include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13
  14 #include <linux/pid.h>
  15 #include <linux/sem.h>
  16 #include <linux/shm.h>
  17 #include <linux/mutex.h>
  18 #include <linux/plist.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/irqflags.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/nodemask.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/refcount.h>
  25 #include <linux/resource.h>
  26 #include <linux/latencytop.h>
  27 #include <linux/sched/prio.h>
  28 #include <linux/sched/types.h>
  29 #include <linux/signal_types.h>
  30 #include <linux/syscall_user_dispatch.h>
  31 #include <linux/mm_types_task.h>
  32 #include <linux/task_io_accounting.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/rseq.h>
  35 #include <linux/seqlock.h>
  36 #include <linux/kcsan.h>
  37 #include <asm/kmap_size.h>
  38
  39 /* task_struct member predeclarations (sorted alphabetically): */
  40 struct audit_context;
  41 struct backing_dev_info;
  42 struct bio_list;
  43 struct blk_plug;
  44 struct bpf_local_storage;
  45 struct bpf_run_ctx;
  46 struct capture_control;
  47 struct cfs_rq;
  48 struct fs_struct;
  49 struct futex_pi_state;
  50 struct io_context;
  51 struct io_uring_task;
  52 struct mempolicy;
  53 struct nameidata;
  54 struct nsproxy;
  55 struct perf_event_context;
  56 struct pid_namespace;
  57 struct pipe_inode_info;
  58 struct rcu_node;
  59 struct reclaim_state;
  60 struct robust_list_head;
  61 struct root_domain;
  62 struct rq;
  63 struct sched_attr;
  64 struct sched_param;
  65 struct seq_file;
  66 struct sighand_struct;
  67 struct signal_struct;
  68 struct task_delay_info;
  69 struct task_group;
  70
  71 /*
  72  * Task state bitmask. NOTE! These bits are also
  73  * encoded in fs/proc/array.c: get_task_state().
  74  *
  75  * We have two separate sets of flags: task->state
  76  * is about runnability, while task->exit_state are
  77  * about the task exiting. Confusing, but this way
  78  * modifying one set can't modify the other one by
  79  * mistake.
  80  */
  81
  82 /* Used in tsk->state: */
  83 #define TASK_RUNNING                    0x0000
  84 #define TASK_INTERRUPTIBLE              0x0001
  85 #define TASK_UNINTERRUPTIBLE            0x0002
  86 #define __TASK_STOPPED                  0x0004
  87 #define __TASK_TRACED                   0x0008
  88 /* Used in tsk->exit_state: */
  89 #define EXIT_DEAD                       0x0010
  90 #define EXIT_ZOMBIE                     0x0020
  91 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
  92 /* Used in tsk->state again: */
  93 #define TASK_PARKED                     0x0040
  94 #define TASK_DEAD                       0x0080
  95 #define TASK_WAKEKILL                   0x0100
  96 #define TASK_WAKING                     0x0200
  97 #define TASK_NOLOAD                     0x0400
  98 #define TASK_NEW                        0x0800
  99 /* RT specific auxilliary flag to mark RT lock waiters */
 100 #define TASK_RTLOCK_WAIT                0x1000
 101 #define TASK_STATE_MAX                  0x2000
 102
 103 /* Convenience macros for the sake of set_current_state: */
 104 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 105 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 106 #define TASK_TRACED                     __TASK_TRACED
 107
 108 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 109
 110 /* Convenience macros for the sake of wake_up(): */
 111 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 112
 113 /* get_task_state(): */
 114 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 115                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 116                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 117                                          TASK_PARKED)
 118
 119 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 120
 121 #define task_is_traced(task)            ((READ_ONCE(task->jobctl) & JOBCTL_TRACED) != 0)
 122 #define task_is_stopped(task)           ((READ_ONCE(task->jobctl) & JOBCTL_STOPPED) != 0)
 123 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->jobctl) & (JOBCTL_STOPPED | JOBCTL_TRACED)) != 0)
 124
 125 /*
 126  * Special states are those that do not use the normal wait-loop pattern. See
 127  * the comment with set_special_state().
 128  */
 129 #define is_special_task_state(state)                            \
 130         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
 131
 132 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 133 # define debug_normal_state_change(state_value)                         \
 134         do {                                                            \
 135                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 136                 current->task_state_change = _THIS_IP_;                 \
 137         } while (0)
 138
 139 # define debug_special_state_change(state_value)                        \
 140         do {                                                            \
 141                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 142                 current->task_state_change = _THIS_IP_;                 \
 143         } while (0)
 144
 145 # define debug_rtlock_wait_set_state()                                  \
 146         do {                                                             \
 147                 current->saved_state_change = current->task_state_change;\
 148                 current->task_state_change = _THIS_IP_;                  \
 149         } while (0)
 150
 151 # define debug_rtlock_wait_restore_state()                              \
 152         do {                                                             \
 153                 current->task_state_change = current->saved_state_change;\
 154         } while (0)
 155
 156 #else
 157 # define debug_normal_state_change(cond)        do { } while (0)
 158 # define debug_special_state_change(cond)       do { } while (0)
 159 # define debug_rtlock_wait_set_state()          do { } while (0)
 160 # define debug_rtlock_wait_restore_state()      do { } while (0)
 161 #endif
 162
 163 /*
 164  * set_current_state() includes a barrier so that the write of current->state
 165  * is correctly serialised wrt the caller's subsequent test of whether to
 166  * actually sleep:
 167  *
 168  *   for (;;) {
 169  *      set_current_state(TASK_UNINTERRUPTIBLE);
 170  *      if (CONDITION)
 171  *         break;
 172  *
 173  *      schedule();
 174  *   }
 175  *   __set_current_state(TASK_RUNNING);
 176  *
 177  * If the caller does not need such serialisation (because, for instance, the
 178  * CONDITION test and condition change and wakeup are under the same lock) then
 179  * use __set_current_state().
 180  *
 181  * The above is typically ordered against the wakeup, which does:
 182  *
 183  *   CONDITION = 1;
 184  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 185  *
 186  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 187  * accessing p->state.
 188  *
 189  * Wakeup will do: if (@state & p->state) p->state = TASK_RUNNING, that is,
 190  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 191  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 192  *
 193  * However, with slightly different timing the wakeup TASK_RUNNING store can
 194  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 195  * a problem either because that will result in one extra go around the loop
 196  * and our @cond test will save the day.
 197  *
 198  * Also see the comments of try_to_wake_up().
 199  */
 200 #define __set_current_state(state_value)                                \
 201         do {                                                            \
 202                 debug_normal_state_change((state_value));               \
 203                 WRITE_ONCE(current->__state, (state_value));            \
 204         } while (0)
 205
 206 #define set_current_state(state_value)                                  \
 207         do {                                                            \
 208                 debug_normal_state_change((state_value));               \
 209                 smp_store_mb(current->__state, (state_value));          \
 210         } while (0)
 211
 212 /*
 213  * set_special_state() should be used for those states when the blocking task
 214  * can not use the regular condition based wait-loop. In that case we must
 215  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 216  * stores will not collide with our state change.
 217  */
 218 #define set_special_state(state_value)                                  \
 219         do {                                                            \
 220                 unsigned long flags; /* may shadow */                   \
 221                                                                         \
 222                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 223                 debug_special_state_change((state_value));              \
 224                 WRITE_ONCE(current->__state, (state_value));            \
 225                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 226         } while (0)
 227
 228 /*
 229  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 230  *
 231  * RT's spin/rwlock substitutions are state preserving. The state of the
 232  * task when blocking on the lock is saved in task_struct::saved_state and
 233  * restored after the lock has been acquired.  These operations are
 234  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 235  * lock related wakeups while the task is blocked on the lock are
 236  * redirected to operate on task_struct::saved_state to ensure that these
 237  * are not dropped. On restore task_struct::saved_state is set to
 238  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 239  *
 240  * The lock operation looks like this:
 241  *
 242  *      current_save_and_set_rtlock_wait_state();
 243  *      for (;;) {
 244  *              if (try_lock())
 245  *                      break;
 246  *              raw_spin_unlock_irq(&lock->wait_lock);
 247  *              schedule_rtlock();
 248  *              raw_spin_lock_irq(&lock->wait_lock);
 249  *              set_current_state(TASK_RTLOCK_WAIT);
 250  *      }
 251  *      current_restore_rtlock_saved_state();
 252  */
 253 #define current_save_and_set_rtlock_wait_state()                        \
 254         do {                                                            \
 255                 lockdep_assert_irqs_disabled();                         \
 256                 raw_spin_lock(&current->pi_lock);                       \
 257                 current->saved_state = current->__state;                \
 258                 debug_rtlock_wait_set_state();                          \
 259                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 260                 raw_spin_unlock(&current->pi_lock);                     \
 261         } while (0);
 262
 263 #define current_restore_rtlock_saved_state()                            \
 264         do {                                                            \
 265                 lockdep_assert_irqs_disabled();                         \
 266                 raw_spin_lock(&current->pi_lock);                       \
 267                 debug_rtlock_wait_restore_state();                      \
 268                 WRITE_ONCE(current->__state, current->saved_state);     \
 269                 current->saved_state = TASK_RUNNING;                    \
 270                 raw_spin_unlock(&current->pi_lock);                     \
 271         } while (0);
 272
 273 #define get_current_state()     READ_ONCE(current->__state)
 274
 275 /*
 276  * Define the task command name length as enum, then it can be visible to
 277  * BPF programs.
