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_MEM_ALLOC_PROFILING
 774         struct alloc_tag                *alloc_tag;
 775 #endif
 776
 777 #ifdef CONFIG_SMP
 778         int                             on_cpu;
 779         struct __call_single_node       wake_entry;
 780         unsigned int                    wakee_flips;
 781         unsigned long                   wakee_flip_decay_ts;
 782         struct task_struct              *last_wakee;
 783
 784         /*
 785          * recent_used_cpu is initially set as the last CPU used by a task
 786          * that wakes affine another task. Waker/wakee relationships can
 787          * push tasks around a CPU where each wakeup moves to the next one.
 788          * Tracking a recently used CPU allows a quick search for a recently
 789          * used CPU that may be idle.
 790          */
 791         int                             recent_used_cpu;
 792         int                             wake_cpu;
 793 #endif
 794         int                             on_rq;
 795
 796         int                             prio;
 797         int                             static_prio;
 798         int                             normal_prio;
 799         unsigned int                    rt_priority;
 800
 801         struct sched_entity             se;
 802         struct sched_rt_entity          rt;
 803         struct sched_dl_entity          dl;
 804         struct sched_dl_entity          *dl_server;
 805         const struct sched_class        *sched_class;
 806
 807 #ifdef CONFIG_SCHED_CORE
 808         struct rb_node                  core_node;
 809         unsigned long                   core_cookie;
 810         unsigned int                    core_occupation;
 811 #endif
 812
 813 #ifdef CONFIG_CGROUP_SCHED
 814         struct task_group               *sched_task_group;
 815 #endif
 816
 817
 818 #ifdef CONFIG_UCLAMP_TASK
 819         /*
 820          * Clamp values requested for a scheduling entity.
 821          * Must be updated with task_rq_lock() held.
 822          */
 823         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 824         /*
 825          * Effective clamp values used for a scheduling entity.
 826          * Must be updated with task_rq_lock() held.
 827          */
 828         struct uclamp_se                uclamp[UCLAMP_CNT];
 829 #endif
 830
 831         struct sched_statistics         stats;
 832
 833 #ifdef CONFIG_PREEMPT_NOTIFIERS
 834         /* List of struct preempt_notifier: */
 835         struct hlist_head               preempt_notifiers;
 836 #endif
 837
 838 #ifdef CONFIG_BLK_DEV_IO_TRACE
 839         unsigned int                    btrace_seq;
 840 #endif
 841
 842         unsigned int                    policy;
 843         int                             nr_cpus_allowed;
 844         const cpumask_t                 *cpus_ptr;
 845         cpumask_t                       *user_cpus_ptr;
 846         cpumask_t                       cpus_mask;
 847         void                            *migration_pending;
 848 #ifdef CONFIG_SMP
 849         unsigned short                  migration_disabled;
 850 #endif
 851         unsigned short                  migration_flags;
 852
 853 #ifdef CONFIG_PREEMPT_RCU
 854         int                             rcu_read_lock_nesting;
 855         union rcu_special               rcu_read_unlock_special;
 856         struct list_head                rcu_node_entry;
 857         struct rcu_node                 *rcu_blocked_node;
 858 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 859
 860 #ifdef CONFIG_TASKS_RCU
 861         unsigned long                   rcu_tasks_nvcsw;
 862         u8                              rcu_tasks_holdout;
 863         u8                              rcu_tasks_idx;
 864         int                             rcu_tasks_idle_cpu;
 865         struct list_head                rcu_tasks_holdout_list;
 866         int                             rcu_tasks_exit_cpu;
 867         struct list_head                rcu_tasks_exit_list;
 868 #endif /* #ifdef CONFIG_TASKS_RCU */
 869
 870 #ifdef CONFIG_TASKS_TRACE_RCU
 871         int                             trc_reader_nesting;
 872         int                             trc_ipi_to_cpu;
 873         union rcu_special               trc_reader_special;
 874         struct list_head                trc_holdout_list;
 875         struct list_head                trc_blkd_node;
 876         int                             trc_blkd_cpu;
 877 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 878
 879         struct sched_info               sched_info;
 880
 881         struct list_head                tasks;
 882 #ifdef CONFIG_SMP
 883         struct plist_node               pushable_tasks;
 884         struct rb_node                  pushable_dl_tasks;
 885 #endif
 886
 887         struct mm_struct                *mm;
 888         struct mm_struct                *active_mm;
 889         struct address_space            *faults_disabled_mapping;
 890
 891         int                             exit_state;
 892         int                             exit_code;
 893         int                             exit_signal;
 894         /* The signal sent when the parent dies: */
 895         int                             pdeath_signal;
 896         /* JOBCTL_*, siglock protected: */
 897         unsigned long                   jobctl;
 898
 899         /* Used for emulating ABI behavior of previous Linux versions: */
 900         unsigned int                    personality;
 901
 902         /* Scheduler bits, serialized by scheduler locks: */
 903         unsigned                        sched_reset_on_fork:1;
 904         unsigned                        sched_contributes_to_load:1;
 905         unsigned                        sched_migrated:1;
 906
 907         /* Force alignment to the next boundary: */
 908         unsigned                        :0;
 909
 910         /* Unserialized, strictly 'current' */
 911
 912         /*
 913          * This field must not be in the scheduler word above due to wakelist
 914          * queueing no longer being serialized by p->on_cpu. However:
 915          *
 916          * p->XXX = X;                  ttwu()
 917          * schedule()                     if (p->on_rq && ..) // false
 918          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 919          *   deactivate_task()                ttwu_queue_wakelist())
 920          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 921          *
 922          * guarantees all stores of 'current' are visible before
 923          * ->sched_remote_wakeup gets used, so it can be in this word.
