include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13 #include <asm/processor.h>
  14 #include <linux/thread_info.h>
  15 #include <linux/preempt.h>
  16 #include <linux/cpumask_types.h>
  17
  18 #include <linux/cache.h>
  19 #include <linux/irqflags_types.h>
  20 #include <linux/smp_types.h>
  21 #include <linux/pid_types.h>
  22 #include <linux/sem_types.h>
  23 #include <linux/shm.h>
  24 #include <linux/kmsan_types.h>
  25 #include <linux/mutex_types.h>
  26 #include <linux/plist_types.h>
  27 #include <linux/hrtimer_types.h>
  28 #include <linux/timer_types.h>
  29 #include <linux/seccomp_types.h>
  30 #include <linux/nodemask_types.h>
  31 #include <linux/refcount_types.h>
  32 #include <linux/resource.h>
  33 #include <linux/latencytop.h>
  34 #include <linux/sched/prio.h>
  35 #include <linux/sched/types.h>
  36 #include <linux/signal_types.h>
  37 #include <linux/syscall_user_dispatch_types.h>
  38 #include <linux/mm_types_task.h>
  39 #include <linux/netdevice_xmit.h>
  40 #include <linux/task_io_accounting.h>
  41 #include <linux/posix-timers_types.h>
  42 #include <linux/restart_block.h>
  43 #include <uapi/linux/rseq.h>
  44 #include <linux/seqlock_types.h>
  45 #include <linux/kcsan.h>
  46 #include <linux/rv.h>
  47 #include <linux/livepatch_sched.h>
  48 #include <linux/uidgid_types.h>
  49 #include <asm/kmap_size.h>
  50
  51 /* task_struct member predeclarations (sorted alphabetically): */
  52 struct audit_context;
  53 struct bio_list;
  54 struct blk_plug;
  55 struct bpf_local_storage;
  56 struct bpf_run_ctx;
  57 struct bpf_net_context;
  58 struct capture_control;
  59 struct cfs_rq;
  60 struct fs_struct;
  61 struct futex_pi_state;
  62 struct io_context;
  63 struct io_uring_task;
  64 struct mempolicy;
  65 struct nameidata;
  66 struct nsproxy;
  67 struct perf_event_context;
  68 struct pid_namespace;
  69 struct pipe_inode_info;
  70 struct rcu_node;
  71 struct reclaim_state;
  72 struct robust_list_head;
  73 struct root_domain;
  74 struct rq;
  75 struct sched_attr;
  76 struct sched_dl_entity;
  77 struct seq_file;
  78 struct sighand_struct;
  79 struct signal_struct;
  80 struct task_delay_info;
  81 struct task_group;
  82 struct task_struct;
  83 struct user_event_mm;
  84
  85 /*
  86  * Task state bitmask. NOTE! These bits are also
  87  * encoded in fs/proc/array.c: get_task_state().
  88  *
  89  * We have two separate sets of flags: task->__state
  90  * is about runnability, while task->exit_state are
  91  * about the task exiting. Confusing, but this way
  92  * modifying one set can't modify the other one by
  93  * mistake.
  94  */
  95
  96 /* Used in tsk->__state: */
  97 #define TASK_RUNNING                    0x00000000
  98 #define TASK_INTERRUPTIBLE              0x00000001
  99 #define TASK_UNINTERRUPTIBLE            0x00000002
 100 #define __TASK_STOPPED                  0x00000004
 101 #define __TASK_TRACED                   0x00000008
 102 /* Used in tsk->exit_state: */
 103 #define EXIT_DEAD                       0x00000010
 104 #define EXIT_ZOMBIE                     0x00000020
 105 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
 106 /* Used in tsk->__state again: */
 107 #define TASK_PARKED                     0x00000040
 108 #define TASK_DEAD                       0x00000080
 109 #define TASK_WAKEKILL                   0x00000100
 110 #define TASK_WAKING                     0x00000200
 111 #define TASK_NOLOAD                     0x00000400
 112 #define TASK_NEW                        0x00000800
 113 #define TASK_RTLOCK_WAIT                0x00001000
 114 #define TASK_FREEZABLE                  0x00002000
 115 #define __TASK_FREEZABLE_UNSAFE        (0x00004000 * IS_ENABLED(CONFIG_LOCKDEP))
 116 #define TASK_FROZEN                     0x00008000
 117 #define TASK_STATE_MAX                  0x00010000
 118
 119 #define TASK_ANY                        (TASK_STATE_MAX-1)
 120
 121 /*
 122  * DO NOT ADD ANY NEW USERS !
 123  */
 124 #define TASK_FREEZABLE_UNSAFE           (TASK_FREEZABLE | __TASK_FREEZABLE_UNSAFE)
 125
 126 /* Convenience macros for the sake of set_current_state: */
 127 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 128 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 129 #define TASK_TRACED                     __TASK_TRACED
 130
 131 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 132
 133 /* Convenience macros for the sake of wake_up(): */
 134 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 135
 136 /* get_task_state(): */
 137 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 138                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 139                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 140                                          TASK_PARKED)
 141
 142 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 143
 144 #define task_is_traced(task)            ((READ_ONCE(task->jobctl) & JOBCTL_TRACED) != 0)
 145 #define task_is_stopped(task)           ((READ_ONCE(task->jobctl) & JOBCTL_STOPPED) != 0)
 146 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->jobctl) & (JOBCTL_STOPPED | JOBCTL_TRACED)) != 0)
 147
 148 /*
 149  * Special states are those that do not use the normal wait-loop pattern. See
 150  * the comment with set_special_state().
 151  */
 152 #define is_special_task_state(state)                                    \
 153         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED |      \
 154                     TASK_DEAD | TASK_FROZEN))
 155
 156 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 157 # define debug_normal_state_change(state_value)                         \
 158         do {                                                            \
 159                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 160                 current->task_state_change = _THIS_IP_;                 \
 161         } while (0)
 162
 163 # define debug_special_state_change(state_value)                        \
 164         do {                                                            \
 165                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 166                 current->task_state_change = _THIS_IP_;                 \
 167         } while (0)
 168
 169 # define debug_rtlock_wait_set_state()                                  \
 170         do {                                                             \
 171                 current->saved_state_change = current->task_state_change;\
 172                 current->task_state_change = _THIS_IP_;                  \
 173         } while (0)
 174
 175 # define debug_rtlock_wait_restore_state()                              \
 176         do {                                                             \
 177                 current->task_state_change = current->saved_state_change;\
 178         } while (0)
 179
 180 #else
 181 # define debug_normal_state_change(cond)        do { } while (0)
 182 # define debug_special_state_change(cond)       do { } while (0)
 183 # define debug_rtlock_wait_set_state()          do { } while (0)
 184 # define debug_rtlock_wait_restore_state()      do { } while (0)
 185 #endif
 186
 187 /*
 188  * set_current_state() includes a barrier so that the write of current->__state
 189  * is correctly serialised wrt the caller's subsequent test of whether to
 190  * actually sleep:
 191  *
 192  *   for (;;) {
 193  *      set_current_state(TASK_UNINTERRUPTIBLE);
 194  *      if (CONDITION)
 195  *         break;
 196  *
 197  *      schedule();
 198  *   }
 199  *   __set_current_state(TASK_RUNNING);
 200  *
 201  * If the caller does not need such serialisation (because, for instance, the
 202  * CONDITION test and condition change and wakeup are under the same lock) then
 203  * use __set_current_state().
 204  *
 205  * The above is typically ordered against the wakeup, which does:
 206  *
 207  *   CONDITION = 1;
 208  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 209  *
 210  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 211  * accessing p->__state.
 212  *
 213  * Wakeup will do: if (@state & p->__state) p->__state = TASK_RUNNING, that is,
 214  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 215  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 216  *
 217  * However, with slightly different timing the wakeup TASK_RUNNING store can
 218  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 219  * a problem either because that will result in one extra go around the loop
 220  * and our @cond test will save the day.
 221  *
 222  * Also see the comments of try_to_wake_up().
 223  */
 224 #define __set_current_state(state_value)                                \
 225         do {                                                            \
 226                 debug_normal_state_change((state_value));               \
 227                 WRITE_ONCE(current->__state, (state_value));            \
 228         } while (0)
 229
 230 #define set_current_state(state_value)                                  \
 231         do {                                                            \
 232                 debug_normal_state_change((state_value));               \
 233                 smp_store_mb(current->__state, (state_value));          \
 234         } while (0)
 235
 236 /*
 237  * set_special_state() should be used for those states when the blocking task
 238  * can not use the regular condition based wait-loop. In that case we must
 239  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 240  * stores will not collide with our state change.