 278  */
 279 enum {
 280         TASK_COMM_LEN = 16,
 281 };
 282
 283 extern void scheduler_tick(void);
 284
 285 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 286
 287 extern long schedule_timeout(long timeout);
 288 extern long schedule_timeout_interruptible(long timeout);
 289 extern long schedule_timeout_killable(long timeout);
 290 extern long schedule_timeout_uninterruptible(long timeout);
 291 extern long schedule_timeout_idle(long timeout);
 292 asmlinkage void schedule(void);
 293 extern void schedule_preempt_disabled(void);
 294 asmlinkage void preempt_schedule_irq(void);
 295 #ifdef CONFIG_PREEMPT_RT
 296  extern void schedule_rtlock(void);
 297 #endif
 298
 299 extern int __must_check io_schedule_prepare(void);
 300 extern void io_schedule_finish(int token);
 301 extern long io_schedule_timeout(long timeout);
 302 extern void io_schedule(void);
 303
 304 /**
 305  * struct prev_cputime - snapshot of system and user cputime
 306  * @utime: time spent in user mode
 307  * @stime: time spent in system mode
 308  * @lock: protects the above two fields
 309  *
 310  * Stores previous user/system time values such that we can guarantee
 311  * monotonicity.
 312  */
 313 struct prev_cputime {
 314 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 315         u64                             utime;
 316         u64                             stime;
 317         raw_spinlock_t                  lock;
 318 #endif
 319 };
 320
 321 enum vtime_state {
 322         /* Task is sleeping or running in a CPU with VTIME inactive: */
 323         VTIME_INACTIVE = 0,
 324         /* Task is idle */
 325         VTIME_IDLE,
 326         /* Task runs in kernelspace in a CPU with VTIME active: */
 327         VTIME_SYS,
 328         /* Task runs in userspace in a CPU with VTIME active: */
 329         VTIME_USER,
 330         /* Task runs as guests in a CPU with VTIME active: */
 331         VTIME_GUEST,
 332 };
 333
 334 struct vtime {
 335         seqcount_t              seqcount;
 336         unsigned long long      starttime;
 337         enum vtime_state        state;
 338         unsigned int            cpu;
 339         u64                     utime;
 340         u64                     stime;
 341         u64                     gtime;
 342 };
 343
 344 /*
 345  * Utilization clamp constraints.
 346  * @UCLAMP_MIN: Minimum utilization
 347  * @UCLAMP_MAX: Maximum utilization
 348  * @UCLAMP_CNT: Utilization clamp constraints count
 349  */
 350 enum uclamp_id {
 351         UCLAMP_MIN = 0,
 352         UCLAMP_MAX,
 353         UCLAMP_CNT
 354 };
 355
 356 #ifdef CONFIG_SMP
 357 extern struct root_domain def_root_domain;
 358 extern struct mutex sched_domains_mutex;
 359 #endif
 360
 361 struct sched_info {
 362 #ifdef CONFIG_SCHED_INFO
 363         /* Cumulative counters: */
 364
 365         /* # of times we have run on this CPU: */
 366         unsigned long                   pcount;
 367
 368         /* Time spent waiting on a runqueue: */
 369         unsigned long long              run_delay;
 370
 371         /* Timestamps: */
 372
 373         /* When did we last run on a CPU? */
 374         unsigned long long              last_arrival;
 375
 376         /* When were we last queued to run? */
 377         unsigned long long              last_queued;
 378
 379 #endif /* CONFIG_SCHED_INFO */
 380 };
 381
 382 /*
 383  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 384  * has a few: load, load_avg, util_avg, freq, and capacity.
 385  *
 386  * We define a basic fixed point arithmetic range, and then formalize
 387  * all these metrics based on that basic range.
 388  */
 389 # define SCHED_FIXEDPOINT_SHIFT         10
 390 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 391
 392 /* Increase resolution of cpu_capacity calculations */
 393 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 394 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 395
 396 struct load_weight {
 397         unsigned long                   weight;
 398         u32                             inv_weight;
 399 };
 400
 401 /**
 402  * struct util_est - Estimation utilization of FAIR tasks
 403  * @enqueued: instantaneous estimated utilization of a task/cpu
 404  * @ewma:     the Exponential Weighted Moving Average (EWMA)
 405  *            utilization of a task
 406  *
 407  * Support data structure to track an Exponential Weighted Moving Average
 408  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
 409  * average each time a task completes an activation. Sample's weight is chosen
 410  * so that the EWMA will be relatively insensitive to transient changes to the
 411  * task's workload.
 412  *
 413  * The enqueued attribute has a slightly different meaning for tasks and cpus:
 414  * - task:   the task's util_avg at last task dequeue time
 415  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
 416  * Thus, the util_est.enqueued of a task represents the contribution on the
 417  * estimated utilization of the CPU where that task is currently enqueued.
 418  *
 419  * Only for tasks we track a moving average of the past instantaneous
 420  * estimated utilization. This allows to absorb sporadic drops in utilization
 421  * of an otherwise almost periodic task.
 422  *
 423  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 424  * updates. When a task is dequeued, its util_est should not be updated if its
 425  * util_avg has not been updated in the meantime.
 426  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
 427  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
 428  * for a task) it is safe to use MSB.
 429  */
 430 struct util_est {
 431         unsigned int                    enqueued;
 432         unsigned int                    ewma;
 433 #define UTIL_EST_WEIGHT_SHIFT           2
 434 #define UTIL_AVG_UNCHANGED              0x80000000
 435 } __attribute__((__aligned__(sizeof(u64))));
 436
 437 /*
 438  * The load/runnable/util_avg accumulates an infinite geometric series
 439  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 440  *
 441  * [load_avg definition]
 442  *
 443  *   load_avg = runnable% * scale_load_down(load)
 444  *
 445  * [runnable_avg definition]
 446  *
 447  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 448  *
 449  * [util_avg definition]
 450  *
 451  *   util_avg = running% * SCHED_CAPACITY_SCALE
 452  *
 453  * where runnable% is the time ratio that a sched_entity is runnable and
 454  * running% the time ratio that a sched_entity is running.
 455  *
 456  * For cfs_rq, they are the aggregated values of all runnable and blocked
 457  * sched_entities.
 458  *
 459  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 460  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 461  * for computing those signals (see update_rq_clock_pelt())
 462  *
 463  * N.B., the above ratios (runnable% and running%) themselves are in the
 464  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 465  * to as large a range as necessary. This is for example reflected by
 466  * util_avg's SCHED_CAPACITY_SCALE.
 467  *
 468  * [Overflow issue]
 469  *
 470  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 471  * with the highest load (=88761), always runnable on a single cfs_rq,
 472  * and should not overflow as the number already hits PID_MAX_LIMIT.
 473  *
 474  * For all other cases (including 32-bit kernels), struct load_weight's
 475  * weight will overflow first before we do, because:
 476  *
 477  *    Max(load_avg) <= Max(load.weight)
 478  *
 479  * Then it is the load_weight's responsibility to consider overflow
 480  * issues.
 481  */
 482 struct sched_avg {
 483         u64                             last_update_time;
 484         u64                             load_sum;
 485         u64                             runnable_sum;
 486         u32                             util_sum;
 487         u32                             period_contrib;
 488         unsigned long                   load_avg;
 489         unsigned long                   runnable_avg;
 490         unsigned long                   util_avg;
 491         struct util_est                 util_est;
 492 } ____cacheline_aligned;
 493
 494 struct sched_statistics {
 495 #ifdef CONFIG_SCHEDSTATS
 496         u64                             wait_start;
 497         u64                             wait_max;
 498         u64                             wait_count;
 499         u64                             wait_sum;
 500         u64                             iowait_count;
 501         u64                             iowait_sum;
 502
 503         u64                             sleep_start;
 504         u64                             sleep_max;
 505         s64                             sum_sleep_runtime;
 506
 507         u64                             block_start;
 508         u64                             block_max;
 509         s64                             sum_block_runtime;
 510
 511         u64                             exec_max;
 512         u64                             slice_max;
 513
 514         u64                             nr_migrations_cold;
 515         u64                             nr_failed_migrations_affine;
 516         u64                             nr_failed_migrations_running;
 517         u64                             nr_failed_migrations_hot;
 518         u64                             nr_forced_migrations;
 519
 520         u64                             nr_wakeups;
 521         u64                             nr_wakeups_sync;
 522         u64                             nr_wakeups_migrate;
 523         u64                             nr_wakeups_local;
 524         u64                             nr_wakeups_remote;
 525         u64                             nr_wakeups_affine;
 526         u64                             nr_wakeups_affine_attempts;
 527         u64                             nr_wakeups_passive;
 528         u64                             nr_wakeups_idle;
 529
 530 #ifdef CONFIG_SCHED_CORE
 531         u64                             core_forceidle_sum;
 532 #endif
 533 #endif /* CONFIG_SCHEDSTATS */
 534 } ____cacheline_aligned;
 535
 536 struct sched_entity {
 537         /* For load-balancing: */
 538         struct load_weight              load;
 539         struct rb_node                  run_node;
 540         struct list_head                group_node;
 541         unsigned int                    on_rq;
 542
 543         u64                             exec_start;
 544         u64                             sum_exec_runtime;
 545         u64                             vruntime;
 546         u64                             prev_sum_exec_runtime;
 547
 548         u64                             nr_migrations;
 549
 550 #ifdef CONFIG_FAIR_GROUP_SCHED
 551         int                             depth;
 552         struct sched_entity             *parent;
 553         /* rq on which this entity is (to be) queued: */
 554         struct cfs_rq                   *cfs_rq;
 555         /* rq "owned" by this entity/group: */
 556         struct cfs_rq                   *my_q;
 557         /* cached value of my_q->h_nr_running */
 558         unsigned long                   runnable_weight;
 559 #endif
 560
 561 #ifdef CONFIG_SMP
 562         /*
 563          * Per entity load average tracking.
