_MAX                  0x00010000
 116
 117 #define TASK_ANY                        (TASK_STATE_MAX-1)
 118
 119 /*
 120  * DO NOT ADD ANY NEW USERS !
 121  */
 122 #define TASK_FREEZABLE_UNSAFE           (TASK_FREEZABLE | __TASK_FREEZABLE_UNSAFE)
 123
 124 /* Convenience macros for the sake of set_current_state: */
 125 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 126 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 127 #define TASK_TRACED                     __TASK_TRACED
 128
 129 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 130
 131 /* Convenience macros for the sake of wake_up(): */
 132 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 133
 134 /* get_task_state(): */
 135 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 136                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 137                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 138                                          TASK_PARKED)
 139
 140 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 141
 142 #define task_is_traced(task)            ((READ_ONCE(task->jobctl) & JOBCTL_TRACED) != 0)
 143 #define task_is_stopped(task)           ((READ_ONCE(task->jobctl) & JOBCTL_STOPPED) != 0)
 144 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->jobctl) & (JOBCTL_STOPPED | JOBCTL_TRACED)) != 0)
 145
 146 /*
 147  * Special states are those that do not use the normal wait-loop pattern. See
 148  * the comment with set_special_state().
 149  */
 150 #define is_special_task_state(state)                            \
 151         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
 152
 153 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 154 # define debug_normal_state_change(state_value)                         \
 155         do {                                                            \
 156                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 157                 current->task_state_change = _THIS_IP_;                 \
 158         } while (0)
 159
 160 # define debug_special_state_change(state_value)                        \
 161         do {                                                            \
 162                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 163                 current->task_state_change = _THIS_IP_;                 \
 164         } while (0)
 165
 166 # define debug_rtlock_wait_set_state()                                  \
 167         do {                                                             \
 168                 current->saved_state_change = current->task_state_change;\
 169                 current->task_state_change = _THIS_IP_;                  \
 170         } while (0)
 171
 172 # define debug_rtlock_wait_restore_state()                              \
 173         do {                                                             \
 174                 current->task_state_change = current->saved_state_change;\
 175         } while (0)
 176
 177 #else
 178 # define debug_normal_state_change(cond)        do { } while (0)
 179 # define debug_special_state_change(cond)       do { } while (0)
 180 # define debug_rtlock_wait_set_state()          do { } while (0)
 181 # define debug_rtlock_wait_restore_state()      do { } while (0)
 182 #endif
 183
 184 /*
 185  * set_current_state() includes a barrier so that the write of current->__state
 186  * is correctly serialised wrt the caller's subsequent test of whether to
 187  * actually sleep:
 188  *
 189  *   for (;;) {
 190  *      set_current_state(TASK_UNINTERRUPTIBLE);
 191  *      if (CONDITION)
 192  *         break;
 193  *
 194  *      schedule();
 195  *   }
 196  *   __set_current_state(TASK_RUNNING);
 197  *
 198  * If the caller does not need such serialisation (because, for instance, the
 199  * CONDITION test and condition change and wakeup are under the same lock) then
 200  * use __set_current_state().
 201  *
 202  * The above is typically ordered against the wakeup, which does:
 203  *
 204  *   CONDITION = 1;
 205  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 206  *
 207  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 208  * accessing p->__state.
 209  *
 210  * Wakeup will do: if (@state & p->__state) p->__state = TASK_RUNNING, that is,
 211  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 212  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 213  *
 214  * However, with slightly different timing the wakeup TASK_RUNNING store can
 215  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 216  * a problem either because that will result in one extra go around the loop
 217  * and our @cond test will save the day.
 218  *
 219  * Also see the comments of try_to_wake_up().
 220  */
 221 #define __set_current_state(state_value)                                \
 222         do {                                                            \
 223                 debug_normal_state_change((state_value));               \
 224                 WRITE_ONCE(current->__state, (state_value));            \
 225         } while (0)
 226
 227 #define set_current_state(state_value)                                  \
 228         do {                                                            \
 229                 debug_normal_state_change((state_value));               \
 230                 smp_store_mb(current->__state, (state_value));          \
 231         } while (0)
 232
 233 /*
 234  * set_special_state() should be used for those states when the blocking task
 235  * can not use the regular condition based wait-loop. In that case we must
 236  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 237  * stores will not collide with our state change.
 238  */
 239 #define set_special_state(state_value)                                  \
 240         do {                                                            \
 241                 unsigned long flags; /* may shadow */                   \
 242                                                                         \
 243                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 244                 debug_special_state_change((state_value));              \
 245                 WRITE_ONCE(current->__state, (state_value));            \
 246                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 247         } while (0)
 248
 249 /*
 250  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 251  *
 252  * RT's spin/rwlock substitutions are state preserving. The state of the
 253  * task when blocking on the lock is saved in task_struct::saved_state and
 254  * restored after the lock has been acquired.  These operations are
 255  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 256  * lock related wakeups while the task is blocked on the lock are
 257  * redirected to operate on task_struct::saved_state to ensure that these
 258  * are not dropped. On restore task_struct::saved_state is set to
 259  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 260  *
 261  * The lock operation looks like this:
 262  *
 263  *      current_save_and_set_rtlock_wait_state();
 264  *      for (;;) {
 265  *              if (try_lock())
 266  *                      break;
 267  *              raw_spin_unlock_irq(&lock->wait_lock);
 268  *              schedule_rtlock();
 269  *              raw_spin_lock_irq(&lock->wait_lock);
 270  *              set_current_state(TASK_RTLOCK_WAIT);
 271  *      }
 272  *      current_restore_rtlock_saved_state();
 273  */
 274 #define current_save_and_set_rtlock_wait_state()                        \
 275         do {                                                            \
 276                 lockdep_assert_irqs_disabled();                         \
 277                 raw_spin_lock(&current->pi_lock);                       \
 278                 current->saved_state = current->__state;                \
 279                 debug_rtlock_wait_set_state();                          \
 280                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 281                 raw_spin_unlock(&current->pi_lock);                     \
 282         } while (0);
 283
 284 #define current_restore_rtlock_saved_state()                            \
 285         do {                                                            \
 286                 lockdep_assert_irqs_disabled();                         \
 287                 raw_spin_lock(&current->pi_lock);                       \
 288                 debug_rtlock_wait_restore_state();                      \
 289                 WRITE_ONCE(current->__state, current->saved_state);     \
 290                 current->saved_state = TASK_RUNNING;                    \
 291                 raw_spin_unlock(&current->pi_lock);                     \
 292         } while (0);
 293
 294 #define get_current_state()     READ_ONCE(current->__state)
 295
 296 /*
 297  * Define the task command name length as enum, then it can be visible to
 298  * BPF programs.
 299  */
 300 enum {
 301         TASK_COMM_LEN = 16,
 302 };
 303
 304 extern void scheduler_tick(void);
 305
 306 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 307
 308 extern long schedule_timeout(long timeout);
 309 extern long schedule_timeout_interruptible(long timeout);
 310 extern long schedule_timeout_killable(long timeout);
 311 extern long schedule_timeout_uninterruptible(long timeout);
 312 extern long schedule_timeout_idle(long timeout);
 313 asmlinkage void schedule(void);
 314 extern void schedule_preempt_disabled(void);
 315 asmlinkage void preempt_schedule_irq(void);
 316 #ifdef CONFIG_PREEMPT_RT
 317  extern void schedule_rtlock(void);
 318 #endif
 319
 320 extern int __must_check io_schedule_prepare(void);
 321 extern void io_schedule_finish(int token);
 322 extern long io_schedule_timeout(long timeout);
 323 extern void io_schedule(void);
 324
 325 /**
 326  * struct prev_cputime - snapshot of system and user cputime
 327  * @utime: time spent in user mode
 328  * @stime: time spent in system mode
 329  * @lock: protects the above two fields
 330  *
 331  * Stores previous user/system time values such that we can guarantee
 332  * monotonicity.
 333  */
 334 struct prev_cputime {
 335 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 336         u64                             utime;
 337         u64                             stime;
 338         raw_spinlock_t                  lock;
 339 #endif
 340 };
 341
 342 enum vtime_state {
 343         /* Task is sleeping or running in a CPU with VTIME inactive: */
 344         VTIME_INACTIVE = 0,
 345         /* Task is idle */
 346         VTIME_IDLE,
 347         /* Task runs in kernelspace in a CPU with VTIME active: */
 348         VTIME_SYS,
 349         /* Task runs in userspace in a CPU with VTIME active: */
 350         VTIME_USER,
 351         /* Task runs as guests in a CPU with VTIME active: */
 352         VTIME_GUEST,
 353 };
 354
 355 struct vtime {
 356         seqcount_t              seqcount;
 357         unsigned long long      starttime;
 358         enum vtime_state        state;
 359         unsigned int            cpu;
 360         u64                     utime;
 361         u64                     stime;
 362         u64                     gtime;
 363 };
 364
 365 /*
 366  * Utilization clamp constraints.