 241  */
 242 #define set_special_state(state_value)                                  \
 243         do {                                                            \
 244                 unsigned long flags; /* may shadow */                   \
 245                                                                         \
 246                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 247                 debug_special_state_change((state_value));              \
 248                 WRITE_ONCE(current->__state, (state_value));            \
 249                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 250         } while (0)
 251
 252 /*
 253  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 254  *
 255  * RT's spin/rwlock substitutions are state preserving. The state of the
 256  * task when blocking on the lock is saved in task_struct::saved_state and
 257  * restored after the lock has been acquired.  These operations are
 258  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 259  * lock related wakeups while the task is blocked on the lock are
 260  * redirected to operate on task_struct::saved_state to ensure that these
 261  * are not dropped. On restore task_struct::saved_state is set to
 262  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 263  *
 264  * The lock operation looks like this:
 265  *
 266  *      current_save_and_set_rtlock_wait_state();
 267  *      for (;;) {
 268  *              if (try_lock())
 269  *                      break;
 270  *              raw_spin_unlock_irq(&lock->wait_lock);
 271  *              schedule_rtlock();
 272  *              raw_spin_lock_irq(&lock->wait_lock);
 273  *              set_current_state(TASK_RTLOCK_WAIT);
 274  *      }
 275  *      current_restore_rtlock_saved_state();
 276  */
 277 #define current_save_and_set_rtlock_wait_state()                        \
 278         do {                                                            \
 279                 lockdep_assert_irqs_disabled();                         \
 280                 raw_spin_lock(&current->pi_lock);                       \
 281                 current->saved_state = current->__state;                \
 282                 debug_rtlock_wait_set_state();                          \
 283                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 284                 raw_spin_unlock(&current->pi_lock);                     \
 285         } while (0);
 286
 287 #define current_restore_rtlock_saved_state()                            \
 288         do {                                                            \
 289                 lockdep_assert_irqs_disabled();                         \
 290                 raw_spin_lock(&current->pi_lock);                       \
 291                 debug_rtlock_wait_restore_state();                      \
 292                 WRITE_ONCE(current->__state, current->saved_state);     \
 293                 current->saved_state = TASK_RUNNING;                    \
 294                 raw_spin_unlock(&current->pi_lock);                     \
 295         } while (0);
 296
 297 #define get_current_state()     READ_ONCE(current->__state)
 298
 299 /*
 300  * Define the task command name length as enum, then it can be visible to
 301  * BPF programs.
 302  */
 303 enum {
 304         TASK_COMM_LEN = 16,
 305 };
 306
 307 extern void sched_tick(void);
 308
 309 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 310
 311 extern long schedule_timeout(long timeout);
 312 extern long schedule_timeout_interruptible(long timeout);
 313 extern long schedule_timeout_killable(long timeout);
 314 extern long schedule_timeout_uninterruptible(long timeout);
 315 extern long schedule_timeout_idle(long timeout);
 316 asmlinkage void schedule(void);
 317 extern void schedule_preempt_disabled(void);
 318 asmlinkage void preempt_schedule_irq(void);
 319 #ifdef CONFIG_PREEMPT_RT
 320  extern void schedule_rtlock(void);
 321 #endif
 322
 323 extern int __must_check io_schedule_prepare(void);
 324 extern void io_schedule_finish(int token);
 325 extern long io_schedule_timeout(long timeout);
 326 extern void io_schedule(void);
 327
 328 /**
 329  * struct prev_cputime - snapshot of system and user cputime
 330  * @utime: time spent in user mode
 331  * @stime: time spent in system mode
 332  * @lock: protects the above two fields
 333  *
 334  * Stores previous user/system time values such that we can guarantee
 335  * monotonicity.
 336  */
 337 struct prev_cputime {
 338 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 339         u64                             utime;
 340         u64                             stime;
 341         raw_spinlock_t                  lock;
 342 #endif
 343 };
 344
 345 enum vtime_state {
 346         /* Task is sleeping or running in a CPU with VTIME inactive: */
 347         VTIME_INACTIVE = 0,
 348         /* Task is idle */
 349         VTIME_IDLE,
 350         /* Task runs in kernelspace in a CPU with VTIME active: */
 351         VTIME_SYS,
 352         /* Task runs in userspace in a CPU with VTIME active: */
 353         VTIME_USER,
 354         /* Task runs as guests in a CPU with VTIME active: */
 355         VTIME_GUEST,
 356 };
 357
 358 struct vtime {
 359         seqcount_t              seqcount;
 360         unsigned long long      starttime;
 361         enum vtime_state        state;
 362         unsigned int            cpu;
 363         u64                     utime;
 364         u64                     stime;
 365         u64                     gtime;
 366 };
 367
 368 /*
 369  * Utilization clamp constraints.
 370  * @UCLAMP_MIN: Minimum utilization
 371  * @UCLAMP_MAX: Maximum utilization
 372  * @UCLAMP_CNT: Utilization clamp constraints count
 373  */
 374 enum uclamp_id {
 375         UCLAMP_MIN = 0,
 376         UCLAMP_MAX,
 377         UCLAMP_CNT
 378 };
 379
 380 #ifdef CONFIG_SMP
 381 extern struct root_domain def_root_domain;
 382 extern struct mutex sched_domains_mutex;
 383 #endif
 384
 385 struct sched_param {
 386         int sched_priority;
 387 };
 388
 389 struct sched_info {
 390 #ifdef CONFIG_SCHED_INFO
 391         /* Cumulative counters: */
 392
 393         /* # of times we have run on this CPU: */
 394         unsigned long                   pcount;
 395
 396         /* Time spent waiting on a runqueue: */
 397         unsigned long long              run_delay;
 398
 399         /* Timestamps: */
 400
 401         /* When did we last run on a CPU? */
 402         unsigned long long              last_arrival;
 403
 404         /* When were we last queued to run? */
 405         unsigned long long              last_queued;
 406
 407 #endif /* CONFIG_SCHED_INFO */
 408 };
 409
 410 /*
 411  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 412  * has a few: load, load_avg, util_avg, freq, and capacity.
 413  *
 414  * We define a basic fixed point arithmetic range, and then formalize
 415  * all these metrics based on that basic range.
 416  */
 417 # define SCHED_FIXEDPOINT_SHIFT         10
 418 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 419
 420 /* Increase resolution of cpu_capacity calculations */
 421 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 422 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 423
 424 struct load_weight {
 425         unsigned long                   weight;
 426         u32                             inv_weight;
 427 };
 428
 429 /*
 430  * The load/runnable/util_avg accumulates an infinite geometric series
 431  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 432  *
 433  * [load_avg definition]
 434  *
 435  *   load_avg = runnable% * scale_load_down(load)
 436  *
 437  * [runnable_avg definition]
 438  *
 439  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 440  *
 441  * [util_avg definition]
 442  *
 443  *   util_avg = running% * SCHED_CAPACITY_SCALE
 444  *
 445  * where runnable% is the time ratio that a sched_entity is runnable and
 446  * running% the time ratio that a sched_entity is running.
 447  *
 448  * For cfs_rq, they are the aggregated values of all runnable and blocked
 449  * sched_entities.
 450  *
 451  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 452  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 453  * for computing those signals (see update_rq_clock_pelt())
 454  *
 455  * N.B., the above ratios (runnable% and running%) themselves are in the
 456  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 457  * to as large a range as necessary. This is for example reflected by
 458  * util_avg's SCHED_CAPACITY_SCALE.
 459  *
 460  * [Overflow issue]
 461  *
 462  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 463  * with the highest load (=88761), always runnable on a single cfs_rq,
 464  * and should not overflow as the number already hits PID_MAX_LIMIT.
 465  *
 466  * For all other cases (including 32-bit kernels), struct load_weight's
 467  * weight will overflow first before we do, because:
 468  *
 469  *    Max(load_avg) <= Max(load.weight)
 470  *
 471  * Then it is the load_weight's responsibility to consider overflow
 472  * issues.
 473  */
 474 struct sched_avg {
 475         u64                             last_update_time;
 476         u64                             load_sum;
 477         u64                             runnable_sum;
 478         u32                             util_sum;
 479         u32                             period_contrib;
 480         unsigned long                   load_avg;
 481         unsigned long                   runnable_avg;
 482         unsigned long                   util_avg;
 483         unsigned int                    util_est;
 484 } ____cacheline_aligned;
 485
 486 /*
 487  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 488  * updates. When a task is dequeued, its util_est should not be updated if its
 489  * util_avg has not been updated in the meantime.
 490  * This information is mapped into the MSB bit of util_est at dequeue time.
 491  * Since max value of util_est for a task is 1024 (PELT util_avg for a task)
 492  * it is safe to use MSB.