 564          *
 565          * Put into separate cache line so it does not
 566          * collide with read-mostly values above.
 567          */
 568         struct sched_avg                avg;
 569 #endif
 570 };
 571
 572 struct sched_rt_entity {
 573         struct list_head                run_list;
 574         unsigned long                   timeout;
 575         unsigned long                   watchdog_stamp;
 576         unsigned int                    time_slice;
 577         unsigned short                  on_rq;
 578         unsigned short                  on_list;
 579
 580         struct sched_rt_entity          *back;
 581 #ifdef CONFIG_RT_GROUP_SCHED
 582         struct sched_rt_entity          *parent;
 583         /* rq on which this entity is (to be) queued: */
 584         struct rt_rq                    *rt_rq;
 585         /* rq "owned" by this entity/group: */
 586         struct rt_rq                    *my_q;
 587 #endif
 588 } __randomize_layout;
 589
 590 struct sched_dl_entity {
 591         struct rb_node                  rb_node;
 592
 593         /*
 594          * Original scheduling parameters. Copied here from sched_attr
 595          * during sched_setattr(), they will remain the same until
 596          * the next sched_setattr().
 597          */
 598         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 599         u64                             dl_deadline;    /* Relative deadline of each instance   */
 600         u64                             dl_period;      /* Separation of two instances (period) */
 601         u64                             dl_bw;          /* dl_runtime / dl_period               */
 602         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 603
 604         /*
 605          * Actual scheduling parameters. Initialized with the values above,
 606          * they are continuously updated during task execution. Note that
 607          * the remaining runtime could be < 0 in case we are in overrun.
 608          */
 609         s64                             runtime;        /* Remaining runtime for this instance  */
 610         u64                             deadline;       /* Absolute deadline for this instance  */
 611         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 612
 613         /*
 614          * Some bool flags:
 615          *
 616          * @dl_throttled tells if we exhausted the runtime. If so, the
 617          * task has to wait for a replenishment to be performed at the
 618          * next firing of dl_timer.
 619          *
 620          * @dl_yielded tells if task gave up the CPU before consuming
 621          * all its available runtime during the last job.
 622          *
 623          * @dl_non_contending tells if the task is inactive while still
 624          * contributing to the active utilization. In other words, it
 625          * indicates if the inactive timer has been armed and its handler
 626          * has not been executed yet. This flag is useful to avoid race
 627          * conditions between the inactive timer handler and the wakeup
 628          * code.
 629          *
 630          * @dl_overrun tells if the task asked to be informed about runtime
 631          * overruns.
 632          */
 633         unsigned int                    dl_throttled      : 1;
 634         unsigned int                    dl_yielded        : 1;
 635         unsigned int                    dl_non_contending : 1;
 636         unsigned int                    dl_overrun        : 1;
 637
 638         /*
 639          * Bandwidth enforcement timer. Each -deadline task has its
 640          * own bandwidth to be enforced, thus we need one timer per task.
 641          */
 642         struct hrtimer                  dl_timer;
 643
 644         /*
 645          * Inactive timer, responsible for decreasing the active utilization
 646          * at the "0-lag time". When a -deadline task blocks, it contributes
 647          * to GRUB's active utilization until the "0-lag time", hence a
 648          * timer is needed to decrease the active utilization at the correct
 649          * time.
 650          */
 651         struct hrtimer inactive_timer;
 652
 653 #ifdef CONFIG_RT_MUTEXES
 654         /*
 655          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 656          * pi_se points to the donor, otherwise points to the dl_se it belongs
 657          * to (the original one/itself).
 658          */
 659         struct sched_dl_entity *pi_se;
 660 #endif
 661 };
 662
 663 #ifdef CONFIG_UCLAMP_TASK
 664 /* Number of utilization clamp buckets (shorter alias) */
 665 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 666
 667 /*
 668  * Utilization clamp for a scheduling entity
 669  * @value:              clamp value "assigned" to a se
 670  * @bucket_id:          bucket index corresponding to the "assigned" value
 671  * @active:             the se is currently refcounted in a rq's bucket
 672  * @user_defined:       the requested clamp value comes from user-space
 673  *
 674  * The bucket_id is the index of the clamp bucket matching the clamp value
 675  * which is pre-computed and stored to avoid expensive integer divisions from
 676  * the fast path.
 677  *
 678  * The active bit is set whenever a task has got an "effective" value assigned,
 679  * which can be different from the clamp value "requested" from user-space.
 680  * This allows to know a task is refcounted in the rq's bucket corresponding
 681  * to the "effective" bucket_id.
 682  *
 683  * The user_defined bit is set whenever a task has got a task-specific clamp
 684  * value requested from userspace, i.e. the system defaults apply to this task
 685  * just as a restriction. This allows to relax default clamps when a less
 686  * restrictive task-specific value has been requested, thus allowing to
 687  * implement a "nice" semantic. For example, a task running with a 20%
 688  * default boost can still drop its own boosting to 0%.
 689  */
 690 struct uclamp_se {
 691         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 692         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 693         unsigned int active             : 1;
 694         unsigned int user_defined       : 1;
 695 };
 696 #endif /* CONFIG_UCLAMP_TASK */
 697
 698 union rcu_special {
 699         struct {
 700                 u8                      blocked;
 701                 u8                      need_qs;
 702                 u8                      exp_hint; /* Hint for performance. */
 703                 u8                      need_mb; /* Readers need smp_mb(). */
 704         } b; /* Bits. */
 705         u32 s; /* Set of bits. */
 706 };
 707
 708 enum perf_event_task_context {
 709         perf_invalid_context = -1,
 710         perf_hw_context = 0,
 711         perf_sw_context,
 712         perf_nr_task_contexts,
 713 };
 714
 715 struct wake_q_node {
 716         struct wake_q_node *next;
 717 };
 718
 719 struct kmap_ctrl {
 720 #ifdef CONFIG_KMAP_LOCAL
 721         int                             idx;
 722         pte_t                           pteval[KM_MAX_IDX];
 723 #endif
 724 };
 725
 726 struct task_struct {
 727 #ifdef CONFIG_THREAD_INFO_IN_TASK
 728         /*
 729          * For reasons of header soup (see current_thread_info()), this
 730          * must be the first element of task_struct.
 731          */
 732         struct thread_info              thread_info;
 733 #endif
 734         unsigned int                    __state;
 735
 736 #ifdef CONFIG_PREEMPT_RT
 737         /* saved state for "spinlock sleepers" */
 738         unsigned int                    saved_state;
 739 #endif
 740
 741         /*
 742          * This begins the randomizable portion of task_struct. Only
 743          * scheduling-critical items should be added above here.
 744          */
 745         randomized_struct_fields_start
 746
 747         void                            *stack;
 748         refcount_t                      usage;
 749         /* Per task flags (PF_*), defined further below: */
 750         unsigned int                    flags;
 751         unsigned int                    ptrace;
 752
 753 #ifdef CONFIG_SMP
 754         int                             on_cpu;
 755         struct __call_single_node       wake_entry;
 756         unsigned int                    wakee_flips;
 757         unsigned long                   wakee_flip_decay_ts;
 758         struct task_struct              *last_wakee;
 759
 760         /*
 761          * recent_used_cpu is initially set as the last CPU used by a task
 762          * that wakes affine another task. Waker/wakee relationships can
 763          * push tasks around a CPU where each wakeup moves to the next one.
 764          * Tracking a recently used CPU allows a quick search for a recently
 765          * used CPU that may be idle.
 766          */
 767         int                             recent_used_cpu;
 768         int                             wake_cpu;
 769 #endif
 770         int                             on_rq;
 771
 772         int                             prio;
 773         int                             static_prio;
 774         int                             normal_prio;
 775         unsigned int                    rt_priority;
 776
 777         struct sched_entity             se;
 778         struct sched_rt_entity          rt;
 779         struct sched_dl_entity          dl;
 780         const struct sched_class        *sched_class;
 781
 782 #ifdef CONFIG_SCHED_CORE
 783         struct rb_node                  core_node;
 784         unsigned long                   core_cookie;
 785         unsigned int                    core_occupation;
 786 #endif
 787
 788 #ifdef CONFIG_CGROUP_SCHED
 789         struct task_group               *sched_task_group;
 790 #endif
 791
 792 #ifdef CONFIG_UCLAMP_TASK
 793         /*
 794          * Clamp values requested for a scheduling entity.
 795          * Must be updated with task_rq_lock() held.
 796          */
 797         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 798         /*
 799          * Effective clamp values used for a scheduling entity.
 800          * Must be updated with task_rq_lock() held.