 367  * @UCLAMP_MIN: Minimum utilization
 368  * @UCLAMP_MAX: Maximum utilization
 369  * @UCLAMP_CNT: Utilization clamp constraints count
 370  */
 371 enum uclamp_id {
 372         UCLAMP_MIN = 0,
 373         UCLAMP_MAX,
 374         UCLAMP_CNT
 375 };
 376
 377 #ifdef CONFIG_SMP
 378 extern struct root_domain def_root_domain;
 379 extern struct mutex sched_domains_mutex;
 380 #endif
 381
 382 struct sched_param {
 383         int sched_priority;
 384 };
 385
 386 struct sched_info {
 387 #ifdef CONFIG_SCHED_INFO
 388         /* Cumulative counters: */
 389
 390         /* # of times we have run on this CPU: */
 391         unsigned long                   pcount;
 392
 393         /* Time spent waiting on a runqueue: */
 394         unsigned long long              run_delay;
 395
 396         /* Timestamps: */
 397
 398         /* When did we last run on a CPU? */
 399         unsigned long long              last_arrival;
 400
 401         /* When were we last queued to run? */
 402         unsigned long long              last_queued;
 403
 404 #endif /* CONFIG_SCHED_INFO */
 405 };
 406
 407 /*
 408  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 409  * has a few: load, load_avg, util_avg, freq, and capacity.
 410  *
 411  * We define a basic fixed point arithmetic range, and then formalize
 412  * all these metrics based on that basic range.
 413  */
 414 # define SCHED_FIXEDPOINT_SHIFT         10
 415 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 416
 417 /* Increase resolution of cpu_capacity calculations */
 418 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 419 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 420
 421 struct load_weight {
 422         unsigned long                   weight;
 423         u32                             inv_weight;
 424 };
 425
 426 /*
 427  * The load/runnable/util_avg accumulates an infinite geometric series
 428  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 429  *
 430  * [load_avg definition]
 431  *
 432  *   load_avg = runnable% * scale_load_down(load)
 433  *
 434  * [runnable_avg definition]
 435  *
 436  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 437  *
 438  * [util_avg definition]
 439  *
 440  *   util_avg = running% * SCHED_CAPACITY_SCALE
 441  *
 442  * where runnable% is the time ratio that a sched_entity is runnable and
 443  * running% the time ratio that a sched_entity is running.
 444  *
 445  * For cfs_rq, they are the aggregated values of all runnable and blocked
 446  * sched_entities.
 447  *
 448  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 449  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 450  * for computing those signals (see update_rq_clock_pelt())
 451  *
 452  * N.B., the above ratios (runnable% and running%) themselves are in the
 453  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 454  * to as large a range as necessary. This is for example reflected by
 455  * util_avg's SCHED_CAPACITY_SCALE.
 456  *
 457  * [Overflow issue]
 458  *
 459  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 460  * with the highest load (=88761), always runnable on a single cfs_rq,
 461  * and should not overflow as the number already hits PID_MAX_LIMIT.
 462  *
 463  * For all other cases (including 32-bit kernels), struct load_weight's
 464  * weight will overflow first before we do, because:
 465  *
 466  *    Max(load_avg) <= Max(load.weight)
 467  *
 468  * Then it is the load_weight's responsibility to consider overflow
 469  * issues.
 470  */
 471 struct sched_avg {
 472         u64                             last_update_time;
 473         u64                             load_sum;
 474         u64                             runnable_sum;
 475         u32                             util_sum;
 476         u32                             period_contrib;
 477         unsigned long                   load_avg;
 478         unsigned long                   runnable_avg;
 479         unsigned long                   util_avg;
 480         unsigned int                    util_est;
 481 } ____cacheline_aligned;
 482
 483 /*
 484  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 485  * updates. When a task is dequeued, its util_est should not be updated if its
 486  * util_avg has not been updated in the meantime.
 487  * This information is mapped into the MSB bit of util_est at dequeue time.
 488  * Since max value of util_est for a task is 1024 (PELT util_avg for a task)
 489  * it is safe to use MSB.
 490  */
 491 #define UTIL_EST_WEIGHT_SHIFT           2
 492 #define UTIL_AVG_UNCHANGED              0x80000000
 493
 494 struct sched_statistics {
 495 #ifdef CONFIG_SCHEDSTATS
 496         u64                             wait_start;
 497         u64                             wait_max;
 498         u64                             wait_count;
 499         u64                             wait_sum;
 500         u64                             iowait_count;
 501         u64                             iowait_sum;
 502
 503         u64                             sleep_start;
 504         u64                             sleep_max;
 505         s64                             sum_sleep_runtime;
 506
 507         u64                             block_start;
 508         u64                             block_max;
 509         s64                             sum_block_runtime;
 510
 511         s64                             exec_max;
 512         u64                             slice_max;
 513
 514         u64                             nr_migrations_cold;
 515         u64                             nr_failed_migrations_affine;
 516         u64                             nr_failed_migrations_running;
 517         u64                             nr_failed_migrations_hot;
 518         u64                             nr_forced_migrations;
 519
 520         u64                             nr_wakeups;
 521         u64                             nr_wakeups_sync;
 522         u64                             nr_wakeups_migrate;
 523         u64                             nr_wakeups_local;
 524         u64                             nr_wakeups_remote;
 525         u64                             nr_wakeups_affine;
 526         u64                             nr_wakeups_affine_attempts;
 527         u64                             nr_wakeups_passive;
 528         u64                             nr_wakeups_idle;
 529
 530 #ifdef CONFIG_SCHED_CORE
 531         u64                             core_forceidle_sum;
 532 #endif
 533 #endif /* CONFIG_SCHEDSTATS */
 534 } ____cacheline_aligned;
 535
 536 struct sched_entity {
 537         /* For load-balancing: */
 538         struct load_weight              load;
 539         struct rb_node                  run_node;
 540         u64                             deadline;
 541         u64                             min_vruntime;
 542
 543         struct list_head                group_node;
 544         unsigned int                    on_rq;
 545
 546         u64                             exec_start;
 547         u64                             sum_exec_runtime;
 548         u64                             prev_sum_exec_runtime;
 549         u64                             vruntime;
 550         s64                             vlag;
 551         u64                             slice;
 552
 553         u64                             nr_migrations;
 554
 555 #ifdef CONFIG_FAIR_GROUP_SCHED
 556         int                             depth;
 557         struct sched_entity             *parent;
 558         /* rq on which this entity is (to be) queued: */
 559         struct cfs_rq                   *cfs_rq;
 560         /* rq "owned" by this entity/group: */
 561         struct cfs_rq                   *my_q;
 562         /* cached value of my_q->h_nr_running */
 563         unsigned long                   runnable_weight;
 564 #endif
 565
 566 #ifdef CONFIG_SMP
 567         /*
 568          * Per entity load average tracking.
 569          *
 570          * Put into separate cache line so it does not
 571          * collide with read-mostly values above.
 572          */
 573         struct sched_avg                avg;
 574 #endif
 575 };
 576
 577 struct sched_rt_entity {
 578         struct list_head                run_list;
 579         unsigned long                   timeout;
 580         unsigned long                   watchdog_stamp;
 581         unsigned int                    time_slice;
 582         unsigned short                  on_rq;
 583         unsigned short                  on_list;
 584
 585         struct sched_rt_entity          *back;
 586 #ifdef CONFIG_RT_GROUP_SCHED
 587         struct sched_rt_entity          *parent;
 588         /* rq on which this entity is (to be) queued: */
 589         struct rt_rq                    *rt_rq;
 590         /* rq "owned" by this entity/group: */
 591         struct rt_rq                    *my_q;
 592 #endif
 593 } __randomize_layout;
 594
 595 typedef bool (*dl_server_has_tasks_f)(struct sched_dl_entity *);
 596 typedef struct task_struct *(*dl_server_pick_f)(struct sched_dl_entity *);
 597
 598 struct sched_dl_entity {
 599         struct rb_node                  rb_node;
 600
 601         /*
 602          * Original scheduling parameters. Copied here from sched_attr
 603          * during sched_setattr(), they will remain the same until
 604          * the next sched_setattr().