 924          */
 925         unsigned                        sched_remote_wakeup:1;
 926 #ifdef CONFIG_RT_MUTEXES
 927         unsigned                        sched_rt_mutex:1;
 928 #endif
 929
 930         /* Bit to tell TOMOYO we're in execve(): */
 931         unsigned                        in_execve:1;
 932         unsigned                        in_iowait:1;
 933 #ifndef TIF_RESTORE_SIGMASK
 934         unsigned                        restore_sigmask:1;
 935 #endif
 936 #ifdef CONFIG_MEMCG
 937         unsigned                        in_user_fault:1;
 938 #endif
 939 #ifdef CONFIG_LRU_GEN
 940         /* whether the LRU algorithm may apply to this access */
 941         unsigned                        in_lru_fault:1;
 942 #endif
 943 #ifdef CONFIG_COMPAT_BRK
 944         unsigned                        brk_randomized:1;
 945 #endif
 946 #ifdef CONFIG_CGROUPS
 947         /* disallow userland-initiated cgroup migration */
 948         unsigned                        no_cgroup_migration:1;
 949         /* task is frozen/stopped (used by the cgroup freezer) */
 950         unsigned                        frozen:1;
 951 #endif
 952 #ifdef CONFIG_BLK_CGROUP
 953         unsigned                        use_memdelay:1;
 954 #endif
 955 #ifdef CONFIG_PSI
 956         /* Stalled due to lack of memory */
 957         unsigned                        in_memstall:1;
 958 #endif
 959 #ifdef CONFIG_PAGE_OWNER
 960         /* Used by page_owner=on to detect recursion in page tracking. */
 961         unsigned                        in_page_owner:1;
 962 #endif
 963 #ifdef CONFIG_EVENTFD
 964         /* Recursion prevention for eventfd_signal() */
 965         unsigned                        in_eventfd:1;
 966 #endif
 967 #ifdef CONFIG_ARCH_HAS_CPU_PASID
 968         unsigned                        pasid_activated:1;
 969 #endif
 970 #ifdef  CONFIG_CPU_SUP_INTEL
 971         unsigned                        reported_split_lock:1;
 972 #endif
 973 #ifdef CONFIG_TASK_DELAY_ACCT
 974         /* delay due to memory thrashing */
 975         unsigned                        in_thrashing:1;
 976 #endif
 977
 978         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 979
 980         struct restart_block            restart_block;
 981
 982         pid_t                           pid;
 983         pid_t                           tgid;
 984
 985 #ifdef CONFIG_STACKPROTECTOR
 986         /* Canary value for the -fstack-protector GCC feature: */
 987         unsigned long                   stack_canary;
 988 #endif
 989         /*
 990          * Pointers to the (original) parent process, youngest child, younger sibling,
 991          * older sibling, respectively.  (p->father can be replaced with
 992          * p->real_parent->pid)
 993          */
 994
 995         /* Real parent process: */
 996         struct task_struct __rcu        *real_parent;
 997
 998         /* Recipient of SIGCHLD, wait4() reports: */
 999         struct task_struct __rcu        *parent;
1000
1001         /*
1002          * Children/sibling form the list of natural children:
1003          */
1004         struct list_head                children;
1005         struct list_head                sibling;
1006         struct task_struct              *group_leader;
1007
1008         /*
1009          * 'ptraced' is the list of tasks this task is using ptrace() on.
1010          *
1011          * This includes both natural children and PTRACE_ATTACH targets.
1012          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
1013          */
1014         struct list_head                ptraced;
1015         struct list_head                ptrace_entry;
1016
1017         /* PID/PID hash table linkage. */
1018         struct pid                      *thread_pid;
1019         struct hlist_node               pid_links[PIDTYPE_MAX];
1020         struct list_head                thread_node;
1021
1022         struct completion               *vfork_done;
1023
1024         /* CLONE_CHILD_SETTID: */
1025         int __user                      *set_child_tid;
1026
1027         /* CLONE_CHILD_CLEARTID: */
1028         int __user                      *clear_child_tid;
1029
1030         /* PF_KTHREAD | PF_IO_WORKER */
1031         void                            *worker_private;
1032
1033         u64                             utime;
1034         u64                             stime;
1035 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1036         u64                             utimescaled;
1037         u64                             stimescaled;
1038 #endif
1039         u64                             gtime;
1040         struct prev_cputime             prev_cputime;
1041 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1042         struct vtime                    vtime;
1043 #endif
1044
1045 #ifdef CONFIG_NO_HZ_FULL
1046         atomic_t                        tick_dep_mask;
1047 #endif
1048         /* Context switch counts: */
1049         unsigned long                   nvcsw;
1050         unsigned long                   nivcsw;
1051
1052         /* Monotonic time in nsecs: */
1053         u64                             start_time;
1054
1055         /* Boot based time in nsecs: */
1056         u64                             start_boottime;
1057
1058         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1059         unsigned long                   min_flt;
1060         unsigned long                   maj_flt;
1061
1062         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1063         struct posix_cputimers          posix_cputimers;
1064
1065 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1066         struct posix_cputimers_work     posix_cputimers_work;
1067 #endif
1068
1069         /* Process credentials: */
1070
1071         /* Tracer's credentials at attach: */
1072         const struct cred __rcu         *ptracer_cred;
1073
1074         /* Objective and real subjective task credentials (COW): */
1075         const struct cred __rcu         *real_cred;
1076
1077         /* Effective (overridable) subjective task credentials (COW): */
1078         const struct cred __rcu         *cred;
1079
1080 #ifdef CONFIG_KEYS
1081         /* Cached requested key. */
1082         struct key                      *cached_requested_key;
1083 #endif
1084
1085         /*
1086          * executable name, excluding path.
