include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13
  14 #include <linux/pid.h>
  15 #include <linux/sem.h>
  16 #include <linux/shm.h>
  17 #include <linux/mutex.h>
  18 #include <linux/plist.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/irqflags.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/nodemask.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/refcount.h>
  25 #include <linux/resource.h>
  26 #include <linux/latencytop.h>
  27 #include <linux/sched/prio.h>
  28 #include <linux/sched/types.h>
  29 #include <linux/signal_types.h>
  30 #include <linux/syscall_user_dispatch.h>
  31 #include <linux/mm_types_task.h>
  32 #include <linux/task_io_accounting.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/rseq.h>
  35 #include <linux/seqlock.h>
  36 #include <linux/kcsan.h>
  37 #include <asm/kmap_size.h>
  38
  39 /* task_struct member predeclarations (sorted alphabetically): */
  40 struct audit_context;
  41 struct backing_dev_info;
  42 struct bio_list;
  43 struct blk_plug;
  44 struct bpf_local_storage;
  45 struct bpf_run_ctx;
  46 struct capture_control;
  47 struct cfs_rq;
  48 struct fs_struct;
  49 struct futex_pi_state;
  50 struct io_context;
  51 struct io_uring_task;
  52 struct mempolicy;
  53 struct nameidata;
  54 struct nsproxy;
  55 struct perf_event_context;
  56 struct pid_namespace;
  57 struct pipe_inode_info;
  58 struct rcu_node;
  59 struct reclaim_state;
  60 struct robust_list_head;
  61 struct root_domain;
  62 struct rq;
  63 struct sched_attr;
  64 struct sched_param;
  65 struct seq_file;
  66 struct sighand_struct;
  67 struct signal_struct;
  68 struct task_delay_info;
  69 struct task_group;
  70
  71 /*
  72  * Task state bitmask. NOTE! These bits are also
  73  * encoded in fs/proc/array.c: get_task_state().
  74  *
  75  * We have two separate sets of flags: task->state
  76  * is about runnability, while task->exit_state are
  77  * about the task exiting. Confusing, but this way
  78  * modifying one set can't modify the other one by
  79  * mistake.
  80  */
  81
  82 /* Used in tsk->state: */
  83 #define TASK_RUNNING                    0x0000
  84 #define TASK_INTERRUPTIBLE              0x0001
  85 #define TASK_UNINTERRUPTIBLE            0x0002
  86 #define __TASK_STOPPED                  0x0004
  87 #define __TASK_TRACED                   0x0008
  88 /* Used in tsk->exit_state: */
  89 #define EXIT_DEAD                       0x0010
  90 #define EXIT_ZOMBIE                     0x0020
  91 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
  92 /* Used in tsk->state again: */
  93 #define TASK_PARKED                     0x0040
  94 #define TASK_DEAD                       0x0080
  95 #define TASK_WAKEKILL                   0x0100
  96 #define TASK_WAKING                     0x0200
  97 #define TASK_NOLOAD                     0x0400
  98 #define TASK_NEW                        0x0800
  99 /* RT specific auxilliary flag to mark RT lock waiters */
 100 #define TASK_RTLOCK_WAIT                0x1000
 101 #define TASK_STATE_MAX                  0x2000
 102
 103 /* Convenience macros for the sake of set_current_state: */
 104 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 105 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 106 #define TASK_TRACED                     (TASK_WAKEKILL | __TASK_TRACED)
 107
 108 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 109
 110 /* Convenience macros for the sake of wake_up(): */
 111 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 112
 113 /* get_task_state(): */
 114 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 115                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 116                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 117                                          TASK_PARKED)
 118
 119 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 120
 121 #define task_is_traced(task)            ((READ_ONCE(task->__state) & __TASK_TRACED) != 0)
 122
 123 #define task_is_stopped(task)           ((READ_ONCE(task->__state) & __TASK_STOPPED) != 0)
 124
 125 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->__state) & (__TASK_STOPPED | __TASK_TRACED)) != 0)
 126
 127 /*
 128  * Special states are those that do not use the normal wait-loop pattern. See
 129  * the comment with set_special_state().
 130  */
 131 #define is_special_task_state(state)                            \
 132         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
 133
 134 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 135 # define debug_normal_state_change(state_value)                         \
 136         do {                                                            \
 137                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 138                 current->task_state_change = _THIS_IP_;                 \
 139         } while (0)
 140
 141 # define debug_special_state_change(state_value)                        \
 142         do {                                                            \
 143                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 144                 current->task_state_change = _THIS_IP_;                 \
 145         } while (0)
 146
 147 # define debug_rtlock_wait_set_state()                                  \
 148         do {                                                             \
 149                 current->saved_state_change = current->task_state_change;\
 150                 current->task_state_change = _THIS_IP_;                  \
 151         } while (0)
 152
 153 # define debug_rtlock_wait_restore_state()                              \
 154         do {                                                             \
 155                 current->task_state_change = current->saved_state_change;\
 156         } while (0)
 157
 158 #else
 159 # define debug_normal_state_change(cond)        do { } while (0)
 160 # define debug_special_state_change(cond)       do { } while (0)
 161 # define debug_rtlock_wait_set_state()          do { } while (0)
 162 # define debug_rtlock_wait_restore_state()      do { } while (0)
 163 #endif
 164
 165 /*
 166  * set_current_state() includes a barrier so that the write of current->state
 167  * is correctly serialised wrt the caller's subsequent test of whether to
 168  * actually sleep:
 169  *
 170  *   for (;;) {
 171  *      set_current_state(TASK_UNINTERRUPTIBLE);
 172  *      if (CONDITION)
 173  *         break;
 174  *
 175  *      schedule();
 176  *   }
 177  *   __set_current_state(TASK_RUNNING);
 178  *
 179  * If the caller does not need such serialisation (because, for instance, the
 180  * CONDITION test and condition change and wakeup are under the same lock) then
 181  * use __set_current_state().
 182  *
 183  * The above is typically ordered against the wakeup, which does:
 184  *
 185  *   CONDITION = 1;
 186  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 187  *
 188  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 189  * accessing p->state.
 190  *
 191  * Wakeup will do: if (@state & p->state) p->state = TASK_RUNNING, that is,
 192  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 193  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 194  *
 195  * However, with slightly different timing the wakeup TASK_RUNNING store can
 196  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 197  * a problem either because that will result in one extra go around the loop
 198  * and our @cond test will save the day.
 199  *
 200  * Also see the comments of try_to_wake_up().
 201  */
 202 #define __set_current_state(state_value)                                \
 203         do {                                                            \
 204                 debug_normal_state_change((state_value));               \
 205                 WRITE_ONCE(current->__state, (state_value));            \
 206         } while (0)
 207
 208 #define set_current_state(state_value)                                  \
 209         do {                                                            \
 210                 debug_normal_state_change((state_value));               \
 211                 smp_store_mb(current->__state, (state_value));          \
 212         } while (0)
 213
 214 /*
 215  * set_special_state() should be used for those states when the blocking task
 216  * can not use the regular condition based wait-loop. In that case we must
 217  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 218  * stores will not collide with our state change.
 219  */
 220 #define set_special_state(state_value)                                  \
 221         do {                                                            \
 222                 unsigned long flags; /* may shadow */                   \
 223                                                                         \
 224                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 225                 debug_special_state_change((state_value));              \
 226                 WRITE_ONCE(current->__state, (state_value));            \
 227                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 228         } while (0)
 229
 230 /*
 231  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 232  *
 233  * RT's spin/rwlock substitutions are state preserving. The state of the
 234  * task when blocking on the lock is saved in task_struct::saved_state and
 235  * restored after the lock has been acquired.  These operations are
 236  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 237  * lock related wakeups while the task is blocked on the lock are
 238  * redirected to operate on task_struct::saved_state to ensure that these
 239  * are not dropped. On restore task_struct::saved_state is set to
 240  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 241  *
 242  * The lock operation looks like this:
 243  *
 244  *      current_save_and_set_rtlock_wait_state();
 245  *      for (;;) {
 246  *              if (try_lock())
 247  *                      break;
 248  *              raw_spin_unlock_irq(&lock->wait_lock);
 249  *              schedule_rtlock();
 250  *              raw_spin_lock_irq(&lock->wait_lock);
 251  *              set_current_state(TASK_RTLOCK_WAIT);
 252  *      }
 253  *      current_restore_rtlock_saved_state();
 254  */
 255 #define current_save_and_set_rtlock_wait_state()                        \
 256         do {                                                            \
 257                 lockdep_assert_irqs_disabled();                         \
 258                 raw_spin_lock(&current->pi_lock);                       \
 259                 current->saved_state = current->__state;                \
 260                 debug_rtlock_wait_set_state();                          \
 261                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 262                 raw_spin_unlock(&current->pi_lock);                     \
 263         } while (0);
 264
 265 #define current_restore_rtlock_saved_state()                            \
 266         do {                                                            \
 267                 lockdep_assert_irqs_disabled();                         \
 268                 raw_spin_lock(&current->pi_lock);                       \
 269                 debug_rtlock_wait_restore_state();                      \
 270                 WRITE_ONCE(current->__state, current->saved_state);     \
 271                 current->saved_state = TASK_RUNNING;                    \
 272                 raw_spin_unlock(&current->pi_lock);                     \
 273         } while (0);
 274
 275 #define get_current_state()     READ_ONCE(current->__state)
 276
 277 /* Task command name length: */
 278 #define TASK_COMM_LEN                   16
 279
 280 extern void scheduler_tick(void);
 281
 282 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 283
 284 extern long schedule_timeout(long timeout);
 285 extern long schedule_timeout_interruptible(long timeout);
 286 extern long schedule_timeout_killable(long timeout);
 287 extern long schedule_timeout_uninterruptible(long timeout);
 288 extern long schedule_timeout_idle(long timeout);
 289 asmlinkage void schedule(void);
 290 extern void schedule_preempt_disabled(void);
 291 asmlinkage void preempt_schedule_irq(void);
 292 #ifdef CONFIG_PREEMPT_RT
 293  extern void schedule_rtlock(void);
 294 #endif
 295
 296 extern int __must_check io_schedule_prepare(void);
 297 extern void io_schedule_finish(int token);
 298 extern long io_schedule_timeout(long timeout);
 299 extern void io_schedule(void);
 300
 301 /**
 302  * struct prev_cputime - snapshot of system and user cputime
 303  * @utime: time spent in user mode
 304  * @stime: time spent in system mode
 305  * @lock: protects the above two fields
 306  *
 307  * Stores previous user/system time values such that we can guarantee
 308  * monotonicity.