 493  */
 494 #define UTIL_EST_WEIGHT_SHIFT           2
 495 #define UTIL_AVG_UNCHANGED              0x80000000
 496
 497 struct sched_statistics {
 498 #ifdef CONFIG_SCHEDSTATS
 499         u64                             wait_start;
 500         u64                             wait_max;
 501         u64                             wait_count;
 502         u64                             wait_sum;
 503         u64                             iowait_count;
 504         u64                             iowait_sum;
 505
 506         u64                             sleep_start;
 507         u64                             sleep_max;
 508         s64                             sum_sleep_runtime;
 509
 510         u64                             block_start;
 511         u64                             block_max;
 512         s64                             sum_block_runtime;
 513
 514         s64                             exec_max;
 515         u64                             slice_max;
 516
 517         u64                             nr_migrations_cold;
 518         u64                             nr_failed_migrations_affine;
 519         u64                             nr_failed_migrations_running;
 520         u64                             nr_failed_migrations_hot;
 521         u64                             nr_forced_migrations;
 522
 523         u64                             nr_wakeups;
 524         u64                             nr_wakeups_sync;
 525         u64                             nr_wakeups_migrate;
 526         u64                             nr_wakeups_local;
 527         u64                             nr_wakeups_remote;
 528         u64                             nr_wakeups_affine;
 529         u64                             nr_wakeups_affine_attempts;
 530         u64                             nr_wakeups_passive;
 531         u64                             nr_wakeups_idle;
 532
 533 #ifdef CONFIG_SCHED_CORE
 534         u64                             core_forceidle_sum;
 535 #endif
 536 #endif /* CONFIG_SCHEDSTATS */
 537 } ____cacheline_aligned;
 538
 539 struct sched_entity {
 540         /* For load-balancing: */
 541         struct load_weight              load;
 542         struct rb_node                  run_node;
 543         u64                             deadline;
 544         u64                             min_vruntime;
 545         u64                             min_slice;
 546
 547         struct list_head                group_node;
 548         unsigned char                   on_rq;
 549         unsigned char                   sched_delayed;
 550         unsigned char                   rel_deadline;
 551         unsigned char                   custom_slice;
 552                                         /* hole */
 553
 554         u64                             exec_start;
 555         u64                             sum_exec_runtime;
 556         u64                             prev_sum_exec_runtime;
 557         u64                             vruntime;
 558         s64                             vlag;
 559         u64                             slice;
 560
 561         u64                             nr_migrations;
 562
 563 #ifdef CONFIG_FAIR_GROUP_SCHED
 564         int                             depth;
 565         struct sched_entity             *parent;
 566         /* rq on which this entity is (to be) queued: */
 567         struct cfs_rq                   *cfs_rq;
 568         /* rq "owned" by this entity/group: */
 569         struct cfs_rq                   *my_q;
 570         /* cached value of my_q->h_nr_running */
 571         unsigned long                   runnable_weight;
 572 #endif
 573
 574 #ifdef CONFIG_SMP
 575         /*
 576          * Per entity load average tracking.
 577          *
 578          * Put into separate cache line so it does not
 579          * collide with read-mostly values above.
 580          */
 581         struct sched_avg                avg;
 582 #endif
 583 };
 584
 585 struct sched_rt_entity {
 586         struct list_head                run_list;
 587         unsigned long                   timeout;
 588         unsigned long                   watchdog_stamp;
 589         unsigned int                    time_slice;
 590         unsigned short                  on_rq;
 591         unsigned short                  on_list;
 592
 593         struct sched_rt_entity          *back;
 594 #ifdef CONFIG_RT_GROUP_SCHED
 595         struct sched_rt_entity          *parent;
 596         /* rq on which this entity is (to be) queued: */
 597         struct rt_rq                    *rt_rq;
 598         /* rq "owned" by this entity/group: */
 599         struct rt_rq                    *my_q;
 600 #endif
 601 } __randomize_layout;
 602
 603 typedef bool (*dl_server_has_tasks_f)(struct sched_dl_entity *);
 604 typedef struct task_struct *(*dl_server_pick_f)(struct sched_dl_entity *);
 605
 606 struct sched_dl_entity {
 607         struct rb_node                  rb_node;
 608
 609         /*
 610          * Original scheduling parameters. Copied here from sched_attr
 611          * during sched_setattr(), they will remain the same until
 612          * the next sched_setattr().
 613          */
 614         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 615         u64                             dl_deadline;    /* Relative deadline of each instance   */
 616         u64                             dl_period;      /* Separation of two instances (period) */
 617         u64                             dl_bw;          /* dl_runtime / dl_period               */
 618         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 619
 620         /*
 621          * Actual scheduling parameters. Initialized with the values above,
 622          * they are continuously updated during task execution. Note that
 623          * the remaining runtime could be < 0 in case we are in overrun.
 624          */
 625         s64                             runtime;        /* Remaining runtime for this instance  */
 626         u64                             deadline;       /* Absolute deadline for this instance  */
 627         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 628
 629         /*
 630          * Some bool flags:
 631          *
 632          * @dl_throttled tells if we exhausted the runtime. If so, the
 633          * task has to wait for a replenishment to be performed at the
 634          * next firing of dl_timer.
 635          *
 636          * @dl_yielded tells if task gave up the CPU before consuming
 637          * all its available runtime during the last job.
 638          *
 639          * @dl_non_contending tells if the task is inactive while still
 640          * contributing to the active utilization. In other words, it
 641          * indicates if the inactive timer has been armed and its handler
 642          * has not been executed yet. This flag is useful to avoid race
 643          * conditions between the inactive timer handler and the wakeup
 644          * code.
 645          *
 646          * @dl_overrun tells if the task asked to be informed about runtime
 647          * overruns.
 648          *
 649          * @dl_server tells if this is a server entity.
 650          *
 651          * @dl_defer tells if this is a deferred or regular server. For
 652          * now only defer server exists.
 653          *
 654          * @dl_defer_armed tells if the deferrable server is waiting
 655          * for the replenishment timer to activate it.
 656          *
 657          * @dl_defer_running tells if the deferrable server is actually
 658          * running, skipping the defer phase.
 659          */
 660         unsigned int                    dl_throttled      : 1;
 661         unsigned int                    dl_yielded        : 1;
 662         unsigned int                    dl_non_contending : 1;
 663         unsigned int                    dl_overrun        : 1;
 664         unsigned int                    dl_server         : 1;
 665         unsigned int                    dl_defer          : 1;
 666         unsigned int                    dl_defer_armed    : 1;
 667         unsigned int                    dl_defer_running  : 1;
 668
 669         /*
 670          * Bandwidth enforcement timer. Each -deadline task has its
 671          * own bandwidth to be enforced, thus we need one timer per task.
 672          */
 673         struct hrtimer                  dl_timer;
 674
 675         /*
 676          * Inactive timer, responsible for decreasing the active utilization
 677          * at the "0-lag time". When a -deadline task blocks, it contributes
 678          * to GRUB's active utilization until the "0-lag time", hence a
 679          * timer is needed to decrease the active utilization at the correct
 680          * time.
 681          */
 682         struct hrtimer                  inactive_timer;
 683
 684         /*
 685          * Bits for DL-server functionality. Also see the comment near
 686          * dl_server_update().
 687          *
 688          * @rq the runqueue this server is for
 689          *
 690          * @server_has_tasks() returns true if @server_pick return a
 691          * runnable task.
 692          */
 693         struct rq                       *rq;
 694         dl_server_has_tasks_f           server_has_tasks;
 695         dl_server_pick_f                server_pick_task;
 696
 697 #ifdef CONFIG_RT_MUTEXES
 698         /*
 699          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 700          * pi_se points to the donor, otherwise points to the dl_se it belongs
 701          * to (the original one/itself).
 702          */
 703         struct sched_dl_entity *pi_se;
 704 #endif
 705 };
 706
 707 #ifdef CONFIG_UCLAMP_TASK
 708 /* Number of utilization clamp buckets (shorter alias) */
 709 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 710
 711 /*
 712  * Utilization clamp for a scheduling entity
 713  * @value:              clamp value "assigned" to a se
 714  * @bucket_id:          bucket index corresponding to the "assigned" value
 715  * @active:             the se is currently refcounted in a rq's bucket
 716  * @user_defined:       the requested clamp value comes from user-space
 717  *
 718  * The bucket_id is the index of the clamp bucket matching the clamp value
 719  * which is pre-computed and stored to avoid expensive integer divisions from
 720  * the fast path.
 721  *
 722  * The active bit is set whenever a task has got an "effective" value assigned,
 723  * which can be different from the clamp value "requested" from user-space.
 724  * This allows to know a task is refcounted in the rq's bucket corresponding
 725  * to the "effective" bucket_id.
 726  *
 727  * The user_defined bit is set whenever a task has got a task-specific clamp
 728  * value requested from userspace, i.e. the system defaults apply to this task
 729  * just as a restriction. This allows to relax default clamps when a less
 730  * restrictive task-specific value has been requested, thus allowing to
 731  * implement a "nice" semantic. For example, a task running with a 20%
 732  * default boost can still drop its own boosting to 0%.
 733  */
 734 struct uclamp_se {
 735         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 736         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 737         unsigned int active             : 1;
 738         unsigned int user_defined       : 1;
 739 };
 740 #endif /* CONFIG_UCLAMP_TASK */
 741
 742 union rcu_special {
 743         struct {
 744                 u8                      blocked;
 745                 u8                      need_qs;
 746                 u8                      exp_hint; /* Hint for performance. */
 747                 u8                      need_mb; /* Readers need smp_mb(). */
 748         } b; /* Bits. */
 749         u32 s; /* Set of bits. */
 750 };
 751
 752 enum perf_event_task_context {
 753         perf_invalid_context = -1,
 754         perf_hw_context = 0,
 755         perf_sw_context,
 756         perf_nr_task_contexts,
 757 };
 758
 759 /*
 760  * Number of contexts where an event can trigger:
 761  *      task, softirq, hardirq, nmi.
 762  */
 763 #define PERF_NR_CONTEXTS        4
 764
 765 struct wake_q_node {
 766         struct wake_q_node *next;
 767 };
 768
 769 struct kmap_ctrl {
 770 #ifdef CONFIG_KMAP_LOCAL
 771         int                             idx;
 772         pte_t                           pteval[KM_MAX_IDX];
 773 #endif
 774 };
 775
 776 struct task_struct {
 777 #ifdef CONFIG_THREAD_INFO_IN_TASK
 778         /*
 779          * For reasons of header soup (see current_thread_info()), this
 780          * must be the first element of task_struct.
 781          */
 782         struct thread_info              thread_info;
 783 #endif
 784         unsigned int                    __state;
 785
 786         /* saved state for "spinlock sleepers" */
 787         unsigned int                    saved_state;
 788
 789         /*
 790          * This begins the randomizable portion of task_struct. Only
 791          * scheduling-critical items should be added above here.