 605          */
 606         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 607         u64                             dl_deadline;    /* Relative deadline of each instance   */
 608         u64                             dl_period;      /* Separation of two instances (period) */
 609         u64                             dl_bw;          /* dl_runtime / dl_period               */
 610         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 611
 612         /*
 613          * Actual scheduling parameters. Initialized with the values above,
 614          * they are continuously updated during task execution. Note that
 615          * the remaining runtime could be < 0 in case we are in overrun.
 616          */
 617         s64                             runtime;        /* Remaining runtime for this instance  */
 618         u64                             deadline;       /* Absolute deadline for this instance  */
 619         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 620
 621         /*
 622          * Some bool flags:
 623          *
 624          * @dl_throttled tells if we exhausted the runtime. If so, the
 625          * task has to wait for a replenishment to be performed at the
 626          * next firing of dl_timer.
 627          *
 628          * @dl_yielded tells if task gave up the CPU before consuming
 629          * all its available runtime during the last job.
 630          *
 631          * @dl_non_contending tells if the task is inactive while still
 632          * contributing to the active utilization. In other words, it
 633          * indicates if the inactive timer has been armed and its handler
 634          * has not been executed yet. This flag is useful to avoid race
 635          * conditions between the inactive timer handler and the wakeup
 636          * code.
 637          *
 638          * @dl_overrun tells if the task asked to be informed about runtime
 639          * overruns.
 640          */
 641         unsigned int                    dl_throttled      : 1;
 642         unsigned int                    dl_yielded        : 1;
 643         unsigned int                    dl_non_contending : 1;
 644         unsigned int                    dl_overrun        : 1;
 645         unsigned int                    dl_server         : 1;
 646
 647         /*
 648          * Bandwidth enforcement timer. Each -deadline task has its
 649          * own bandwidth to be enforced, thus we need one timer per task.
 650          */
 651         struct hrtimer                  dl_timer;
 652
 653         /*
 654          * Inactive timer, responsible for decreasing the active utilization
 655          * at the "0-lag time". When a -deadline task blocks, it contributes
 656          * to GRUB's active utilization until the "0-lag time", hence a
 657          * timer is needed to decrease the active utilization at the correct
 658          * time.
 659          */
 660         struct hrtimer                  inactive_timer;
 661
 662         /*
 663          * Bits for DL-server functionality. Also see the comment near
 664          * dl_server_update().
 665          *
 666          * @rq the runqueue this server is for
 667          *
 668          * @server_has_tasks() returns true if @server_pick return a
 669          * runnable task.
 670          */
 671         struct rq                       *rq;
 672         dl_server_has_tasks_f           server_has_tasks;
 673         dl_server_pick_f                server_pick;
 674
 675 #ifdef CONFIG_RT_MUTEXES
 676         /*
 677          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 678          * pi_se points to the donor, otherwise points to the dl_se it belongs
 679          * to (the original one/itself).
 680          */
 681         struct sched_dl_entity *pi_se;
 682 #endif
 683 };
 684
 685 #ifdef CONFIG_UCLAMP_TASK
 686 /* Number of utilization clamp buckets (shorter alias) */
 687 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 688
 689 /*
 690  * Utilization clamp for a scheduling entity
 691  * @value:              clamp value "assigned" to a se
 692  * @bucket_id:          bucket index corresponding to the "assigned" value
 693  * @active:             the se is currently refcounted in a rq's bucket
 694  * @user_defined:       the requested clamp value comes from user-space
 695  *
 696  * The bucket_id is the index of the clamp bucket matching the clamp value
 697  * which is pre-computed and stored to avoid expensive integer divisions from
 698  * the fast path.
 699  *
 700  * The active bit is set whenever a task has got an "effective" value assigned,
 701  * which can be different from the clamp value "requested" from user-space.
 702  * This allows to know a task is refcounted in the rq's bucket corresponding
 703  * to the "effective" bucket_id.
 704  *
 705  * The user_defined bit is set whenever a task has got a task-specific clamp
 706  * value requested from userspace, i.e. the system defaults apply to this task
 707  * just as a restriction. This allows to relax default clamps when a less
 708  * restrictive task-specific value has been requested, thus allowing to
 709  * implement a "nice" semantic. For example, a task running with a 20%
 710  * default boost can still drop its own boosting to 0%.
 711  */
 712 struct uclamp_se {
 713         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 714         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 715         unsigned int active             : 1;
 716         unsigned int user_defined       : 1;
 717 };
 718 #endif /* CONFIG_UCLAMP_TASK */
 719
 720 union rcu_special {
 721         struct {
 722                 u8                      blocked;
 723                 u8                      need_qs;
 724                 u8                      exp_hint; /* Hint for performance. */
 725                 u8                      need_mb; /* Readers need smp_mb(). */
 726         } b; /* Bits. */
 727         u32 s; /* Set of bits. */
 728 };
 729
 730 enum perf_event_task_context {
 731         perf_invalid_context = -1,
 732         perf_hw_context = 0,
 733         perf_sw_context,
 734         perf_nr_task_contexts,
 735 };
 736
 737 struct wake_q_node {
 738         struct wake_q_node *next;
 739 };
 740
 741 struct kmap_ctrl {
 742 #ifdef CONFIG_KMAP_LOCAL
 743         int                             idx;
 744         pte_t                           pteval[KM_MAX_IDX];
 745 #endif
 746 };
 747
 748 struct task_struct {
 749 #ifdef CONFIG_THREAD_INFO_IN_TASK
 750         /*
 751          * For reasons of header soup (see current_thread_info()), this
 752          * must be the first element of task_struct.
 753          */
 754         struct thread_info              thread_info;
 755 #endif
 756         unsigned int                    __state;
 757
 758         /* saved state for "spinlock sleepers" */
 759         unsigned int                    saved_state;
 760
 761         /*
 762          * This begins the randomizable portion of task_struct. Only
 763          * scheduling-critical items should be added above here.
 764          */
 765         randomized_struct_fields_start
 766
 767         void                            *stack;
 768         refcount_t                      usage;
 769         /* Per task flags (PF_*), defined further below: */
 770         unsigned int                    flags;
 771         unsigned int                    ptrace;
 772
 773 #ifdef CONFIG_SMP
 774         int                             on_cpu;
 775         struct __call_single_node       wake_entry;
 776         unsigned int                    wakee_flips;
 777         unsigned long                   wakee_flip_decay_ts;
 778         struct task_struct              *last_wakee;
 779
 780         /*
 781          * recent_used_cpu is initially set as the last CPU used by a task
 782          * that wakes affine another task. Waker/wakee relationships can
 783          * push tasks around a CPU where each wakeup moves to the next one.
 784          * Tracking a recently used CPU allows a quick search for a recently
 785          * used CPU that may be idle.
 786          */
 787         int                             recent_used_cpu;
 788         int                             wake_cpu;
 789 #endif
 790         int                             on_rq;
 791
 792         int                             prio;
 793         int                             static_prio;
 794         int                             normal_prio;
 795         unsigned int                    rt_priority;
 796
 797         struct sched_entity             se;
 798         struct sched_rt_entity          rt;
 799         struct sched_dl_entity          dl;
 800         struct sched_dl_entity          *dl_server;
 801         const struct sched_class        *sched_class;
 802
 803 #ifdef CONFIG_SCHED_CORE
 804         struct rb_node                  core_node;
 805         unsigned long                   core_cookie;
 806         unsigned int                    core_occupation;
 807 #endif
 808
 809 #ifdef CONFIG_CGROUP_SCHED
 810         struct task_group               *sched_task_group;
 811 #endif
 812
 813 #ifdef CONFIG_UCLAMP_TASK
 814         /*
 815          * Clamp values requested for a scheduling entity.
 816          * Must be updated with task_rq_lock() held.
 817          */
 818         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 819         /*
 820          * Effective clamp values used for a scheduling entity.
 821          * Must be updated with task_rq_lock() held.
