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         u64                             exec_max;
 507         u64                             slice_max;
 508
 509         u64                             nr_migrations_cold;
 510         u64                             nr_failed_migrations_affine;
 511         u64                             nr_failed_migrations_running;
 512         u64                             nr_failed_migrations_hot;
 513         u64                             nr_forced_migrations;
 514
 515         u64                             nr_wakeups;
 516         u64                             nr_wakeups_sync;
 517         u64                             nr_wakeups_migrate;
 518         u64                             nr_wakeups_local;
 519         u64                             nr_wakeups_remote;
 520         u64                             nr_wakeups_affine;
 521         u64                             nr_wakeups_affine_attempts;
 522         u64                             nr_wakeups_passive;
 523         u64                             nr_wakeups_idle;
 524 #endif
 525 };
 526
 527 struct sched_entity {
 528         /* For load-balancing: */
 529         struct load_weight              load;
 530         struct rb_node                  run_node;
 531         struct list_head                group_node;
 532         unsigned int                    on_rq;
 533
 534         u64                             exec_start;
 535         u64                             sum_exec_runtime;
 536         u64                             vruntime;
 537         u64                             prev_sum_exec_runtime;
 538
 539         u64                             nr_migrations;
 540
 541         struct sched_statistics         statistics;
 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 #ifdef CONFIG_THREAD_INFO_IN_TASK
 754         /* Current CPU: */
 755         unsigned int                    cpu;
 756 #endif
 757         unsigned int                    wakee_flips;
 758         unsigned long                   wakee_flip_decay_ts;
 759         struct task_struct              *last_wakee;
 760
 761         /*
 762          * recent_used_cpu is initially set as the last CPU used by a task
 763          * that wakes affine another task. Waker/wakee relationships can
 764          * push tasks around a CPU where each wakeup moves to the next one.
 765          * Tracking a recently used CPU allows a quick search for a recently
 766          * used CPU that may be idle.
 767          */
 768         int                             recent_used_cpu;
 769         int                             wake_cpu;
 770 #endif
 771         int                             on_rq;
 772
 773         int                             prio;
 774         int                             static_prio;
 775         int                             normal_prio;
 776         unsigned int                    rt_priority;
 777
 778         const struct sched_class        *sched_class;
 779         struct sched_entity             se;
 780         struct sched_rt_entity          rt;
 781         struct sched_dl_entity          dl;
 782
 783 #ifdef CONFIG_SCHED_CORE
 784         struct rb_node                  core_node;
 785         unsigned long                   core_cookie;
 786         unsigned int                    core_occupation;
 787 #endif
 788
 789 #ifdef CONFIG_CGROUP_SCHED
 790         struct task_group               *sched_task_group;
 791 #endif
 792
 793 #ifdef CONFIG_UCLAMP_TASK
 794         /*
 795          * Clamp values requested for a scheduling entity.
 796          * Must be updated with task_rq_lock() held.
 797          */
 798         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 799         /*
 800          * Effective clamp values used for a scheduling entity.
 801          * Must be updated with task_rq_lock() held.
 802          */
 803         struct uclamp_se                uclamp[UCLAMP_CNT];
 804 #endif
 805
 806 #ifdef CONFIG_PREEMPT_NOTIFIERS
 807         /* List of struct preempt_notifier: */
 808         struct hlist_head               preempt_notifiers;
 809 #endif
 810
 811 #ifdef CONFIG_BLK_DEV_IO_TRACE
 812         unsigned int                    btrace_seq;
 813 #endif
 814
 815         unsigned int                    policy;
 816         int                             nr_cpus_allowed;
 817         const cpumask_t                 *cpus_ptr;
 818         cpumask_t                       *user_cpus_ptr;
 819         cpumask_t                       cpus_mask;
 820         void                            *migration_pending;
 821 #ifdef CONFIG_SMP
 822         unsigned short                  migration_disabled;
 823 #endif
 824         unsigned short                  migration_flags;
 825
 826 #ifdef CONFIG_PREEMPT_RCU
 827         int                             rcu_read_lock_nesting;
 828         union rcu_special               rcu_read_unlock_special;
 829         struct list_head                rcu_node_entry;
 830         struct rcu_node                 *rcu_blocked_node;
 831 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 832
 833 #ifdef CONFIG_TASKS_RCU
 834         unsigned long                   rcu_tasks_nvcsw;
 835         u8                              rcu_tasks_holdout;
 836         u8                              rcu_tasks_idx;
 837         int                             rcu_tasks_idle_cpu;
 838         struct list_head                rcu_tasks_holdout_list;
 839 #endif /* #ifdef CONFIG_TASKS_RCU */
 840
 841 #ifdef CONFIG_TASKS_TRACE_RCU
 842         int                             trc_reader_nesting;
 843         int                             trc_ipi_to_cpu;
 844         union rcu_special               trc_reader_special;
 845         bool                            trc_reader_checked;
 846         struct list_head                trc_holdout_list;
 847 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 848
 849         struct sched_info               sched_info;
 850
 851         struct list_head                tasks;
 852 #ifdef CONFIG_SMP
 853         struct plist_node               pushable_tasks;
 854         struct rb_node                  pushable_dl_tasks;
 855 #endif
 856
 857         struct mm_struct                *mm;
 858         struct mm_struct                *active_mm;
 859
 860         /* Per-thread vma caching: */
 861         struct vmacache                 vmacache;
 862
 863 #ifdef SPLIT_RSS_COUNTING
 864         struct task_rss_stat            rss_stat;
 865 #endif
 866         int                             exit_state;
 867         int                             exit_code;
 868         int                             exit_signal;
 869         /* The signal sent when the parent dies: */
 870         int                             pdeath_signal;
 871         /* JOBCTL_*, siglock protected: */
 872         unsigned long                   jobctl;
 873
 874         /* Used for emulating ABI behavior of previous Linux versions: */
 875         unsigned int                    personality;
 876
 877         /* Scheduler bits, serialized by scheduler locks: */
 878         unsigned                        sched_reset_on_fork:1;
 879         unsigned                        sched_contributes_to_load:1;
 880         unsigned                        sched_migrated:1;
 881 #ifdef CONFIG_PSI
 882         unsigned                        sched_psi_wake_requeue:1;
 883 #endif
 884
 885         /* Force alignment to the next boundary: */
 886         unsigned                        :0;
 887
 888         /* Unserialized, strictly 'current' */
 889
 890         /*
 891          * This field must not be in the scheduler word above due to wakelist
 892          * queueing no longer being serialized by p->on_cpu. However:
 893          *
 894          * p->XXX = X;                  ttwu()
 895          * schedule()                     if (p->on_rq && ..) // false
 896          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 897          *   deactivate_task()                ttwu_queue_wakelist())
 898          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 899          *
 900          * guarantees all stores of 'current' are visible before
 901          * ->sched_remote_wakeup gets used, so it can be in this word.