1087          *
1088          * - normally initialized setup_new_exec()
1089          * - access it with [gs]et_task_comm()
1090          * - lock it with task_lock()
1091          */
1092         char                            comm[TASK_COMM_LEN];
1093
1094         struct nameidata                *nameidata;
1095
1096 #ifdef CONFIG_SYSVIPC
1097         struct sysv_sem                 sysvsem;
1098         struct sysv_shm                 sysvshm;
1099 #endif
1100 #ifdef CONFIG_DETECT_HUNG_TASK
1101         unsigned long                   last_switch_count;
1102         unsigned long                   last_switch_time;
1103 #endif
1104         /* Filesystem information: */
1105         struct fs_struct                *fs;
1106
1107         /* Open file information: */
1108         struct files_struct             *files;
1109
1110 #ifdef CONFIG_IO_URING
1111         struct io_uring_task            *io_uring;
1112 #endif
1113
1114         /* Namespaces: */
1115         struct nsproxy                  *nsproxy;
1116
1117         /* Signal handlers: */
1118         struct signal_struct            *signal;
1119         struct sighand_struct __rcu             *sighand;
1120         sigset_t                        blocked;
1121         sigset_t                        real_blocked;
1122         /* Restored if set_restore_sigmask() was used: */
1123         sigset_t                        saved_sigmask;
1124         struct sigpending               pending;
1125         unsigned long                   sas_ss_sp;
1126         size_t                          sas_ss_size;
1127         unsigned int                    sas_ss_flags;
1128
1129         struct callback_head            *task_works;
1130
1131 #ifdef CONFIG_AUDIT
1132 #ifdef CONFIG_AUDITSYSCALL
1133         struct audit_context            *audit_context;
1134 #endif
1135         kuid_t                          loginuid;
1136         unsigned int                    sessionid;
1137 #endif
1138         struct seccomp                  seccomp;
1139         struct syscall_user_dispatch    syscall_dispatch;
1140
1141         /* Thread group tracking: */
1142         u64                             parent_exec_id;
1143         u64                             self_exec_id;
1144
1145         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1146         spinlock_t                      alloc_lock;
1147
1148         /* Protection of the PI data structures: */
1149         raw_spinlock_t                  pi_lock;
1150
1151         struct wake_q_node              wake_q;
1152
1153 #ifdef CONFIG_RT_MUTEXES
1154         /* PI waiters blocked on a rt_mutex held by this task: */
1155         struct rb_root_cached           pi_waiters;
1156         /* Updated under owner's pi_lock and rq lock */
1157         struct task_struct              *pi_top_task;
1158         /* Deadlock detection and priority inheritance handling: */
1159         struct rt_mutex_waiter          *pi_blocked_on;
1160 #endif
1161
1162 #ifdef CONFIG_DEBUG_MUTEXES
1163         /* Mutex deadlock detection: */
1164         struct mutex_waiter             *blocked_on;
1165 #endif
1166
1167 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1168         int                             non_block_count;
1169 #endif
1170
1171 #ifdef CONFIG_TRACE_IRQFLAGS
1172         struct irqtrace_events          irqtrace;
1173         unsigned int                    hardirq_threaded;
1174         u64                             hardirq_chain_key;
1175         int                             softirqs_enabled;
1176         int                             softirq_context;
1177         int                             irq_config;
1178 #endif
1179 #ifdef CONFIG_PREEMPT_RT
1180         int                             softirq_disable_cnt;
1181 #endif
1182
1183 #ifdef CONFIG_LOCKDEP
1184 # define MAX_LOCK_DEPTH                 48UL
1185         u64                             curr_chain_key;
1186         int                             lockdep_depth;
1187         unsigned int                    lockdep_recursion;
1188         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1189 #endif
1190
1191 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1192         unsigned int                    in_ubsan;
1193 #endif
1194
1195         /* Journalling filesystem info: */
1196         void                            *journal_info;
1197
1198         /* Stacked block device info: */
1199         struct bio_list                 *bio_list;
1200
1201         /* Stack plugging: */
1202         struct blk_plug                 *plug;
1203
1204         /* VM state: */
1205         struct reclaim_state            *reclaim_state;
1206
1207         struct io_context               *io_context;
1208
1209 #ifdef CONFIG_COMPACTION
1210         struct capture_control          *capture_control;
1211 #endif
1212         /* Ptrace state: */
1213         unsigned long                   ptrace_message;
1214         kernel_siginfo_t                *last_siginfo;
1215
1216         struct task_io_accounting       ioac;
1217 #ifdef CONFIG_PSI
1218         /* Pressure stall state */
1219         unsigned int                    psi_flags;
1220 #endif
1221 #ifdef CONFIG_TASK_XACCT
1222         /* Accumulated RSS usage: */
1223         u64                             acct_rss_mem1;
1224         /* Accumulated virtual memory usage: */
1225         u64                             acct_vm_mem1;
1226         /* stime + utime since last update: */
1227         u64                             acct_timexpd;
1228 #endif
1229 #ifdef CONFIG_CPUSETS
1230         /* Protected by ->alloc_lock: */
1231         nodemask_t                      mems_allowed;
1232         /* Sequence number to catch updates: */
1233         seqcount_spinlock_t             mems_allowed_seq;
1234         int                             cpuset_mem_spread_rotor;
1235         int                             cpuset_slab_spread_rotor;
1236 #endif
1237 #ifdef CONFIG_CGROUPS
1238         /* Control Group info protected by css_set_lock: */
1239         struct css_set __rcu            *cgroups;
1240         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1241         struct list_head                cg_list;
1242 #endif
1243 #ifdef CONFIG_X86_CPU_RESCTRL
1244         u32                             closid;
1245         u32                             rmid;
1246 #endif
1247 #ifdef CONFIG_FUTEX
1248         struct robust_list_head __user  *robust_list;
1249 #ifdef CONFIG_COMPAT
1250         struct compat_robust_list_head __user *compat_robust_list;
1251 #endif
1252         struct list_head                pi_state_list;
1253         struct futex_pi_state           *pi_state_cache;
1254         struct mutex                    futex_exit_mutex;
1255         unsigned int                    futex_state;
1256 #endif
1257 #ifdef CONFIG_PERF_EVENTS
1258         struct perf_event_context       *perf_event_ctxp;
1259         struct mutex                    perf_event_mutex;
1260         struct list_head                perf_event_list;
1261 #endif
1262 #ifdef CONFIG_DEBUG_PREEMPT
1263         unsigned long                   preempt_disable_ip;
1264 #endif
1265 #ifdef CONFIG_NUMA
1266         /* Protected by alloc_lock: */
1267         struct mempolicy                *mempolicy;
1268         short                           il_prev;
1269         u8                              il_weight;
1270         short                           pref_node_fork;
1271 #endif
1272 #ifdef CONFIG_NUMA_BALANCING
1273         int                             numa_scan_seq;
1274         unsigned int                    numa_scan_period;
1275         unsigned int                    numa_scan_period_max;
1276         int                             numa_preferred_nid;
1277         unsigned long                   numa_migrate_retry;
1278         /* Migration stamp: */
1279         u64                             node_stamp;
1280         u64                             last_task_numa_placement;
1281         u64                             last_sum_exec_runtime;
1282         struct callback_head            numa_work;
1283
1284         /*
1285          * This pointer is only modified for current in syscall and
1286          * pagefault context (and for tasks being destroyed), so it can be read
1287          * from any of the following contexts:
1288          *  - RCU read-side critical section
1289          *  - current->numa_group from everywhere
1290          *  - task's runqueue locked, task not running
1291          */
1292         struct numa_group __rcu         *numa_group;
1293
1294         /*
1295          * numa_faults is an array split into four regions:
1296          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1297          * in this precise order.