 822          */
 823         struct uclamp_se                uclamp[UCLAMP_CNT];
 824 #endif
 825
 826         struct sched_statistics         stats;
 827
 828 #ifdef CONFIG_PREEMPT_NOTIFIERS
 829         /* List of struct preempt_notifier: */
 830         struct hlist_head               preempt_notifiers;
 831 #endif
 832
 833 #ifdef CONFIG_BLK_DEV_IO_TRACE
 834         unsigned int                    btrace_seq;
 835 #endif
 836
 837         unsigned int                    policy;
 838         int                             nr_cpus_allowed;
 839         const cpumask_t                 *cpus_ptr;
 840         cpumask_t                       *user_cpus_ptr;
 841         cpumask_t                       cpus_mask;
 842         void                            *migration_pending;
 843 #ifdef CONFIG_SMP
 844         unsigned short                  migration_disabled;
 845 #endif
 846         unsigned short                  migration_flags;
 847
 848 #ifdef CONFIG_PREEMPT_RCU
 849         int                             rcu_read_lock_nesting;
 850         union rcu_special               rcu_read_unlock_special;
 851         struct list_head                rcu_node_entry;
 852         struct rcu_node                 *rcu_blocked_node;
 853 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 854
 855 #ifdef CONFIG_TASKS_RCU
 856         unsigned long                   rcu_tasks_nvcsw;
 857         u8                              rcu_tasks_holdout;
 858         u8                              rcu_tasks_idx;
 859         int                             rcu_tasks_idle_cpu;
 860         struct list_head                rcu_tasks_holdout_list;
 861 #endif /* #ifdef CONFIG_TASKS_RCU */
 862
 863 #ifdef CONFIG_TASKS_TRACE_RCU
 864         int                             trc_reader_nesting;
 865         int                             trc_ipi_to_cpu;
 866         union rcu_special               trc_reader_special;
 867         struct list_head                trc_holdout_list;
 868         struct list_head                trc_blkd_node;
 869         int                             trc_blkd_cpu;
 870 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 871
 872         struct sched_info               sched_info;
 873
 874         struct list_head                tasks;
 875 #ifdef CONFIG_SMP
 876         struct plist_node               pushable_tasks;
 877         struct rb_node                  pushable_dl_tasks;
 878 #endif
 879
 880         struct mm_struct                *mm;
 881         struct mm_struct                *active_mm;
 882         struct address_space            *faults_disabled_mapping;
 883
 884         int                             exit_state;
 885         int                             exit_code;
 886         int                             exit_signal;
 887         /* The signal sent when the parent dies: */
 888         int                             pdeath_signal;
 889         /* JOBCTL_*, siglock protected: */
 890         unsigned long                   jobctl;
 891
 892         /* Used for emulating ABI behavior of previous Linux versions: */
 893         unsigned int                    personality;
 894
 895         /* Scheduler bits, serialized by scheduler locks: */
 896         unsigned                        sched_reset_on_fork:1;
 897         unsigned                        sched_contributes_to_load:1;
 898         unsigned                        sched_migrated:1;
 899
 900         /* Force alignment to the next boundary: */
 901         unsigned                        :0;
 902
 903         /* Unserialized, strictly 'current' */
 904
 905         /*
 906          * This field must not be in the scheduler word above due to wakelist
 907          * queueing no longer being serialized by p->on_cpu. However:
 908          *
 909          * p->XXX = X;                  ttwu()
 910          * schedule()                     if (p->on_rq && ..) // false
 911          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 912          *   deactivate_task()                ttwu_queue_wakelist())
 913          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 914          *
 915          * guarantees all stores of 'current' are visible before
 916          * ->sched_remote_wakeup gets used, so it can be in this word.
 917          */
 918         unsigned                        sched_remote_wakeup:1;
 919 #ifdef CONFIG_RT_MUTEXES
 920         unsigned                        sched_rt_mutex:1;
 921 #endif
 922
 923         /* Bit to tell TOMOYO we're in execve(): */
 924         unsigned                        in_execve:1;
 925         unsigned                        in_iowait:1;
 926 #ifndef TIF_RESTORE_SIGMASK
 927         unsigned                        restore_sigmask:1;
 928 #endif
 929 #ifdef CONFIG_MEMCG
 930         unsigned                        in_user_fault:1;
 931 #endif
 932 #ifdef CONFIG_LRU_GEN
 933         /* whether the LRU algorithm may apply to this access */
 934         unsigned                        in_lru_fault:1;
 935 #endif
 936 #ifdef CONFIG_COMPAT_BRK
 937         unsigned                        brk_randomized:1;
 938 #endif
 939 #ifdef CONFIG_CGROUPS
 940         /* disallow userland-initiated cgroup migration */
 941         unsigned                        no_cgroup_migration:1;
 942         /* task is frozen/stopped (used by the cgroup freezer) */
 943         unsigned                        frozen:1;
 944 #endif
 945 #ifdef CONFIG_BLK_CGROUP
 946         unsigned                        use_memdelay:1;
 947 #endif
 948 #ifdef CONFIG_PSI
 949         /* Stalled due to lack of memory */
 950         unsigned                        in_memstall:1;
 951 #endif
 952 #ifdef CONFIG_PAGE_OWNER
 953         /* Used by page_owner=on to detect recursion in page tracking. */
 954         unsigned                        in_page_owner:1;
 955 #endif
 956 #ifdef CONFIG_EVENTFD
 957         /* Recursion prevention for eventfd_signal() */
 958         unsigned                        in_eventfd:1;
 959 #endif
 960 #ifdef CONFIG_ARCH_HAS_CPU_PASID
 961         unsigned                        pasid_activated:1;
 962 #endif
 963 #ifdef  CONFIG_CPU_SUP_INTEL
 964         unsigned                        reported_split_lock:1;
 965 #endif
 966 #ifdef CONFIG_TASK_DELAY_ACCT
 967         /* delay due to memory thrashing */
 968         unsigned                        in_thrashing:1;
 969 #endif
 970
 971         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 972
 973         struct restart_block            restart_block;
 974
 975         pid_t                           pid;
 976         pid_t                           tgid;
 977
 978 #ifdef CONFIG_STACKPROTECTOR
 979         /* Canary value for the -fstack-protector GCC feature: */
 980         unsigned long                   stack_canary;
 981 #endif
 982         /*
 983          * Pointers to the (original) parent process, youngest child, younger sibling,
 984          * older sibling, respectively.  (p->father can be replaced with
 985          * p->real_parent->pid)
 986          */
 987
 988         /* Real parent process: */
 989         struct task_struct __rcu        *real_parent;
 990
 991         /* Recipient of SIGCHLD, wait4() reports: */
 992         struct task_struct __rcu        *parent;
 993
 994         /*
 995          * Children/sibling form the list of natural children:
 996          */
 997         struct list_head                children;
 998         struct list_head                sibling;
 999         struct task_struct              *group_leader;
1000
1001         /*
1002          * 'ptraced' is the list of tasks this task is using ptrace() on.
1003          *
1004          * This includes both natural children and PTRACE_ATTACH targets.
1005          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
1006          */
1007         struct list_head                ptraced;
1008         struct list_head                ptrace_entry;
1009
1010         /* PID/PID hash table linkage. */
1011         struct pid                      *thread_pid;
1012         struct hlist_node               pid_links[PIDTYPE_MAX];
1013         struct list_head                thread_node;
1014
1015         struct completion               *vfork_done;
1016
1017         /* CLONE_CHILD_SETTID: */
1018         int __user                      *set_child_tid;
1019
1020         /* CLONE_CHILD_CLEARTID: */
1021         int __user                      *clear_child_tid;
1022
1023         /* PF_KTHREAD | PF_IO_WORKER */
1024         void                            *worker_private;
1025
1026         u64                             utime;
1027         u64                             stime;
1028 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1029         u64                             utimescaled;
1030         u64                             stimescaled;
1031 #endif
1032         u64                             gtime;
1033         struct prev_cputime             prev_cputime;
1034 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1035         struct vtime                    vtime;
1036 #endif
1037
1038 #ifdef CONFIG_NO_HZ_FULL
1039         atomic_t                        tick_dep_mask;
1040 #endif
1041         /* Context switch counts: */
1042         unsigned long                   nvcsw;
1043         unsigned long                   nivcsw;
1044
1045         /* Monotonic time in nsecs: */
1046         u64                             start_time;
1047
1048         /* Boot based time in nsecs: */
1049         u64                             start_boottime;
1050
1051         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1052         unsigned long                   min_flt;
1053         unsigned long                   maj_flt;
1054
1055         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1056         struct posix_cputimers          posix_cputimers;
1057
1058 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1059         struct posix_cputimers_work     posix_cputimers_work;
1060 #endif
1061
1062         /* Process credentials: */
1063
1064         /* Tracer's credentials at attach: */
1065         const struct cred __rcu         *ptracer_cred;
1066
1067         /* Objective and real subjective task credentials (COW): */
1068         const struct cred __rcu         *real_cred;
1069
1070         /* Effective (overridable) subjective task credentials (COW): */
1071         const struct cred __rcu         *cred;
1072
1073 #ifdef CONFIG_KEYS
1074         /* Cached requested key. */
1075         struct key                      *cached_requested_key;
1076 #endif
1077
1078         /*
1079          * executable name, excluding path.