 309  */
 310 struct prev_cputime {
 311 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 312         u64                             utime;
 313         u64                             stime;
 314         raw_spinlock_t                  lock;
 315 #endif
 316 };
 317
 318 enum vtime_state {
 319         /* Task is sleeping or running in a CPU with VTIME inactive: */
 320         VTIME_INACTIVE = 0,
 321         /* Task is idle */
 322         VTIME_IDLE,
 323         /* Task runs in kernelspace in a CPU with VTIME active: */
 324         VTIME_SYS,
 325         /* Task runs in userspace in a CPU with VTIME active: */
 326         VTIME_USER,
 327         /* Task runs as guests in a CPU with VTIME active: */
 328         VTIME_GUEST,
 329 };
 330
 331 struct vtime {
 332         seqcount_t              seqcount;
 333         unsigned long long      starttime;
 334         enum vtime_state        state;
 335         unsigned int            cpu;
 336         u64                     utime;
 337         u64                     stime;
 338         u64                     gtime;
 339 };
 340
 341 /*
 342  * Utilization clamp constraints.
 343  * @UCLAMP_MIN: Minimum utilization
 344  * @UCLAMP_MAX: Maximum utilization
 345  * @UCLAMP_CNT: Utilization clamp constraints count
 346  */
 347 enum uclamp_id {
 348         UCLAMP_MIN = 0,
 349         UCLAMP_MAX,
 350         UCLAMP_CNT
 351 };
 352
 353 #ifdef CONFIG_SMP
 354 extern struct root_domain def_root_domain;
 355 extern struct mutex sched_domains_mutex;
 356 #endif
 357
 358 struct sched_info {
 359 #ifdef CONFIG_SCHED_INFO
 360         /* Cumulative counters: */
 361
 362         /* # of times we have run on this CPU: */
 363         unsigned long                   pcount;
 364
 365         /* Time spent waiting on a runqueue: */
 366         unsigned long long              run_delay;
 367
 368         /* Timestamps: */
 369
 370         /* When did we last run on a CPU? */
 371         unsigned long long              last_arrival;
 372
 373         /* When were we last queued to run? */
 374         unsigned long long              last_queued;
 375
 376 #endif /* CONFIG_SCHED_INFO */
 377 };
 378
 379 /*
 380  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 381  * has a few: load, load_avg, util_avg, freq, and capacity.
 382  *
 383  * We define a basic fixed point arithmetic range, and then formalize
 384  * all these metrics based on that basic range.
 385  */
 386 # define SCHED_FIXEDPOINT_SHIFT         10
 387 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 388
 389 /* Increase resolution of cpu_capacity calculations */
 390 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 391 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 392
 393 struct load_weight {
 394         unsigned long                   weight;
 395         u32                             inv_weight;
 396 };
 397
 398 /**
 399  * struct util_est - Estimation utilization of FAIR tasks
 400  * @enqueued: instantaneous estimated utilization of a task/cpu
 401  * @ewma:     the Exponential Weighted Moving Average (EWMA)
 402  *            utilization of a task
 403  *
 404  * Support data structure to track an Exponential Weighted Moving Average
 405  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
 406  * average each time a task completes an activation. Sample's weight is chosen
 407  * so that the EWMA will be relatively insensitive to transient changes to the
 408  * task's workload.
 409  *
 410  * The enqueued attribute has a slightly different meaning for tasks and cpus:
 411  * - task:   the task's util_avg at last task dequeue time
 412  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
 413  * Thus, the util_est.enqueued of a task represents the contribution on the
 414  * estimated utilization of the CPU where that task is currently enqueued.
 415  *
 416  * Only for tasks we track a moving average of the past instantaneous
 417  * estimated utilization. This allows to absorb sporadic drops in utilization
 418  * of an otherwise almost periodic task.
 419  *
 420  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 421  * updates. When a task is dequeued, its util_est should not be updated if its
 422  * util_avg has not been updated in the meantime.
 423  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
 424  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
 425  * for a task) it is safe to use MSB.
 426  */
 427 struct util_est {
 428         unsigned int                    enqueued;
 429         unsigned int                    ewma;
 430 #define UTIL_EST_WEIGHT_SHIFT           2
 431 #define UTIL_AVG_UNCHANGED              0x80000000
 432 } __attribute__((__aligned__(sizeof(u64))));
 433
 434 /*
 435  * The load/runnable/util_avg accumulates an infinite geometric series
 436  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 437  *
 438  * [load_avg definition]
 439  *
 440  *   load_avg = runnable% * scale_load_down(load)
 441  *
 442  * [runnable_avg definition]
 443  *
 444  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 445  *
 446  * [util_avg definition]
 447  *
 448  *   util_avg = running% * SCHED_CAPACITY_SCALE
 449  *
 450  * where runnable% is the time ratio that a sched_entity is runnable and
 451  * running% the time ratio that a sched_entity is running.
 452  *
 453  * For cfs_rq, they are the aggregated values of all runnable and blocked
 454  * sched_entities.
 455  *
 456  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 457  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 458  * for computing those signals (see update_rq_clock_pelt())
 459  *
 460  * N.B., the above ratios (runnable% and running%) themselves are in the
 461  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 462  * to as large a range as necessary. This is for example reflected by
 463  * util_avg's SCHED_CAPACITY_SCALE.
 464  *
 465  * [Overflow issue]
 466  *
 467  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 468  * with the highest load (=88761), always runnable on a single cfs_rq,
 469  * and should not overflow as the number already hits PID_MAX_LIMIT.
 470  *
 471  * For all other cases (including 32-bit kernels), struct load_weight's
 472  * weight will overflow first before we do, because:
 473  *
 474  *    Max(load_avg) <= Max(load.weight)
 475  *
 476  * Then it is the load_weight's responsibility to consider overflow
 477  * issues.
 478  */
 479 struct sched_avg {
 480         u64                             last_update_time;
 481         u64                             load_sum;
 482         u64                             runnable_sum;
 483         u32                             util_sum;
 484         u32                             period_contrib;
 485         unsigned long                   load_avg;
 486         unsigned long                   runnable_avg;
 487         unsigned long                   util_avg;
 488         struct util_est                 util_est;
 489 } ____cacheline_aligned;
 490
 491 struct sched_statistics {
 492 #ifdef CONFIG_SCHEDSTATS
 493         u64                             wait_start;
 494         u64                             wait_max;
 495         u64                             wait_count;
 496         u64                             wait_sum;
 497         u64                             iowait_count;
 498         u64                             iowait_sum;
 499
 500         u64                             sleep_start;
 501         u64                             sleep_max;
 502         s64                             sum_sleep_runtime;
 503
 504         u64                             block_start;
 505         u64                             block_max;
 506         s64                             sum_block_runtime;
 507
 508         u64                             exec_max;
 509         u64                             slice_max;
 510
 511         u64                             nr_migrations_cold;
 512         u64                             nr_failed_migrations_affine;
 513         u64                             nr_failed_migrations_running;
 514         u64                             nr_failed_migrations_hot;
 515         u64                             nr_forced_migrations;
 516
 517         u64                             nr_wakeups;
 518         u64                             nr_wakeups_sync;
 519         u64                             nr_wakeups_migrate;
 520         u64                             nr_wakeups_local;
 521         u64                             nr_wakeups_remote;
 522         u64                             nr_wakeups_affine;
 523         u64                             nr_wakeups_affine_attempts;
 524         u64                             nr_wakeups_passive;
 525         u64                             nr_wakeups_idle;
 526 #endif
 527 } ____cacheline_aligned;
 528
 529 struct sched_entity {
 530         /* For load-balancing: */
 531         struct load_weight              load;
 532         struct rb_node                  run_node;
 533         struct list_head                group_node;
 534         unsigned int                    on_rq;
 535
 536         u64                             exec_start;
 537         u64                             sum_exec_runtime;
 538         u64                             vruntime;
 539         u64                             prev_sum_exec_runtime;
 540
 541         u64                             nr_migrations;
 542
 543 #ifdef CONFIG_FAIR_GROUP_SCHED
 544         int                             depth;
 545         struct sched_entity             *parent;
 546         /* rq on which this entity is (to be) queued: */
 547         struct cfs_rq                   *cfs_rq;
 548         /* rq "owned" by this entity/group: */
 549         struct cfs_rq                   *my_q;
 550         /* cached value of my_q->h_nr_running */
 551         unsigned long                   runnable_weight;
 552 #endif
 553
 554 #ifdef CONFIG_SMP
 555         /*
 556          * Per entity load average tracking.