1298          *
1299          * faults_memory: Exponential decaying average of faults on a per-node
1300          * basis. Scheduling placement decisions are made based on these
1301          * counts. The values remain static for the duration of a PTE scan.
1302          * faults_cpu: Track the nodes the process was running on when a NUMA
1303          * hinting fault was incurred.
1304          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1305          * during the current scan window. When the scan completes, the counts
1306          * in faults_memory and faults_cpu decay and these values are copied.
1307          */
1308         unsigned long                   *numa_faults;
1309         unsigned long                   total_numa_faults;
1310
1311         /*
1312          * numa_faults_locality tracks if faults recorded during the last
1313          * scan window were remote/local or failed to migrate. The task scan
1314          * period is adapted based on the locality of the faults with different
1315          * weights depending on whether they were shared or private faults
1316          */
1317         unsigned long                   numa_faults_locality[3];
1318
1319         unsigned long                   numa_pages_migrated;
1320 #endif /* CONFIG_NUMA_BALANCING */
1321
1322 #ifdef CONFIG_RSEQ
1323         struct rseq __user *rseq;
1324         u32 rseq_len;
1325         u32 rseq_sig;
1326         /*
1327          * RmW on rseq_event_mask must be performed atomically
1328          * with respect to preemption.
1329          */
1330         unsigned long rseq_event_mask;
1331 #endif
1332
1333 #ifdef CONFIG_SCHED_MM_CID
1334         int                             mm_cid;         /* Current cid in mm */
1335         int                             last_mm_cid;    /* Most recent cid in mm */
1336         int                             migrate_from_cpu;
1337         int                             mm_cid_active;  /* Whether cid bitmap is active */
1338         struct callback_head            cid_work;
1339 #endif
1340
1341         struct tlbflush_unmap_batch     tlb_ubc;
1342
1343         /* Cache last used pipe for splice(): */
1344         struct pipe_inode_info          *splice_pipe;
1345
1346         struct page_frag                task_frag;
1347
1348 #ifdef CONFIG_TASK_DELAY_ACCT
1349         struct task_delay_info          *delays;
1350 #endif
1351
1352 #ifdef CONFIG_FAULT_INJECTION
1353         int                             make_it_fail;
1354         unsigned int                    fail_nth;
1355 #endif
1356         /*
1357          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1358          * balance_dirty_pages() for a dirty throttling pause:
1359          */
1360         int                             nr_dirtied;
1361         int                             nr_dirtied_pause;
1362         /* Start of a write-and-pause period: */
1363         unsigned long                   dirty_paused_when;
1364
1365 #ifdef CONFIG_LATENCYTOP
1366         int                             latency_record_count;
1367         struct latency_record           latency_record[LT_SAVECOUNT];
1368 #endif
1369         /*
1370          * Time slack values; these are used to round up poll() and
1371          * select() etc timeout values. These are in nanoseconds.
1372          */
1373         u64                             timer_slack_ns;
1374         u64                             default_timer_slack_ns;
1375
1376 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1377         unsigned int                    kasan_depth;
1378 #endif
1379
1380 #ifdef CONFIG_KCSAN
1381         struct kcsan_ctx                kcsan_ctx;
1382 #ifdef CONFIG_TRACE_IRQFLAGS
1383         struct irqtrace_events          kcsan_save_irqtrace;
1384 #endif
1385 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1386         int                             kcsan_stack_depth;
1387 #endif
1388 #endif
1389
1390 #ifdef CONFIG_KMSAN
1391         struct kmsan_ctx                kmsan_ctx;
1392 #endif
1393
1394 #if IS_ENABLED(CONFIG_KUNIT)
1395         struct kunit                    *kunit_test;
1396 #endif
1397
1398 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1399         /* Index of current stored address in ret_stack: */
1400         int                             curr_ret_stack;
1401         int                             curr_ret_depth;
1402
1403         /* Stack of return addresses for return function tracing: */
1404         struct ftrace_ret_stack         *ret_stack;
1405
1406         /* Timestamp for last schedule: */
1407         unsigned long long              ftrace_timestamp;
1408
1409         /*
1410          * Number of functions that haven't been traced
1411          * because of depth overrun:
1412          */
1413         atomic_t                        trace_overrun;
1414
1415         /* Pause tracing: */
1416         atomic_t                        tracing_graph_pause;
1417 #endif
1418
1419 #ifdef CONFIG_TRACING
1420         /* Bitmask and counter of trace recursion: */
1421         unsigned long                   trace_recursion;
1422 #endif /* CONFIG_TRACING */
1423
1424 #ifdef CONFIG_KCOV
1425         /* See kernel/kcov.c for more details. */
1426
1427         /* Coverage collection mode enabled for this task (0 if disabled): */
1428         unsigned int                    kcov_mode;
1429
1430         /* Size of the kcov_area: */
1431         unsigned int                    kcov_size;
1432
1433         /* Buffer for coverage collection: */
1434         void                            *kcov_area;
1435
1436         /* KCOV descriptor wired with this task or NULL: */
1437         struct kcov                     *kcov;
1438
1439         /* KCOV common handle for remote coverage collection: */
1440         u64                             kcov_handle;