 792          */
 793         randomized_struct_fields_start
 794
 795         void                            *stack;
 796         refcount_t                      usage;
 797         /* Per task flags (PF_*), defined further below: */
 798         unsigned int                    flags;
 799         unsigned int                    ptrace;
 800
 801 #ifdef CONFIG_MEM_ALLOC_PROFILING
 802         struct alloc_tag                *alloc_tag;
 803 #endif
 804
 805 #ifdef CONFIG_SMP
 806         int                             on_cpu;
 807         struct __call_single_node       wake_entry;
 808         unsigned int                    wakee_flips;
 809         unsigned long                   wakee_flip_decay_ts;
 810         struct task_struct              *last_wakee;
 811
 812         /*
 813          * recent_used_cpu is initially set as the last CPU used by a task
 814          * that wakes affine another task. Waker/wakee relationships can
 815          * push tasks around a CPU where each wakeup moves to the next one.
 816          * Tracking a recently used CPU allows a quick search for a recently
 817          * used CPU that may be idle.
 818          */
 819         int                             recent_used_cpu;
 820         int                             wake_cpu;
 821 #endif
 822         int                             on_rq;
 823
 824         int                             prio;
 825         int                             static_prio;
 826         int                             normal_prio;
 827         unsigned int                    rt_priority;
 828
 829         struct sched_entity             se;
 830         struct sched_rt_entity          rt;
 831         struct sched_dl_entity          dl;
 832         struct sched_dl_entity          *dl_server;
 833         const struct sched_class        *sched_class;
 834
 835 #ifdef CONFIG_SCHED_CORE
 836         struct rb_node                  core_node;
 837         unsigned long                   core_cookie;
 838         unsigned int                    core_occupation;
 839 #endif
 840
 841 #ifdef CONFIG_CGROUP_SCHED
 842         struct task_group               *sched_task_group;
 843 #endif
 844
 845
 846 #ifdef CONFIG_UCLAMP_TASK
 847         /*
 848          * Clamp values requested for a scheduling entity.
 849          * Must be updated with task_rq_lock() held.
 850          */
 851         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 852         /*
 853          * Effective clamp values used for a scheduling entity.
 854          * Must be updated with task_rq_lock() held.
 855          */
 856         struct uclamp_se                uclamp[UCLAMP_CNT];
 857 #endif
 858
 859         struct sched_statistics         stats;
 860
 861 #ifdef CONFIG_PREEMPT_NOTIFIERS
 862         /* List of struct preempt_notifier: */
 863         struct hlist_head               preempt_notifiers;
 864 #endif
 865
 866 #ifdef CONFIG_BLK_DEV_IO_TRACE
 867         unsigned int                    btrace_seq;
 868 #endif
 869
 870         unsigned int                    policy;
 871         unsigned long                   max_allowed_capacity;
 872         int                             nr_cpus_allowed;
 873         const cpumask_t                 *cpus_ptr;
 874         cpumask_t                       *user_cpus_ptr;
 875         cpumask_t                       cpus_mask;
 876         void                            *migration_pending;
 877 #ifdef CONFIG_SMP
 878         unsigned short                  migration_disabled;
 879 #endif
 880         unsigned short                  migration_flags;
 881
 882 #ifdef CONFIG_PREEMPT_RCU
 883         int                             rcu_read_lock_nesting;
 884         union rcu_special               rcu_read_unlock_special;
 885         struct list_head                rcu_node_entry;
 886         struct rcu_node                 *rcu_blocked_node;
 887 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 888
 889 #ifdef CONFIG_TASKS_RCU
 890         unsigned long                   rcu_tasks_nvcsw;
 891         u8                              rcu_tasks_holdout;
 892         u8                              rcu_tasks_idx;
 893         int                             rcu_tasks_idle_cpu;
 894         struct list_head                rcu_tasks_holdout_list;
 895         int                             rcu_tasks_exit_cpu;
 896         struct list_head                rcu_tasks_exit_list;
 897 #endif /* #ifdef CONFIG_TASKS_RCU */
 898
 899 #ifdef CONFIG_TASKS_TRACE_RCU
 900         int                             trc_reader_nesting;
 901         int                             trc_ipi_to_cpu;
 902         union rcu_special               trc_reader_special;
 903         struct list_head                trc_holdout_list;
 904         struct list_head                trc_blkd_node;
 905         int                             trc_blkd_cpu;
 906 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 907
 908         struct sched_info               sched_info;
 909
 910         struct list_head                tasks;
 911 #ifdef CONFIG_SMP
 912         struct plist_node               pushable_tasks;
 913         struct rb_node                  pushable_dl_tasks;
 914 #endif
 915
 916         struct mm_struct                *mm;
 917         struct mm_struct                *active_mm;
 918         struct address_space            *faults_disabled_mapping;
 919
 920         int                             exit_state;
 921         int                             exit_code;
 922         int                             exit_signal;
 923         /* The signal sent when the parent dies: */
 924         int                             pdeath_signal;
 925         /* JOBCTL_*, siglock protected: */
 926         unsigned long                   jobctl;
 927
 928         /* Used for emulating ABI behavior of previous Linux versions: */
 929         unsigned int                    personality;
 930
 931         /* Scheduler bits, serialized by scheduler locks: */
 932         unsigned                        sched_reset_on_fork:1;
 933         unsigned                        sched_contributes_to_load:1;
 934         unsigned                        sched_migrated:1;
 935
 936         /* Force alignment to the next boundary: */
 937         unsigned                        :0;
 938
 939         /* Unserialized, strictly 'current' */
 940
 941         /*
 942          * This field must not be in the scheduler word above due to wakelist
 943          * queueing no longer being serialized by p->on_cpu. However:
 944          *
 945          * p->XXX = X;                  ttwu()
 946          * schedule()                     if (p->on_rq && ..) // false
 947          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 948          *   deactivate_task()                ttwu_queue_wakelist())
 949          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 950          *
 951          * guarantees all stores of 'current' are visible before
 952          * ->sched_remote_wakeup gets used, so it can be in this word.
 953          */
 954         unsigned                        sched_remote_wakeup:1;
 955 #ifdef CONFIG_RT_MUTEXES
 956         unsigned                        sched_rt_mutex:1;
 957 #endif
 958
 959         /* Bit to tell TOMOYO we're in execve(): */
 960         unsigned                        in_execve:1;
 961         unsigned                        in_iowait:1;
 962 #ifndef TIF_RESTORE_SIGMASK
 963         unsigned                        restore_sigmask:1;
 964 #endif
 965 #ifdef CONFIG_MEMCG_V1
 966         unsigned                        in_user_fault:1;
 967 #endif
 968 #ifdef CONFIG_LRU_GEN
 969         /* whether the LRU algorithm may apply to this access */
 970         unsigned                        in_lru_fault:1;
 971 #endif
 972 #ifdef CONFIG_COMPAT_BRK
 973         unsigned                        brk_randomized:1;
 974 #endif
 975 #ifdef CONFIG_CGROUPS
 976         /* disallow userland-initiated cgroup migration */
 977         unsigned                        no_cgroup_migration:1;
 978         /* task is frozen/stopped (used by the cgroup freezer) */
 979         unsigned                        frozen:1;
 980 #endif
 981 #ifdef CONFIG_BLK_CGROUP
 982         unsigned                        use_memdelay:1;
 983 #endif
 984 #ifdef CONFIG_PSI
 985         /* Stalled due to lack of memory */
 986         unsigned                        in_memstall:1;
 987 #endif
 988 #ifdef CONFIG_PAGE_OWNER
 989         /* Used by page_owner=on to detect recursion in page tracking. */
 990         unsigned                        in_page_owner:1;
 991 #endif
 992 #ifdef CONFIG_EVENTFD
 993         /* Recursion prevention for eventfd_signal() */
 994         unsigned                        in_eventfd:1;
 995 #endif
 996 #ifdef CONFIG_ARCH_HAS_CPU_PASID
 997         unsigned                        pasid_activated:1;
 998 #endif
 999 #ifdef  CONFIG_CPU_SUP_INTEL
1000         unsigned                        reported_split_lock:1;
1001 #endif
1002 #ifdef CONFIG_TASK_DELAY_ACCT
1003         /* delay due to memory thrashing */
1004         unsigned                        in_thrashing:1;
1005 #endif
1006 #ifdef CONFIG_PREEMPT_RT
1007         struct netdev_xmit              net_xmit;
1008 #endif
1009         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
1010
1011         struct restart_block            restart_block;
1012
1013         pid_t                           pid;
1014         pid_t                           tgid;
1015
1016 #ifdef CONFIG_STACKPROTECTOR
1017         /* Canary value for the -fstack-protector GCC feature: */
1018         unsigned long                   stack_canary;
1019 #endif
1020         /*
1021          * Pointers to the (original) parent process, youngest child, younger sibling,
1022          * older sibling, respectively.  (p->father can be replaced with
1023          * p->real_parent->pid)
1024          */
1025
1026         /* Real parent process: */
1027         struct task_struct __rcu        *real_parent;
1028
1029         /* Recipient of SIGCHLD, wait4() reports: */
1030         struct task_struct __rcu        *parent;
1031
1032         /*
1033          * Children/sibling form the list of natural children:
1034          */
1035         struct list_head                children;
1036         struct list_head                sibling;
1037         struct task_struct              *group_leader;
1038
1039         /*
1040          * 'ptraced' is the list of tasks this task is using ptrace() on.