 801          */
 802         struct uclamp_se                uclamp[UCLAMP_CNT];
 803 #endif
 804
 805         struct sched_statistics         stats;
 806
 807 #ifdef CONFIG_PREEMPT_NOTIFIERS
 808         /* List of struct preempt_notifier: */
 809         struct hlist_head               preempt_notifiers;
 810 #endif
 811
 812 #ifdef CONFIG_BLK_DEV_IO_TRACE
 813         unsigned int                    btrace_seq;
 814 #endif
 815
 816         unsigned int                    policy;
 817         int                             nr_cpus_allowed;
 818         const cpumask_t                 *cpus_ptr;
 819         cpumask_t                       *user_cpus_ptr;
 820         cpumask_t                       cpus_mask;
 821         void                            *migration_pending;
 822 #ifdef CONFIG_SMP
 823         unsigned short                  migration_disabled;
 824 #endif
 825         unsigned short                  migration_flags;
 826
 827 #ifdef CONFIG_PREEMPT_RCU
 828         int                             rcu_read_lock_nesting;
 829         union rcu_special               rcu_read_unlock_special;
 830         struct list_head                rcu_node_entry;
 831         struct rcu_node                 *rcu_blocked_node;
 832 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 833
 834 #ifdef CONFIG_TASKS_RCU
 835         unsigned long                   rcu_tasks_nvcsw;
 836         u8                              rcu_tasks_holdout;
 837         u8                              rcu_tasks_idx;
 838         int                             rcu_tasks_idle_cpu;
 839         struct list_head                rcu_tasks_holdout_list;
 840 #endif /* #ifdef CONFIG_TASKS_RCU */
 841
 842 #ifdef CONFIG_TASKS_TRACE_RCU
 843         int                             trc_reader_nesting;
 844         int                             trc_ipi_to_cpu;
 845         union rcu_special               trc_reader_special;
 846         struct list_head                trc_holdout_list;
 847         struct list_head                trc_blkd_node;
 848         int                             trc_blkd_cpu;
 849 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 850
 851         struct sched_info               sched_info;
 852
 853         struct list_head                tasks;
 854 #ifdef CONFIG_SMP
 855         struct plist_node               pushable_tasks;
 856         struct rb_node                  pushable_dl_tasks;
 857 #endif
 858
 859         struct mm_struct                *mm;
 860         struct mm_struct                *active_mm;
 861
 862         /* Per-thread vma caching: */
 863         struct vmacache                 vmacache;
 864
 865 #ifdef SPLIT_RSS_COUNTING
 866         struct task_rss_stat            rss_stat;
 867 #endif
 868         int                             exit_state;
 869         int                             exit_code;
 870         int                             exit_signal;
 871         /* The signal sent when the parent dies: */
 872         int                             pdeath_signal;
 873         /* JOBCTL_*, siglock protected: */
 874         unsigned long                   jobctl;
 875
 876         /* Used for emulating ABI behavior of previous Linux versions: */
 877         unsigned int                    personality;
 878
 879         /* Scheduler bits, serialized by scheduler locks: */
 880         unsigned                        sched_reset_on_fork:1;
 881         unsigned                        sched_contributes_to_load:1;
 882         unsigned                        sched_migrated:1;
 883 #ifdef CONFIG_PSI
 884         unsigned                        sched_psi_wake_requeue:1;
 885 #endif
 886
 887         /* Force alignment to the next boundary: */
 888         unsigned                        :0;
 889
 890         /* Unserialized, strictly 'current' */
 891
 892         /*
 893          * This field must not be in the scheduler word above due to wakelist
 894          * queueing no longer being serialized by p->on_cpu. However:
 895          *
 896          * p->XXX = X;                  ttwu()
 897          * schedule()                     if (p->on_rq && ..) // false
 898          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 899          *   deactivate_task()                ttwu_queue_wakelist())
 900          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 901          *
 902          * guarantees all stores of 'current' are visible before
 903          * ->sched_remote_wakeup gets used, so it can be in this word.
 904          */
 905         unsigned                        sched_remote_wakeup:1;
 906
 907         /* Bit to tell LSMs we're in execve(): */
 908         unsigned                        in_execve:1;
 909         unsigned                        in_iowait:1;
 910 #ifndef TIF_RESTORE_SIGMASK
 911         unsigned                        restore_sigmask:1;
 912 #endif
 913 #ifdef CONFIG_MEMCG
 914         unsigned                        in_user_fault:1;
 915 #endif
 916 #ifdef CONFIG_COMPAT_BRK
 917         unsigned                        brk_randomized:1;
 918 #endif
 919 #ifdef CONFIG_CGROUPS
 920         /* disallow userland-initiated cgroup migration */
 921         unsigned                        no_cgroup_migration:1;
 922         /* task is frozen/stopped (used by the cgroup freezer) */
 923         unsigned                        frozen:1;
 924 #endif
 925 #ifdef CONFIG_BLK_CGROUP
 926         unsigned                        use_memdelay:1;
 927 #endif
 928 #ifdef CONFIG_PSI
 929         /* Stalled due to lack of memory */
 930         unsigned                        in_memstall:1;
 931 #endif
 932 #ifdef CONFIG_PAGE_OWNER
 933         /* Used by page_owner=on to detect recursion in page tracking. */
 934         unsigned                        in_page_owner:1;
 935 #endif
 936 #ifdef CONFIG_EVENTFD
 937         /* Recursion prevention for eventfd_signal() */
 938         unsigned                        in_eventfd_signal:1;
 939 #endif
 940 #ifdef CONFIG_IOMMU_SVA
 941         unsigned                        pasid_activated:1;
 942 #endif
 943 #ifdef  CONFIG_CPU_SUP_INTEL
 944         unsigned                        reported_split_lock:1;
 945 #endif
 946
 947         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 948
 949         struct restart_block            restart_block;
 950
 951         pid_t                           pid;
 952         pid_t                           tgid;
 953
 954 #ifdef CONFIG_STACKPROTECTOR
 955         /* Canary value for the -fstack-protector GCC feature: */
 956         unsigned long                   stack_canary;
 957 #endif
 958         /*
 959          * Pointers to the (original) parent process, youngest child, younger sibling,
 960          * older sibling, respectively.  (p->father can be replaced with
 961          * p->real_parent->pid)
 962          */
 963
 964         /* Real parent process: */
 965         struct task_struct __rcu        *real_parent;
 966
 967         /* Recipient of SIGCHLD, wait4() reports: */
 968         struct task_struct __rcu        *parent;
 969
 970         /*
 971          * Children/sibling form the list of natural children:
 972          */
 973         struct list_head                children;
 974         struct list_head                sibling;
 975         struct task_struct              *group_leader;
 976
 977         /*
 978          * 'ptraced' is the list of tasks this task is using ptrace() on.
 979          *
 980          * This includes both natural children and PTRACE_ATTACH targets.
 981          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 982          */
 983         struct list_head                ptraced;
 984         struct list_head                ptrace_entry;
 985
 986         /* PID/PID hash table linkage. */
 987         struct pid                      *thread_pid;
 988         struct hlist_node               pid_links[PIDTYPE_MAX];
 989         struct list_head                thread_group;
 990         struct list_head                thread_node;
 991
 992         struct completion               *vfork_done;
 993
 994         /* CLONE_CHILD_SETTID: */
 995         int __user                      *set_child_tid;
 996
 997         /* CLONE_CHILD_CLEARTID: */
 998         int __user                      *clear_child_tid;
 999
1000         /* PF_KTHREAD | PF_IO_WORKER */
1001         void                            *worker_private;
1002
1003         u64                             utime;
1004         u64                             stime;
1005 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1006         u64                             utimescaled;
1007         u64                             stimescaled;
1008 #endif
1009         u64                             gtime;
1010         struct prev_cputime             prev_cputime;
1011 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1012         struct vtime                    vtime;
1013 #endif
1014
1015 #ifdef CONFIG_NO_HZ_FULL
1016         atomic_t                        tick_dep_mask;
1017 #endif
1018         /* Context switch counts: */
1019         unsigned long                   nvcsw;
1020         unsigned long                   nivcsw;
1021
1022         /* Monotonic time in nsecs: */
1023         u64                             start_time;
1024
1025         /* Boot based time in nsecs: */
1026         u64                             start_boottime;
1027
1028         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1029         unsigned long                   min_flt;
1030         unsigned long                   maj_flt;
1031
1032         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1033         struct posix_cputimers          posix_cputimers;
1034
1035 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1036         struct posix_cputimers_work     posix_cputimers_work;
1037 #endif
1038
1039         /* Process credentials: */
1040
1041         /* Tracer's credentials at attach: */
1042         const struct cred __rcu         *ptracer_cred;
1043
1044         /* Objective and real subjective task credentials (COW): */
1045         const struct cred __rcu         *real_cred;
1046
1047         /* Effective (overridable) subjective task credentials (COW): */
1048         const struct cred __rcu         *cred;
1049
1050 #ifdef CONFIG_KEYS
1051         /* Cached requested key. */
1052         struct key                      *cached_requested_key;
1053 #endif
1054
1055         /*
1056          * executable name, excluding path.