 902          */
 903         unsigned                        sched_remote_wakeup:1;
 904
 905         /* Bit to tell LSMs we're in execve(): */
 906         unsigned                        in_execve:1;
 907         unsigned                        in_iowait:1;
 908 #ifndef TIF_RESTORE_SIGMASK
 909         unsigned                        restore_sigmask:1;
 910 #endif
 911 #ifdef CONFIG_MEMCG
 912         unsigned                        in_user_fault:1;
 913 #endif
 914 #ifdef CONFIG_COMPAT_BRK
 915         unsigned                        brk_randomized:1;
 916 #endif
 917 #ifdef CONFIG_CGROUPS
 918         /* disallow userland-initiated cgroup migration */
 919         unsigned                        no_cgroup_migration:1;
 920         /* task is frozen/stopped (used by the cgroup freezer) */
 921         unsigned                        frozen:1;
 922 #endif
 923 #ifdef CONFIG_BLK_CGROUP
 924         unsigned                        use_memdelay:1;
 925 #endif
 926 #ifdef CONFIG_PSI
 927         /* Stalled due to lack of memory */
 928         unsigned                        in_memstall:1;
 929 #endif
 930 #ifdef CONFIG_PAGE_OWNER
 931         /* Used by page_owner=on to detect recursion in page tracking. */
 932         unsigned                        in_page_owner:1;
 933 #endif
 934 #ifdef CONFIG_EVENTFD
 935         /* Recursion prevention for eventfd_signal() */
 936         unsigned                        in_eventfd_signal:1;
 937 #endif
 938
 939         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 940
 941         struct restart_block            restart_block;
 942
 943         pid_t                           pid;
 944         pid_t                           tgid;
 945
 946 #ifdef CONFIG_STACKPROTECTOR
 947         /* Canary value for the -fstack-protector GCC feature: */
 948         unsigned long                   stack_canary;
 949 #endif
 950         /*
 951          * Pointers to the (original) parent process, youngest child, younger sibling,
 952          * older sibling, respectively.  (p->father can be replaced with
 953          * p->real_parent->pid)
 954          */
 955
 956         /* Real parent process: */
 957         struct task_struct __rcu        *real_parent;
 958
 959         /* Recipient of SIGCHLD, wait4() reports: */
 960         struct task_struct __rcu        *parent;
 961
 962         /*
 963          * Children/sibling form the list of natural children:
 964          */
 965         struct list_head                children;
 966         struct list_head                sibling;
 967         struct task_struct              *group_leader;
 968
 969         /*
 970          * 'ptraced' is the list of tasks this task is using ptrace() on.
 971          *
 972          * This includes both natural children and PTRACE_ATTACH targets.
 973          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 974          */
 975         struct list_head                ptraced;
 976         struct list_head                ptrace_entry;
 977
 978         /* PID/PID hash table linkage. */
 979         struct pid                      *thread_pid;
 980         struct hlist_node               pid_links[PIDTYPE_MAX];
 981         struct list_head                thread_group;
 982         struct list_head                thread_node;
 983
 984         struct completion               *vfork_done;
 985
 986         /* CLONE_CHILD_SETTID: */
 987         int __user                      *set_child_tid;
 988
 989         /* CLONE_CHILD_CLEARTID: */
 990         int __user                      *clear_child_tid;
 991
 992         /* PF_IO_WORKER */
 993         void                            *pf_io_worker;
 994
 995         u64                             utime;
 996         u64                             stime;
 997 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 998         u64                             utimescaled;
 999         u64                             stimescaled;
1000 #endif
1001         u64                             gtime;
1002         struct prev_cputime             prev_cputime;
1003 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1004         struct vtime                    vtime;
1005 #endif
1006
1007 #ifdef CONFIG_NO_HZ_FULL
1008         atomic_t                        tick_dep_mask;
1009 #endif
1010         /* Context switch counts: */
1011         unsigned long                   nvcsw;
1012         unsigned long                   nivcsw;
1013
1014         /* Monotonic time in nsecs: */
1015         u64                             start_time;
1016
1017         /* Boot based time in nsecs: */
1018         u64                             start_boottime;
1019
1020         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1021         unsigned long                   min_flt;
1022         unsigned long                   maj_flt;
1023
1024         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1025         struct posix_cputimers          posix_cputimers;
1026
1027 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1028         struct posix_cputimers_work     posix_cputimers_work;
1029 #endif
1030
1031         /* Process credentials: */
1032
1033         /* Tracer's credentials at attach: */
1034         const struct cred __rcu         *ptracer_cred;
1035
1036         /* Objective and real subjective task credentials (COW): */
1037         const struct cred __rcu         *real_cred;
1038
1039         /* Effective (overridable) subjective task credentials (COW): */
1040         const struct cred __rcu         *cred;
1041
1042 #ifdef CONFIG_KEYS
1043         /* Cached requested key. */
1044         struct key                      *cached_requested_key;
1045 #endif
1046
1047         /*
1048          * executable name, excluding path.