1041          *
1042          * This includes both natural children and PTRACE_ATTACH targets.
1043          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
1044          */
1045         struct list_head                ptraced;
1046         struct list_head                ptrace_entry;
1047
1048         /* PID/PID hash table linkage. */
1049         struct pid                      *thread_pid;
1050         struct hlist_node               pid_links[PIDTYPE_MAX];
1051         struct list_head                thread_node;
1052
1053         struct completion               *vfork_done;
1054
1055         /* CLONE_CHILD_SETTID: */
1056         int __user                      *set_child_tid;
1057
1058         /* CLONE_CHILD_CLEARTID: */
1059         int __user                      *clear_child_tid;
1060
1061         /* PF_KTHREAD | PF_IO_WORKER */
1062         void                            *worker_private;
1063
1064         u64                             utime;
1065         u64                             stime;
1066 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1067         u64                             utimescaled;
1068         u64                             stimescaled;
1069 #endif
1070         u64                             gtime;
1071         struct prev_cputime             prev_cputime;
1072 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1073         struct vtime                    vtime;
1074 #endif
1075
1076 #ifdef CONFIG_NO_HZ_FULL
1077         atomic_t                        tick_dep_mask;
1078 #endif
1079         /* Context switch counts: */
1080         unsigned long                   nvcsw;
1081         unsigned long                   nivcsw;
1082
1083         /* Monotonic time in nsecs: */
1084         u64                             start_time;
1085
1086         /* Boot based time in nsecs: */
1087         u64                             start_boottime;
1088
1089         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1090         unsigned long                   min_flt;
1091         unsigned long                   maj_flt;
1092
1093         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1094         struct posix_cputimers          posix_cputimers;
1095
1096 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1097         struct posix_cputimers_work     posix_cputimers_work;
1098 #endif
1099
1100         /* Process credentials: */
1101
1102         /* Tracer's credentials at attach: */
1103         const struct cred __rcu         *ptracer_cred;
1104
1105         /* Objective and real subjective task credentials (COW): */
1106         const struct cred __rcu         *real_cred;
1107
1108         /* Effective (overridable) subjective task credentials (COW): */
1109         const struct cred __rcu         *cred;
1110
1111 #ifdef CONFIG_KEYS
1112         /* Cached requested key. */
1113         struct key                      *cached_requested_key;
1114 #endif
1115
1116         /*
1117          * executable name, excluding path.
1118          *
1119          * - normally initialized setup_new_exec()
1120          * - access it with [gs]et_task_comm()
1121          * - lock it with task_lock()
1122          */
1123         char                            comm[TASK_COMM_LEN];
1124
1125         struct nameidata                *nameidata;
1126
1127 #ifdef CONFIG_SYSVIPC
1128         struct sysv_sem                 sysvsem;
1129         struct sysv_shm                 sysvshm;
1130 #endif
1131 #ifdef CONFIG_DETECT_HUNG_TASK
1132         unsigned long                   last_switch_count;
1133         unsigned long                   last_switch_time;
1134 #endif
1135         /* Filesystem information: */
1136         struct fs_struct                *fs;
1137
1138         /* Open file information: */
1139         struct files_struct             *files;
1140
1141 #ifdef CONFIG_IO_URING
1142         struct io_uring_task            *io_uring;
1143 #endif
1144
1145         /* Namespaces: */
1146         struct nsproxy                  *nsproxy;
1147
1148         /* Signal handlers: */
1149         struct signal_struct            *signal;
1150         struct sighand_struct __rcu             *sighand;
1151         sigset_t                        blocked;
1152         sigset_t                        real_blocked;
1153         /* Restored if set_restore_sigmask() was used: */
1154         sigset_t                        saved_sigmask;
1155         struct sigpending               pending;
1156         unsigned long                   sas_ss_sp;
1157         size_t                          sas_ss_size;
1158         unsigned int                    sas_ss_flags;
1159
1160         struct callback_head            *task_works;
1161
1162 #ifdef CONFIG_AUDIT
1163 #ifdef CONFIG_AUDITSYSCALL
1164         struct audit_context            *audit_context;
1165 #endif
1166         kuid_t                          loginuid;
1167         unsigned int                    sessionid;
1168 #endif
1169         struct seccomp                  seccomp;
1170         struct syscall_user_dispatch    syscall_dispatch;
1171
1172         /* Thread group tracking: */
1173         u64                             parent_exec_id;
1174         u64                             self_exec_id;
1175
1176         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1177         spinlock_t                      alloc_lock;
1178
1179         /* Protection of the PI data structures: */
1180         raw_spinlock_t                  pi_lock;
1181
1182         struct wake_q_node              wake_q;
1183
1184 #ifdef CONFIG_RT_MUTEXES
1185         /* PI waiters blocked on a rt_mutex held by this task: */
1186         struct rb_root_cached           pi_waiters;
1187         /* Updated under owner's pi_lock and rq lock */
1188         struct task_struct              *pi_top_task;
1189         /* Deadlock detection and priority inheritance handling: */
1190         struct rt_mutex_waiter          *pi_blocked_on;
1191 #endif
1192
1193 #ifdef CONFIG_DEBUG_MUTEXES
1194         /* Mutex deadlock detection: */
1195         struct mutex_waiter             *blocked_on;
1196 #endif
1197
1198 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1199         int                             non_block_count;
1200 #endif
1201
1202 #ifdef CONFIG_TRACE_IRQFLAGS
1203         struct irqtrace_events          irqtrace;
1204         unsigned int                    hardirq_threaded;
1205         u64                             hardirq_chain_key;
1206         int                             softirqs_enabled;
1207         int                             softirq_context;
1208         int                             irq_config;
1209 #endif
1210 #ifdef CONFIG_PREEMPT_RT
1211         int                             softirq_disable_cnt;
1212 #endif
1213
1214 #ifdef CONFIG_LOCKDEP
1215 # define MAX_LOCK_DEPTH                 48UL
1216         u64                             curr_chain_key;
1217         int                             lockdep_depth;
1218         unsigned int                    lockdep_recursion;
1219         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1220 #endif
1221
1222 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1223         unsigned int                    in_ubsan;
1224 #endif
1225
1226         /* Journalling filesystem info: */
1227         void                            *journal_info;
1228
1229         /* Stacked block device info: */
1230         struct bio_list                 *bio_list;
1231
1232         /* Stack plugging: */
1233         struct blk_plug                 *plug;
1234
1235         /* VM state: */
1236         struct reclaim_state            *reclaim_state;
1237
1238         struct io_context               *io_context;
1239
1240 #ifdef CONFIG_COMPACTION
1241         struct capture_control          *capture_control;
1242 #endif
1243         /* Ptrace state: */
1244         unsigned long                   ptrace_message;
1245         kernel_siginfo_t                *last_siginfo;
1246
1247         struct task_io_accounting       ioac;
1248 #ifdef CONFIG_PSI
1249         /* Pressure stall state */
1250         unsigned int                    psi_flags;
1251 #endif
1252 #ifdef CONFIG_TASK_XACCT
1253         /* Accumulated RSS usage: */
1254         u64                             acct_rss_mem1;
1255         /* Accumulated virtual memory usage: */
1256         u64                             acct_vm_mem1;
1257         /* stime + utime since last update: */
1258         u64                             acct_timexpd;
1259 #endif
1260 #ifdef CONFIG_CPUSETS
1261         /* Protected by ->alloc_lock: */
1262         nodemask_t                      mems_allowed;
1263         /* Sequence number to catch updates: */
1264         seqcount_spinlock_t             mems_allowed_seq;
1265         int                             cpuset_mem_spread_rotor;
1266 #endif
1267 #ifdef CONFIG_CGROUPS
1268         /* Control Group info protected by css_set_lock: */
1269         struct css_set __rcu            *cgroups;
1270         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1271         struct list_head                cg_list;
1272 #endif
1273 #ifdef CONFIG_X86_CPU_RESCTRL
1274         u32                             closid;
1275         u32                             rmid;
1276 #endif
1277 #ifdef CONFIG_FUTEX
1278         struct robust_list_head __user  *robust_list;
1279 #ifdef CONFIG_COMPAT
1280         struct compat_robust_list_head __user *compat_robust_list;
1281 #endif
1282         struct list_head                pi_state_list;
1283         struct futex_pi_state           *pi_state_cache;
1284         struct mutex                    futex_exit_mutex;
1285         unsigned int                    futex_state;
1286 #endif
1287 #ifdef CONFIG_PERF_EVENTS
1288         u8                              perf_recursion[PERF_NR_CONTEXTS];
1289         struct perf_event_context       *perf_event_ctxp;
1290         struct mutex                    perf_event_mutex;
1291         struct list_head                perf_event_list;
1292 #endif
1293 #ifdef CONFIG_DEBUG_PREEMPT
1294         unsigned long                   preempt_disable_ip;
1295 #endif
1296 #ifdef CONFIG_NUMA
1297         /* Protected by alloc_lock: */
1298         struct mempolicy                *mempolicy;
1299         short                           il_prev;
1300         u8                              il_weight;
1301         short                           pref_node_fork;
1302 #endif
1303 #ifdef CONFIG_NUMA_BALANCING
1304         int                             numa_scan_seq;
1305         unsigned int                    numa_scan_period;
1306         unsigned int                    numa_scan_period_max;
1307         int                             numa_preferred_nid;
1308         unsigned long                   numa_migrate_retry;
1309         /* Migration stamp: */
1310         u64                             node_stamp;
1311         u64                             last_task_numa_placement;
1312         u64                             last_sum_exec_runtime;
1313         struct callback_head            numa_work;
1314
1315         /*
1316          * This pointer is only modified for current in syscall and
1317          * pagefault context (and for tasks being destroyed), so it can be read
1318          * from any of the following contexts:
1319          *  - RCU read-side critical section
1320          *  - current->numa_group from everywhere
1321          *  - task's runqueue locked, task not running
1322          */
1323         struct numa_group __rcu         *numa_group;
1324
1325         /*
1326          * numa_faults is an array split into four regions:
1327          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1328          * in this precise order.