 557          *
 558          * Put into separate cache line so it does not
 559          * collide with read-mostly values above.
 560          */
 561         struct sched_avg                avg;
 562 #endif
 563 };
 564
 565 struct sched_rt_entity {
 566         struct list_head                run_list;
 567         unsigned long                   timeout;
 568         unsigned long                   watchdog_stamp;
 569         unsigned int                    time_slice;
 570         unsigned short                  on_rq;
 571         unsigned short                  on_list;
 572
 573         struct sched_rt_entity          *back;
 574 #ifdef CONFIG_RT_GROUP_SCHED
 575         struct sched_rt_entity          *parent;
 576         /* rq on which this entity is (to be) queued: */
 577         struct rt_rq                    *rt_rq;
 578         /* rq "owned" by this entity/group: */
 579         struct rt_rq                    *my_q;
 580 #endif
 581 } __randomize_layout;
 582
 583 struct sched_dl_entity {
 584         struct rb_node                  rb_node;
 585
 586         /*
 587          * Original scheduling parameters. Copied here from sched_attr
 588          * during sched_setattr(), they will remain the same until
 589          * the next sched_setattr().
 590          */
 591         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 592         u64                             dl_deadline;    /* Relative deadline of each instance   */
 593         u64                             dl_period;      /* Separation of two instances (period) */
 594         u64                             dl_bw;          /* dl_runtime / dl_period               */
 595         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 596
 597         /*
 598          * Actual scheduling parameters. Initialized with the values above,
 599          * they are continuously updated during task execution. Note that
 600          * the remaining runtime could be < 0 in case we are in overrun.
 601          */
 602         s64                             runtime;        /* Remaining runtime for this instance  */
 603         u64                             deadline;       /* Absolute deadline for this instance  */
 604         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 605
 606         /*
 607          * Some bool flags:
 608          *
 609          * @dl_throttled tells if we exhausted the runtime. If so, the
 610          * task has to wait for a replenishment to be performed at the
 611          * next firing of dl_timer.
 612          *
 613          * @dl_boosted tells if we are boosted due to DI. If so we are
 614          * outside bandwidth enforcement mechanism (but only until we
 615          * exit the critical section);
 616          *
 617          * @dl_yielded tells if task gave up the CPU before consuming
 618          * all its available runtime during the last job.
 619          *
 620          * @dl_non_contending tells if the task is inactive while still
 621          * contributing to the active utilization. In other words, it
 622          * indicates if the inactive timer has been armed and its handler
 623          * has not been executed yet. This flag is useful to avoid race
 624          * conditions between the inactive timer handler and the wakeup
 625          * code.
 626          *
 627          * @dl_overrun tells if the task asked to be informed about runtime
 628          * overruns.
 629          */
 630         unsigned int                    dl_throttled      : 1;
 631         unsigned int                    dl_yielded        : 1;
 632         unsigned int                    dl_non_contending : 1;
 633         unsigned int                    dl_overrun        : 1;
 634
 635         /*
 636          * Bandwidth enforcement timer. Each -deadline task has its
 637          * own bandwidth to be enforced, thus we need one timer per task.
 638          */
 639         struct hrtimer                  dl_timer;
 640
 641         /*
 642          * Inactive timer, responsible for decreasing the active utilization
 643          * at the "0-lag time". When a -deadline task blocks, it contributes
 644          * to GRUB's active utilization until the "0-lag time", hence a
 645          * timer is needed to decrease the active utilization at the correct
 646          * time.
 647          */
 648         struct hrtimer inactive_timer;
 649
 650 #ifdef CONFIG_RT_MUTEXES
 651         /*
 652          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 653          * pi_se points to the donor, otherwise points to the dl_se it belongs
 654          * to (the original one/itself).
 655          */
 656         struct sched_dl_entity *pi_se;
 657 #endif
 658 };
 659
 660 #ifdef CONFIG_UCLAMP_TASK
 661 /* Number of utilization clamp buckets (shorter alias) */
 662 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 663
 664 /*
 665  * Utilization clamp for a scheduling entity
 666  * @value:              clamp value "assigned" to a se
 667  * @bucket_id:          bucket index corresponding to the "assigned" value
 668  * @active:             the se is currently refcounted in a rq's bucket
 669  * @user_defined:       the requested clamp value comes from user-space
 670  *
 671  * The bucket_id is the index of the clamp bucket matching the clamp value
 672  * which is pre-computed and stored to avoid expensive integer divisions from
 673  * the fast path.
 674  *
 675  * The active bit is set whenever a task has got an "effective" value assigned,
 676  * which can be different from the clamp value "requested" from user-space.
 677  * This allows to know a task is refcounted in the rq's bucket corresponding
 678  * to the "effective" bucket_id.
 679  *
 680  * The user_defined bit is set whenever a task has got a task-specific clamp
 681  * value requested from userspace, i.e. the system defaults apply to this task
 682  * just as a restriction. This allows to relax default clamps when a less
 683  * restrictive task-specific value has been requested, thus allowing to
 684  * implement a "nice" semantic. For example, a task running with a 20%
 685  * default boost can still drop its own boosting to 0%.
 686  */
 687 struct uclamp_se {
 688         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 689         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 690         unsigned int active             : 1;
 691         unsigned int user_defined       : 1;
 692 };
 693 #endif /* CONFIG_UCLAMP_TASK */
 694
 695 union rcu_special {
 696         struct {
 697                 u8                      blocked;
 698                 u8                      need_qs;
 699                 u8                      exp_hint; /* Hint for performance. */
 700                 u8                      need_mb; /* Readers need smp_mb(). */
 701         } b; /* Bits. */
 702         u32 s; /* Set of bits. */
 703 };
 704
 705 enum perf_event_task_context {
 706         perf_invalid_context = -1,
 707         perf_hw_context = 0,
 708         perf_sw_context,
 709         perf_nr_task_contexts,
 710 };
 711
 712 struct wake_q_node {
 713         struct wake_q_node *next;
 714 };
 715
 716 struct kmap_ctrl {
 717 #ifdef CONFIG_KMAP_LOCAL
 718         int                             idx;
 719         pte_t                           pteval[KM_MAX_IDX];
 720 #endif
 721 };
 722
 723 struct task_struct {
 724 #ifdef CONFIG_THREAD_INFO_IN_TASK
 725         /*
 726          * For reasons of header soup (see current_thread_info()), this
 727          * must be the first element of task_struct.
 728          */
 729         struct thread_info              thread_info;
 730 #endif
 731         unsigned int                    __state;
 732
 733 #ifdef CONFIG_PREEMPT_RT
 734         /* saved state for "spinlock sleepers" */
 735         unsigned int                    saved_state;
 736 #endif
 737
 738         /*
 739          * This begins the randomizable portion of task_struct. Only
 740          * scheduling-critical items should be added above here.
 741          */
 742         randomized_struct_fields_start
 743
 744         void                            *stack;
 745         refcount_t                      usage;
 746         /* Per task flags (PF_*), defined further below: */
 747         unsigned int                    flags;
 748         unsigned int                    ptrace;
 749
 750 #ifdef CONFIG_SMP
 751         int                             on_cpu;
 752         struct __call_single_node       wake_entry;
 753         unsigned int                    wakee_flips;
 754         unsigned long                   wakee_flip_decay_ts;
 755         struct task_struct              *last_wakee;
 756
 757         /*
 758          * recent_used_cpu is initially set as the last CPU used by a task
 759          * that wakes affine another task. Waker/wakee relationships can
 760          * push tasks around a CPU where each wakeup moves to the next one.
 761          * Tracking a recently used CPU allows a quick search for a recently
 762          * used CPU that may be idle.
 763          */
 764         int                             recent_used_cpu;
 765         int                             wake_cpu;
 766 #endif
 767         int                             on_rq;
 768
 769         int                             prio;
 770         int                             static_prio;
 771         int                             normal_prio;
 772         unsigned int                    rt_priority;
 773
 774         struct sched_entity             se;
 775         struct sched_rt_entity          rt;
 776         struct sched_dl_entity          dl;
 777         const struct sched_class        *sched_class;
 778
 779 #ifdef CONFIG_SCHED_CORE
 780         struct rb_node                  core_node;
 781         unsigned long                   core_cookie;
 782         unsigned int                    core_occupation;
 783 #endif
 784
 785 #ifdef CONFIG_CGROUP_SCHED
 786         struct task_group               *sched_task_group;
 787 #endif
 788
 789 #ifdef CONFIG_UCLAMP_TASK
 790         /*
 791          * Clamp values requested for a scheduling entity.
 792          * Must be updated with task_rq_lock() held.
 793          */
 794         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 795         /*
 796          * Effective clamp values used for a scheduling entity.