1441
1442         /* KCOV sequence number: */
1443         int                             kcov_sequence;
1444
1445         /* Collect coverage from softirq context: */
1446         unsigned int                    kcov_softirq;
1447 #endif
1448
1449 #ifdef CONFIG_MEMCG
1450         struct mem_cgroup               *memcg_in_oom;
1451         gfp_t                           memcg_oom_gfp_mask;
1452         int                             memcg_oom_order;
1453
1454         /* Number of pages to reclaim on returning to userland: */
1455         unsigned int                    memcg_nr_pages_over_high;
1456
1457         /* Used by memcontrol for targeted memcg charge: */
1458         struct mem_cgroup               *active_memcg;
1459 #endif
1460
1461 #ifdef CONFIG_MEMCG_KMEM
1462         struct obj_cgroup               *objcg;
1463 #endif
1464
1465 #ifdef CONFIG_BLK_CGROUP
1466         struct gendisk                  *throttle_disk;
1467 #endif
1468
1469 #ifdef CONFIG_UPROBES
1470         struct uprobe_task              *utask;
1471 #endif
1472 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1473         unsigned int                    sequential_io;
1474         unsigned int                    sequential_io_avg;
1475 #endif
1476         struct kmap_ctrl                kmap_ctrl;
1477 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1478         unsigned long                   task_state_change;
1479 # ifdef CONFIG_PREEMPT_RT
1480         unsigned long                   saved_state_change;
1481 # endif
1482 #endif
1483         struct rcu_head                 rcu;
1484         refcount_t                      rcu_users;
1485         int                             pagefault_disabled;
1486 #ifdef CONFIG_MMU
1487         struct task_struct              *oom_reaper_list;
1488         struct timer_list               oom_reaper_timer;
1489 #endif
1490 #ifdef CONFIG_VMAP_STACK
1491         struct vm_struct                *stack_vm_area;
1492 #endif
1493 #ifdef CONFIG_THREAD_INFO_IN_TASK
1494         /* A live task holds one reference: */
1495         refcount_t                      stack_refcount;
1496 #endif
1497 #ifdef CONFIG_LIVEPATCH
1498         int patch_state;
1499 #endif
1500 #ifdef CONFIG_SECURITY
1501         /* Used by LSM modules for access restriction: */
1502         void                            *security;
1503 #endif
1504 #ifdef CONFIG_BPF_SYSCALL
1505         /* Used by BPF task local storage */
1506         struct bpf_local_storage __rcu  *bpf_storage;
1507         /* Used for BPF run context */
1508         struct bpf_run_ctx              *bpf_ctx;
1509 #endif
1510
1511 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1512         unsigned long                   lowest_stack;
1513         unsigned long                   prev_lowest_stack;
1514 #endif
1515
1516 #ifdef CONFIG_X86_MCE
1517         void __user                     *mce_vaddr;
1518         __u64                           mce_kflags;
1519         u64                             mce_addr;
1520         __u64                           mce_ripv : 1,
1521                                         mce_whole_page : 1,
1522                                         __mce_reserved : 62;
1523         struct callback_head            mce_kill_me;
1524         int                             mce_count;
1525 #endif
1526
1527 #ifdef CONFIG_KRETPROBES
1528         struct llist_head               kretprobe_instances;
1529 #endif
1530 #ifdef CONFIG_RETHOOK
1531         struct llist_head               rethooks;
1532 #endif
1533
1534 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1535         /*
1536          * If L1D flush is supported on mm context switch
1537          * then we use this callback head to queue kill work
1538          * to kill tasks that are not running on SMT disabled
1539          * cores
1540          */
1541         struct callback_head            l1d_flush_kill;
1542 #endif
1543
1544 #ifdef CONFIG_RV
1545         /*
1546          * Per-task RV monitor. Nowadays fixed in RV_PER_TASK_MONITORS.
1547          * If we find justification for more monitors, we can think
1548          * about adding more or developing a dynamic method. So far,
1549          * none of these are justified.
1550          */
1551         union rv_task_monitor           rv[RV_PER_TASK_MONITORS];
1552 #endif
1553
1554 #ifdef CONFIG_USER_EVENTS
1555         struct user_event_mm            *user_event_mm;
1556 #endif
1557
1558         /*
1559          * New fields for task_struct should be added above here, so that
1560          * they are included in the randomized portion of task_struct.
1561          */
1562         randomized_struct_fields_end
1563
1564         /* CPU-specific state of this task: */
1565         struct thread_struct            thread;
1566
1567         /*
1568          * WARNING: on x86, 'thread_struct' contains a variable-sized
1569          * structure.  It *MUST* be at the end of 'task_struct'.
1570          *
1571          * Do not put anything below here!
1572          */
1573 };
1574
1575 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1576 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1577
1578 static inline unsigned int __task_state_index(unsigned int tsk_state,
1579                                               unsigned int tsk_exit_state)
1580 {
1581         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1582
1583         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1584
1585         if ((tsk_state & TASK_IDLE) == TASK_IDLE)
1586                 state = TASK_REPORT_IDLE;
1587
1588         /*
1589          * We're lying here, but rather than expose a completely new task state
1590          * to userspace, we can make this appear as if the task has gone through
1591          * a regular rt_mutex_lock() call.