1057          *
1058          * - normally initialized setup_new_exec()
1059          * - access it with [gs]et_task_comm()
1060          * - lock it with task_lock()
1061          */
1062         char                            comm[TASK_COMM_LEN];
1063
1064         struct nameidata                *nameidata;
1065
1066 #ifdef CONFIG_SYSVIPC
1067         struct sysv_sem                 sysvsem;
1068         struct sysv_shm                 sysvshm;
1069 #endif
1070 #ifdef CONFIG_DETECT_HUNG_TASK
1071         unsigned long                   last_switch_count;
1072         unsigned long                   last_switch_time;
1073 #endif
1074         /* Filesystem information: */
1075         struct fs_struct                *fs;
1076
1077         /* Open file information: */
1078         struct files_struct             *files;
1079
1080 #ifdef CONFIG_IO_URING
1081         struct io_uring_task            *io_uring;
1082 #endif
1083
1084         /* Namespaces: */
1085         struct nsproxy                  *nsproxy;
1086
1087         /* Signal handlers: */
1088         struct signal_struct            *signal;
1089         struct sighand_struct __rcu             *sighand;
1090         sigset_t                        blocked;
1091         sigset_t                        real_blocked;
1092         /* Restored if set_restore_sigmask() was used: */
1093         sigset_t                        saved_sigmask;
1094         struct sigpending               pending;
1095         unsigned long                   sas_ss_sp;
1096         size_t                          sas_ss_size;
1097         unsigned int                    sas_ss_flags;
1098
1099         struct callback_head            *task_works;
1100
1101 #ifdef CONFIG_AUDIT
1102 #ifdef CONFIG_AUDITSYSCALL
1103         struct audit_context            *audit_context;
1104 #endif
1105         kuid_t                          loginuid;
1106         unsigned int                    sessionid;
1107 #endif
1108         struct seccomp                  seccomp;
1109         struct syscall_user_dispatch    syscall_dispatch;
1110
1111         /* Thread group tracking: */
1112         u64                             parent_exec_id;
1113         u64                             self_exec_id;
1114
1115         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1116         spinlock_t                      alloc_lock;
1117
1118         /* Protection of the PI data structures: */
1119         raw_spinlock_t                  pi_lock;
1120
1121         struct wake_q_node              wake_q;
1122
1123 #ifdef CONFIG_RT_MUTEXES
1124         /* PI waiters blocked on a rt_mutex held by this task: */
1125         struct rb_root_cached           pi_waiters;
1126         /* Updated under owner's pi_lock and rq lock */
1127         struct task_struct              *pi_top_task;
1128         /* Deadlock detection and priority inheritance handling: */
1129         struct rt_mutex_waiter          *pi_blocked_on;
1130 #endif
1131
1132 #ifdef CONFIG_DEBUG_MUTEXES
1133         /* Mutex deadlock detection: */
1134         struct mutex_waiter             *blocked_on;
1135 #endif
1136
1137 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1138         int                             non_block_count;
1139 #endif
1140
1141 #ifdef CONFIG_TRACE_IRQFLAGS
1142         struct irqtrace_events          irqtrace;
1143         unsigned int                    hardirq_threaded;
1144         u64                             hardirq_chain_key;
1145         int                             softirqs_enabled;
1146         int                             softirq_context;
1147         int                             irq_config;
1148 #endif
1149 #ifdef CONFIG_PREEMPT_RT
1150         int                             softirq_disable_cnt;
1151 #endif
1152
1153 #ifdef CONFIG_LOCKDEP
1154 # define MAX_LOCK_DEPTH                 48UL
1155         u64                             curr_chain_key;
1156         int                             lockdep_depth;
1157         unsigned int                    lockdep_recursion;
1158         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1159 #endif
1160
1161 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1162         unsigned int                    in_ubsan;
1163 #endif
1164
1165         /* Journalling filesystem info: */
1166         void                            *journal_info;
1167
1168         /* Stacked block device info: */
1169         struct bio_list                 *bio_list;
1170
1171         /* Stack plugging: */
1172         struct blk_plug                 *plug;
1173
1174         /* VM state: */
1175         struct reclaim_state            *reclaim_state;
1176
1177         struct backing_dev_info         *backing_dev_info;
1178
1179         struct io_context               *io_context;
1180
1181 #ifdef CONFIG_COMPACTION
1182         struct capture_control          *capture_control;
1183 #endif
1184         /* Ptrace state: */
1185         unsigned long                   ptrace_message;
1186         kernel_siginfo_t                *last_siginfo;
1187
1188         struct task_io_accounting       ioac;
1189 #ifdef CONFIG_PSI
1190         /* Pressure stall state */
1191         unsigned int                    psi_flags;
1192 #endif
1193 #ifdef CONFIG_TASK_XACCT
1194         /* Accumulated RSS usage: */
1195         u64                             acct_rss_mem1;
1196         /* Accumulated virtual memory usage: */
1197         u64                             acct_vm_mem1;
1198         /* stime + utime since last update: */
1199         u64                             acct_timexpd;
1200 #endif
1201 #ifdef CONFIG_CPUSETS
1202         /* Protected by ->alloc_lock: */
1203         nodemask_t                      mems_allowed;
1204         /* Sequence number to catch updates: */
1205         seqcount_spinlock_t             mems_allowed_seq;
1206         int                             cpuset_mem_spread_rotor;
1207         int                             cpuset_slab_spread_rotor;
1208 #endif
1209 #ifdef CONFIG_CGROUPS
1210         /* Control Group info protected by css_set_lock: */
1211         struct css_set __rcu            *cgroups;
1212         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1213         struct list_head                cg_list;
1214 #endif
1215 #ifdef CONFIG_X86_CPU_RESCTRL
1216         u32                             closid;
1217         u32                             rmid;
1218 #endif
1219 #ifdef CONFIG_FUTEX
1220         struct robust_list_head __user  *robust_list;
1221 #ifdef CONFIG_COMPAT
1222         struct compat_robust_list_head __user *compat_robust_list;
1223 #endif
1224         struct list_head                pi_state_list;
1225         struct futex_pi_state           *pi_state_cache;
1226         struct mutex                    futex_exit_mutex;
1227         unsigned int                    futex_state;
1228 #endif
1229 #ifdef CONFIG_PERF_EVENTS
1230         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1231         struct mutex                    perf_event_mutex;
1232         struct list_head                perf_event_list;
1233 #endif
1234 #ifdef CONFIG_DEBUG_PREEMPT
1235         unsigned long                   preempt_disable_ip;
1236 #endif
1237 #ifdef CONFIG_NUMA
1238         /* Protected by alloc_lock: */
1239         struct mempolicy                *mempolicy;
1240         short                           il_prev;
1241         short                           pref_node_fork;
1242 #endif
1243 #ifdef CONFIG_NUMA_BALANCING
1244         int                             numa_scan_seq;
1245         unsigned int                    numa_scan_period;
1246         unsigned int                    numa_scan_period_max;
1247         int                             numa_preferred_nid;
1248         unsigned long                   numa_migrate_retry;
1249         /* Migration stamp: */
1250         u64                             node_stamp;
1251         u64                             last_task_numa_placement;
1252         u64                             last_sum_exec_runtime;
1253         struct callback_head            numa_work;
1254
1255         /*
1256          * This pointer is only modified for current in syscall and
1257          * pagefault context (and for tasks being destroyed), so it can be read
1258          * from any of the following contexts:
1259          *  - RCU read-side critical section
1260          *  - current->numa_group from everywhere
1261          *  - task's runqueue locked, task not running
1262          */
1263         struct numa_group __rcu         *numa_group;
1264
1265         /*
1266          * numa_faults is an array split into four regions:
1267          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1268          * in this precise order.
1269          *
1270          * faults_memory: Exponential decaying average of faults on a per-node
1271          * basis. Scheduling placement decisions are made based on these
1272          * counts. The values remain static for the duration of a PTE scan.
1273          * faults_cpu: Track the nodes the process was running on when a NUMA
1274          * hinting fault was incurred.
1275          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1276          * during the current scan window. When the scan completes, the counts
1277          * in faults_memory and faults_cpu decay and these values are copied.
1278          */
1279         unsigned long                   *numa_faults;
1280         unsigned long                   total_numa_faults;
1281
1282         /*
1283          * numa_faults_locality tracks if faults recorded during the last
1284          * scan window were remote/local or failed to migrate. The task scan
1285          * period is adapted based on the locality of the faults with different
1286          * weights depending on whether they were shared or private faults
1287          */
1288         unsigned long                   numa_faults_locality[3];
1289
1290         unsigned long                   numa_pages_migrated;
1291 #endif /* CONFIG_NUMA_BALANCING */
1292
1293 #ifdef CONFIG_RSEQ
1294         struct rseq __user *rseq;
1295         u32 rseq_sig;
1296         /*
1297          * RmW on rseq_event_mask must be performed atomically
1298          * with respect to preemption.