1329          *
1330          * faults_memory: Exponential decaying average of faults on a per-node
1331          * basis. Scheduling placement decisions are made based on these
1332          * counts. The values remain static for the duration of a PTE scan.
1333          * faults_cpu: Track the nodes the process was running on when a NUMA
1334          * hinting fault was incurred.
1335          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1336          * during the current scan window. When the scan completes, the counts
1337          * in faults_memory and faults_cpu decay and these values are copied.
1338          */
1339         unsigned long                   *numa_faults;
1340         unsigned long                   total_numa_faults;
1341
1342         /*
1343          * numa_faults_locality tracks if faults recorded during the last
1344          * scan window were remote/local or failed to migrate. The task scan
1345          * period is adapted based on the locality of the faults with different
1346          * weights depending on whether they were shared or private faults
1347          */
1348         unsigned long                   numa_faults_locality[3];
1349
1350         unsigned long                   numa_pages_migrated;
1351 #endif /* CONFIG_NUMA_BALANCING */
1352
1353 #ifdef CONFIG_RSEQ
1354         struct rseq __user *rseq;
1355         u32 rseq_len;
1356         u32 rseq_sig;
1357         /*
1358          * RmW on rseq_event_mask must be performed atomically
1359          * with respect to preemption.
1360          */
1361         unsigned long rseq_event_mask;
1362 #endif
1363
1364 #ifdef CONFIG_SCHED_MM_CID
1365         int                             mm_cid;         /* Current cid in mm */
1366         int                             last_mm_cid;    /* Most recent cid in mm */
1367         int                             migrate_from_cpu;
1368         int                             mm_cid_active;  /* Whether cid bitmap is active */
1369         struct callback_head            cid_work;
1370 #endif
1371
1372         struct tlbflush_unmap_batch     tlb_ubc;
1373
1374         /* Cache last used pipe for splice(): */
1375         struct pipe_inode_info          *splice_pipe;
1376
1377         struct page_frag                task_frag;
1378
1379 #ifdef CONFIG_TASK_DELAY_ACCT
1380         struct task_delay_info          *delays;
1381 #endif
1382
1383 #ifdef CONFIG_FAULT_INJECTION
1384         int                             make_it_fail;
1385         unsigned int                    fail_nth;
1386 #endif
1387         /*
1388          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1389          * balance_dirty_pages() for a dirty throttling pause:
1390          */
1391         int                             nr_dirtied;
1392         int                             nr_dirtied_pause;
1393         /* Start of a write-and-pause period: */
1394         unsigned long                   dirty_paused_when;
1395
1396 #ifdef CONFIG_LATENCYTOP
1397         int                             latency_record_count;
1398         struct latency_record           latency_record[LT_SAVECOUNT];
1399 #endif
1400         /*
1401          * Time slack values; these are used to round up poll() and
1402          * select() etc timeout values. These are in nanoseconds.
1403          */
1404         u64                             timer_slack_ns;
1405         u64                             default_timer_slack_ns;
1406
1407 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1408         unsigned int                    kasan_depth;
1409 #endif
1410
1411 #ifdef CONFIG_KCSAN
1412         struct kcsan_ctx                kcsan_ctx;
1413 #ifdef CONFIG_TRACE_IRQFLAGS
1414         struct irqtrace_events          kcsan_save_irqtrace;
1415 #endif
1416 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1417         int                             kcsan_stack_depth;
1418 #endif
1419 #endif
1420
1421 #ifdef CONFIG_KMSAN
1422         struct kmsan_ctx                kmsan_ctx;
1423 #endif
1424
1425 #if IS_ENABLED(CONFIG_KUNIT)
1426         struct kunit                    *kunit_test;
1427 #endif
1428
1429 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1430         /* Index of current stored address in ret_stack: */
1431         int                             curr_ret_stack;
1432         int                             curr_ret_depth;
1433
1434         /* Stack of return addresses for return function tracing: */
1435         unsigned long                   *ret_stack;
1436
1437         /* Timestamp for last schedule: */
1438         unsigned long long              ftrace_timestamp;
1439
1440         /*
1441          * Number of functions that haven't been traced
1442          * because of depth overrun:
1443          */
1444         atomic_t                        trace_overrun;
1445
1446         /* Pause tracing: */
1447         atomic_t                        tracing_graph_pause;
1448 #endif
1449
1450 #ifdef CONFIG_TRACING
1451         /* Bitmask and counter of trace recursion: */
1452         unsigned long                   trace_recursion;
1453 #endif /* CONFIG_TRACING */
1454
1455 #ifdef CONFIG_KCOV
1456         /* See kernel/kcov.c for more details. */
1457
1458         /* Coverage collection mode enabled for this task (0 if disabled): */
1459         unsigned int                    kcov_mode;
1460
1461         /* Size of the kcov_area: */
1462         unsigned int                    kcov_size;
1463
1464         /* Buffer for coverage collection: */
1465         void                            *kcov_area;
1466
1467         /* KCOV descriptor wired with this task or NULL: */
1468         struct kcov                     *kcov;
1469
1470         /* KCOV common handle for remote coverage collection: */
1471         u64                             kcov_handle;
1472
1473         /* KCOV sequence number: */
1474         int                             kcov_sequence;
1475
1476         /* Collect coverage from softirq context: */
1477         unsigned int                    kcov_softirq;
1478 #endif
1479
1480 #ifdef CONFIG_MEMCG_V1
1481         struct mem_cgroup               *memcg_in_oom;
1482 #endif
1483
1484 #ifdef CONFIG_MEMCG
1485         /* Number of pages to reclaim on returning to userland: */
1486         unsigned int                    memcg_nr_pages_over_high;
1487
1488         /* Used by memcontrol for targeted memcg charge: */
1489         struct mem_cgroup               *active_memcg;
1490
1491         /* Cache for current->cgroups->memcg->objcg lookups: */
1492         struct obj_cgroup               *objcg;
1493 #endif
1494
1495 #ifdef CONFIG_BLK_CGROUP
1496         struct gendisk                  *throttle_disk;
1497 #endif
1498
1499 #ifdef CONFIG_UPROBES
1500         struct uprobe_task              *utask;
1501 #endif
1502 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1503         unsigned int                    sequential_io;
1504         unsigned int                    sequential_io_avg;
1505 #endif
1506         struct kmap_ctrl                kmap_ctrl;
1507 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1508         unsigned long                   task_state_change;
1509 # ifdef CONFIG_PREEMPT_RT
1510         unsigned long                   saved_state_change;
1511 # endif
1512 #endif
1513         struct rcu_head                 rcu;
1514         refcount_t                      rcu_users;
1515         int                             pagefault_disabled;
1516 #ifdef CONFIG_MMU
1517         struct task_struct              *oom_reaper_list;
1518         struct timer_list               oom_reaper_timer;
1519 #endif
1520 #ifdef CONFIG_VMAP_STACK
1521         struct vm_struct                *stack_vm_area;
1522 #endif
1523 #ifdef CONFIG_THREAD_INFO_IN_TASK
1524         /* A live task holds one reference: */
1525         refcount_t                      stack_refcount;
1526 #endif
1527 #ifdef CONFIG_LIVEPATCH
1528         int patch_state;
1529 #endif
1530 #ifdef CONFIG_SECURITY
1531         /* Used by LSM modules for access restriction: */
1532         void                            *security;
1533 #endif
1534 #ifdef CONFIG_BPF_SYSCALL
1535         /* Used by BPF task local storage */
1536         struct bpf_local_storage __rcu  *bpf_storage;
1537         /* Used for BPF run context */
1538         struct bpf_run_ctx              *bpf_ctx;
1539 #endif
1540         /* Used by BPF for per-TASK xdp storage */
1541         struct bpf_net_context          *bpf_net_context;
1542
1543 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1544         unsigned long                   lowest_stack;
1545         unsigned long                   prev_lowest_stack;
1546 #endif
1547
1548 #ifdef CONFIG_X86_MCE
1549         void __user                     *mce_vaddr;
1550         __u64                           mce_kflags;
1551         u64                             mce_addr;
1552         __u64                           mce_ripv : 1,
1553                                         mce_whole_page : 1,
1554                                         __mce_reserved : 62;
1555         struct callback_head            mce_kill_me;
1556         int                             mce_count;
1557 #endif
1558
1559 #ifdef CONFIG_KRETPROBES
1560         struct llist_head               kretprobe_instances;
1561 #endif
1562 #ifdef CONFIG_RETHOOK
1563         struct llist_head               rethooks;
1564 #endif
1565
1566 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1567         /*
1568          * If L1D flush is supported on mm context switch
1569          * then we use this callback head to queue kill work
1570          * to kill tasks that are not running on SMT disabled
1571          * cores
1572          */
1573         struct callback_head            l1d_flush_kill;
1574 #endif
1575
1576 #ifdef CONFIG_RV
1577         /*
1578          * Per-task RV monitor. Nowadays fixed in RV_PER_TASK_MONITORS.