1080          *
1081          * - normally initialized setup_new_exec()
1082          * - access it with [gs]et_task_comm()
1083          * - lock it with task_lock()
1084          */
1085         char                            comm[TASK_COMM_LEN];
1086
1087         struct nameidata                *nameidata;
1088
1089 #ifdef CONFIG_SYSVIPC
1090         struct sysv_sem                 sysvsem;
1091         struct sysv_shm                 sysvshm;
1092 #endif
1093 #ifdef CONFIG_DETECT_HUNG_TASK
1094         unsigned long                   last_switch_count;
1095         unsigned long                   last_switch_time;
1096 #endif
1097         /* Filesystem information: */
1098         struct fs_struct                *fs;
1099
1100         /* Open file information: */
1101         struct files_struct             *files;
1102
1103 #ifdef CONFIG_IO_URING
1104         struct io_uring_task            *io_uring;
1105 #endif
1106
1107         /* Namespaces: */
1108         struct nsproxy                  *nsproxy;
1109
1110         /* Signal handlers: */
1111         struct signal_struct            *signal;
1112         struct sighand_struct __rcu             *sighand;
1113         sigset_t                        blocked;
1114         sigset_t                        real_blocked;
1115         /* Restored if set_restore_sigmask() was used: */
1116         sigset_t                        saved_sigmask;
1117         struct sigpending               pending;
1118         unsigned long                   sas_ss_sp;
1119         size_t                          sas_ss_size;
1120         unsigned int                    sas_ss_flags;
1121
1122         struct callback_head            *task_works;
1123
1124 #ifdef CONFIG_AUDIT
1125 #ifdef CONFIG_AUDITSYSCALL
1126         struct audit_context            *audit_context;
1127 #endif
1128         kuid_t                          loginuid;
1129         unsigned int                    sessionid;
1130 #endif
1131         struct seccomp                  seccomp;
1132         struct syscall_user_dispatch    syscall_dispatch;
1133
1134         /* Thread group tracking: */
1135         u64                             parent_exec_id;
1136         u64                             self_exec_id;
1137
1138         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1139         spinlock_t                      alloc_lock;
1140
1141         /* Protection of the PI data structures: */
1142         raw_spinlock_t                  pi_lock;
1143
1144         struct wake_q_node              wake_q;
1145
1146 #ifdef CONFIG_RT_MUTEXES
1147         /* PI waiters blocked on a rt_mutex held by this task: */
1148         struct rb_root_cached           pi_waiters;
1149         /* Updated under owner's pi_lock and rq lock */
1150         struct task_struct              *pi_top_task;
1151         /* Deadlock detection and priority inheritance handling: */
1152         struct rt_mutex_waiter          *pi_blocked_on;
1153 #endif
1154
1155 #ifdef CONFIG_DEBUG_MUTEXES
1156         /* Mutex deadlock detection: */
1157         struct mutex_waiter             *blocked_on;
1158 #endif
1159
1160 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1161         int                             non_block_count;
1162 #endif
1163
1164 #ifdef CONFIG_TRACE_IRQFLAGS
1165         struct irqtrace_events          irqtrace;
1166         unsigned int                    hardirq_threaded;
1167         u64                             hardirq_chain_key;
1168         int                             softirqs_enabled;
1169         int                             softirq_context;
1170         int                             irq_config;
1171 #endif
1172 #ifdef CONFIG_PREEMPT_RT
1173         int                             softirq_disable_cnt;
1174 #endif
1175
1176 #ifdef CONFIG_LOCKDEP
1177 # define MAX_LOCK_DEPTH                 48UL
1178         u64                             curr_chain_key;
1179         int                             lockdep_depth;
1180         unsigned int                    lockdep_recursion;
1181         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1182 #endif
1183
1184 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1185         unsigned int                    in_ubsan;
1186 #endif
1187
1188         /* Journalling filesystem info: */
1189         void                            *journal_info;
1190
1191         /* Stacked block device info: */
1192         struct bio_list                 *bio_list;
1193
1194         /* Stack plugging: */
1195         struct blk_plug                 *plug;
1196
1197         /* VM state: */
1198         struct reclaim_state            *reclaim_state;
1199
1200         struct io_context               *io_context;
1201
1202 #ifdef CONFIG_COMPACTION
1203         struct capture_control          *capture_control;
1204 #endif
1205         /* Ptrace state: */
1206         unsigned long                   ptrace_message;
1207         kernel_siginfo_t                *last_siginfo;
1208
1209         struct task_io_accounting       ioac;
1210 #ifdef CONFIG_PSI
1211         /* Pressure stall state */
1212         unsigned int                    psi_flags;
1213 #endif
1214 #ifdef CONFIG_TASK_XACCT
1215         /* Accumulated RSS usage: */
1216         u64                             acct_rss_mem1;
1217         /* Accumulated virtual memory usage: */
1218         u64                             acct_vm_mem1;
1219         /* stime + utime since last update: */
1220         u64                             acct_timexpd;
1221 #endif
1222 #ifdef CONFIG_CPUSETS
1223         /* Protected by ->alloc_lock: */
1224         nodemask_t                      mems_allowed;
1225         /* Sequence number to catch updates: */
1226         seqcount_spinlock_t             mems_allowed_seq;
1227         int                             cpuset_mem_spread_rotor;
1228         int                             cpuset_slab_spread_rotor;
1229 #endif
1230 #ifdef CONFIG_CGROUPS
1231         /* Control Group info protected by css_set_lock: */
1232         struct css_set __rcu            *cgroups;
1233         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1234         struct list_head                cg_list;
1235 #endif
1236 #ifdef CONFIG_X86_CPU_RESCTRL
1237         u32                             closid;
1238         u32                             rmid;
1239 #endif
1240 #ifdef CONFIG_FUTEX
1241         struct robust_list_head __user  *robust_list;
1242 #ifdef CONFIG_COMPAT
1243         struct compat_robust_list_head __user *compat_robust_list;
1244 #endif
1245         struct list_head                pi_state_list;
1246         struct futex_pi_state           *pi_state_cache;
1247         struct mutex                    futex_exit_mutex;
1248         unsigned int                    futex_state;
1249 #endif
1250 #ifdef CONFIG_PERF_EVENTS
1251         struct perf_event_context       *perf_event_ctxp;
1252         struct mutex                    perf_event_mutex;
1253         struct list_head                perf_event_list;
1254 #endif
1255 #ifdef CONFIG_DEBUG_PREEMPT
1256         unsigned long                   preempt_disable_ip;
1257 #endif
1258 #ifdef CONFIG_NUMA
1259         /* Protected by alloc_lock: */
1260         struct mempolicy                *mempolicy;
1261         short                           il_prev;
1262         short                           pref_node_fork;
1263 #endif
1264 #ifdef CONFIG_NUMA_BALANCING
1265         int                             numa_scan_seq;
1266         unsigned int                    numa_scan_period;
1267         unsigned int                    numa_scan_period_max;
1268         int                             numa_preferred_nid;
1269         unsigned long                   numa_migrate_retry;
1270         /* Migration stamp: */
1271         u64                             node_stamp;
1272         u64                             last_task_numa_placement;
1273         u64                             last_sum_exec_runtime;
1274         struct callback_head            numa_work;
1275
1276         /*
1277          * This pointer is only modified for current in syscall and
1278          * pagefault context (and for tasks being destroyed), so it can be read
1279          * from any of the following contexts:
1280          *  - RCU read-side critical section
1281          *  - current->numa_group from everywhere
1282          *  - task's runqueue locked, task not running
1283          */
1284         struct numa_group __rcu         *numa_group;
1285
1286         /*
1287          * numa_faults is an array split into four regions:
1288          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1289          * in this precise order.
1290          *
1291          * faults_memory: Exponential decaying average of faults on a per-node
1292          * basis. Scheduling placement decisions are made based on these
1293          * counts. The values remain static for the duration of a PTE scan.
1294          * faults_cpu: Track the nodes the process was running on when a NUMA
1295          * hinting fault was incurred.
1296          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1297          * during the current scan window. When the scan completes, the counts
1298          * in faults_memory and faults_cpu decay and these values are copied.
1299          */
1300         unsigned long                   *numa_faults;
1301         unsigned long                   total_numa_faults;
1302
1303         /*
1304          * numa_faults_locality tracks if faults recorded during the last
1305          * scan window were remote/local or failed to migrate. The task scan
1306          * period is adapted based on the locality of the faults with different
1307          * weights depending on whether they were shared or private faults
1308          */
1309         unsigned long                   numa_faults_locality[3];
1310
1311         unsigned long                   numa_pages_migrated;
1312 #endif /* CONFIG_NUMA_BALANCING */
1313
1314 #ifdef CONFIG_RSEQ
1315         struct rseq __user *rseq;
1316         u32 rseq_len;
1317         u32 rseq_sig;
1318         /*
1319          * RmW on rseq_event_mask must be performed atomically
1320          * with respect to preemption.