 797          * Must be updated with task_rq_lock() held.
 798          */
 799         struct uclamp_se                uclamp[UCLAMP_CNT];
 800 #endif
 801
 802         struct sched_statistics         stats;
 803
 804 #ifdef CONFIG_PREEMPT_NOTIFIERS
 805         /* List of struct preempt_notifier: */
 806         struct hlist_head               preempt_notifiers;
 807 #endif
 808
 809 #ifdef CONFIG_BLK_DEV_IO_TRACE
 810         unsigned int                    btrace_seq;
 811 #endif
 812
 813         unsigned int                    policy;
 814         int                             nr_cpus_allowed;
 815         const cpumask_t                 *cpus_ptr;
 816         cpumask_t                       *user_cpus_ptr;
 817         cpumask_t                       cpus_mask;
 818         void                            *migration_pending;
 819 #ifdef CONFIG_SMP
 820         unsigned short                  migration_disabled;
 821 #endif
 822         unsigned short                  migration_flags;
 823
 824 #ifdef CONFIG_PREEMPT_RCU
 825         int                             rcu_read_lock_nesting;
 826         union rcu_special               rcu_read_unlock_special;
 827         struct list_head                rcu_node_entry;
 828         struct rcu_node                 *rcu_blocked_node;
 829 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 830
 831 #ifdef CONFIG_TASKS_RCU
 832         unsigned long                   rcu_tasks_nvcsw;
 833         u8                              rcu_tasks_holdout;
 834         u8                              rcu_tasks_idx;
 835         int                             rcu_tasks_idle_cpu;
 836         struct list_head                rcu_tasks_holdout_list;
 837 #endif /* #ifdef CONFIG_TASKS_RCU */
 838
 839 #ifdef CONFIG_TASKS_TRACE_RCU
 840         int                             trc_reader_nesting;
 841         int                             trc_ipi_to_cpu;
 842         union rcu_special               trc_reader_special;
 843         bool                            trc_reader_checked;
 844         struct list_head                trc_holdout_list;
 845 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 846
 847         struct sched_info               sched_info;
 848
 849         struct list_head                tasks;
 850 #ifdef CONFIG_SMP
 851         struct plist_node               pushable_tasks;
 852         struct rb_node                  pushable_dl_tasks;
 853 #endif
 854
 855         struct mm_struct                *mm;
 856         struct mm_struct                *active_mm;
 857
 858         /* Per-thread vma caching: */
 859         struct vmacache                 vmacache;
 860
 861 #ifdef SPLIT_RSS_COUNTING
 862         struct task_rss_stat            rss_stat;
 863 #endif
 864         int                             exit_state;
 865         int                             exit_code;
 866         int                             exit_signal;
 867         /* The signal sent when the parent dies: */
 868         int                             pdeath_signal;
 869         /* JOBCTL_*, siglock protected: */
 870         unsigned long                   jobctl;
 871
 872         /* Used for emulating ABI behavior of previous Linux versions: */
 873         unsigned int                    personality;
 874
 875         /* Scheduler bits, serialized by scheduler locks: */
 876         unsigned                        sched_reset_on_fork:1;
 877         unsigned                        sched_contributes_to_load:1;
 878         unsigned                        sched_migrated:1;
 879 #ifdef CONFIG_PSI
 880         unsigned                        sched_psi_wake_requeue:1;
 881 #endif
 882
 883         /* Force alignment to the next boundary: */
 884         unsigned                        :0;
 885
 886         /* Unserialized, strictly 'current' */
 887
 888         /*
 889          * This field must not be in the scheduler word above due to wakelist
 890          * queueing no longer being serialized by p->on_cpu. However:
 891          *
 892          * p->XXX = X;                  ttwu()
 893          * schedule()                     if (p->on_rq && ..) // false
 894          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 895          *   deactivate_task()                ttwu_queue_wakelist())
 896          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 897          *
 898          * guarantees all stores of 'current' are visible before
 899          * ->sched_remote_wakeup gets used, so it can be in this word.
 900          */
 901         unsigned                        sched_remote_wakeup:1;
 902
 903         /* Bit to tell LSMs we're in execve(): */
 904         unsigned                        in_execve:1;
 905         unsigned                        in_iowait:1;
 906 #ifndef TIF_RESTORE_SIGMASK
 907         unsigned                        restore_sigmask:1;
 908 #endif
 909 #ifdef CONFIG_MEMCG
 910         unsigned                        in_user_fault:1;
 911 #endif
 912 #ifdef CONFIG_COMPAT_BRK
 913         unsigned                        brk_randomized:1;
 914 #endif
 915 #ifdef CONFIG_CGROUPS
 916         /* disallow userland-initiated cgroup migration */
 917         unsigned                        no_cgroup_migration:1;
 918         /* task is frozen/stopped (used by the cgroup freezer) */
 919         unsigned                        frozen:1;
 920 #endif
 921 #ifdef CONFIG_BLK_CGROUP
 922         unsigned                        use_memdelay:1;
 923 #endif
 924 #ifdef CONFIG_PSI
 925         /* Stalled due to lack of memory */
 926         unsigned                        in_memstall:1;
 927 #endif
 928 #ifdef CONFIG_PAGE_OWNER
 929         /* Used by page_owner=on to detect recursion in page tracking. */
 930         unsigned                        in_page_owner:1;
 931 #endif
 932 #ifdef CONFIG_EVENTFD
 933         /* Recursion prevention for eventfd_signal() */
 934         unsigned                        in_eventfd_signal:1;
 935 #endif
 936
 937         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 938
 939         struct restart_block            restart_block;
 940
 941         pid_t                           pid;
 942         pid_t                           tgid;
 943
 944 #ifdef CONFIG_STACKPROTECTOR
 945         /* Canary value for the -fstack-protector GCC feature: */
 946         unsigned long                   stack_canary;
 947 #endif
 948         /*
 949          * Pointers to the (original) parent process, youngest child, younger sibling,
 950          * older sibling, respectively.  (p->father can be replaced with
 951          * p->real_parent->pid)
 952          */
 953
 954         /* Real parent process: */
 955         struct task_struct __rcu        *real_parent;
 956
 957         /* Recipient of SIGCHLD, wait4() reports: */
 958         struct task_struct __rcu        *parent;
 959
 960         /*
 961          * Children/sibling form the list of natural children:
 962          */
 963         struct list_head                children;
 964         struct list_head                sibling;
 965         struct task_struct              *group_leader;
 966
 967         /*
 968          * 'ptraced' is the list of tasks this task is using ptrace() on.
 969          *
 970          * This includes both natural children and PTRACE_ATTACH targets.
 971          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 972          */
 973         struct list_head                ptraced;
 974         struct list_head                ptrace_entry;
 975
 976         /* PID/PID hash table linkage. */
 977         struct pid                      *thread_pid;
 978         struct hlist_node               pid_links[PIDTYPE_MAX];
 979         struct list_head                thread_group;
 980         struct list_head                thread_node;
 981
 982         struct completion               *vfork_done;
 983
 984         /* CLONE_CHILD_SETTID: */
 985         int __user                      *set_child_tid;
 986
 987         /* CLONE_CHILD_CLEARTID: */
 988         int __user                      *clear_child_tid;
 989
 990         /* PF_IO_WORKER */
 991         void                            *pf_io_worker;
 992
 993         u64                             utime;
 994         u64                             stime;
 995 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 996         u64                             utimescaled;
 997         u64                             stimescaled;
 998 #endif
 999         u64                             gtime;
1000         struct prev_cputime             prev_cputime;
1001 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1002         struct vtime                    vtime;
1003 #endif
1004
1005 #ifdef CONFIG_NO_HZ_FULL
1006         atomic_t                        tick_dep_mask;
1007 #endif
1008         /* Context switch counts: */
1009         unsigned long                   nvcsw;
1010         unsigned long                   nivcsw;
1011
1012         /* Monotonic time in nsecs: */
1013         u64                             start_time;
1014
1015         /* Boot based time in nsecs: */
1016         u64                             start_boottime;
1017
1018         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1019         unsigned long                   min_flt;
1020         unsigned long                   maj_flt;
1021
1022         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1023         struct posix_cputimers          posix_cputimers;
1024
1025 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1026         struct posix_cputimers_work     posix_cputimers_work;
1027 #endif
1028
1029         /* Process credentials: */
1030
1031         /* Tracer's credentials at attach: */
1032         const struct cred __rcu         *ptracer_cred;
1033
1034         /* Objective and real subjective task credentials (COW): */
1035         const struct cred __rcu         *real_cred;
1036
1037         /* Effective (overridable) subjective task credentials (COW): */
1038         const struct cred __rcu         *cred;
1039
1040 #ifdef CONFIG_KEYS
1041         /* Cached requested key. */
1042         struct key                      *cached_requested_key;
1043 #endif
1044
1045         /*
1046          * executable name, excluding path.