1579          * If we find justification for more monitors, we can think
1580          * about adding more or developing a dynamic method. So far,
1581          * none of these are justified.
1582          */
1583         union rv_task_monitor           rv[RV_PER_TASK_MONITORS];
1584 #endif
1585
1586 #ifdef CONFIG_USER_EVENTS
1587         struct user_event_mm            *user_event_mm;
1588 #endif
1589
1590         /*
1591          * New fields for task_struct should be added above here, so that
1592          * they are included in the randomized portion of task_struct.
1593          */
1594         randomized_struct_fields_end
1595
1596         /* CPU-specific state of this task: */
1597         struct thread_struct            thread;
1598
1599         /*
1600          * WARNING: on x86, 'thread_struct' contains a variable-sized
1601          * structure.  It *MUST* be at the end of 'task_struct'.
1602          *
1603          * Do not put anything below here!
1604          */
1605 };
1606
1607 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1608 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1609
1610 static inline unsigned int __task_state_index(unsigned int tsk_state,
1611                                               unsigned int tsk_exit_state)
1612 {
1613         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1614
1615         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1616
1617         if ((tsk_state & TASK_IDLE) == TASK_IDLE)
1618                 state = TASK_REPORT_IDLE;
1619
1620         /*
1621          * We're lying here, but rather than expose a completely new task state
1622          * to userspace, we can make this appear as if the task has gone through
1623          * a regular rt_mutex_lock() call.
1624          */
1625         if (tsk_state & TASK_RTLOCK_WAIT)
1626                 state = TASK_UNINTERRUPTIBLE;
1627
1628         return fls(state);
1629 }
1630
1631 static inline unsigned int task_state_index(struct task_struct *tsk)
1632 {
1633         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1634 }
1635
1636 static inline char task_index_to_char(unsigned int state)
1637 {
1638         static const char state_char[] = "RSDTtXZPI";
1639
1640         BUILD_BUG_ON(TASK_REPORT_MAX * 2 != 1 << (sizeof(state_char) - 1));
1641
1642         return state_char[state];
1643 }
1644
1645 static inline char task_state_to_char(struct task_struct *tsk)
1646 {
1647         return task_index_to_char(task_state_index(tsk));
1648 }
1649
1650 extern struct pid *cad_pid;
1651
1652 /*
1653  * Per process flags
1654  */
1655 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1656 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1657 #define PF_EXITING              0x00000004      /* Getting shut down */
1658 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1659 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1660 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1661 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1662 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1663 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1664 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1665 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1666 #define PF_MEMALLOC             0x00000800      /* Allocating memory to free memory. See memalloc_noreclaim_save() */
1667 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1668 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1669 #define PF_USER_WORKER          0x00004000      /* Kernel thread cloned from userspace thread */
1670 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1671 #define PF__HOLE__00010000      0x00010000
1672 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1673 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocations inherit GFP_NOFS. See memalloc_nfs_save() */
1674 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocations inherit GFP_NOIO. See memalloc_noio_save() */
1675 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1676                                                  * I am cleaning dirty pages from some other bdi. */
1677 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1678 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1679 #define PF_MEMALLOC_NORECLAIM   0x00800000      /* All allocation requests will clear __GFP_DIRECT_RECLAIM */
1680 #define PF_MEMALLOC_NOWARN      0x01000000      /* All allocation requests will inherit __GFP_NOWARN */
1681 #define PF__HOLE__02000000      0x02000000
1682 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1683 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1684 #define PF_MEMALLOC_PIN         0x10000000      /* Allocations constrained to zones which allow long term pinning.
1685                                                  * See memalloc_pin_save() */
1686 #define PF_BLOCK_TS             0x20000000      /* plug has ts that needs updating */
1687 #define PF__HOLE__40000000      0x40000000
1688 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1689
1690 /*
1691  * Only the _current_ task can read/write to tsk->flags, but other
1692  * tasks can access tsk->flags in readonly mode for example
1693  * with tsk_used_math (like during threaded core dumping).
1694  * There is however an exception to this rule during ptrace
1695  * or during fork: the ptracer task is allowed to write to the
1696  * child->flags of its traced child (same goes for fork, the parent
1697  * can write to the child->flags), because we're guaranteed the
1698  * child is not running and in turn not changing child->flags
1699  * at the same time the parent does it.
1700  */
1701 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1702 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1703 #define clear_used_math()                       clear_stopped_child_used_math(current)
1704 #define set_used_math()                         set_stopped_child_used_math(current)
1705
1706 #define conditional_stopped_child_used_math(condition, child) \
1707         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1708
1709 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1710
1711 #define copy_to_stopped_child_used_math(child) \
1712         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1713
1714 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1715 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1716 #define used_math()                             tsk_used_math(current)
1717
1718 static __always_inline bool is_percpu_thread(void)
1719 {
1720 #ifdef CONFIG_SMP
1721         return (current->flags & PF_NO_SETAFFINITY) &&
1722                 (current->nr_cpus_allowed  == 1);
1723 #else
1724         return true;
1725 #endif
1726 }
1727
1728 /* Per-process atomic flags. */
1729 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1730 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1731 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1732 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1733 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1734 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1735 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1736 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1737
1738 #define TASK_PFA_TEST(name, func)                                       \
1739         static inline bool task_##func(struct task_struct *p)           \
1740         { return test_bit(PFA_##name, &p->atomic_flags); }
1741
1742 #define TASK_PFA_SET(name, func)                                        \
1743         static inline void task_set_##func(struct task_struct *p)       \
1744         { set_bit(PFA_##name, &p->atomic_flags); }
1745
1746 #define TASK_PFA_CLEAR(name, func)                                      \
1747         static inline void task_clear_##func(struct task_struct *p)     \
1748         { clear_bit(PFA_##name, &p->atomic_flags); }
1749
1750 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1751 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1752
1753 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1754 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1755 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1756
1757 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1758 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1759 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1760
1761 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1762 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1763 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1764
1765 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1766 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1767 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1768
1769 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1770 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1771
1772 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1773 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1774 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1775
1776 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1777 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1778
1779 static inline void
1780 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1781 {
1782         current->flags &= ~flags;
1783         current->flags |= orig_flags & flags;
1784 }
1785
1786 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1787 extern int task_can_attach(struct task_struct *p);
1788 extern int dl_bw_alloc(int cpu, u64 dl_bw);
1789 extern void dl_bw_free(int cpu, u64 dl_bw);
1790 #ifdef CONFIG_SMP
1791
1792 /* do_set_cpus_allowed() - consider using set_cpus_allowed_ptr() instead */
1793 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1794
1795 /**
1796  * set_cpus_allowed_ptr - set CPU affinity mask of a task
1797  * @p: the task
1798  * @new_mask: CPU affinity mask
1799  *
1800  * Return: zero if successful, or a negative error code
1801  */
1802 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1803 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1804 extern void release_user_cpus_ptr(struct task_struct *p);
1805 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1806 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1807 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1808 #else
1809 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1810 {
1811 }
1812 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1813 {
1814         /* Opencoded cpumask_test_cpu(0, new_mask) to avoid dependency on cpumask.h */
1815         if ((*cpumask_bits(new_mask) & 1) == 0)
1816                 return -EINVAL;
1817         return 0;
1818 }
1819 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1820 {
1821         if (src->user_cpus_ptr)
1822                 return -EINVAL;
1823         return 0;
1824 }
1825 static inline void release_user_cpus_ptr(struct task_struct *p)
1826 {
1827         WARN_ON(p->user_cpus_ptr);
1828 }
1829
1830 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1831 {
1832         return 0;
1833 }
1834 #endif
1835
1836 extern int yield_to(struct task_struct *p, bool preempt);
1837 extern void set_user_nice(struct task_struct *p, long nice);
1838 extern int task_prio(const struct task_struct *p);
1839
1840 /**
1841  * task_nice - return the nice value of a given task.
1842  * @p: the task in question.
1843  *
1844  * Return: The nice value [ -20 ... 0 ... 19 ].
1845  */
1846 static inline int task_nice(const struct task_struct *p)
1847 {
1848         return PRIO_TO_NICE((p)->static_prio);
1849 }
1850
1851 extern int can_nice(const struct task_struct *p, const int nice);
1852 extern int task_curr(const struct task_struct *p);
1853 extern int idle_cpu(int cpu);
1854 extern int available_idle_cpu(int cpu);
1855 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1856 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1857 extern void sched_set_fifo(struct task_struct *p);
1858 extern void sched_set_fifo_low(struct task_struct *p);
1859 extern void sched_set_normal(struct task_struct *p, int nice);
1860 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1861 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1862 extern struct task_struct *idle_task(int cpu);
1863
1864 /**
1865  * is_idle_task - is the specified task an idle task?
1866  * @p: the task in question.
1867  *
1868  * Return: 1 if @p is an idle task. 0 otherwise.