1321          */
1322         unsigned long rseq_event_mask;
1323 #endif
1324
1325 #ifdef CONFIG_SCHED_MM_CID
1326         int                             mm_cid;         /* Current cid in mm */
1327         int                             last_mm_cid;    /* Most recent cid in mm */
1328         int                             migrate_from_cpu;
1329         int                             mm_cid_active;  /* Whether cid bitmap is active */
1330         struct callback_head            cid_work;
1331 #endif
1332
1333         struct tlbflush_unmap_batch     tlb_ubc;
1334
1335         /* Cache last used pipe for splice(): */
1336         struct pipe_inode_info          *splice_pipe;
1337
1338         struct page_frag                task_frag;
1339
1340 #ifdef CONFIG_TASK_DELAY_ACCT
1341         struct task_delay_info          *delays;
1342 #endif
1343
1344 #ifdef CONFIG_FAULT_INJECTION
1345         int                             make_it_fail;
1346         unsigned int                    fail_nth;
1347 #endif
1348         /*
1349          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1350          * balance_dirty_pages() for a dirty throttling pause:
1351          */
1352         int                             nr_dirtied;
1353         int                             nr_dirtied_pause;
1354         /* Start of a write-and-pause period: */
1355         unsigned long                   dirty_paused_when;
1356
1357 #ifdef CONFIG_LATENCYTOP
1358         int                             latency_record_count;
1359         struct latency_record           latency_record[LT_SAVECOUNT];
1360 #endif
1361         /*
1362          * Time slack values; these are used to round up poll() and
1363          * select() etc timeout values. These are in nanoseconds.
1364          */
1365         u64                             timer_slack_ns;
1366         u64                             default_timer_slack_ns;
1367
1368 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1369         unsigned int                    kasan_depth;
1370 #endif
1371
1372 #ifdef CONFIG_KCSAN
1373         struct kcsan_ctx                kcsan_ctx;
1374 #ifdef CONFIG_TRACE_IRQFLAGS
1375         struct irqtrace_events          kcsan_save_irqtrace;
1376 #endif
1377 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1378         int                             kcsan_stack_depth;
1379 #endif
1380 #endif
1381
1382 #ifdef CONFIG_KMSAN
1383         struct kmsan_ctx                kmsan_ctx;
1384 #endif
1385
1386 #if IS_ENABLED(CONFIG_KUNIT)
1387         struct kunit                    *kunit_test;
1388 #endif
1389
1390 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1391         /* Index of current stored address in ret_stack: */
1392         int                             curr_ret_stack;
1393         int                             curr_ret_depth;
1394
1395         /* Stack of return addresses for return function tracing: */
1396         struct ftrace_ret_stack         *ret_stack;
1397
1398         /* Timestamp for last schedule: */
1399         unsigned long long              ftrace_timestamp;
1400
1401         /*
1402          * Number of functions that haven't been traced
1403          * because of depth overrun:
1404          */
1405         atomic_t                        trace_overrun;
1406
1407         /* Pause tracing: */
1408         atomic_t                        tracing_graph_pause;
1409 #endif
1410
1411 #ifdef CONFIG_TRACING
1412         /* Bitmask and counter of trace recursion: */
1413         unsigned long                   trace_recursion;
1414 #endif /* CONFIG_TRACING */
1415
1416 #ifdef CONFIG_KCOV
1417         /* See kernel/kcov.c for more details. */
1418
1419         /* Coverage collection mode enabled for this task (0 if disabled): */
1420         unsigned int                    kcov_mode;
1421
1422         /* Size of the kcov_area: */
1423         unsigned int                    kcov_size;
1424
1425         /* Buffer for coverage collection: */
1426         void                            *kcov_area;
1427
1428         /* KCOV descriptor wired with this task or NULL: */
1429         struct kcov                     *kcov;
1430
1431         /* KCOV common handle for remote coverage collection: */
1432         u64                             kcov_handle;
1433
1434         /* KCOV sequence number: */
1435         int                             kcov_sequence;
1436
1437         /* Collect coverage from softirq context: */
1438         unsigned int                    kcov_softirq;
1439 #endif
1440
1441 #ifdef CONFIG_MEMCG
1442         struct mem_cgroup               *memcg_in_oom;
1443         gfp_t                           memcg_oom_gfp_mask;
1444         int                             memcg_oom_order;
1445
1446         /* Number of pages to reclaim on returning to userland: */
1447         unsigned int                    memcg_nr_pages_over_high;
1448
1449         /* Used by memcontrol for targeted memcg charge: */
1450         struct mem_cgroup               *active_memcg;
1451 #endif
1452
1453 #ifdef CONFIG_MEMCG_KMEM
1454         struct obj_cgroup               *objcg;
1455 #endif
1456
1457 #ifdef CONFIG_BLK_CGROUP
1458         struct gendisk                  *throttle_disk;
1459 #endif
1460
1461 #ifdef CONFIG_UPROBES
1462         struct uprobe_task              *utask;
1463 #endif
1464 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1465         unsigned int                    sequential_io;
1466         unsigned int                    sequential_io_avg;
1467 #endif
1468         struct kmap_ctrl                kmap_ctrl;
1469 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1470         unsigned long                   task_state_change;
1471 # ifdef CONFIG_PREEMPT_RT
1472         unsigned long                   saved_state_change;
1473 # endif
1474 #endif
1475         struct rcu_head                 rcu;
1476         refcount_t                      rcu_users;
1477         int                             pagefault_disabled;
1478 #ifdef CONFIG_MMU
1479         struct task_struct              *oom_reaper_list;
1480         struct timer_list               oom_reaper_timer;
1481 #endif
1482 #ifdef CONFIG_VMAP_STACK
1483         struct vm_struct                *stack_vm_area;
1484 #endif
1485 #ifdef CONFIG_THREAD_INFO_IN_TASK
1486         /* A live task holds one reference: */
1487         refcount_t                      stack_refcount;
1488 #endif
1489 #ifdef CONFIG_LIVEPATCH
1490         int patch_state;
1491 #endif
1492 #ifdef CONFIG_SECURITY
1493         /* Used by LSM modules for access restriction: */
1494         void                            *security;
1495 #endif
1496 #ifdef CONFIG_BPF_SYSCALL
1497         /* Used by BPF task local storage */
1498         struct bpf_local_storage __rcu  *bpf_storage;
1499         /* Used for BPF run context */
1500         struct bpf_run_ctx              *bpf_ctx;
1501 #endif
1502
1503 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1504         unsigned long                   lowest_stack;
1505         unsigned long                   prev_lowest_stack;
1506 #endif
1507
1508 #ifdef CONFIG_X86_MCE
1509         void __user                     *mce_vaddr;
1510         __u64                           mce_kflags;
1511         u64                             mce_addr;
1512         __u64                           mce_ripv : 1,
1513                                         mce_whole_page : 1,
1514                                         __mce_reserved : 62;
1515         struct callback_head            mce_kill_me;
1516         int                             mce_count;
1517 #endif
1518
1519 #ifdef CONFIG_KRETPROBES
1520         struct llist_head               kretprobe_instances;
1521 #endif
1522 #ifdef CONFIG_RETHOOK
1523         struct llist_head               rethooks;
1524 #endif
1525
1526 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1527         /*
1528          * If L1D flush is supported on mm context switch
1529          * then we use this callback head to queue kill work
1530          * to kill tasks that are not running on SMT disabled
1531          * cores
1532          */
1533         struct callback_head            l1d_flush_kill;
1534 #endif
1535
1536 #ifdef CONFIG_RV
1537         /*
1538          * Per-task RV monitor. Nowadays fixed in RV_PER_TASK_MONITORS.
1539          * If we find justification for more monitors, we can think
1540          * about adding more or developing a dynamic method. So far,
1541          * none of these are justified.
1542          */
1543         union rv_task_monitor           rv[RV_PER_TASK_MONITORS];
1544 #endif
1545
1546 #ifdef CONFIG_USER_EVENTS
1547         struct user_event_mm            *user_event_mm;
1548 #endif
1549
1550         /*
1551          * New fields for task_struct should be added above here, so that
1552          * they are included in the randomized portion of task_struct.
1553          */
1554         randomized_struct_fields_end
1555
1556         /* CPU-specific state of this task: */
1557         struct thread_struct            thread;
1558
1559         /*
1560          * WARNING: on x86, 'thread_struct' contains a variable-sized
1561          * structure.  It *MUST* be at the end of 'task_struct'.
1562          *
1563          * Do not put anything below here!