1049          *
1050          * - normally initialized setup_new_exec()
1051          * - access it with [gs]et_task_comm()
1052          * - lock it with task_lock()
1053          */
1054         char                            comm[TASK_COMM_LEN];
1055
1056         struct nameidata                *nameidata;
1057
1058 #ifdef CONFIG_SYSVIPC
1059         struct sysv_sem                 sysvsem;
1060         struct sysv_shm                 sysvshm;
1061 #endif
1062 #ifdef CONFIG_DETECT_HUNG_TASK
1063         unsigned long                   last_switch_count;
1064         unsigned long                   last_switch_time;
1065 #endif
1066         /* Filesystem information: */
1067         struct fs_struct                *fs;
1068
1069         /* Open file information: */
1070         struct files_struct             *files;
1071
1072 #ifdef CONFIG_IO_URING
1073         struct io_uring_task            *io_uring;
1074 #endif
1075
1076         /* Namespaces: */
1077         struct nsproxy                  *nsproxy;
1078
1079         /* Signal handlers: */
1080         struct signal_struct            *signal;
1081         struct sighand_struct __rcu             *sighand;
1082         sigset_t                        blocked;
1083         sigset_t                        real_blocked;
1084         /* Restored if set_restore_sigmask() was used: */
1085         sigset_t                        saved_sigmask;
1086         struct sigpending               pending;
1087         unsigned long                   sas_ss_sp;
1088         size_t                          sas_ss_size;
1089         unsigned int                    sas_ss_flags;
1090
1091         struct callback_head            *task_works;
1092
1093 #ifdef CONFIG_AUDIT
1094 #ifdef CONFIG_AUDITSYSCALL
1095         struct audit_context            *audit_context;
1096 #endif
1097         kuid_t                          loginuid;
1098         unsigned int                    sessionid;
1099 #endif
1100         struct seccomp                  seccomp;
1101         struct syscall_user_dispatch    syscall_dispatch;
1102
1103         /* Thread group tracking: */
1104         u64                             parent_exec_id;
1105         u64                             self_exec_id;
1106
1107         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1108         spinlock_t                      alloc_lock;
1109
1110         /* Protection of the PI data structures: */
1111         raw_spinlock_t                  pi_lock;
1112
1113         struct wake_q_node              wake_q;
1114
1115 #ifdef CONFIG_RT_MUTEXES
1116         /* PI waiters blocked on a rt_mutex held by this task: */
1117         struct rb_root_cached           pi_waiters;
1118         /* Updated under owner's pi_lock and rq lock */
1119         struct task_struct              *pi_top_task;
1120         /* Deadlock detection and priority inheritance handling: */
1121         struct rt_mutex_waiter          *pi_blocked_on;
1122 #endif
1123
1124 #ifdef CONFIG_DEBUG_MUTEXES
1125         /* Mutex deadlock detection: */
1126         struct mutex_waiter             *blocked_on;
1127 #endif
1128
1129 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1130         int                             non_block_count;
1131 #endif
1132
1133 #ifdef CONFIG_TRACE_IRQFLAGS
1134         struct irqtrace_events          irqtrace;
1135         unsigned int                    hardirq_threaded;
1136         u64                             hardirq_chain_key;
1137         int                             softirqs_enabled;
1138         int                             softirq_context;
1139         int                             irq_config;
1140 #endif
1141 #ifdef CONFIG_PREEMPT_RT
1142         int                             softirq_disable_cnt;
1143 #endif
1144
1145 #ifdef CONFIG_LOCKDEP
1146 # define MAX_LOCK_DEPTH                 48UL
1147         u64                             curr_chain_key;
1148         int                             lockdep_depth;
1149         unsigned int                    lockdep_recursion;
1150         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1151 #endif
1152
1153 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1154         unsigned int                    in_ubsan;
1155 #endif
1156
1157         /* Journalling filesystem info: */
1158         void                            *journal_info;
1159
1160         /* Stacked block device info: */
1161         struct bio_list                 *bio_list;
1162
1163 #ifdef CONFIG_BLOCK
1164         /* Stack plugging: */
1165         struct blk_plug                 *plug;
1166 #endif
1167
1168         /* VM state: */
1169         struct reclaim_state            *reclaim_state;
1170
1171         struct backing_dev_info         *backing_dev_info;
1172
1173         struct io_context               *io_context;
1174
1175 #ifdef CONFIG_COMPACTION
1176         struct capture_control          *capture_control;
1177 #endif
1178         /* Ptrace state: */
1179         unsigned long                   ptrace_message;
1180         kernel_siginfo_t                *last_siginfo;
1181
1182         struct task_io_accounting       ioac;
1183 #ifdef CONFIG_PSI
1184         /* Pressure stall state */
1185         unsigned int                    psi_flags;
1186 #endif
1187 #ifdef CONFIG_TASK_XACCT
1188         /* Accumulated RSS usage: */
1189         u64                             acct_rss_mem1;
1190         /* Accumulated virtual memory usage: */
1191         u64                             acct_vm_mem1;
1192         /* stime + utime since last update: */
1193         u64                             acct_timexpd;
1194 #endif
1195 #ifdef CONFIG_CPUSETS
1196         /* Protected by ->alloc_lock: */
1197         nodemask_t                      mems_allowed;
1198         /* Sequence number to catch updates: */
1199         seqcount_spinlock_t             mems_allowed_seq;
1200         int                             cpuset_mem_spread_rotor;
1201         int                             cpuset_slab_spread_rotor;
1202 #endif
1203 #ifdef CONFIG_CGROUPS
1204         /* Control Group info protected by css_set_lock: */
1205         struct css_set __rcu            *cgroups;
1206         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1207         struct list_head                cg_list;
1208 #endif
1209 #ifdef CONFIG_X86_CPU_RESCTRL
1210         u32                             closid;
1211         u32                             rmid;
1212 #endif
1213 #ifdef CONFIG_FUTEX
1214         struct robust_list_head __user  *robust_list;
1215 #ifdef CONFIG_COMPAT
1216         struct compat_robust_list_head __user *compat_robust_list;
1217 #endif
1218         struct list_head                pi_state_list;
1219         struct futex_pi_state           *pi_state_cache;
1220         struct mutex                    futex_exit_mutex;
1221         unsigned int                    futex_state;
1222 #endif
1223 #ifdef CONFIG_PERF_EVENTS
1224         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1225         struct mutex                    perf_event_mutex;
1226         struct list_head                perf_event_list;
1227 #endif
1228 #ifdef CONFIG_DEBUG_PREEMPT
1229         unsigned long                   preempt_disable_ip;
1230 #endif
1231 #ifdef CONFIG_NUMA
1232         /* Protected by alloc_lock: */
1233         struct mempolicy                *mempolicy;
1234         short                           il_prev;
1235         short                           pref_node_fork;
1236 #endif
1237 #ifdef CONFIG_NUMA_BALANCING
1238         int                             numa_scan_seq;
1239         unsigned int                    numa_scan_period;
1240         unsigned int                    numa_scan_period_max;
1241         int                             numa_preferred_nid;
1242         unsigned long                   numa_migrate_retry;
1243         /* Migration stamp: */
1244         u64                             node_stamp;
1245         u64                             last_task_numa_placement;
1246         u64                             last_sum_exec_runtime;
1247         struct callback_head            numa_work;
1248
1249         /*
1250          * This pointer is only modified for current in syscall and
1251          * pagefault context (and for tasks being destroyed), so it can be read
1252          * from any of the following contexts:
1253          *  - RCU read-side critical section
1254          *  - current->numa_group from everywhere
1255          *  - task's runqueue locked, task not running
1256          */
1257         struct numa_group __rcu         *numa_group;
1258
1259         /*
1260          * numa_faults is an array split into four regions:
1261          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1262          * in this precise order.