1592          */
1593         if (tsk_state & TASK_RTLOCK_WAIT)
1594                 state = TASK_UNINTERRUPTIBLE;
1595
1596         return fls(state);
1597 }
1598
1599 static inline unsigned int task_state_index(struct task_struct *tsk)
1600 {
1601         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1602 }
1603
1604 static inline char task_index_to_char(unsigned int state)
1605 {
1606         static const char state_char[] = "RSDTtXZPI";
1607
1608         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1609
1610         return state_char[state];
1611 }
1612
1613 static inline char task_state_to_char(struct task_struct *tsk)
1614 {
1615         return task_index_to_char(task_state_index(tsk));
1616 }
1617
1618 extern struct pid *cad_pid;
1619
1620 /*
1621  * Per process flags
1622  */
1623 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1624 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1625 #define PF_EXITING              0x00000004      /* Getting shut down */
1626 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1627 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1628 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1629 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1630 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1631 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1632 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1633 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1634 #define PF_MEMALLOC             0x00000800      /* Allocating memory to free memory. See memalloc_noreclaim_save() */
1635 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1636 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1637 #define PF_USER_WORKER          0x00004000      /* Kernel thread cloned from userspace thread */
1638 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1639 #define PF__HOLE__00010000      0x00010000
1640 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1641 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocations inherit GFP_NOFS. See memalloc_nfs_save() */
1642 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocations inherit GFP_NOIO. See memalloc_noio_save() */
1643 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1644                                                  * I am cleaning dirty pages from some other bdi. */
1645 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1646 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1647 #define PF_MEMALLOC_NORECLAIM   0x00800000      /* All allocation requests will clear __GFP_DIRECT_RECLAIM */
1648 #define PF_MEMALLOC_NOWARN      0x01000000      /* All allocation requests will inherit __GFP_NOWARN */
1649 #define PF__HOLE__02000000      0x02000000
1650 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1651 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1652 #define PF_MEMALLOC_PIN         0x10000000      /* Allocations constrained to zones which allow long term pinning.
1653                                                  * See memalloc_pin_save() */
1654 #define PF_BLOCK_TS             0x20000000      /* plug has ts that needs updating */
1655 #define PF__HOLE__40000000      0x40000000
1656 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1657
1658 /*
1659  * Only the _current_ task can read/write to tsk->flags, but other
1660  * tasks can access tsk->flags in readonly mode for example
1661  * with tsk_used_math (like during threaded core dumping).
1662  * There is however an exception to this rule during ptrace
1663  * or during fork: the ptracer task is allowed to write to the
1664  * child->flags of its traced child (same goes for fork, the parent
1665  * can write to the child->flags), because we're guaranteed the
1666  * child is not running and in turn not changing child->flags
1667  * at the same time the parent does it.
1668  */
1669 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1670 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1671 #define clear_used_math()                       clear_stopped_child_used_math(current)
1672 #define set_used_math()                         set_stopped_child_used_math(current)
1673
1674 #define conditional_stopped_child_used_math(condition, child) \
1675         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1676
1677 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1678
1679 #define copy_to_stopped_child_used_math(child) \
1680         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1681
1682 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1683 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1684 #define used_math()                             tsk_used_math(current)
1685
1686 static __always_inline bool is_percpu_thread(void)
1687 {
1688 #ifdef CONFIG_SMP
1689         return (current->flags & PF_NO_SETAFFINITY) &&
1690                 (current->nr_cpus_allowed  == 1);
1691 #else
1692         return true;
1693 #endif
1694 }
1695
1696 /* Per-process atomic flags. */
1697 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1698 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1699 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1700 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1701 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1702 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1703 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1704 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1705
1706 #define TASK_PFA_TEST(name, func)                                       \
1707         static inline bool task_##func(struct task_struct *p)           \
1708         { return test_bit(PFA_##name, &p->atomic_flags); }
1709
1710 #define TASK_PFA_SET(name, func)                                        \
1711         static inline void task_set_##func(struct task_struct *p)       \
1712         { set_bit(PFA_##name, &p->atomic_flags); }
1713
1714 #define TASK_PFA_CLEAR(name, func)                                      \
1715         static inline void task_clear_##func(struct task_struct *p)     \
1716         { clear_bit(PFA_##name, &p->atomic_flags); }
1717
1718 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1719 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1720
1721 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1722 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1723 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1724
1725 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1726 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1727 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1728
1729 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1730 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1731 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1732
1733 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1734 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1735 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1736
1737 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1738 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1739
1740 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1741 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1742 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1743
1744 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1745 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1746
1747 static inline void
1748 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1749 {
1750         current->flags &= ~flags;
1751         current->flags |= orig_flags & flags;
1752 }
1753
1754 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1755 extern int task_can_attach(struct task_struct *p);
1756 extern int dl_bw_alloc(int cpu, u64 dl_bw);
1757 extern void dl_bw_free(int cpu, u64 dl_bw);
1758 #ifdef CONFIG_SMP
1759
1760 /* do_set_cpus_allowed() - consider using set_cpus_allowed_ptr() instead */
1761 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1762
1763 /**
1764  * set_cpus_allowed_ptr - set CPU affinity mask of a task
1765  * @p: the task
1766  * @new_mask: CPU affinity mask
1767  *
1768  * Return: zero if successful, or a negative error code
1769  */
1770 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1771 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1772 extern void release_user_cpus_ptr(struct task_struct *p);
1773 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1774 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1775 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1776 #else
1777 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1778 {
1779 }
1780 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1781 {
1782         if (!cpumask_test_cpu(0, new_mask))
1783                 return -EINVAL;
1784         return 0;
1785 }
1786 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1787 {
1788         if (src->user_cpus_ptr)
1789                 return -EINVAL;
1790         return 0;
1791 }
1792 static inline void release_user_cpus_ptr(struct task_struct *p)
1793 {
1794         WARN_ON(p->user_cpus_ptr);
1795 }
1796
1797 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1798 {
1799         return 0;
1800 }
1801 #endif
1802
1803 extern int yield_to(struct task_struct *p, bool preempt);
1804 extern void set_user_nice(struct task_struct *p, long nice);
1805 extern int task_prio(const struct task_struct *p);
1806
1807 /**
1808  * task_nice - return the nice value of a given task.
1809  * @p: the task in question.
1810  *
1811  * Return: The nice value [ -20 ... 0 ... 19 ].