1047          *
1048          * - normally initialized setup_new_exec()
1049          * - access it with [gs]et_task_comm()
1050          * - lock it with task_lock()
1051          */
1052         char                            comm[TASK_COMM_LEN];
1053
1054         struct nameidata                *nameidata;
1055
1056 #ifdef CONFIG_SYSVIPC
1057         struct sysv_sem                 sysvsem;
1058         struct sysv_shm                 sysvshm;
1059 #endif
1060 #ifdef CONFIG_DETECT_HUNG_TASK
1061         unsigned long                   last_switch_count;
1062         unsigned long                   last_switch_time;
1063 #endif
1064         /* Filesystem information: */
1065         struct fs_struct                *fs;
1066
1067         /* Open file information: */
1068         struct files_struct             *files;
1069
1070 #ifdef CONFIG_IO_URING
1071         struct io_uring_task            *io_uring;
1072 #endif
1073
1074         /* Namespaces: */
1075         struct nsproxy                  *nsproxy;
1076
1077         /* Signal handlers: */
1078         struct signal_struct            *signal;
1079         struct sighand_struct __rcu             *sighand;
1080         sigset_t                        blocked;
1081         sigset_t                        real_blocked;
1082         /* Restored if set_restore_sigmask() was used: */
1083         sigset_t                        saved_sigmask;
1084         struct sigpending               pending;
1085         unsigned long                   sas_ss_sp;
1086         size_t                          sas_ss_size;
1087         unsigned int                    sas_ss_flags;
1088
1089         struct callback_head            *task_works;
1090
1091 #ifdef CONFIG_AUDIT
1092 #ifdef CONFIG_AUDITSYSCALL
1093         struct audit_context            *audit_context;
1094 #endif
1095         kuid_t                          loginuid;
1096         unsigned int                    sessionid;
1097 #endif
1098         struct seccomp                  seccomp;
1099         struct syscall_user_dispatch    syscall_dispatch;
1100
1101         /* Thread group tracking: */
1102         u64                             parent_exec_id;
1103         u64                             self_exec_id;
1104
1105         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1106         spinlock_t                      alloc_lock;
1107
1108         /* Protection of the PI data structures: */
1109         raw_spinlock_t                  pi_lock;
1110
1111         struct wake_q_node              wake_q;
1112
1113 #ifdef CONFIG_RT_MUTEXES
1114         /* PI waiters blocked on a rt_mutex held by this task: */
1115         struct rb_root_cached           pi_waiters;
1116         /* Updated under owner's pi_lock and rq lock */
1117         struct task_struct              *pi_top_task;
1118         /* Deadlock detection and priority inheritance handling: */
1119         struct rt_mutex_waiter          *pi_blocked_on;
1120 #endif
1121
1122 #ifdef CONFIG_DEBUG_MUTEXES
1123         /* Mutex deadlock detection: */
1124         struct mutex_waiter             *blocked_on;
1125 #endif
1126
1127 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1128         int                             non_block_count;
1129 #endif
1130
1131 #ifdef CONFIG_TRACE_IRQFLAGS
1132         struct irqtrace_events          irqtrace;
1133         unsigned int                    hardirq_threaded;
1134         u64                             hardirq_chain_key;
1135         int                             softirqs_enabled;
1136         int                             softirq_context;
1137         int                             irq_config;
1138 #endif
1139 #ifdef CONFIG_PREEMPT_RT
1140         int                             softirq_disable_cnt;
1141 #endif
1142
1143 #ifdef CONFIG_LOCKDEP
1144 # define MAX_LOCK_DEPTH                 48UL
1145         u64                             curr_chain_key;
1146         int                             lockdep_depth;
1147         unsigned int                    lockdep_recursion;
1148         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1149 #endif
1150
1151 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1152         unsigned int                    in_ubsan;
1153 #endif
1154
1155         /* Journalling filesystem info: */
1156         void                            *journal_info;
1157
1158         /* Stacked block device info: */
1159         struct bio_list                 *bio_list;
1160
1161         /* Stack plugging: */
1162         struct blk_plug                 *plug;
1163
1164         /* VM state: */
1165         struct reclaim_state            *reclaim_state;
1166
1167         struct backing_dev_info         *backing_dev_info;
1168
1169         struct io_context               *io_context;
1170
1171 #ifdef CONFIG_COMPACTION
1172         struct capture_control          *capture_control;
1173 #endif
1174         /* Ptrace state: */
1175         unsigned long                   ptrace_message;
1176         kernel_siginfo_t                *last_siginfo;
1177
1178         struct task_io_accounting       ioac;
1179 #ifdef CONFIG_PSI
1180         /* Pressure stall state */
1181         unsigned int                    psi_flags;
1182 #endif
1183 #ifdef CONFIG_TASK_XACCT
1184         /* Accumulated RSS usage: */
1185         u64                             acct_rss_mem1;
1186         /* Accumulated virtual memory usage: */
1187         u64                             acct_vm_mem1;
1188         /* stime + utime since last update: */
1189         u64                             acct_timexpd;
1190 #endif
1191 #ifdef CONFIG_CPUSETS
1192         /* Protected by ->alloc_lock: */
1193         nodemask_t                      mems_allowed;
1194         /* Sequence number to catch updates: */
1195         seqcount_spinlock_t             mems_allowed_seq;
1196         int                             cpuset_mem_spread_rotor;
1197         int                             cpuset_slab_spread_rotor;
1198 #endif
1199 #ifdef CONFIG_CGROUPS
1200         /* Control Group info protected by css_set_lock: */
1201         struct css_set __rcu            *cgroups;
1202         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1203         struct list_head                cg_list;
1204 #endif
1205 #ifdef CONFIG_X86_CPU_RESCTRL
1206         u32                             closid;
1207         u32                             rmid;
1208 #endif
1209 #ifdef CONFIG_FUTEX
1210         struct robust_list_head __user  *robust_list;
1211 #ifdef CONFIG_COMPAT
1212         struct compat_robust_list_head __user *compat_robust_list;
1213 #endif
1214         struct list_head                pi_state_list;
1215         struct futex_pi_state           *pi_state_cache;
1216         struct mutex                    futex_exit_mutex;
1217         unsigned int                    futex_state;
1218 #endif
1219 #ifdef CONFIG_PERF_EVENTS
1220         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1221         struct mutex                    perf_event_mutex;
1222         struct list_head                perf_event_list;
1223 #endif
1224 #ifdef CONFIG_DEBUG_PREEMPT
1225         unsigned long                   preempt_disable_ip;
1226 #endif
1227 #ifdef CONFIG_NUMA
1228         /* Protected by alloc_lock: */
1229         struct mempolicy                *mempolicy;
1230         short                           il_prev;
1231         short                           pref_node_fork;
1232 #endif
1233 #ifdef CONFIG_NUMA_BALANCING
1234         int                             numa_scan_seq;
1235         unsigned int                    numa_scan_period;
1236         unsigned int                    numa_scan_period_max;
1237         int                             numa_preferred_nid;
1238         unsigned long                   numa_migrate_retry;
1239         /* Migration stamp: */
1240         u64                             node_stamp;
1241         u64                             last_task_numa_placement;
1242         u64                             last_sum_exec_runtime;
1243         struct callback_head            numa_work;
1244
1245         /*
1246          * This pointer is only modified for current in syscall and
1247          * pagefault context (and for tasks being destroyed), so it can be read
1248          * from any of the following contexts:
1249          *  - RCU read-side critical section
1250          *  - current->numa_group from everywhere
1251          *  - task's runqueue locked, task not running
1252          */
1253         struct numa_group __rcu         *numa_group;
1254
1255         /*
1256          * numa_faults is an array split into four regions:
1257          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1258          * in this precise order.
1259          *
1260          * faults_memory: Exponential decaying average of faults on a per-node
1261          * basis. Scheduling placement decisions are made based on these
1262          * counts. The values remain static for the duration of a PTE scan.
1263          * faults_cpu: Track the nodes the process was running on when a NUMA
1264          * hinting fault was incurred.
1265          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1266          * during the current scan window. When the scan completes, the counts
1267          * in faults_memory and faults_cpu decay and these values are copied.
1268          */
1269         unsigned long                   *numa_faults;
1270         unsigned long                   total_numa_faults;
1271
1272         /*
1273          * numa_faults_locality tracks if faults recorded during the last
1274          * scan window were remote/local or failed to migrate. The task scan
1275          * period is adapted based on the locality of the faults with different
1276          * weights depending on whether they were shared or private faults
1277          */
1278         unsigned long                   numa_faults_locality[3];
1279
1280         unsigned long                   numa_pages_migrated;
1281 #endif /* CONFIG_NUMA_BALANCING */
1282
1283 #ifdef CONFIG_RSEQ
1284         struct rseq __user *rseq;
1285         u32 rseq_sig;
1286         /*
1287          * RmW on rseq_event_mask must be performed atomically
1288          * with respect to preemption.
1289          */
1290         unsigned long rseq_event_mask;
1291 #endif
1292
1293         struct tlbflush_unmap_batch     tlb_ubc;
1294
1295         union {
1296                 refcount_t              rcu_users;
1297                 struct rcu_head         rcu;
1298         };
1299
1300         /* Cache last used pipe for splice(): */
1301         struct pipe_inode_info          *splice_pipe;
1302
1303         struct page_frag                task_frag;
1304
1305 #ifdef CONFIG_TASK_DELAY_ACCT
1306         struct task_delay_info          *delays;
1307 #endif
1308
1309 #ifdef CONFIG_FAULT_INJECTION
1310         int                             make_it_fail;
1311         unsigned int                    fail_nth;
1312 #endif
1313         /*
1314          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1315          * balance_dirty_pages() for a dirty throttling pause:
1316          */
1317         int                             nr_dirtied;
1318         int                             nr_dirtied_pause;
1319         /* Start of a write-and-pause period: */
1320         unsigned long                   dirty_paused_when;
1321
1322 #ifdef CONFIG_LATENCYTOP
1323         int                             latency_record_count;
1324         struct latency_record           latency_record[LT_SAVECOUNT];
1325 #endif
1326         /*
1327          * Time slack values; these are used to round up poll() and
1328          * select() etc timeout values. These are in nanoseconds.