1263          *
1264          * faults_memory: Exponential decaying average of faults on a per-node
1265          * basis. Scheduling placement decisions are made based on these
1266          * counts. The values remain static for the duration of a PTE scan.
1267          * faults_cpu: Track the nodes the process was running on when a NUMA
1268          * hinting fault was incurred.
1269          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1270          * during the current scan window. When the scan completes, the counts
1271          * in faults_memory and faults_cpu decay and these values are copied.
1272          */
1273         unsigned long                   *numa_faults;
1274         unsigned long                   total_numa_faults;
1275
1276         /*
1277          * numa_faults_locality tracks if faults recorded during the last
1278          * scan window were remote/local or failed to migrate. The task scan
1279          * period is adapted based on the locality of the faults with different
1280          * weights depending on whether they were shared or private faults
1281          */
1282         unsigned long                   numa_faults_locality[3];
1283
1284         unsigned long                   numa_pages_migrated;
1285 #endif /* CONFIG_NUMA_BALANCING */
1286
1287 #ifdef CONFIG_RSEQ
1288         struct rseq __user *rseq;
1289         u32 rseq_sig;
1290         /*
1291          * RmW on rseq_event_mask must be performed atomically
1292          * with respect to preemption.
1293          */
1294         unsigned long rseq_event_mask;
1295 #endif
1296
1297         struct tlbflush_unmap_batch     tlb_ubc;
1298
1299         union {
1300                 refcount_t              rcu_users;
1301                 struct rcu_head         rcu;
1302         };
1303
1304         /* Cache last used pipe for splice(): */
1305         struct pipe_inode_info          *splice_pipe;
1306
1307         struct page_frag                task_frag;
1308
1309 #ifdef CONFIG_TASK_DELAY_ACCT
1310         struct task_delay_info          *delays;
1311 #endif
1312
1313 #ifdef CONFIG_FAULT_INJECTION
1314         int                             make_it_fail;
1315         unsigned int                    fail_nth;
1316 #endif
1317         /*
1318          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1319          * balance_dirty_pages() for a dirty throttling pause:
1320          */
1321         int                             nr_dirtied;
1322         int                             nr_dirtied_pause;
1323         /* Start of a write-and-pause period: */
1324         unsigned long                   dirty_paused_when;
1325
1326 #ifdef CONFIG_LATENCYTOP
1327         int                             latency_record_count;
1328         struct latency_record           latency_record[LT_SAVECOUNT];
1329 #endif
1330         /*
1331          * Time slack values; these are used to round up poll() and
1332          * select() etc timeout values. These are in nanoseconds.
1333          */
1334         u64                             timer_slack_ns;
1335         u64                             default_timer_slack_ns;
1336
1337 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1338         unsigned int                    kasan_depth;
1339 #endif
1340
1341 #ifdef CONFIG_KCSAN
1342         struct kcsan_ctx                kcsan_ctx;
1343 #ifdef CONFIG_TRACE_IRQFLAGS
1344         struct irqtrace_events          kcsan_save_irqtrace;
1345 #endif
1346 #endif
1347
1348 #if IS_ENABLED(CONFIG_KUNIT)
1349         struct kunit                    *kunit_test;
1350 #endif
1351
1352 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1353         /* Index of current stored address in ret_stack: */
1354         int                             curr_ret_stack;
1355         int                             curr_ret_depth;
1356
1357         /* Stack of return addresses for return function tracing: */
1358         struct ftrace_ret_stack         *ret_stack;
1359
1360         /* Timestamp for last schedule: */
1361         unsigned long long              ftrace_timestamp;
1362
1363         /*
1364          * Number of functions that haven't been traced
1365          * because of depth overrun:
1366          */
1367         atomic_t                        trace_overrun;
1368
1369         /* Pause tracing: */
1370         atomic_t                        tracing_graph_pause;
1371 #endif
1372
1373 #ifdef CONFIG_TRACING
1374         /* State flags for use by tracers: */
1375         unsigned long                   trace;
1376
1377         /* Bitmask and counter of trace recursion: */
1378         unsigned long                   trace_recursion;
1379 #endif /* CONFIG_TRACING */
1380
1381 #ifdef CONFIG_KCOV
1382         /* See kernel/kcov.c for more details. */
1383
1384         /* Coverage collection mode enabled for this task (0 if disabled): */
1385         unsigned int                    kcov_mode;
1386
1387         /* Size of the kcov_area: */
1388         unsigned int                    kcov_size;
1389
1390         /* Buffer for coverage collection: */
1391         void                            *kcov_area;
1392
1393         /* KCOV descriptor wired with this task or NULL: */
1394         struct kcov                     *kcov;
1395
1396         /* KCOV common handle for remote coverage collection: */
1397         u64                             kcov_handle;
1398
1399         /* KCOV sequence number: */
1400         int                             kcov_sequence;
1401
1402         /* Collect coverage from softirq context: */
1403         unsigned int                    kcov_softirq;
1404 #endif
1405
1406 #ifdef CONFIG_MEMCG
1407         struct mem_cgroup               *memcg_in_oom;
1408         gfp_t                           memcg_oom_gfp_mask;
1409         int                             memcg_oom_order;
1410
1411         /* Number of pages to reclaim on returning to userland: */
1412         unsigned int                    memcg_nr_pages_over_high;
1413
1414         /* Used by memcontrol for targeted memcg charge: */
1415         struct mem_cgroup               *active_memcg;
1416 #endif
1417
1418 #ifdef CONFIG_BLK_CGROUP
1419         struct request_queue            *throttle_queue;
1420 #endif
1421
1422 #ifdef CONFIG_UPROBES
1423         struct uprobe_task              *utask;
1424 #endif
1425 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1426         unsigned int                    sequential_io;
1427         unsigned int                    sequential_io_avg;
1428 #endif
1429         struct kmap_ctrl                kmap_ctrl;
1430 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1431         unsigned long                   task_state_change;
1432 # ifdef CONFIG_PREEMPT_RT
1433         unsigned long                   saved_state_change;
1434 # endif
1435 #endif