1812  */
1813 static inline int task_nice(const struct task_struct *p)
1814 {
1815         return PRIO_TO_NICE((p)->static_prio);
1816 }
1817
1818 extern int can_nice(const struct task_struct *p, const int nice);
1819 extern int task_curr(const struct task_struct *p);
1820 extern int idle_cpu(int cpu);
1821 extern int available_idle_cpu(int cpu);
1822 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1823 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1824 extern void sched_set_fifo(struct task_struct *p);
1825 extern void sched_set_fifo_low(struct task_struct *p);
1826 extern void sched_set_normal(struct task_struct *p, int nice);
1827 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1828 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1829 extern struct task_struct *idle_task(int cpu);
1830
1831 /**
1832  * is_idle_task - is the specified task an idle task?
1833  * @p: the task in question.
1834  *
1835  * Return: 1 if @p is an idle task. 0 otherwise.
1836  */
1837 static __always_inline bool is_idle_task(const struct task_struct *p)
1838 {
1839         return !!(p->flags & PF_IDLE);
1840 }
1841
1842 extern struct task_struct *curr_task(int cpu);
1843 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1844
1845 void yield(void);
1846
1847 union thread_union {
1848         struct task_struct task;
1849 #ifndef CONFIG_THREAD_INFO_IN_TASK
1850         struct thread_info thread_info;
1851 #endif
1852         unsigned long stack[THREAD_SIZE/sizeof(long)];
1853 };
1854
1855 #ifndef CONFIG_THREAD_INFO_IN_TASK
1856 extern struct thread_info init_thread_info;
1857 #endif
1858
1859 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1860
1861 #ifdef CONFIG_THREAD_INFO_IN_TASK
1862 # define task_thread_info(task) (&(task)->thread_info)
1863 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1864 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1865 #endif
1866
1867 /*
1868  * find a task by one of its numerical ids
1869  *
1870  * find_task_by_pid_ns():
1871  *      finds a task by its pid in the specified namespace
1872  * find_task_by_vpid():
1873  *      finds a task by its virtual pid
1874  *
1875  * see also find_vpid() etc in include/linux/pid.h
1876  */
1877
1878 extern struct task_struct *find_task_by_vpid(pid_t nr);
1879 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1880
1881 /*
1882  * find a task by its virtual pid and get the task struct
1883  */
1884 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1885
1886 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1887 extern int wake_up_process(struct task_struct *tsk);
1888 extern void wake_up_new_task(struct task_struct *tsk);
1889
1890 #ifdef CONFIG_SMP
1891 extern void kick_process(struct task_struct *tsk);
1892 #else
1893 static inline void kick_process(struct task_struct *tsk) { }
1894 #endif
1895
1896 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1897
1898 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1899 {
1900         __set_task_comm(tsk, from, false);
1901 }
1902
1903 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1904 #define get_task_comm(buf, tsk) ({                      \
1905         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1906         __get_task_comm(buf, sizeof(buf), tsk);         \
1907 })
1908
1909 #ifdef CONFIG_SMP
1910 static __always_inline void scheduler_ipi(void)
1911 {
1912         /*
1913          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1914          * TIF_NEED_RESCHED remotely (for the first time) will also send
1915          * this IPI.
1916          */
1917         preempt_fold_need_resched();
1918 }
1919 #else
1920 static inline void scheduler_ipi(void) { }
1921 #endif
1922
1923 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1924
1925 /*
1926  * Set thread flags in other task's structures.
1927  * See asm/thread_info.h for TIF_xxxx flags available:
1928  */
1929 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1930 {
1931         set_ti_thread_flag(task_thread_info(tsk), flag);
1932 }
1933
1934 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1935 {
1936         clear_ti_thread_flag(task_thread_info(tsk), flag);
1937 }
1938
1939 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1940                                           bool value)
1941 {
1942         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1943 }
1944
1945 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1946 {
1947         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1948 }
1949
1950 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1951 {
1952         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1953 }
1954
1955 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1956 {
1957         return test_ti_thread_flag(task_thread_info(tsk), flag);
1958 }
1959
1960 static inline void set_tsk_need_resched(struct task_struct *tsk)
1961 {
1962         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1963 }
1964
1965 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1966 {
1967         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1968 }
1969
1970 static inline int test_tsk_need_resched(struct task_struct *tsk)
1971 {
1972         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
1973 }
1974
1975 /*
1976  * cond_resched() and cond_resched_lock(): latency reduction via
1977  * explicit rescheduling in places that are safe. The return
1978  * value indicates whether a reschedule was done in fact.
1979  * cond_resched_lock() will drop the spinlock before scheduling,
1980  */
1981 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
1982 extern int __cond_resched(void);
1983
1984 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
1985
1986 void sched_dynamic_klp_enable(void);
1987 void sched_dynamic_klp_disable(void);
1988
1989 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
1990
1991 static __always_inline int _cond_resched(void)
1992 {
1993         return static_call_mod(cond_resched)();
1994 }
1995
1996 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
1997
1998 extern int dynamic_cond_resched(void);
1999
2000 static __always_inline int _cond_resched(void)
2001 {
2002         return dynamic_cond_resched();
2003 }
2004
2005 #else /* !CONFIG_PREEMPTION */
2006
2007 static inline int _cond_resched(void)
2008 {
2009         klp_sched_try_switch();
2010         return __cond_resched();
2011 }
2012
2013 #endif /* PREEMPT_DYNAMIC && CONFIG_HAVE_PREEMPT_DYNAMIC_CALL */
2014
2015 #else /* CONFIG_PREEMPTION && !CONFIG_PREEMPT_DYNAMIC */
2016
2017 static inline int _cond_resched(void)
2018 {
2019         klp_sched_try_switch();
2020         return 0;
2021 }
2022
2023 #endif /* !CONFIG_PREEMPTION || CONFIG_PREEMPT_DYNAMIC */
2024
2025 #define cond_resched() ({                       \
2026         __might_resched(__FILE__, __LINE__, 0); \
2027         _cond_resched();                        \
2028 })
2029
2030 extern int __cond_resched_lock(spinlock_t *lock);
2031 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2032 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2033
2034 #define MIGHT_RESCHED_RCU_SHIFT         8
2035 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2036
2037 #ifndef CONFIG_PREEMPT_RT
2038 /*
2039  * Non RT kernels have an elevated preempt count due to the held lock,
2040  * but are not allowed to be inside a RCU read side critical section
2041  */
2042 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2043 #else
2044 /*
2045  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2046  * cond_resched*lock() has to take that into account because it checks for
2047  * preempt_count() and rcu_preempt_depth().