1329          */
1330         u64                             timer_slack_ns;
1331         u64                             default_timer_slack_ns;
1332
1333 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1334         unsigned int                    kasan_depth;
1335 #endif
1336
1337 #ifdef CONFIG_KCSAN
1338         struct kcsan_ctx                kcsan_ctx;
1339 #ifdef CONFIG_TRACE_IRQFLAGS
1340         struct irqtrace_events          kcsan_save_irqtrace;
1341 #endif
1342 #endif
1343
1344 #if IS_ENABLED(CONFIG_KUNIT)
1345         struct kunit                    *kunit_test;
1346 #endif
1347
1348 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1349         /* Index of current stored address in ret_stack: */
1350         int                             curr_ret_stack;
1351         int                             curr_ret_depth;
1352
1353         /* Stack of return addresses for return function tracing: */
1354         struct ftrace_ret_stack         *ret_stack;
1355
1356         /* Timestamp for last schedule: */
1357         unsigned long long              ftrace_timestamp;
1358
1359         /*
1360          * Number of functions that haven't been traced
1361          * because of depth overrun:
1362          */
1363         atomic_t                        trace_overrun;
1364
1365         /* Pause tracing: */
1366         atomic_t                        tracing_graph_pause;
1367 #endif
1368
1369 #ifdef CONFIG_TRACING
1370         /* State flags for use by tracers: */
1371         unsigned long                   trace;
1372
1373         /* Bitmask and counter of trace recursion: */
1374         unsigned long                   trace_recursion;
1375 #endif /* CONFIG_TRACING */
1376
1377 #ifdef CONFIG_KCOV
1378         /* See kernel/kcov.c for more details. */
1379
1380         /* Coverage collection mode enabled for this task (0 if disabled): */
1381         unsigned int                    kcov_mode;
1382
1383         /* Size of the kcov_area: */
1384         unsigned int                    kcov_size;
1385
1386         /* Buffer for coverage collection: */
1387         void                            *kcov_area;
1388
1389         /* KCOV descriptor wired with this task or NULL: */
1390         struct kcov                     *kcov;
1391
1392         /* KCOV common handle for remote coverage collection: */
1393         u64                             kcov_handle;
1394
1395         /* KCOV sequence number: */
1396         int                             kcov_sequence;
1397
1398         /* Collect coverage from softirq context: */
1399         unsigned int                    kcov_softirq;
1400 #endif
1401
1402 #ifdef CONFIG_MEMCG
1403         struct mem_cgroup               *memcg_in_oom;
1404         gfp_t                           memcg_oom_gfp_mask;
1405         int                             memcg_oom_order;
1406
1407         /* Number of pages to reclaim on returning to userland: */
1408         unsigned int                    memcg_nr_pages_over_high;
1409
1410         /* Used by memcontrol for targeted memcg charge: */
1411         struct mem_cgroup               *active_memcg;
1412 #endif
1413
1414 #ifdef CONFIG_BLK_CGROUP
1415         struct request_queue            *throttle_queue;
1416 #endif
1417
1418 #ifdef CONFIG_UPROBES
1419         struct uprobe_task              *utask;
1420 #endif
1421 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1422         unsigned int                    sequential_io;
1423         unsigned int                    sequential_io_avg;
1424 #endif
1425         struct kmap_ctrl                kmap_ctrl;
1426 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1427         unsigned long                   task_state_change;
1428 # ifdef CONFIG_PREEMPT_RT
1429         unsigned long                   saved_state_change;
1430 # endif
1431 #endif
1432         int                             pagefault_disabled;
1433 #ifdef CONFIG_MMU
1434         struct task_struct              *oom_reaper_list;
1435 #endif
1436 #ifdef CONFIG_VMAP_STACK
1437         struct vm_struct                *stack_vm_area;
1438 #endif
1439 #ifdef CONFIG_THREAD_INFO_IN_TASK
1440         /* A live task holds one reference: */
1441         refcount_t                      stack_refcount;
1442 #endif
1443 #ifdef CONFIG_LIVEPATCH
1444         int patch_state;
1445 #endif
1446 #ifdef CONFIG_SECURITY
1447         /* Used by LSM modules for access restriction: */
1448         void                            *security;
1449 #endif
1450 #ifdef CONFIG_BPF_SYSCALL
1451         /* Used by BPF task local storage */
1452         struct bpf_local_storage __rcu  *bpf_storage;
1453         /* Used for BPF run context */
1454         struct bpf_run_ctx              *bpf_ctx;
1455 #endif
1456
1457 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1458         unsigned long                   lowest_stack;
1459         unsigned long                   prev_lowest_stack;
1460 #endif
1461
1462 #ifdef CONFIG_X86_MCE
1463         void __user                     *mce_vaddr;
1464         __u64                           mce_kflags;
1465         u64                             mce_addr;
1466         __u64                           mce_ripv : 1,
1467                                         mce_whole_page : 1,
1468                                         __mce_reserved : 62;
1469         struct callback_head            mce_kill_me;
1470         int                             mce_count;
1471 #endif
1472
1473 #ifdef CONFIG_KRETPROBES
1474         struct llist_head               kretprobe_instances;
1475 #endif
1476
1477 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1478         /*
1479          * If L1D flush is supported on mm context switch
1480          * then we use this callback head to queue kill work
1481          * to kill tasks that are not running on SMT disabled
1482          * cores
1483          */
1484         struct callback_head            l1d_flush_kill;
1485 #endif
1486
1487         /*
1488          * New fields for task_struct should be added above here, so that
1489          * they are included in the randomized portion of task_struct.
1490          */
1491         randomized_struct_fields_end
1492
1493         /* CPU-specific state of this task: */
1494         struct thread_struct            thread;
1495
1496         /*
1497          * WARNING: on x86, 'thread_struct' contains a variable-sized
1498          * structure.  It *MUST* be at the end of 'task_struct'.
1499          *
1500          * Do not put anything below here!
1501          */
1502 };
1503
1504 static inline struct pid *task_pid(struct task_struct *task)
1505 {
1506         return task->thread_pid;
1507 }
1508
1509 /*
1510  * the helpers to get the task's different pids as they are seen
1511  * from various namespaces
1512  *
1513  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1514  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1515  *                     current.
1516  * task_xid_nr_ns()  : id seen from the ns specified;
1517  *
1518  * see also pid_nr() etc in include/linux/pid.h
1519  */
1520 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1521
1522 static inline pid_t task_pid_nr(struct task_struct *tsk)
1523 {
1524         return tsk->pid;
1525 }
1526
1527 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1528 {
1529         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1530 }
1531
1532 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1533 {
1534         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1535 }
1536
1537
1538 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1539 {
1540         return tsk->tgid;
1541 }
1542
1543 /**
1544  * pid_alive - check that a task structure is not stale
1545  * @p: Task structure to be checked.
1546  *
1547  * Test if a process is not yet dead (at most zombie state)
1548  * If pid_alive fails, then pointers within the task structure
1549  * can be stale and must not be dereferenced.
1550  *
1551  * Return: 1 if the process is alive. 0 otherwise.
1552  */
1553 static inline int pid_alive(const struct task_struct *p)
1554 {
1555         return p->thread_pid != NULL;
1556 }
1557
1558 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1559 {
1560         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1561 }
1562
1563 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1564 {
1565         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1566 }
1567
1568
1569 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1570 {
1571         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1572 }
1573
1574 static inline pid_t task_session_vnr(struct task_struct *tsk)
1575 {
1576         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1577 }
1578
1579 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1580 {
1581         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1582 }
1583
1584 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1585 {
1586         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1587 }
1588
1589 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1590 {
1591         pid_t pid = 0;
1592
1593         rcu_read_lock();
1594         if (pid_alive(tsk))
1595                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1596         rcu_read_unlock();
1597
1598         return pid;
1599 }
1600
1601 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1602 {
1603         return task_ppid_nr_ns(tsk, &init_pid_ns);
1604 }
1605
1606 /* Obsolete, do not use: */
1607 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1608 {
1609         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1610 }
1611
1612 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1613 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1614
1615 static inline unsigned int task_state_index(struct task_struct *tsk)
1616 {
1617         unsigned int tsk_state = READ_ONCE(tsk->__state);
1618         unsigned int state = (tsk_state | tsk->exit_state) & TASK_REPORT;
1619
1620         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1621
1622         if (tsk_state == TASK_IDLE)
1623                 state = TASK_REPORT_IDLE;
1624
1625         return fls(state);
1626 }
1627
1628 static inline char task_index_to_char(unsigned int state)
1629 {
1630         static const char state_char[] = "RSDTtXZPI";
1631
1632         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1633
1634         return state_char[state];
1635 }
1636
1637 static inline char task_state_to_char(struct task_struct *tsk)
1638 {
1639         return task_index_to_char(task_state_index(tsk));
1640 }
1641
1642 /**
1643  * is_global_init - check if a task structure is init. Since init
1644  * is free to have sub-threads we need to check tgid.