1436         int                             pagefault_disabled;
1437 #ifdef CONFIG_MMU
1438         struct task_struct              *oom_reaper_list;
1439 #endif
1440 #ifdef CONFIG_VMAP_STACK
1441         struct vm_struct                *stack_vm_area;
1442 #endif
1443 #ifdef CONFIG_THREAD_INFO_IN_TASK
1444         /* A live task holds one reference: */
1445         refcount_t                      stack_refcount;
1446 #endif
1447 #ifdef CONFIG_LIVEPATCH
1448         int patch_state;
1449 #endif
1450 #ifdef CONFIG_SECURITY
1451         /* Used by LSM modules for access restriction: */
1452         void                            *security;
1453 #endif
1454 #ifdef CONFIG_BPF_SYSCALL
1455         /* Used by BPF task local storage */
1456         struct bpf_local_storage __rcu  *bpf_storage;
1457         /* Used for BPF run context */
1458         struct bpf_run_ctx              *bpf_ctx;
1459 #endif
1460
1461 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1462         unsigned long                   lowest_stack;
1463         unsigned long                   prev_lowest_stack;
1464 #endif
1465
1466 #ifdef CONFIG_X86_MCE
1467         void __user                     *mce_vaddr;
1468         __u64                           mce_kflags;
1469         u64                             mce_addr;
1470         __u64                           mce_ripv : 1,
1471                                         mce_whole_page : 1,
1472                                         __mce_reserved : 62;
1473         struct callback_head            mce_kill_me;
1474 #endif
1475
1476 #ifdef CONFIG_KRETPROBES
1477         struct llist_head               kretprobe_instances;
1478 #endif
1479
1480 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1481         /*
1482          * If L1D flush is supported on mm context switch
1483          * then we use this callback head to queue kill work
1484          * to kill tasks that are not running on SMT disabled
1485          * cores
1486          */
1487         struct callback_head            l1d_flush_kill;
1488 #endif
1489
1490         /*
1491          * New fields for task_struct should be added above here, so that
1492          * they are included in the randomized portion of task_struct.
1493          */
1494         randomized_struct_fields_end
1495
1496         /* CPU-specific state of this task: */
1497         struct thread_struct            thread;
1498
1499         /*
1500          * WARNING: on x86, 'thread_struct' contains a variable-sized
1501          * structure.  It *MUST* be at the end of 'task_struct'.
1502          *
1503          * Do not put anything below here!
1504          */
1505 };
1506
1507 static inline struct pid *task_pid(struct task_struct *task)
1508 {
1509         return task->thread_pid;
1510 }
1511
1512 /*
1513  * the helpers to get the task's different pids as they are seen
1514  * from various namespaces
1515  *
1516  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1517  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1518  *                     current.
1519  * task_xid_nr_ns()  : id seen from the ns specified;
1520  *
1521  * see also pid_nr() etc in include/linux/pid.h
1522  */
1523 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1524
1525 static inline pid_t task_pid_nr(struct task_struct *tsk)
1526 {
1527         return tsk->pid;
1528 }
1529
1530 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1531 {
1532         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1533 }
1534
1535 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1536 {
1537         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1538 }
1539
1540
1541 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1542 {
1543         return tsk->tgid;
1544 }
1545
1546 /**
1547  * pid_alive - check that a task structure is not stale
1548  * @p: Task structure to be checked.
1549  *
1550  * Test if a process is not yet dead (at most zombie state)
1551  * If pid_alive fails, then pointers within the task structure
1552  * can be stale and must not be dereferenced.
1553  *
1554  * Return: 1 if the process is alive. 0 otherwise.
1555  */
1556 static inline int pid_alive(const struct task_struct *p)
1557 {
1558         return p->thread_pid != NULL;
1559 }
1560
1561 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1562 {
1563         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1564 }
1565
1566 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1567 {
1568         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1569 }
1570
1571
1572 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1573 {
1574         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1575 }
1576
1577 static inline pid_t task_session_vnr(struct task_struct *tsk)
1578 {
1579         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1580 }
1581
1582 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1583 {
1584         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1585 }
1586
1587 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1588 {
1589         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1590 }
1591
1592 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1593 {
1594         pid_t pid = 0;
1595
1596         rcu_read_lock();
1597         if (pid_alive(tsk))
1598                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1599         rcu_read_unlock();
1600
1601         return pid;
1602 }
1603
1604 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1605 {
1606         return task_ppid_nr_ns(tsk, &init_pid_ns);
1607 }
1608
1609 /* Obsolete, do not use: */
1610 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1611 {
1612         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1613 }
1614
1615 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1616 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1617
1618 static inline unsigned int task_state_index(struct task_struct *tsk)
1619 {
1620         unsigned int tsk_state = READ_ONCE(tsk->__state);
1621         unsigned int state = (tsk_state | tsk->exit_state) & TASK_REPORT;
1622
1623         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1624
1625         if (tsk_state == TASK_IDLE)
1626                 state = TASK_REPORT_IDLE;
1627
1628         return fls(state);
1629 }
1630
1631 static inline char task_index_to_char(unsigned int state)
1632 {
1633         static const char state_char[] = "RSDTtXZPI";
1634
1635         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1636
1637         return state_char[state];
1638 }
1639
1640 static inline char task_state_to_char(struct task_struct *tsk)
1641 {
1642         return task_index_to_char(task_state_index(tsk));
1643 }
1644
1645 /**
1646  * is_global_init - check if a task structure is init. Since init
1647  * is free to have sub-threads we need to check tgid.