1299          */
1300         unsigned long rseq_event_mask;
1301 #endif
1302
1303         struct tlbflush_unmap_batch     tlb_ubc;
1304
1305         union {
1306                 refcount_t              rcu_users;
1307                 struct rcu_head         rcu;
1308         };
1309
1310         /* Cache last used pipe for splice(): */
1311         struct pipe_inode_info          *splice_pipe;
1312
1313         struct page_frag                task_frag;
1314
1315 #ifdef CONFIG_TASK_DELAY_ACCT
1316         struct task_delay_info          *delays;
1317 #endif
1318
1319 #ifdef CONFIG_FAULT_INJECTION
1320         int                             make_it_fail;
1321         unsigned int                    fail_nth;
1322 #endif
1323         /*
1324          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1325          * balance_dirty_pages() for a dirty throttling pause:
1326          */
1327         int                             nr_dirtied;
1328         int                             nr_dirtied_pause;
1329         /* Start of a write-and-pause period: */
1330         unsigned long                   dirty_paused_when;
1331
1332 #ifdef CONFIG_LATENCYTOP
1333         int                             latency_record_count;
1334         struct latency_record           latency_record[LT_SAVECOUNT];
1335 #endif
1336         /*
1337          * Time slack values; these are used to round up poll() and
1338          * select() etc timeout values. These are in nanoseconds.
1339          */
1340         u64                             timer_slack_ns;
1341         u64                             default_timer_slack_ns;
1342
1343 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1344         unsigned int                    kasan_depth;
1345 #endif
1346
1347 #ifdef CONFIG_KCSAN
1348         struct kcsan_ctx                kcsan_ctx;
1349 #ifdef CONFIG_TRACE_IRQFLAGS
1350         struct irqtrace_events          kcsan_save_irqtrace;
1351 #endif
1352 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1353         int                             kcsan_stack_depth;
1354 #endif
1355 #endif
1356
1357 #if IS_ENABLED(CONFIG_KUNIT)
1358         struct kunit                    *kunit_test;
1359 #endif
1360
1361 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1362         /* Index of current stored address in ret_stack: */
1363         int                             curr_ret_stack;
1364         int                             curr_ret_depth;
1365
1366         /* Stack of return addresses for return function tracing: */
1367         struct ftrace_ret_stack         *ret_stack;
1368
1369         /* Timestamp for last schedule: */
1370         unsigned long long              ftrace_timestamp;
1371
1372         /*
1373          * Number of functions that haven't been traced
1374          * because of depth overrun:
1375          */
1376         atomic_t                        trace_overrun;
1377
1378         /* Pause tracing: */
1379         atomic_t                        tracing_graph_pause;
1380 #endif
1381
1382 #ifdef CONFIG_TRACING
1383         /* State flags for use by tracers: */
1384         unsigned long                   trace;
1385
1386         /* Bitmask and counter of trace recursion: */
1387         unsigned long                   trace_recursion;
1388 #endif /* CONFIG_TRACING */
1389
1390 #ifdef CONFIG_KCOV
1391         /* See kernel/kcov.c for more details. */
1392
1393         /* Coverage collection mode enabled for this task (0 if disabled): */
1394         unsigned int                    kcov_mode;
1395
1396         /* Size of the kcov_area: */
1397         unsigned int                    kcov_size;
1398
1399         /* Buffer for coverage collection: */
1400         void                            *kcov_area;
1401
1402         /* KCOV descriptor wired with this task or NULL: */
1403         struct kcov                     *kcov;
1404
1405         /* KCOV common handle for remote coverage collection: */
1406         u64                             kcov_handle;
1407
1408         /* KCOV sequence number: */
1409         int                             kcov_sequence;
1410
1411         /* Collect coverage from softirq context: */
1412         unsigned int                    kcov_softirq;
1413 #endif
1414
1415 #ifdef CONFIG_MEMCG
1416         struct mem_cgroup               *memcg_in_oom;
1417         gfp_t                           memcg_oom_gfp_mask;
1418         int                             memcg_oom_order;
1419
1420         /* Number of pages to reclaim on returning to userland: */
1421         unsigned int                    memcg_nr_pages_over_high;
1422
1423         /* Used by memcontrol for targeted memcg charge: */
1424         struct mem_cgroup               *active_memcg;
1425 #endif
1426
1427 #ifdef CONFIG_BLK_CGROUP
1428         struct request_queue            *throttle_queue;
1429 #endif
1430
1431 #ifdef CONFIG_UPROBES
1432         struct uprobe_task              *utask;
1433 #endif
1434 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1435         unsigned int                    sequential_io;
1436         unsigned int                    sequential_io_avg;
1437 #endif
1438         struct kmap_ctrl                kmap_ctrl;
1439 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1440         unsigned long                   task_state_change;
1441 # ifdef CONFIG_PREEMPT_RT
1442         unsigned long                   saved_state_change;
1443 # endif
1444 #endif
1445         int                             pagefault_disabled;
1446 #ifdef CONFIG_MMU
1447         struct task_struct              *oom_reaper_list;
1448         struct timer_list               oom_reaper_timer;
1449 #endif
1450 #ifdef CONFIG_VMAP_STACK
1451         struct vm_struct                *stack_vm_area;
1452 #endif
1453 #ifdef CONFIG_THREAD_INFO_IN_TASK
1454         /* A live task holds one reference: */
1455         refcount_t                      stack_refcount;
1456 #endif
1457 #ifdef CONFIG_LIVEPATCH
1458         int patch_state;
1459 #endif
1460 #ifdef CONFIG_SECURITY
1461         /* Used by LSM modules for access restriction: */
1462         void                            *security;
1463 #endif
1464 #ifdef CONFIG_BPF_SYSCALL
1465         /* Used by BPF task local storage */
1466         struct bpf_local_storage __rcu  *bpf_storage;
1467         /* Used for BPF run context */
1468         struct bpf_run_ctx              *bpf_ctx;
1469 #endif
1470
1471 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1472         unsigned long                   lowest_stack;
1473         unsigned long                   prev_lowest_stack;
1474 #endif
1475
1476 #ifdef CONFIG_X86_MCE
1477         void __user                     *mce_vaddr;
1478         __u64                           mce_kflags;
1479         u64                             mce_addr;
1480         __u64                           mce_ripv : 1,
1481                                         mce_whole_page : 1,
1482                                         __mce_reserved : 62;
1483         struct callback_head            mce_kill_me;
1484         int                             mce_count;
1485 #endif
1486
1487 #ifdef CONFIG_KRETPROBES
1488         struct llist_head               kretprobe_instances;
1489 #endif
1490 #ifdef CONFIG_RETHOOK
1491         struct llist_head               rethooks;
1492 #endif
1493
1494 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1495         /*
1496          * If L1D flush is supported on mm context switch
1497          * then we use this callback head to queue kill work
1498          * to kill tasks that are not running on SMT disabled
1499          * cores
1500          */
1501         struct callback_head            l1d_flush_kill;
1502 #endif
1503
1504         /*
1505          * New fields for task_struct should be added above here, so that
1506          * they are included in the randomized portion of task_struct.
1507          */
1508         randomized_struct_fields_end
1509
1510         /* CPU-specific state of this task: */
1511         struct thread_struct            thread;
1512
1513         /*
1514          * WARNING: on x86, 'thread_struct' contains a variable-sized
1515          * structure.  It *MUST* be at the end of 'task_struct'.
1516          *
1517          * Do not put anything below here!
1518          */
1519 };
1520
1521 static inline struct pid *task_pid(struct task_struct *task)
1522 {
1523         return task->thread_pid;
1524 }
1525
1526 /*
1527  * the helpers to get the task's different pids as they are seen
1528  * from various namespaces
1529  *
1530  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1531  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1532  *                     current.
1533  * task_xid_nr_ns()  : id seen from the ns specified;
1534  *
1535  * see also pid_nr() etc in include/linux/pid.h
1536  */
1537 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1538
1539 static inline pid_t task_pid_nr(struct task_struct *tsk)
1540 {
1541         return tsk->pid;
1542 }
1543
1544 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1545 {
1546         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1547 }
1548
1549 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1550 {
1551         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1552 }
1553
1554
1555 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1556 {
1557         return tsk->tgid;
1558 }
1559
1560 /**
1561  * pid_alive - check that a task structure is not stale
1562  * @p: Task structure to be checked.
1563  *
1564  * Test if a process is not yet dead (at most zombie state)
1565  * If pid_alive fails, then pointers within the task structure
1566  * can be stale and must not be dereferenced.
1567  *
1568  * Return: 1 if the process is alive. 0 otherwise.
1569  */
1570 static inline int pid_alive(const struct task_struct *p)
1571 {
1572         return p->thread_pid != NULL;
1573 }
1574
1575 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1576 {
1577         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1578 }
1579
1580 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1581 {
1582         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1583 }
1584
1585
1586 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1587 {
1588         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1589 }
1590
1591 static inline pid_t task_session_vnr(struct task_struct *tsk)
1592 {
1593         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1594 }
1595
1596 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1597 {
1598         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1599 }
1600
1601 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1602 {
1603         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1604 }
1605
1606 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1607 {
1608         pid_t pid = 0;
1609
1610         rcu_read_lock();
1611         if (pid_alive(tsk))
1612                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1613         rcu_read_unlock();
1614
1615         return pid;
1616 }
1617
1618 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1619 {
1620         return task_ppid_nr_ns(tsk, &init_pid_ns);
1621 }
1622
1623 /* Obsolete, do not use: */
1624 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1625 {
1626         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1627 }
1628
1629 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1630 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1631
1632 static inline unsigned int __task_state_index(unsigned int tsk_state,
1633                                               unsigned int tsk_exit_state)
1634 {
1635         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1636
1637         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1638
1639         if (tsk_state == TASK_IDLE)
1640                 state = TASK_REPORT_IDLE;
1641
1642         /*
1643          * We're lying here, but rather than expose a completely new task state
1644          * to userspace, we can make this appear as if the task has gone through
1645          * a regular rt_mutex_lock() call.