1645  * @tsk: Task structure to be checked.
1646  *
1647  * Check if a task structure is the first user space task the kernel created.
1648  *
1649  * Return: 1 if the task structure is init. 0 otherwise.
1650  */
1651 static inline int is_global_init(struct task_struct *tsk)
1652 {
1653         return task_tgid_nr(tsk) == 1;
1654 }
1655
1656 extern struct pid *cad_pid;
1657
1658 /*
1659  * Per process flags
1660  */
1661 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1662 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1663 #define PF_EXITING              0x00000004      /* Getting shut down */
1664 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1665 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1666 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1667 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1668 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1669 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1670 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1671 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1672 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1673 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1674 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1675 #define PF_USED_ASYNC           0x00004000      /* Used async_schedule*(), used by module init */
1676 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1677 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1678 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1679 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1680 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1681 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1682                                                  * I am cleaning dirty pages from some other bdi. */
1683 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1684 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1685 #define PF_SWAPWRITE            0x00800000      /* Allowed to write to swap */
1686 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1687 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1688 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1689 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1690 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1691
1692 /*
1693  * Only the _current_ task can read/write to tsk->flags, but other
1694  * tasks can access tsk->flags in readonly mode for example
1695  * with tsk_used_math (like during threaded core dumping).
1696  * There is however an exception to this rule during ptrace
1697  * or during fork: the ptracer task is allowed to write to the
1698  * child->flags of its traced child (same goes for fork, the parent
1699  * can write to the child->flags), because we're guaranteed the
1700  * child is not running and in turn not changing child->flags
1701  * at the same time the parent does it.
1702  */
1703 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1704 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1705 #define clear_used_math()                       clear_stopped_child_used_math(current)
1706 #define set_used_math()                         set_stopped_child_used_math(current)
1707
1708 #define conditional_stopped_child_used_math(condition, child) \
1709         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1710
1711 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1712
1713 #define copy_to_stopped_child_used_math(child) \
1714         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1715
1716 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1717 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1718 #define used_math()                             tsk_used_math(current)
1719
1720 static __always_inline bool is_percpu_thread(void)
1721 {
1722 #ifdef CONFIG_SMP
1723         return (current->flags & PF_NO_SETAFFINITY) &&
1724                 (current->nr_cpus_allowed  == 1);
1725 #else
1726         return true;
1727 #endif
1728 }
1729
1730 /* Per-process atomic flags. */
1731 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1732 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1733 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1734 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1735 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1736 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1737 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1738 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1739
1740 #define TASK_PFA_TEST(name, func)                                       \
1741         static inline bool task_##func(struct task_struct *p)           \
1742         { return test_bit(PFA_##name, &p->atomic_flags); }
1743
1744 #define TASK_PFA_SET(name, func)                                        \
1745         static inline void task_set_##func(struct task_struct *p)       \
1746         { set_bit(PFA_##name, &p->atomic_flags); }
1747
1748 #define TASK_PFA_CLEAR(name, func)                                      \
1749         static inline void task_clear_##func(struct task_struct *p)     \
1750         { clear_bit(PFA_##name, &p->atomic_flags); }
1751
1752 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1753 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1754
1755 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1756 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1757 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1758
1759 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1760 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1761 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1762
1763 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1764 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1765 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1766
1767 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1768 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1769 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1770
1771 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1772 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1773
1774 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1775 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1776 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1777
1778 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1779 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1780
1781 static inline void
1782 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1783 {
1784         current->flags &= ~flags;
1785         current->flags |= orig_flags & flags;
1786 }
1787
1788 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1789 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1790 #ifdef CONFIG_SMP
1791 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1792 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1793 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1794 extern void release_user_cpus_ptr(struct task_struct *p);
1795 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1796 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1797 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1798 #else
1799 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1800 {
1801 }
1802 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1803 {
1804         if (!cpumask_test_cpu(0, new_mask))
1805                 return -EINVAL;
1806         return 0;
1807 }
1808 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1809 {
1810         if (src->user_cpus_ptr)
1811                 return -EINVAL;
1812         return 0;
1813 }
1814 static inline void release_user_cpus_ptr(struct task_struct *p)
1815 {
1816         WARN_ON(p->user_cpus_ptr);
1817 }
1818
1819 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1820 {
1821         return 0;
1822 }
1823 #endif
1824
1825 extern int yield_to(struct task_struct *p, bool preempt);
1826 extern void set_user_nice(struct task_struct *p, long nice);
1827 extern int task_prio(const struct task_struct *p);
1828
1829 /**
1830  * task_nice - return the nice value of a given task.
1831  * @p: the task in question.
1832  *
1833  * Return: The nice value [ -20 ... 0 ... 19 ].
1834  */
1835 static inline int task_nice(const struct task_struct *p)
1836 {
1837         return PRIO_TO_NICE((p)->static_prio);
1838 }
1839
1840 extern int can_nice(const struct task_struct *p, const int nice);
1841 extern int task_curr(const struct task_struct *p);
1842 extern int idle_cpu(int cpu);
1843 extern int available_idle_cpu(int cpu);
1844 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1845 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1846 extern void sched_set_fifo(struct task_struct *p);
1847 extern void sched_set_fifo_low(struct task_struct *p);
1848 extern void sched_set_normal(struct task_struct *p, int nice);
1849 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1850 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1851 extern struct task_struct *idle_task(int cpu);
1852
1853 /**
1854  * is_idle_task - is the specified task an idle task?
1855  * @p: the task in question.
1856  *
1857  * Return: 1 if @p is an idle task. 0 otherwise.
1858  */
1859 static __always_inline bool is_idle_task(const struct task_struct *p)
1860 {
1861         return !!(p->flags & PF_IDLE);
1862 }
1863
1864 extern struct task_struct *curr_task(int cpu);
1865 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1866
1867 void yield(void);
1868
1869 union thread_union {
1870 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1871         struct task_struct task;
1872 #endif
1873 #ifndef CONFIG_THREAD_INFO_IN_TASK
1874         struct thread_info thread_info;
1875 #endif
1876         unsigned long stack[THREAD_SIZE/sizeof(long)];
1877 };
1878
1879 #ifndef CONFIG_THREAD_INFO_IN_TASK
1880 extern struct thread_info init_thread_info;
1881 #endif
1882
1883 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1884
1885 #ifdef CONFIG_THREAD_INFO_IN_TASK
1886 # define task_thread_info(task) (&(task)->thread_info)
1887 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1888 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1889 #endif
1890
1891 /*
1892  * find a task by one of its numerical ids
1893  *
1894  * find_task_by_pid_ns():
1895  *      finds a task by its pid in the specified namespace
1896  * find_task_by_vpid():
1897  *      finds a task by its virtual pid
1898  *
1899  * see also find_vpid() etc in include/linux/pid.h
1900  */
1901
1902 extern struct task_struct *find_task_by_vpid(pid_t nr);
1903 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1904
1905 /*
1906  * find a task by its virtual pid and get the task struct
1907  */
1908 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1909
1910 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1911 extern int wake_up_process(struct task_struct *tsk);
1912 extern void wake_up_new_task(struct task_struct *tsk);
1913
1914 #ifdef CONFIG_SMP
1915 extern void kick_process(struct task_struct *tsk);
1916 #else
1917 static inline void kick_process(struct task_struct *tsk) { }
1918 #endif
1919
1920 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1921
1922 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1923 {
1924         __set_task_comm(tsk, from, false);
1925 }
1926
1927 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1928 #define get_task_comm(buf, tsk) ({                      \
1929         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1930         __get_task_comm(buf, sizeof(buf), tsk);         \
1931 })
1932
1933 #ifdef CONFIG_SMP
1934 static __always_inline void scheduler_ipi(void)
1935 {
1936         /*
1937          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1938          * TIF_NEED_RESCHED remotely (for the first time) will also send
1939          * this IPI.
1940          */
1941         preempt_fold_need_resched();
1942 }
1943 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1944 #else
1945 static inline void scheduler_ipi(void) { }
1946 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1947 {
1948         return 1;
1949 }
1950 #endif
1951
1952 /*
1953  * Set thread flags in other task's structures.
1954  * See asm/thread_info.h for TIF_xxxx flags available:
1955  */
1956 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1957 {
1958         set_ti_thread_flag(task_thread_info(tsk), flag);
1959 }
1960
1961 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1962 {
1963         clear_ti_thread_flag(task_thread_info(tsk), flag);
1964 }
1965
1966 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1967                                           bool value)
1968 {
1969         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1970 }
1971
1972 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1973 {
1974         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1975 }
1976
1977 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1978 {
1979         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1980 }
1981
1982 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1983 {
1984         return test_ti_thread_flag(task_thread_info(tsk), flag);
1985 }
1986
1987 static inline void set_tsk_need_resched(struct task_struct *tsk)
1988 {
1989         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1990 }
1991
1992 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1993 {
1994         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1995 }
1996
1997 static inline int test_tsk_need_resched(struct task_struct *tsk)
1998 {
1999         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2000 }
2001
2002 /*
2003  * cond_resched() and cond_resched_lock(): latency reduction via
2004  * explicit rescheduling in places that are safe. The return
2005  * value indicates whether a reschedule was done in fact.