1648  * @tsk: Task structure to be checked.
1649  *
1650  * Check if a task structure is the first user space task the kernel created.
1651  *
1652  * Return: 1 if the task structure is init. 0 otherwise.
1653  */
1654 static inline int is_global_init(struct task_struct *tsk)
1655 {
1656         return task_tgid_nr(tsk) == 1;
1657 }
1658
1659 extern struct pid *cad_pid;
1660
1661 /*
1662  * Per process flags
1663  */
1664 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1665 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1666 #define PF_EXITING              0x00000004      /* Getting shut down */
1667 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1668 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1669 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1670 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1671 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1672 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1673 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1674 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1675 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1676 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1677 #define PF_USED_ASYNC           0x00004000      /* Used async_schedule*(), used by module init */
1678 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1679 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1680 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1681 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1682 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1683 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1684                                                  * I am cleaning dirty pages from some other bdi. */
1685 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1686 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1687 #define PF_SWAPWRITE            0x00800000      /* Allowed to write to swap */
1688 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1689 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1690 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1691 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1692 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1693
1694 /*
1695  * Only the _current_ task can read/write to tsk->flags, but other
1696  * tasks can access tsk->flags in readonly mode for example
1697  * with tsk_used_math (like during threaded core dumping).
1698  * There is however an exception to this rule during ptrace
1699  * or during fork: the ptracer task is allowed to write to the
1700  * child->flags of its traced child (same goes for fork, the parent
1701  * can write to the child->flags), because we're guaranteed the
1702  * child is not running and in turn not changing child->flags
1703  * at the same time the parent does it.
1704  */
1705 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1706 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1707 #define clear_used_math()                       clear_stopped_child_used_math(current)
1708 #define set_used_math()                         set_stopped_child_used_math(current)
1709
1710 #define conditional_stopped_child_used_math(condition, child) \
1711         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1712
1713 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1714
1715 #define copy_to_stopped_child_used_math(child) \
1716         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1717
1718 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1719 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1720 #define used_math()                             tsk_used_math(current)
1721
1722 static inline bool is_percpu_thread(void)
1723 {
1724 #ifdef CONFIG_SMP
1725         return (current->flags & PF_NO_SETAFFINITY) &&
1726                 (current->nr_cpus_allowed  == 1);
1727 #else
1728         return true;
1729 #endif
1730 }
1731
1732 /* Per-process atomic flags. */
1733 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1734 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1735 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1736 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1737 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1738 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1739 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1740 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1741
1742 #define TASK_PFA_TEST(name, func)                                       \
1743         static inline bool task_##func(struct task_struct *p)           \
1744         { return test_bit(PFA_##name, &p->atomic_flags); }
1745
1746 #define TASK_PFA_SET(name, func)                                        \
1747         static inline void task_set_##func(struct task_struct *p)       \
1748         { set_bit(PFA_##name, &p->atomic_flags); }
1749
1750 #define TASK_PFA_CLEAR(name, func)                                      \
1751         static inline void task_clear_##func(struct task_struct *p)     \
1752         { clear_bit(PFA_##name, &p->atomic_flags); }
1753
1754 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1755 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1756
1757 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1758 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1759 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1760
1761 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1762 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1763 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1764
1765 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1766 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1767 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1768
1769 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1770 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1771 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1772
1773 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1774 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1775
1776 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1777 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1778 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1779
1780 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1781 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1782
1783 static inline void
1784 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1785 {
1786         current->flags &= ~flags;
1787         current->flags |= orig_flags & flags;
1788 }
1789
1790 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1791 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1792 #ifdef CONFIG_SMP
1793 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1794 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1795 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1796 extern void release_user_cpus_ptr(struct task_struct *p);
1797 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1798 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1799 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1800 #else
1801 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1802 {
1803 }
1804 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1805 {
1806         if (!cpumask_test_cpu(0, new_mask))
1807                 return -EINVAL;
1808         return 0;
1809 }
1810 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1811 {
1812         if (src->user_cpus_ptr)
1813                 return -EINVAL;
1814         return 0;
1815 }
1816 static inline void release_user_cpus_ptr(struct task_struct *p)
1817 {
1818         WARN_ON(p->user_cpus_ptr);
1819 }
1820
1821 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1822 {
1823         return 0;
1824 }
1825 #endif
1826
1827 extern int yield_to(struct task_struct *p, bool preempt);
1828 extern void set_user_nice(struct task_struct *p, long nice);
1829 extern int task_prio(const struct task_struct *p);
1830
1831 /**
1832  * task_nice - return the nice value of a given task.
1833  * @p: the task in question.
1834  *
1835  * Return: The nice value [ -20 ... 0 ... 19 ].
1836  */
1837 static inline int task_nice(const struct task_struct *p)
1838 {
1839         return PRIO_TO_NICE((p)->static_prio);
1840 }
1841
1842 extern int can_nice(const struct task_struct *p, const int nice);
1843 extern int task_curr(const struct task_struct *p);
1844 extern int idle_cpu(int cpu);
1845 extern int available_idle_cpu(int cpu);
1846 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1847 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1848 extern void sched_set_fifo(struct task_struct *p);
1849 extern void sched_set_fifo_low(struct task_struct *p);
1850 extern void sched_set_normal(struct task_struct *p, int nice);
1851 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1852 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1853 extern struct task_struct *idle_task(int cpu);
1854
1855 /**
1856  * is_idle_task - is the specified task an idle task?
1857  * @p: the task in question.
1858  *
1859  * Return: 1 if @p is an idle task. 0 otherwise.