1869  */
1870 static __always_inline bool is_idle_task(const struct task_struct *p)
1871 {
1872         return !!(p->flags & PF_IDLE);
1873 }
1874
1875 extern struct task_struct *curr_task(int cpu);
1876 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1877
1878 void yield(void);
1879
1880 union thread_union {
1881         struct task_struct task;
1882 #ifndef CONFIG_THREAD_INFO_IN_TASK
1883         struct thread_info thread_info;
1884 #endif
1885         unsigned long stack[THREAD_SIZE/sizeof(long)];
1886 };
1887
1888 #ifndef CONFIG_THREAD_INFO_IN_TASK
1889 extern struct thread_info init_thread_info;
1890 #endif
1891
1892 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1893
1894 #ifdef CONFIG_THREAD_INFO_IN_TASK
1895 # define task_thread_info(task) (&(task)->thread_info)
1896 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1897 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1898 #endif
1899
1900 /*
1901  * find a task by one of its numerical ids
1902  *
1903  * find_task_by_pid_ns():
1904  *      finds a task by its pid in the specified namespace
1905  * find_task_by_vpid():
1906  *      finds a task by its virtual pid
1907  *
1908  * see also find_vpid() etc in include/linux/pid.h
1909  */
1910
1911 extern struct task_struct *find_task_by_vpid(pid_t nr);
1912 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1913
1914 /*
1915  * find a task by its virtual pid and get the task struct
1916  */
1917 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1918
1919 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1920 extern int wake_up_process(struct task_struct *tsk);
1921 extern void wake_up_new_task(struct task_struct *tsk);
1922
1923 #ifdef CONFIG_SMP
1924 extern void kick_process(struct task_struct *tsk);
1925 #else
1926 static inline void kick_process(struct task_struct *tsk) { }
1927 #endif
1928
1929 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1930
1931 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1932 {
1933         __set_task_comm(tsk, from, false);
1934 }
1935
1936 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1937 #define get_task_comm(buf, tsk) ({                      \
1938         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1939         __get_task_comm(buf, sizeof(buf), tsk);         \
1940 })
1941
1942 #ifdef CONFIG_SMP
1943 static __always_inline void scheduler_ipi(void)
1944 {
1945         /*
1946          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1947          * TIF_NEED_RESCHED remotely (for the first time) will also send
1948          * this IPI.
1949          */
1950         preempt_fold_need_resched();
1951 }
1952 #else
1953 static inline void scheduler_ipi(void) { }
1954 #endif
1955
1956 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1957
1958 /*
1959  * Set thread flags in other task's structures.
1960  * See asm/thread_info.h for TIF_xxxx flags available:
1961  */
1962 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1963 {
1964         set_ti_thread_flag(task_thread_info(tsk), flag);
1965 }
1966
1967 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1968 {
1969         clear_ti_thread_flag(task_thread_info(tsk), flag);
1970 }
1971
1972 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1973                                           bool value)
1974 {
1975         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1976 }
1977
1978 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1979 {
1980         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
1981 }
1982
1983 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1984 {
1985         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
1986 }
1987
1988 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
1989 {
1990         return test_ti_thread_flag(task_thread_info(tsk), flag);
1991 }
1992
1993 static inline void set_tsk_need_resched(struct task_struct *tsk)
1994 {
1995         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
1996 }
1997
1998 static inline void clear_tsk_need_resched(struct task_struct *tsk)
1999 {
2000         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2001 }
2002
2003 static inline int test_tsk_need_resched(struct task_struct *tsk)
2004 {
2005         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2006 }
2007
2008 /*
2009  * cond_resched() and cond_resched_lock(): latency reduction via
2010  * explicit rescheduling in places that are safe. The return
2011  * value indicates whether a reschedule was done in fact.
2012  * cond_resched_lock() will drop the spinlock before scheduling,
2013  */
2014 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2015 extern int __cond_resched(void);
2016
2017 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
2018
2019 void sched_dynamic_klp_enable(void);
2020 void sched_dynamic_klp_disable(void);
2021
2022 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2023
2024 static __always_inline int _cond_resched(void)
2025 {
2026         return static_call_mod(cond_resched)();
2027 }
2028
2029 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
2030
2031 extern int dynamic_cond_resched(void);
2032
2033 static __always_inline int _cond_resched(void)
2034 {
2035         return dynamic_cond_resched();
2036 }
2037
2038 #else /* !CONFIG_PREEMPTION */
2039
2040 static inline int _cond_resched(void)
2041 {
2042         klp_sched_try_switch();
2043         return __cond_resched();
2044 }
2045
2046 #endif /* PREEMPT_DYNAMIC && CONFIG_HAVE_PREEMPT_DYNAMIC_CALL */
2047
2048 #else /* CONFIG_PREEMPTION && !CONFIG_PREEMPT_DYNAMIC */
2049
2050 static inline int _cond_resched(void)
2051 {
2052         klp_sched_try_switch();
2053         return 0;
2054 }
2055
2056 #endif /* !CONFIG_PREEMPTION || CONFIG_PREEMPT_DYNAMIC */
2057
2058 #define cond_resched() ({                       \
2059         __might_resched(__FILE__, __LINE__, 0); \
2060         _cond_resched();                        \
2061 })
2062
2063 extern int __cond_resched_lock(spinlock_t *lock);
2064 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2065 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2066
2067 #define MIGHT_RESCHED_RCU_SHIFT         8
2068 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2069
2070 #ifndef CONFIG_PREEMPT_RT
2071 /*
2072  * Non RT kernels have an elevated preempt count due to the held lock,
2073  * but are not allowed to be inside a RCU read side critical section
2074  */
2075 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2076 #else
2077 /*
2078  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2079  * cond_resched*lock() has to take that into account because it checks for
2080  * preempt_count() and rcu_preempt_depth().
2081  */
2082 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2083         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2084 #endif
2085
2086 #define cond_resched_lock(lock) ({                                              \
2087         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2088         __cond_resched_lock(lock);                                              \
2089 })
2090
2091 #define cond_resched_rwlock_read(lock) ({                                       \
2092         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2093         __cond_resched_rwlock_read(lock);                                       \
2094 })
2095
2096 #define cond_resched_rwlock_write(lock) ({                                      \
2097         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2098         __cond_resched_rwlock_write(lock);                                      \
2099 })
2100
2101 static __always_inline bool need_resched(void)
2102 {
2103         return unlikely(tif_need_resched());
2104 }
2105
2106 /*
2107  * Wrappers for p->thread_info->cpu access. No-op on UP.
2108  */
2109 #ifdef CONFIG_SMP
2110
2111 static inline unsigned int task_cpu(const struct task_struct *p)
2112 {
2113         return READ_ONCE(task_thread_info(p)->cpu);
2114 }
2115
2116 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2117
2118 #else
2119
2120 static inline unsigned int task_cpu(const struct task_struct *p)
2121 {
2122         return 0;
2123 }
2124
2125 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2126 {
2127 }
2128
2129 #endif /* CONFIG_SMP */
2130
2131 extern bool sched_task_on_rq(struct task_struct *p);
2132 extern unsigned long get_wchan(struct task_struct *p);
2133 extern struct task_struct *cpu_curr_snapshot(int cpu);
2134
2135 #include <linux/spinlock.h>
2136
2137 /*
2138  * In order to reduce various lock holder preemption latencies provide an
2139  * interface to see if a vCPU is currently running or not.
2140  *
2141  * This allows us to terminate optimistic spin loops and block, analogous to
2142  * the native optimistic spin heuristic of testing if the lock owner task is
2143  * running or not.
2144  */
2145 #ifndef vcpu_is_preempted
2146 static inline bool vcpu_is_preempted(int cpu)
2147 {
2148         return false;
2149 }
2150 #endif
2151
2152 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2153 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2154
2155 #ifndef TASK_SIZE_OF
2156 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2157 #endif
2158
2159 #ifdef CONFIG_SMP
2160 static inline bool owner_on_cpu(struct task_struct *owner)
2161 {
2162         /*
2163          * As lock holder preemption issue, we both skip spinning if
2164          * task is not on cpu or its cpu is preempted
2165          */
2166         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2167 }
2168
2169 /* Returns effective CPU energy utilization, as seen by the scheduler */
2170 unsigned long sched_cpu_util(int cpu);
2171 #endif /* CONFIG_SMP */
2172
2173 #ifdef CONFIG_SCHED_CORE
2174 extern void sched_core_free(struct task_struct *tsk);
2175 extern void sched_core_fork(struct task_struct *p);
2176 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2177                                 unsigned long uaddr);
2178 extern int sched_core_idle_cpu(int cpu);
2179 #else
2180 static inline void sched_core_free(struct task_struct *tsk) { }
2181 static inline void sched_core_fork(struct task_struct *p) { }
2182 static inline int sched_core_idle_cpu(int cpu) { return idle_cpu(cpu); }
2183 #endif
2184
2185 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2186
2187 #ifdef CONFIG_MEM_ALLOC_PROFILING
2188 static __always_inline struct alloc_tag *alloc_tag_save(struct alloc_tag *tag)
2189 {
2190         swap(current->alloc_tag, tag);
2191         return tag;
2192 }
2193
2194 static __always_inline void alloc_tag_restore(struct alloc_tag *tag, struct alloc_tag *old)
2195 {
2196 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
2197         WARN(current->alloc_tag != tag, "current->alloc_tag was changed:\n");
2198 #endif
2199         current->alloc_tag = old;
2200 }
2201 #else
2202 #define alloc_tag_save(_tag)                    NULL
2203 #define alloc_tag_restore(_tag, _old)           do {} while (0)
2204 #endif
2205
2206 #endif