2006  * cond_resched_lock() will drop the spinlock before scheduling,
2007  */
2008 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2009 extern int __cond_resched(void);
2010
2011 #ifdef CONFIG_PREEMPT_DYNAMIC
2012
2013 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2014
2015 static __always_inline int _cond_resched(void)
2016 {
2017         return static_call_mod(cond_resched)();
2018 }
2019
2020 #else
2021
2022 static inline int _cond_resched(void)
2023 {
2024         return __cond_resched();
2025 }
2026
2027 #endif /* CONFIG_PREEMPT_DYNAMIC */
2028
2029 #else
2030
2031 static inline int _cond_resched(void) { return 0; }
2032
2033 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2034
2035 #define cond_resched() ({                       \
2036         __might_resched(__FILE__, __LINE__, 0); \
2037         _cond_resched();                        \
2038 })
2039
2040 extern int __cond_resched_lock(spinlock_t *lock);
2041 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2042 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2043
2044 #define MIGHT_RESCHED_RCU_SHIFT         8
2045 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2046
2047 #ifndef CONFIG_PREEMPT_RT
2048 /*
2049  * Non RT kernels have an elevated preempt count due to the held lock,
2050  * but are not allowed to be inside a RCU read side critical section
2051  */
2052 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2053 #else
2054 /*
2055  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2056  * cond_resched*lock() has to take that into account because it checks for
2057  * preempt_count() and rcu_preempt_depth().
2058  */
2059 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2060         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2061 #endif
2062
2063 #define cond_resched_lock(lock) ({                                              \
2064         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2065         __cond_resched_lock(lock);                                              \
2066 })
2067
2068 #define cond_resched_rwlock_read(lock) ({                                       \
2069         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2070         __cond_resched_rwlock_read(lock);                                       \
2071 })
2072
2073 #define cond_resched_rwlock_write(lock) ({                                      \
2074         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2075         __cond_resched_rwlock_write(lock);                                      \
2076 })
2077
2078 static inline void cond_resched_rcu(void)
2079 {
2080 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2081         rcu_read_unlock();
2082         cond_resched();
2083         rcu_read_lock();
2084 #endif
2085 }
2086
2087 /*
2088  * Does a critical section need to be broken due to another
2089  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2090  * but a general need for low latency)
2091  */
2092 static inline int spin_needbreak(spinlock_t *lock)
2093 {
2094 #ifdef CONFIG_PREEMPTION
2095         return spin_is_contended(lock);
2096 #else
2097         return 0;
2098 #endif
2099 }
2100
2101 /*
2102  * Check if a rwlock is contended.
2103  * Returns non-zero if there is another task waiting on the rwlock.
2104  * Returns zero if the lock is not contended or the system / underlying
2105  * rwlock implementation does not support contention detection.
2106  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2107  * for low latency.
2108  */
2109 static inline int rwlock_needbreak(rwlock_t *lock)
2110 {
2111 #ifdef CONFIG_PREEMPTION
2112         return rwlock_is_contended(lock);
2113 #else
2114         return 0;
2115 #endif
2116 }
2117
2118 static __always_inline bool need_resched(void)
2119 {
2120         return unlikely(tif_need_resched());
2121 }
2122
2123 /*
2124  * Wrappers for p->thread_info->cpu access. No-op on UP.
2125  */
2126 #ifdef CONFIG_SMP
2127
2128 static inline unsigned int task_cpu(const struct task_struct *p)
2129 {
2130         return READ_ONCE(task_thread_info(p)->cpu);
2131 }
2132
2133 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2134
2135 #else
2136
2137 static inline unsigned int task_cpu(const struct task_struct *p)
2138 {
2139         return 0;
2140 }
2141
2142 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2143 {
2144 }
2145
2146 #endif /* CONFIG_SMP */
2147
2148 extern bool sched_task_on_rq(struct task_struct *p);
2149 extern unsigned long get_wchan(struct task_struct *p);
2150
2151 /*
2152  * In order to reduce various lock holder preemption latencies provide an
2153  * interface to see if a vCPU is currently running or not.
2154  *
2155  * This allows us to terminate optimistic spin loops and block, analogous to
2156  * the native optimistic spin heuristic of testing if the lock owner task is
2157  * running or not.
2158  */
2159 #ifndef vcpu_is_preempted
2160 static inline bool vcpu_is_preempted(int cpu)
2161 {
2162         return false;
2163 }
2164 #endif
2165
2166 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2167 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2168
2169 #ifndef TASK_SIZE_OF
2170 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2171 #endif
2172
2173 #ifdef CONFIG_SMP
2174 /* Returns effective CPU energy utilization, as seen by the scheduler */
2175 unsigned long sched_cpu_util(int cpu, unsigned long max);
2176 #endif /* CONFIG_SMP */
2177
2178 #ifdef CONFIG_RSEQ
2179
2180 /*
2181  * Map the event mask on the user-space ABI enum rseq_cs_flags
2182  * for direct mask checks.
2183  */
2184 enum rseq_event_mask_bits {
2185         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2186         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2187         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2188 };
2189
2190 enum rseq_event_mask {
2191         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2192         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2193         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2194 };
2195
2196 static inline void rseq_set_notify_resume(struct task_struct *t)
2197 {
2198         if (t->rseq)
2199                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2200 }
2201
2202 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2203
2204 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2205                                              struct pt_regs *regs)
2206 {
2207         if (current->rseq)
2208                 __rseq_handle_notify_resume(ksig, regs);
2209 }
2210
2211 static inline void rseq_signal_deliver(struct ksignal *ksig,
2212                                        struct pt_regs *regs)
2213 {
2214         preempt_disable();
2215         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2216         preempt_enable();
2217         rseq_handle_notify_resume(ksig, regs);
2218 }
2219
2220 /* rseq_preempt() requires preemption to be disabled. */
2221 static inline void rseq_preempt(struct task_struct *t)
2222 {
2223         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2224         rseq_set_notify_resume(t);
2225 }
2226
2227 /* rseq_migrate() requires preemption to be disabled. */
2228 static inline void rseq_migrate(struct task_struct *t)
2229 {
2230         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2231         rseq_set_notify_resume(t);
2232 }
2233
2234 /*
2235  * If parent process has a registered restartable sequences area, the
2236  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2237  */
2238 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2239 {
2240         if (clone_flags & CLONE_VM) {
2241                 t->rseq = NULL;
2242                 t->rseq_sig = 0;
2243                 t->rseq_event_mask = 0;
2244         } else {
2245                 t->rseq = current->rseq;
2246                 t->rseq_sig = current->rseq_sig;
2247                 t->rseq_event_mask = current->rseq_event_mask;
2248         }
2249 }
2250
2251 static inline void rseq_execve(struct task_struct *t)
2252 {
2253         t->rseq = NULL;
2254         t->rseq_sig = 0;
2255         t->rseq_event_mask = 0;
2256 }
2257
2258 #else
2259
2260 static inline void rseq_set_notify_resume(struct task_struct *t)
2261 {
2262 }
2263 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2264                                              struct pt_regs *regs)
2265 {
2266 }
2267 static inline void rseq_signal_deliver(struct ksignal *ksig,
2268                                        struct pt_regs *regs)
2269 {
2270 }
2271 static inline void rseq_preempt(struct task_struct *t)
2272 {
2273 }
2274 static inline void rseq_migrate(struct task_struct *t)
2275 {
2276 }
2277 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2278 {
2279 }
2280 static inline void rseq_execve(struct task_struct *t)
2281 {
2282 }
2283
2284 #endif
2285
2286 #ifdef CONFIG_DEBUG_RSEQ
2287
2288 void rseq_syscall(struct pt_regs *regs);
2289
2290 #else
2291
2292 static inline void rseq_syscall(struct pt_regs *regs)
2293 {
2294 }
2295
2296 #endif
2297
2298 const struct sched_avg *sched_trace_cfs_rq_avg(struct cfs_rq *cfs_rq);
2299 char *sched_trace_cfs_rq_path(struct cfs_rq *cfs_rq, char *str, int len);
2300 int sched_trace_cfs_rq_cpu(struct cfs_rq *cfs_rq);
2301
2302 const struct sched_avg *sched_trace_rq_avg_rt(struct rq *rq);
2303 const struct sched_avg *sched_trace_rq_avg_dl(struct rq *rq);
2304 const struct sched_avg *sched_trace_rq_avg_irq(struct rq *rq);
2305
2306 int sched_trace_rq_cpu(struct rq *rq);
2307 int sched_trace_rq_cpu_capacity(struct rq *rq);
2308 int sched_trace_rq_nr_running(struct rq *rq);
2309
2310 const struct cpumask *sched_trace_rd_span(struct root_domain *rd);
2311
2312 #ifdef CONFIG_SCHED_CORE
2313 extern void sched_core_free(struct task_struct *tsk);
2314 extern void sched_core_fork(struct task_struct *p);
2315 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2316                                 unsigned long uaddr);
2317 #else
2318 static inline void sched_core_free(struct task_struct *tsk) { }
2319 static inline void sched_core_fork(struct task_struct *p) { }
2320 #endif
2321
2322 #endif