1564          */
1565 };
1566
1567 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1568 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1569
1570 static inline unsigned int __task_state_index(unsigned int tsk_state,
1571                                               unsigned int tsk_exit_state)
1572 {
1573         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1574
1575         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1576
1577         if ((tsk_state & TASK_IDLE) == TASK_IDLE)
1578                 state = TASK_REPORT_IDLE;
1579
1580         /*
1581          * We're lying here, but rather than expose a completely new task state
1582          * to userspace, we can make this appear as if the task has gone through
1583          * a regular rt_mutex_lock() call.
1584          */
1585         if (tsk_state & TASK_RTLOCK_WAIT)
1586                 state = TASK_UNINTERRUPTIBLE;
1587
1588         return fls(state);
1589 }
1590
1591 static inline unsigned int task_state_index(struct task_struct *tsk)
1592 {
1593         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1594 }
1595
1596 static inline char task_index_to_char(unsigned int state)
1597 {
1598         static const char state_char[] = "RSDTtXZPI";
1599
1600         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1601
1602         return state_char[state];
1603 }
1604
1605 static inline char task_state_to_char(struct task_struct *tsk)
1606 {
1607         return task_index_to_char(task_state_index(tsk));
1608 }
1609
1610 extern struct pid *cad_pid;
1611
1612 /*
1613  * Per process flags
1614  */
1615 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1616 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1617 #define PF_EXITING              0x00000004      /* Getting shut down */
1618 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1619 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1620 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1621 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1622 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1623 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1624 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1625 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1626 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1627 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1628 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1629 #define PF_USER_WORKER          0x00004000      /* Kernel thread cloned from userspace thread */
1630 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1631 #define PF__HOLE__00010000      0x00010000
1632 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1633 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1634 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1635 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1636                                                  * I am cleaning dirty pages from some other bdi. */
1637 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1638 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1639 #define PF__HOLE__00800000      0x00800000
1640 #define PF__HOLE__01000000      0x01000000
1641 #define PF__HOLE__02000000      0x02000000
1642 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1643 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1644 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1645 #define PF__HOLE__20000000      0x20000000
1646 #define PF__HOLE__40000000      0x40000000
1647 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1648
1649 /*
1650  * Only the _current_ task can read/write to tsk->flags, but other
1651  * tasks can access tsk->flags in readonly mode for example
1652  * with tsk_used_math (like during threaded core dumping).
1653  * There is however an exception to this rule during ptrace
1654  * or during fork: the ptracer task is allowed to write to the
1655  * child->flags of its traced child (same goes for fork, the parent
1656  * can write to the child->flags), because we're guaranteed the
1657  * child is not running and in turn not changing child->flags
1658  * at the same time the parent does it.
1659  */
1660 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1661 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1662 #define clear_used_math()                       clear_stopped_child_used_math(current)
1663 #define set_used_math()                         set_stopped_child_used_math(current)
1664
1665 #define conditional_stopped_child_used_math(condition, child) \
1666         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1667
1668 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1669
1670 #define copy_to_stopped_child_used_math(child) \
1671         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1672
1673 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1674 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1675 #define used_math()                             tsk_used_math(current)
1676
1677 static __always_inline bool is_percpu_thread(void)
1678 {
1679 #ifdef CONFIG_SMP
1680         return (current->flags & PF_NO_SETAFFINITY) &&
1681                 (current->nr_cpus_allowed  == 1);
1682 #else
1683         return true;
1684 #endif
1685 }
1686
1687 /* Per-process atomic flags. */
1688 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1689 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1690 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1691 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1692 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1693 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1694 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1695 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1696
1697 #define TASK_PFA_TEST(name, func)                                       \
1698         static inline bool task_##func(struct task_struct *p)           \
1699         { return test_bit(PFA_##name, &p->atomic_flags); }
1700
1701 #define TASK_PFA_SET(name, func)                                        \
1702         static inline void task_set_##func(struct task_struct *p)       \
1703         { set_bit(PFA_##name, &p->atomic_flags); }
1704
1705 #define TASK_PFA_CLEAR(name, func)                                      \
1706         static inline void task_clear_##func(struct task_struct *p)     \
1707         { clear_bit(PFA_##name, &p->atomic_flags); }
1708
1709 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1710 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1711
1712 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1713 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1714 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1715
1716 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1717 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1718 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1719
1720 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1721 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1722 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1723
1724 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1725 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1726 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1727
1728 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1729 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1730
1731 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1732 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1733 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1734
1735 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1736 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1737
1738 static inline void
1739 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1740 {
1741         current->flags &= ~flags;
1742         current->flags |= orig_flags & flags;
1743 }
1744
1745 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1746 extern int task_can_attach(struct task_struct *p);
1747 extern int dl_bw_alloc(int cpu, u64 dl_bw);
1748 extern void dl_bw_free(int cpu, u64 dl_bw);
1749 #ifdef CONFIG_SMP
1750
1751 /* do_set_cpus_allowed() - consider using set_cpus_allowed_ptr() instead */
1752 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1753
1754 /**
1755  * set_cpus_allowed_ptr - set CPU affinity mask of a task
1756  * @p: the task
1757  * @new_mask: CPU affinity mask
1758  *
1759  * Return: zero if successful, or a negative error code
1760  */
1761 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1762 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1763 extern void release_user_cpus_ptr(struct task_struct *p);
1764 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1765 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1766 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1767 #else
1768 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1769 {
1770 }
1771 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1772 {
1773         if (!cpumask_test_cpu(0, new_mask))
1774                 return -EINVAL;
1775         return 0;
1776 }
1777 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1778 {
1779         if (src->user_cpus_ptr)
1780                 return -EINVAL;
1781         return 0;
1782 }
1783 static inline void release_user_cpus_ptr(struct task_struct *p)
1784 {
1785         WARN_ON(p->user_cpus_ptr);
1786 }
1787
1788 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1789 {
1790         return 0;
1791 }
1792 #endif
1793
1794 extern int yield_to(struct task_struct *p, bool preempt);
1795 extern void set_user_nice(struct task_struct *p, long nice);
1796 extern int task_prio(const struct task_struct *p);
1797
1798 /**
1799  * task_nice - return the nice value of a given task.
1800  * @p: the task in question.
1801  *
1802  * Return: The nice value [ -20 ... 0 ... 19 ].
1803  */
1804 static inline int task_nice(const struct task_struct *p)
1805 {
1806         return PRIO_TO_NICE((p)->static_prio);
1807 }
1808
1809 extern int can_nice(const struct task_struct *p, const int nice);
1810 extern int task_curr(const struct task_struct *p);
1811 extern int idle_cpu(int cpu);
1812 extern int available_idle_cpu(int cpu);
1813 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1814 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1815 extern void sched_set_fifo(struct task_struct *p);
1816 extern void sched_set_fifo_low(struct task_struct *p);
1817 extern void sched_set_normal(struct task_struct *p, int nice);
1818 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1819 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1820 extern struct task_struct *idle_task(int cpu);
1821
1822 /**
1823  * is_idle_task - is the specified task an idle task?
1824  * @p: the task in question.
1825  *
1826  * Return: 1 if @p is an idle task. 0 otherwise.
1827  */
1828 static __always_inline bool is_idle_task(const struct task_struct *p)
1829 {
1830         return !!(p->flags & PF_IDLE);
1831 }
1832
1833 extern struct task_struct *curr_task(int cpu);
1834 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1835
1836 void yield(void);
1837
1838 union thread_union {
1839         struct task_struct task;
1840 #ifndef CONFIG_THREAD_INFO_IN_TASK
1841         struct thread_info thread_info;
1842 #endif
1843         unsigned long stack[THREAD_SIZE/sizeof(long)];
1844 };
1845
1846 #ifndef CONFIG_THREAD_INFO_IN_TASK
1847 extern struct thread_info init_thread_info;
1848 #endif
1849
1850 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1851
1852 #ifdef CONFIG_THREAD_INFO_IN_TASK
1853 # define task_thread_info(task) (&(task)->thread_info)
1854 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1855 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1856 #endif
1857
1858 /*
1859  * find a task by one of its numerical ids
1860  *
1861  * find_task_by_pid_ns():
1862  *      finds a task by its pid in the specified namespace
1863  * find_task_by_vpid():
1864  *      finds a task by its virtual pid
1865  *
1866  * see also find_vpid() etc in include/linux/pid.h
1867  */
1868
1869 extern struct task_struct *find_task_by_vpid(pid_t nr);
1870 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1871
1872 /*
1873  * find a task by its virtual pid and get the task struct
1874  */
1875 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1876
1877 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1878 extern int wake_up_process(struct task_struct *tsk);
1879 extern void wake_up_new_task(struct task_struct *tsk);
1880
1881 #ifdef CONFIG_SMP
1882 extern void kick_process(struct task_struct *tsk);
1883 #else
1884 static inline void kick_process(struct task_struct *tsk) { }
1885 #endif
1886
1887 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1888
1889 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1890 {
1891         __set_task_comm(tsk, from, false);
1892 }
1893
1894 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1895 #define get_task_comm(buf, tsk) ({                      \
1896         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1897         __get_task_comm(buf, sizeof(buf), tsk);         \
1898 })
1899
1900 #ifdef CONFIG_SMP
1901 static __always_inline void scheduler_ipi(void)
1902 {
1903         /*
1904          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1905          * TIF_NEED_RESCHED remotely (for the first time) will also send
1906          * this IPI.