1646          */
1647         if (tsk_state == TASK_RTLOCK_WAIT)
1648                 state = TASK_UNINTERRUPTIBLE;
1649
1650         return fls(state);
1651 }
1652
1653 static inline unsigned int task_state_index(struct task_struct *tsk)
1654 {
1655         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1656 }
1657
1658 static inline char task_index_to_char(unsigned int state)
1659 {
1660         static const char state_char[] = "RSDTtXZPI";
1661
1662         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1663
1664         return state_char[state];
1665 }
1666
1667 static inline char task_state_to_char(struct task_struct *tsk)
1668 {
1669         return task_index_to_char(task_state_index(tsk));
1670 }
1671
1672 /**
1673  * is_global_init - check if a task structure is init. Since init
1674  * is free to have sub-threads we need to check tgid.
1675  * @tsk: Task structure to be checked.
1676  *
1677  * Check if a task structure is the first user space task the kernel created.
1678  *
1679  * Return: 1 if the task structure is init. 0 otherwise.
1680  */
1681 static inline int is_global_init(struct task_struct *tsk)
1682 {
1683         return task_tgid_nr(tsk) == 1;
1684 }
1685
1686 extern struct pid *cad_pid;
1687
1688 /*
1689  * Per process flags
1690  */
1691 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1692 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1693 #define PF_EXITING              0x00000004      /* Getting shut down */
1694 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1695 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1696 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1697 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1698 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1699 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1700 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1701 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1702 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1703 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1704 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1705 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1706 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1707 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1708 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1709 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1710 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1711                                                  * I am cleaning dirty pages from some other bdi. */
1712 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1713 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1714 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1715 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1716 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1717 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1718 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1719
1720 /*
1721  * Only the _current_ task can read/write to tsk->flags, but other
1722  * tasks can access tsk->flags in readonly mode for example
1723  * with tsk_used_math (like during threaded core dumping).
1724  * There is however an exception to this rule during ptrace
1725  * or during fork: the ptracer task is allowed to write to the
1726  * child->flags of its traced child (same goes for fork, the parent
1727  * can write to the child->flags), because we're guaranteed the
1728  * child is not running and in turn not changing child->flags
1729  * at the same time the parent does it.
1730  */
1731 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1732 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1733 #define clear_used_math()                       clear_stopped_child_used_math(current)
1734 #define set_used_math()                         set_stopped_child_used_math(current)
1735
1736 #define conditional_stopped_child_used_math(condition, child) \
1737         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1738
1739 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1740
1741 #define copy_to_stopped_child_used_math(child) \
1742         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1743
1744 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1745 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1746 #define used_math()                             tsk_used_math(current)
1747
1748 static __always_inline bool is_percpu_thread(void)
1749 {
1750 #ifdef CONFIG_SMP
1751         return (current->flags & PF_NO_SETAFFINITY) &&
1752                 (current->nr_cpus_allowed  == 1);
1753 #else
1754         return true;
1755 #endif
1756 }
1757
1758 /* Per-process atomic flags. */
1759 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1760 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1761 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1762 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1763 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1764 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1765 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1766 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1767
1768 #define TASK_PFA_TEST(name, func)                                       \
1769         static inline bool task_##func(struct task_struct *p)           \
1770         { return test_bit(PFA_##name, &p->atomic_flags); }
1771
1772 #define TASK_PFA_SET(name, func)                                        \
1773         static inline void task_set_##func(struct task_struct *p)       \
1774         { set_bit(PFA_##name, &p->atomic_flags); }
1775
1776 #define TASK_PFA_CLEAR(name, func)                                      \
1777         static inline void task_clear_##func(struct task_struct *p)     \
1778         { clear_bit(PFA_##name, &p->atomic_flags); }
1779
1780 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1781 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1782
1783 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1784 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1785 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1786
1787 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1788 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1789 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1790
1791 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1792 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1793 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1794
1795 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1796 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1797 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1798
1799 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1800 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1801
1802 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1803 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1804 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1805
1806 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1807 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1808
1809 static inline void
1810 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1811 {
1812         current->flags &= ~flags;
1813         current->flags |= orig_flags & flags;
1814 }
1815
1816 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1817 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1818 #ifdef CONFIG_SMP
1819 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1820 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1821 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1822 extern void release_user_cpus_ptr(struct task_struct *p);
1823 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1824 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1825 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1826 #else
1827 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1828 {
1829 }
1830 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1831 {
1832         if (!cpumask_test_cpu(0, new_mask))
1833                 return -EINVAL;
1834         return 0;
1835 }
1836 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1837 {
1838         if (src->user_cpus_ptr)
1839                 return -EINVAL;
1840         return 0;
1841 }
1842 static inline void release_user_cpus_ptr(struct task_struct *p)
1843 {
1844         WARN_ON(p->user_cpus_ptr);
1845 }
1846
1847 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1848 {
1849         return 0;
1850 }
1851 #endif
1852
1853 extern int yield_to(struct task_struct *p, bool preempt);
1854 extern void set_user_nice(struct task_struct *p, long nice);
1855 extern int task_prio(const struct task_struct *p);
1856
1857 /**
1858  * task_nice - return the nice value of a given task.
1859  * @p: the task in question.
1860  *
1861  * Return: The nice value [ -20 ... 0 ... 19 ].
1862  */
1863 static inline int task_nice(const struct task_struct *p)
1864 {
1865         return PRIO_TO_NICE((p)->static_prio);
1866 }
1867
1868 extern int can_nice(const struct task_struct *p, const int nice);
1869 extern int task_curr(const struct task_struct *p);
1870 extern int idle_cpu(int cpu);
1871 extern int available_idle_cpu(int cpu);
1872 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1873 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1874 extern void sched_set_fifo(struct task_struct *p);
1875 extern void sched_set_fifo_low(struct task_struct *p);
1876 extern void sched_set_normal(struct task_struct *p, int nice);
1877 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1878 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1879 extern struct task_struct *idle_task(int cpu);
1880
1881 /**
1882  * is_idle_task - is the specified task an idle task?
1883  * @p: the task in question.
1884  *
1885  * Return: 1 if @p is an idle task. 0 otherwise.
1886  */
1887 static __always_inline bool is_idle_task(const struct task_struct *p)
1888 {
1889         return !!(p->flags & PF_IDLE);
1890 }
1891
1892 extern struct task_struct *curr_task(int cpu);
1893 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1894
1895 void yield(void);
1896
1897 union thread_union {
1898 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1899         struct task_struct task;
1900 #endif
1901 #ifndef CONFIG_THREAD_INFO_IN_TASK
1902         struct thread_info thread_info;
1903 #endif
1904         unsigned long stack[THREAD_SIZE/sizeof(long)];
1905 };
1906
1907 #ifndef CONFIG_THREAD_INFO_IN_TASK
1908 extern struct thread_info init_thread_info;
1909 #endif
1910
1911 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1912
1913 #ifdef CONFIG_THREAD_INFO_IN_TASK
1914 # define task_thread_info(task) (&(task)->thread_info)
1915 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1916 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1917 #endif
1918
1919 /*
1920  * find a task by one of its numerical ids
1921  *
1922  * find_task_by_pid_ns():
1923  *      finds a task by its pid in the specified namespace
1924  * find_task_by_vpid():
1925  *      finds a task by its virtual pid
1926  *
1927  * see also find_vpid() etc in include/linux/pid.h
1928  */
1929
1930 extern struct task_struct *find_task_by_vpid(pid_t nr);
1931 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1932
1933 /*
1934  * find a task by its virtual pid and get the task struct
1935  */
1936 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1937
1938 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1939 extern int wake_up_process(struct task_struct *tsk);
1940 extern void wake_up_new_task(struct task_struct *tsk);
1941
1942 #ifdef CONFIG_SMP
1943 extern void kick_process(struct task_struct *tsk);
1944 #else
1945 static inline void kick_process(struct task_struct *tsk) { }
1946 #endif
1947
1948 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1949
1950 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1951 {
1952         __set_task_comm(tsk, from, false);
1953 }
1954
1955 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1956 #define get_task_comm(buf, tsk) ({                      \
1957         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1958         __get_task_comm(buf, sizeof(buf), tsk);         \
1959 })
1960
1961 #ifdef CONFIG_SMP
1962 static __always_inline void scheduler_ipi(void)
1963 {
1964         /*
1965          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1966          * TIF_NEED_RESCHED remotely (for the first time) will also send
1967          * this IPI.
1968          */
1969         preempt_fold_need_resched();
1970 }
1971 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1972 #else
1973 static inline void scheduler_ipi(void) { }
1974 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1975 {
1976         return 1;
1977 }
1978 #endif
1979
1980 /*
1981  * Set thread flags in other task's structures.