2048  */
2049 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2050         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2051 #endif
2052
2053 #define cond_resched_lock(lock) ({                                              \
2054         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2055         __cond_resched_lock(lock);                                              \
2056 })
2057
2058 #define cond_resched_rwlock_read(lock) ({                                       \
2059         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2060         __cond_resched_rwlock_read(lock);                                       \
2061 })
2062
2063 #define cond_resched_rwlock_write(lock) ({                                      \
2064         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2065         __cond_resched_rwlock_write(lock);                                      \
2066 })
2067
2068 #ifdef CONFIG_PREEMPT_DYNAMIC
2069
2070 extern bool preempt_model_none(void);
2071 extern bool preempt_model_voluntary(void);
2072 extern bool preempt_model_full(void);
2073
2074 #else
2075
2076 static inline bool preempt_model_none(void)
2077 {
2078         return IS_ENABLED(CONFIG_PREEMPT_NONE);
2079 }
2080 static inline bool preempt_model_voluntary(void)
2081 {
2082         return IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY);
2083 }
2084 static inline bool preempt_model_full(void)
2085 {
2086         return IS_ENABLED(CONFIG_PREEMPT);
2087 }
2088
2089 #endif
2090
2091 static inline bool preempt_model_rt(void)
2092 {
2093         return IS_ENABLED(CONFIG_PREEMPT_RT);
2094 }
2095
2096 /*
2097  * Does the preemption model allow non-cooperative preemption?
2098  *
2099  * For !CONFIG_PREEMPT_DYNAMIC kernels this is an exact match with
2100  * CONFIG_PREEMPTION; for CONFIG_PREEMPT_DYNAMIC this doesn't work as the
2101  * kernel is *built* with CONFIG_PREEMPTION=y but may run with e.g. the
2102  * PREEMPT_NONE model.
2103  */
2104 static inline bool preempt_model_preemptible(void)
2105 {
2106         return preempt_model_full() || preempt_model_rt();
2107 }
2108
2109 static __always_inline bool need_resched(void)
2110 {
2111         return unlikely(tif_need_resched());
2112 }
2113
2114 /*
2115  * Wrappers for p->thread_info->cpu access. No-op on UP.
2116  */
2117 #ifdef CONFIG_SMP
2118
2119 static inline unsigned int task_cpu(const struct task_struct *p)
2120 {
2121         return READ_ONCE(task_thread_info(p)->cpu);
2122 }
2123
2124 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2125
2126 #else
2127
2128 static inline unsigned int task_cpu(const struct task_struct *p)
2129 {
2130         return 0;
2131 }
2132
2133 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2134 {
2135 }
2136
2137 #endif /* CONFIG_SMP */
2138
2139 extern bool sched_task_on_rq(struct task_struct *p);
2140 extern unsigned long get_wchan(struct task_struct *p);
2141 extern struct task_struct *cpu_curr_snapshot(int cpu);
2142
2143 #include <linux/spinlock.h>
2144
2145 /*
2146  * In order to reduce various lock holder preemption latencies provide an
2147  * interface to see if a vCPU is currently running or not.
2148  *
2149  * This allows us to terminate optimistic spin loops and block, analogous to
2150  * the native optimistic spin heuristic of testing if the lock owner task is
2151  * running or not.
2152  */
2153 #ifndef vcpu_is_preempted
2154 static inline bool vcpu_is_preempted(int cpu)
2155 {
2156         return false;
2157 }
2158 #endif
2159
2160 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2161 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2162
2163 #ifndef TASK_SIZE_OF
2164 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2165 #endif
2166
2167 #ifdef CONFIG_SMP
2168 static inline bool owner_on_cpu(struct task_struct *owner)
2169 {
2170         /*
2171          * As lock holder preemption issue, we both skip spinning if
2172          * task is not on cpu or its cpu is preempted
2173          */
2174         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2175 }
2176
2177 /* Returns effective CPU energy utilization, as seen by the scheduler */
2178 unsigned long sched_cpu_util(int cpu);
2179 #endif /* CONFIG_SMP */
2180
2181 #ifdef CONFIG_SCHED_CORE
2182 extern void sched_core_free(struct task_struct *tsk);
2183 extern void sched_core_fork(struct task_struct *p);
2184 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2185                                 unsigned long uaddr);
2186 extern int sched_core_idle_cpu(int cpu);
2187 #else
2188 static inline void sched_core_free(struct task_struct *tsk) { }
2189 static inline void sched_core_fork(struct task_struct *p) { }
2190 static inline int sched_core_idle_cpu(int cpu) { return idle_cpu(cpu); }
2191 #endif
2192
2193 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2194
2195 #ifdef CONFIG_MEM_ALLOC_PROFILING
2196 static inline struct alloc_tag *alloc_tag_save(struct alloc_tag *tag)
2197 {
2198         swap(current->alloc_tag, tag);
2199         return tag;
2200 }
2201
2202 static inline void alloc_tag_restore(struct alloc_tag *tag, struct alloc_tag *old)
2203 {
2204 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
2205         WARN(current->alloc_tag != tag, "current->alloc_tag was changed:\n");
2206 #endif
2207         current->alloc_tag = old;
2208 }
2209 #else
2210 #define alloc_tag_save(_tag)                    NULL
2211 #define alloc_tag_restore(_tag, _old)           do {} while (0)
2212 #endif
2213
2214 #endif