1907          */
1908         preempt_fold_need_resched();
1909 }
1910 #else
1911 static inline void scheduler_ipi(void) { }
1912 #endif
1913
1914 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1915
1916 /*
1917  * Set thread flags in other task's structures.
1918  * See asm/thread_info.h for TIF_xxxx flags available:
1919  */
1920 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1921 {
1922         set_ti_thread_flag(task_thread_info(tsk), flag);
1923 }
1924
1925 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1926 {
1927         clear_ti_thread_flag(task_thread_info(tsk), flag);
1928 }
1929
1930 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1931                                           bool value)
1932 {
1933         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1934 }
1935
1936 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1937 {
1938         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1939 }
1940
1941 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1942 {
1943         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1944 }
1945
1946 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1947 {
1948         return test_ti_thread_flag(task_thread_info(tsk), flag);
1949 }
1950
1951 static inline void set_tsk_need_resched(struct task_struct *tsk)
1952 {
1953         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1954 }
1955
1956 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1957 {
1958         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1959 }
1960
1961 static inline int test_tsk_need_resched(struct task_struct *tsk)
1962 {
1963         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
1964 }
1965
1966 /*
1967  * cond_resched() and cond_resched_lock(): latency reduction via
1968  * explicit rescheduling in places that are safe. The return
1969  * value indicates whether a reschedule was done in fact.
1970  * cond_resched_lock() will drop the spinlock before scheduling,
1971  */
1972 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
1973 extern int __cond_resched(void);
1974
1975 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
1976
1977 void sched_dynamic_klp_enable(void);
1978 void sched_dynamic_klp_disable(void);
1979
1980 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
1981
1982 static __always_inline int _cond_resched(void)
1983 {
1984         return static_call_mod(cond_resched)();
1985 }
1986
1987 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
1988
1989 extern int dynamic_cond_resched(void);
1990
1991 static __always_inline int _cond_resched(void)
1992 {
1993         return dynamic_cond_resched();
1994 }
1995
1996 #else /* !CONFIG_PREEMPTION */
1997
1998 static inline int _cond_resched(void)
1999 {
2000         klp_sched_try_switch();
2001         return __cond_resched();
2002 }
2003
2004 #endif /* PREEMPT_DYNAMIC && CONFIG_HAVE_PREEMPT_DYNAMIC_CALL */
2005
2006 #else /* CONFIG_PREEMPTION && !CONFIG_PREEMPT_DYNAMIC */
2007
2008 static inline int _cond_resched(void)
2009 {
2010         klp_sched_try_switch();
2011         return 0;
2012 }
2013
2014 #endif /* !CONFIG_PREEMPTION || CONFIG_PREEMPT_DYNAMIC */
2015
2016 #define cond_resched() ({                       \
2017         __might_resched(__FILE__, __LINE__, 0); \
2018         _cond_resched();                        \
2019 })
2020
2021 extern int __cond_resched_lock(spinlock_t *lock);
2022 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2023 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2024
2025 #define MIGHT_RESCHED_RCU_SHIFT         8
2026 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2027
2028 #ifndef CONFIG_PREEMPT_RT
2029 /*
2030  * Non RT kernels have an elevated preempt count due to the held lock,
2031  * but are not allowed to be inside a RCU read side critical section
2032  */
2033 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2034 #else
2035 /*
2036  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2037  * cond_resched*lock() has to take that into account because it checks for
2038  * preempt_count() and rcu_preempt_depth().
2039  */
2040 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2041         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2042 #endif
2043
2044 #define cond_resched_lock(lock) ({                                              \
2045         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2046         __cond_resched_lock(lock);                                              \
2047 })
2048
2049 #define cond_resched_rwlock_read(lock) ({                                       \
2050         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2051         __cond_resched_rwlock_read(lock);                                       \
2052 })
2053
2054 #define cond_resched_rwlock_write(lock) ({                                      \
2055         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2056         __cond_resched_rwlock_write(lock);                                      \
2057 })
2058
2059 #ifdef CONFIG_PREEMPT_DYNAMIC
2060
2061 extern bool preempt_model_none(void);
2062 extern bool preempt_model_voluntary(void);
2063 extern bool preempt_model_full(void);
2064
2065 #else
2066
2067 static inline bool preempt_model_none(void)
2068 {
2069         return IS_ENABLED(CONFIG_PREEMPT_NONE);
2070 }
2071 static inline bool preempt_model_voluntary(void)
2072 {
2073         return IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY);
2074 }
2075 static inline bool preempt_model_full(void)
2076 {
2077         return IS_ENABLED(CONFIG_PREEMPT);
2078 }
2079
2080 #endif
2081
2082 static inline bool preempt_model_rt(void)
2083 {
2084         return IS_ENABLED(CONFIG_PREEMPT_RT);
2085 }
2086
2087 /*
2088  * Does the preemption model allow non-cooperative preemption?
2089  *
2090  * For !CONFIG_PREEMPT_DYNAMIC kernels this is an exact match with
2091  * CONFIG_PREEMPTION; for CONFIG_PREEMPT_DYNAMIC this doesn't work as the
2092  * kernel is *built* with CONFIG_PREEMPTION=y but may run with e.g. the
2093  * PREEMPT_NONE model.
2094  */
2095 static inline bool preempt_model_preemptible(void)
2096 {
2097         return preempt_model_full() || preempt_model_rt();
2098 }
2099
2100 static __always_inline bool need_resched(void)
2101 {
2102         return unlikely(tif_need_resched());
2103 }
2104
2105 /*
2106  * Wrappers for p->thread_info->cpu access. No-op on UP.
2107  */
2108 #ifdef CONFIG_SMP
2109
2110 static inline unsigned int task_cpu(const struct task_struct *p)
2111 {
2112         return READ_ONCE(task_thread_info(p)->cpu);
2113 }
2114
2115 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2116
2117 #else
2118
2119 static inline unsigned int task_cpu(const struct task_struct *p)
2120 {
2121         return 0;
2122 }
2123
2124 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2125 {
2126 }
2127
2128 #endif /* CONFIG_SMP */
2129
2130 extern bool sched_task_on_rq(struct task_struct *p);
2131 extern unsigned long get_wchan(struct task_struct *p);
2132 extern struct task_struct *cpu_curr_snapshot(int cpu);
2133
2134 #include <linux/spinlock.h>
2135
2136 /*
2137  * In order to reduce various lock holder preemption latencies provide an
2138  * interface to see if a vCPU is currently running or not.
2139  *
2140  * This allows us to terminate optimistic spin loops and block, analogous to
2141  * the native optimistic spin heuristic of testing if the lock owner task is
2142  * running or not.
2143  */
2144 #ifndef vcpu_is_preempted
2145 static inline bool vcpu_is_preempted(int cpu)
2146 {
2147         return false;
2148 }
2149 #endif
2150
2151 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2152 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2153
2154 #ifndef TASK_SIZE_OF
2155 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2156 #endif
2157
2158 #ifdef CONFIG_SMP
2159 static inline bool owner_on_cpu(struct task_struct *owner)
2160 {
2161         /*
2162          * As lock holder preemption issue, we both skip spinning if
2163          * task is not on cpu or its cpu is preempted
2164          */
2165         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2166 }
2167
2168 /* Returns effective CPU energy utilization, as seen by the scheduler */
2169 unsigned long sched_cpu_util(int cpu);
2170 #endif /* CONFIG_SMP */
2171
2172 #ifdef CONFIG_SCHED_CORE
2173 extern void sched_core_free(struct task_struct *tsk);
2174 extern void sched_core_fork(struct task_struct *p);
2175 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2176                                 unsigned long uaddr);
2177 extern int sched_core_idle_cpu(int cpu);
2178 #else
2179 static inline void sched_core_free(struct task_struct *tsk) { }
2180 static inline void sched_core_fork(struct task_struct *p) { }
2181 static inline int sched_core_idle_cpu(int cpu) { return idle_cpu(cpu); }
2182 #endif
2183
2184 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2185
2186 #endif