1860  */
1861 static __always_inline bool is_idle_task(const struct task_struct *p)
1862 {
1863         return !!(p->flags & PF_IDLE);
1864 }
1865
1866 extern struct task_struct *curr_task(int cpu);
1867 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1868
1869 void yield(void);
1870
1871 union thread_union {
1872 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1873         struct task_struct task;
1874 #endif
1875 #ifndef CONFIG_THREAD_INFO_IN_TASK
1876         struct thread_info thread_info;
1877 #endif
1878         unsigned long stack[THREAD_SIZE/sizeof(long)];
1879 };
1880
1881 #ifndef CONFIG_THREAD_INFO_IN_TASK
1882 extern struct thread_info init_thread_info;
1883 #endif
1884
1885 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1886
1887 #ifdef CONFIG_THREAD_INFO_IN_TASK
1888 static inline struct thread_info *task_thread_info(struct task_struct *task)
1889 {
1890         return &task->thread_info;
1891 }
1892 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1893 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1894 #endif
1895
1896 /*
1897  * find a task by one of its numerical ids
1898  *
1899  * find_task_by_pid_ns():
1900  *      finds a task by its pid in the specified namespace
1901  * find_task_by_vpid():
1902  *      finds a task by its virtual pid
1903  *
1904  * see also find_vpid() etc in include/linux/pid.h
1905  */
1906
1907 extern struct task_struct *find_task_by_vpid(pid_t nr);
1908 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1909
1910 /*
1911  * find a task by its virtual pid and get the task struct
1912  */
1913 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1914
1915 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1916 extern int wake_up_process(struct task_struct *tsk);
1917 extern void wake_up_new_task(struct task_struct *tsk);
1918
1919 #ifdef CONFIG_SMP
1920 extern void kick_process(struct task_struct *tsk);
1921 #else
1922 static inline void kick_process(struct task_struct *tsk) { }
1923 #endif
1924
1925 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1926
1927 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1928 {
1929         __set_task_comm(tsk, from, false);
1930 }
1931
1932 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1933 #define get_task_comm(buf, tsk) ({                      \
1934         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1935         __get_task_comm(buf, sizeof(buf), tsk);         \
1936 })
1937
1938 #ifdef CONFIG_SMP
1939 static __always_inline void scheduler_ipi(void)
1940 {
1941         /*
1942          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1943          * TIF_NEED_RESCHED remotely (for the first time) will also send
1944          * this IPI.
1945          */
1946         preempt_fold_need_resched();
1947 }
1948 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1949 #else
1950 static inline void scheduler_ipi(void) { }
1951 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1952 {
1953         return 1;
1954 }
1955 #endif
1956
1957 /*
1958  * Set thread flags in other task's structures.
1959  * See asm/thread_info.h for TIF_xxxx flags available:
1960  */
1961 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1962 {
1963         set_ti_thread_flag(task_thread_info(tsk), flag);
1964 }
1965
1966 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1967 {
1968         clear_ti_thread_flag(task_thread_info(tsk), flag);
1969 }
1970
1971 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1972                                           bool value)
1973 {
1974         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1975 }
1976
1977 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1978 {
1979         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1980 }
1981
1982 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1983 {
1984         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1985 }
1986
1987 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1988 {
1989         return test_ti_thread_flag(task_thread_info(tsk), flag);
1990 }
1991
1992 static inline void set_tsk_need_resched(struct task_struct *tsk)
1993 {
1994         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1995 }
1996
1997 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1998 {
1999         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2000 }
2001
2002 static inline int test_tsk_need_resched(struct task_struct *tsk)
2003 {
2004         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2005 }
2006
2007 /*
2008  * cond_resched() and cond_resched_lock(): latency reduction via
2009  * explicit rescheduling in places that are safe. The return
2010  * value indicates whether a reschedule was done in fact.
2011  * cond_resched_lock() will drop the spinlock before scheduling,
2012  */
2013 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2014 extern int __cond_resched(void);
2015
2016 #ifdef CONFIG_PREEMPT_DYNAMIC
2017
2018 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2019
2020 static __always_inline int _cond_resched(void)
2021 {
2022         return static_call_mod(cond_resched)();
2023 }
2024
2025 #else
2026
2027 static inline int _cond_resched(void)
2028 {
2029         return __cond_resched();
2030 }
2031
2032 #endif /* CONFIG_PREEMPT_DYNAMIC */
2033
2034 #else
2035
2036 static inline int _cond_resched(void) { return 0; }
2037
2038 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2039
2040 #define cond_resched() ({                       \
2041         ___might_sleep(__FILE__, __LINE__, 0);  \
2042         _cond_resched();                        \
2043 })
2044
2045 extern int __cond_resched_lock(spinlock_t *lock);
2046 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2047 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2048
2049 #define cond_resched_lock(lock) ({                              \
2050         ___might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET);\
2051         __cond_resched_lock(lock);                              \
2052 })
2053
2054 #define cond_resched_rwlock_read(lock) ({                       \
2055         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2056         __cond_resched_rwlock_read(lock);                       \
2057 })
2058
2059 #define cond_resched_rwlock_write(lock) ({                      \
2060         __might_sleep(__FILE__, __LINE__, PREEMPT_LOCK_OFFSET); \
2061         __cond_resched_rwlock_write(lock);                      \
2062 })
2063
2064 static inline void cond_resched_rcu(void)
2065 {
2066 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2067         rcu_read_unlock();
2068         cond_resched();
2069         rcu_read_lock();
2070 #endif
2071 }
2072
2073 /*
2074  * Does a critical section need to be broken due to another
2075  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2076  * but a general need for low latency)
2077  */
2078 static inline int spin_needbreak(spinlock_t *lock)
2079 {
2080 #ifdef CONFIG_PREEMPTION
2081         return spin_is_contended(lock);
2082 #else
2083         return 0;
2084 #endif
2085 }
2086
2087 /*
2088  * Check if a rwlock is contended.
2089  * Returns non-zero if there is another task waiting on the rwlock.
2090  * Returns zero if the lock is not contended or the system / underlying
2091  * rwlock implementation does not support contention detection.
2092  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2093  * for low latency.
2094  */
2095 static inline int rwlock_needbreak(rwlock_t *lock)
2096 {
2097 #ifdef CONFIG_PREEMPTION
2098         return rwlock_is_contended(lock);
2099 #else
2100         return 0;
2101 #endif
2102 }
2103
2104 static __always_inline bool need_resched(void)
2105 {
2106         return unlikely(tif_need_resched());
2107 }
2108
2109 /*
2110  * Wrappers for p->thread_info->cpu access. No-op on UP.
2111  */
2112 #ifdef CONFIG_SMP
2113
2114 static inline unsigned int task_cpu(const struct task_struct *p)
2115 {
2116 #ifdef CONFIG_THREAD_INFO_IN_TASK
2117         return READ_ONCE(p->cpu);
2118 #else
2119         return READ_ONCE(task_thread_info(p)->cpu);
2120 #endif
2121 }
2122
2123 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2124
2125 #else
2126
2127 static inline unsigned int task_cpu(const struct task_struct *p)
2128 {
2129         return 0;
2130 }
2131
2132 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2133 {
2134 }
2135
2136 #endif /* CONFIG_SMP */
2137
2138 extern bool sched_task_on_rq(struct task_struct *p);
2139
2140 /*
2141  * In order to reduce various lock holder preemption latencies provide an
2142  * interface to see if a vCPU is currently running or not.
2143  *
2144  * This allows us to terminate optimistic spin loops and block, analogous to
2145  * the native optimistic spin heuristic of testing if the lock owner task is
2146  * running or not.
2147  */
2148 #ifndef vcpu_is_preempted
2149 static inline bool vcpu_is_preempted(int cpu)
2150 {
2151         return false;
2152 }
2153 #endif
2154
2155 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2156 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2157
2158 #ifndef TASK_SIZE_OF
2159 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2160 #endif
2161
2162 #ifdef CONFIG_SMP
2163 /* Returns effective CPU energy utilization, as seen by the scheduler */
2164 unsigned long sched_cpu_util(int cpu, unsigned long max);
2165 #endif /* CONFIG_SMP */
2166
2167 #ifdef CONFIG_RSEQ
2168
2169 /*
2170  * Map the event mask on the user-space ABI enum rseq_cs_flags
2171  * for direct mask checks.
2172  */
2173 enum rseq_event_mask_bits {
2174         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2175         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2176         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2177 };
2178
2179 enum rseq_event_mask {
2180         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2181         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2182         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2183 };
2184
2185 static inline void rseq_set_notify_resume(struct task_struct *t)
2186 {
2187         if (t->rseq)
2188                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2189 }
2190
2191 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2192
2193 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2194                                              struct pt_regs *regs)
2195 {
2196         if (current->rseq)
2197                 __rseq_handle_notify_resume(ksig, regs);
2198 }
2199
2200 static inline void rseq_signal_deliver(struct ksignal *ksig,
2201                                        struct pt_regs *regs)
2202 {
2203         preempt_disable();
2204         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2205         preempt_enable();
2206         rseq_handle_notify_resume(ksig, regs);
2207 }
2208
2209 /* rseq_preempt() requires preemption to be disabled. */
2210 static inline void rseq_preempt(struct task_struct *t)
2211 {
2212         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2213         rseq_set_notify_resume(t);
2214 }
2215
2216 /* rseq_migrate() requires preemption to be disabled. */
2217 static inline void rseq_migrate(struct task_struct *t)
2218 {
2219         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2220         rseq_set_notify_resume(t);
2221 }
2222
2223 /*
2224  * If parent process has a registered restartable sequences area, the
2225  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2226  */
2227 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2228 {
2229         if (clone_flags & CLONE_VM) {
2230                 t->rseq = NULL;
2231                 t->rseq_sig = 0;
2232                 t->rseq_event_mask = 0;
2233         } else {
2234                 t->rseq = current->rseq;
2235                 t->rseq_sig = current->rseq_sig;
2236                 t->rseq_event_mask = current->rseq_event_mask;
2237         }
2238 }
2239
2240 static inline void rseq_execve(struct task_struct *t)
2241 {
2242         t->rseq = NULL;
2243         t->rseq_sig = 0;
2244         t->rseq_event_mask = 0;
2245 }
2246
2247 #else
2248
2249 static inline void rseq_set_notify_resume(struct task_struct *t)
2250 {
2251 }
2252 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2253                                              struct pt_regs *regs)
2254 {
2255 }
2256 static inline void rseq_signal_deliver(struct ksignal *ksig,
2257                                        struct pt_regs *regs)
2258 {
2259 }
2260 static inline void rseq_preempt(struct task_struct *t)
2261 {
2262 }
2263 static inline void rseq_migrate(struct task_struct *t)
2264 {
2265 }
2266 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2267 {
2268 }
2269 static inline void rseq_execve(struct task_struct *t)
2270 {
2271 }
2272
2273 #endif
2274
2275 #ifdef CONFIG_DEBUG_RSEQ
2276
2277 void rseq_syscall(struct pt_regs *regs);
2278
2279 #else
2280
2281 static inline void rseq_syscall(struct pt_regs *regs)
2282 {
2283 }
2284
2285 #endif
2286
2287 const struct sched_avg *sched_trace_cfs_rq_avg(struct cfs_rq *cfs_rq);
2288 char *sched_trace_cfs_rq_path(struct cfs_rq *cfs_rq, char *str, int len);
2289 int sched_trace_cfs_rq_cpu(struct cfs_rq *cfs_rq);
2290
2291 const struct sched_avg *sched_trace_rq_avg_rt(struct rq *rq);
2292 const struct sched_avg *sched_trace_rq_avg_dl(struct rq *rq);
2293 const struct sched_avg *sched_trace_rq_avg_irq(struct rq *rq);
2294
2295 int sched_trace_rq_cpu(struct rq *rq);
2296 int sched_trace_rq_cpu_capacity(struct rq *rq);
2297 int sched_trace_rq_nr_running(struct rq *rq);
2298
2299 const struct cpumask *sched_trace_rd_span(struct root_domain *rd);
2300
2301 #ifdef CONFIG_SCHED_CORE
2302 extern void sched_core_free(struct task_struct *tsk);
2303 extern void sched_core_fork(struct task_struct *p);
2304 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2305                                 unsigned long uaddr);
2306 #else
2307 static inline void sched_core_free(struct task_struct *tsk) { }
2308 static inline void sched_core_fork(struct task_struct *p) { }
2309 #endif
2310
2311 #endif