1982  * See asm/thread_info.h for TIF_xxxx flags available:
1983  */
1984 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1985 {
1986         set_ti_thread_flag(task_thread_info(tsk), flag);
1987 }
1988
1989 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1990 {
1991         clear_ti_thread_flag(task_thread_info(tsk), flag);
1992 }
1993
1994 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1995                                           bool value)
1996 {
1997         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1998 }
1999
2000 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
2001 {
2002         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
2003 }
2004
2005 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
2006 {
2007         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
2008 }
2009
2010 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
2011 {
2012         return test_ti_thread_flag(task_thread_info(tsk), flag);
2013 }
2014
2015 static inline void set_tsk_need_resched(struct task_struct *tsk)
2016 {
2017         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2018 }
2019
2020 static inline void clear_tsk_need_resched(struct task_struct *tsk)
2021 {
2022         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2023 }
2024
2025 static inline int test_tsk_need_resched(struct task_struct *tsk)
2026 {
2027         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2028 }
2029
2030 /*
2031  * cond_resched() and cond_resched_lock(): latency reduction via
2032  * explicit rescheduling in places that are safe. The return
2033  * value indicates whether a reschedule was done in fact.
2034  * cond_resched_lock() will drop the spinlock before scheduling,
2035  */
2036 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2037 extern int __cond_resched(void);
2038
2039 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
2040
2041 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2042
2043 static __always_inline int _cond_resched(void)
2044 {
2045         return static_call_mod(cond_resched)();
2046 }
2047
2048 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
2049 extern int dynamic_cond_resched(void);
2050
2051 static __always_inline int _cond_resched(void)
2052 {
2053         return dynamic_cond_resched();
2054 }
2055
2056 #else
2057
2058 static inline int _cond_resched(void)
2059 {
2060         return __cond_resched();
2061 }
2062
2063 #endif /* CONFIG_PREEMPT_DYNAMIC */
2064
2065 #else
2066
2067 static inline int _cond_resched(void) { return 0; }
2068
2069 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2070
2071 #define cond_resched() ({                       \
2072         __might_resched(__FILE__, __LINE__, 0); \
2073         _cond_resched();                        \
2074 })
2075
2076 extern int __cond_resched_lock(spinlock_t *lock);
2077 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2078 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2079
2080 #define MIGHT_RESCHED_RCU_SHIFT         8
2081 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2082
2083 #ifndef CONFIG_PREEMPT_RT
2084 /*
2085  * Non RT kernels have an elevated preempt count due to the held lock,
2086  * but are not allowed to be inside a RCU read side critical section
2087  */
2088 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2089 #else
2090 /*
2091  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2092  * cond_resched*lock() has to take that into account because it checks for
2093  * preempt_count() and rcu_preempt_depth().
2094  */
2095 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2096         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2097 #endif
2098
2099 #define cond_resched_lock(lock) ({                                              \
2100         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2101         __cond_resched_lock(lock);                                              \
2102 })
2103
2104 #define cond_resched_rwlock_read(lock) ({                                       \
2105         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2106         __cond_resched_rwlock_read(lock);                                       \
2107 })
2108
2109 #define cond_resched_rwlock_write(lock) ({                                      \
2110         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2111         __cond_resched_rwlock_write(lock);                                      \
2112 })
2113
2114 static inline void cond_resched_rcu(void)
2115 {
2116 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2117         rcu_read_unlock();
2118         cond_resched();
2119         rcu_read_lock();
2120 #endif
2121 }
2122
2123 #ifdef CONFIG_PREEMPT_DYNAMIC
2124
2125 extern bool preempt_model_none(void);
2126 extern bool preempt_model_voluntary(void);
2127 extern bool preempt_model_full(void);
2128
2129 #else
2130
2131 static inline bool preempt_model_none(void)
2132 {
2133         return IS_ENABLED(CONFIG_PREEMPT_NONE);
2134 }
2135 static inline bool preempt_model_voluntary(void)
2136 {
2137         return IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY);
2138 }
2139 static inline bool preempt_model_full(void)
2140 {
2141         return IS_ENABLED(CONFIG_PREEMPT);
2142 }
2143
2144 #endif
2145
2146 static inline bool preempt_model_rt(void)
2147 {
2148         return IS_ENABLED(CONFIG_PREEMPT_RT);
2149 }
2150
2151 /*
2152  * Does the preemption model allow non-cooperative preemption?
2153  *
2154  * For !CONFIG_PREEMPT_DYNAMIC kernels this is an exact match with
2155  * CONFIG_PREEMPTION; for CONFIG_PREEMPT_DYNAMIC this doesn't work as the
2156  * kernel is *built* with CONFIG_PREEMPTION=y but may run with e.g. the
2157  * PREEMPT_NONE model.
2158  */
2159 static inline bool preempt_model_preemptible(void)
2160 {
2161         return preempt_model_full() || preempt_model_rt();
2162 }
2163
2164 /*
2165  * Does a critical section need to be broken due to another
2166  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2167  * but a general need for low latency)
2168  */
2169 static inline int spin_needbreak(spinlock_t *lock)
2170 {
2171 #ifdef CONFIG_PREEMPTION
2172         return spin_is_contended(lock);
2173 #else
2174         return 0;
2175 #endif
2176 }
2177
2178 /*
2179  * Check if a rwlock is contended.
2180  * Returns non-zero if there is another task waiting on the rwlock.
2181  * Returns zero if the lock is not contended or the system / underlying
2182  * rwlock implementation does not support contention detection.
2183  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2184  * for low latency.
2185  */
2186 static inline int rwlock_needbreak(rwlock_t *lock)
2187 {
2188 #ifdef CONFIG_PREEMPTION
2189         return rwlock_is_contended(lock);
2190 #else
2191         return 0;
2192 #endif
2193 }
2194
2195 static __always_inline bool need_resched(void)
2196 {
2197         return unlikely(tif_need_resched());
2198 }
2199
2200 /*
2201  * Wrappers for p->thread_info->cpu access. No-op on UP.
2202  */
2203 #ifdef CONFIG_SMP
2204
2205 static inline unsigned int task_cpu(const struct task_struct *p)
2206 {
2207         return READ_ONCE(task_thread_info(p)->cpu);
2208 }
2209
2210 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2211
2212 #else
2213
2214 static inline unsigned int task_cpu(const struct task_struct *p)
2215 {
2216         return 0;
2217 }
2218
2219 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2220 {
2221 }
2222
2223 #endif /* CONFIG_SMP */
2224
2225 extern bool sched_task_on_rq(struct task_struct *p);
2226 extern unsigned long get_wchan(struct task_struct *p);
2227 extern struct task_struct *cpu_curr_snapshot(int cpu);
2228
2229 /*
2230  * In order to reduce various lock holder preemption latencies provide an
2231  * interface to see if a vCPU is currently running or not.
2232  *
2233  * This allows us to terminate optimistic spin loops and block, analogous to
2234  * the native optimistic spin heuristic of testing if the lock owner task is
2235  * running or not.
2236  */
2237 #ifndef vcpu_is_preempted
2238 static inline bool vcpu_is_preempted(int cpu)
2239 {
2240         return false;
2241 }
2242 #endif
2243
2244 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2245 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2246
2247 #ifndef TASK_SIZE_OF
2248 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2249 #endif
2250
2251 #ifdef CONFIG_SMP
2252 static inline bool owner_on_cpu(struct task_struct *owner)
2253 {
2254         /*
2255          * As lock holder preemption issue, we both skip spinning if
2256          * task is not on cpu or its cpu is preempted
2257          */
2258         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2259 }
2260
2261 /* Returns effective CPU energy utilization, as seen by the scheduler */
2262 unsigned long sched_cpu_util(int cpu);
2263 #endif /* CONFIG_SMP */
2264
2265 #ifdef CONFIG_RSEQ
2266
2267 /*
2268  * Map the event mask on the user-space ABI enum rseq_cs_flags
2269  * for direct mask checks.
2270  */
2271 enum rseq_event_mask_bits {
2272         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2273         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2274         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2275 };
2276
2277 enum rseq_event_mask {
2278         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2279         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2280         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2281 };
2282
2283 static inline void rseq_set_notify_resume(struct task_struct *t)
2284 {
2285         if (t->rseq)
2286                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2287 }
2288
2289 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2290
2291 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2292                                              struct pt_regs *regs)
2293 {
2294         if (current->rseq)
2295                 __rseq_handle_notify_resume(ksig, regs);
2296 }
2297
2298 static inline void rseq_signal_deliver(struct ksignal *ksig,
2299                                        struct pt_regs *regs)
2300 {
2301         preempt_disable();
2302         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2303         preempt_enable();
2304         rseq_handle_notify_resume(ksig, regs);
2305 }
2306
2307 /* rseq_preempt() requires preemption to be disabled. */
2308 static inline void rseq_preempt(struct task_struct *t)
2309 {
2310         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2311         rseq_set_notify_resume(t);
2312 }
2313
2314 /* rseq_migrate() requires preemption to be disabled. */
2315 static inline void rseq_migrate(struct task_struct *t)
2316 {
2317         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2318         rseq_set_notify_resume(t);
2319 }
2320
2321 /*
2322  * If parent process has a registered restartable sequences area, the
2323  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2324  */
2325 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2326 {
2327         if (clone_flags & CLONE_VM) {
2328                 t->rseq = NULL;
2329                 t->rseq_sig = 0;
2330                 t->rseq_event_mask = 0;
2331         } else {
2332                 t->rseq = current->rseq;
2333                 t->rseq_sig = current->rseq_sig;
2334                 t->rseq_event_mask = current->rseq_event_mask;
2335         }
2336 }
2337
2338 static inline void rseq_execve(struct task_struct *t)
2339 {
2340         t->rseq = NULL;
2341         t->rseq_sig = 0;
2342         t->rseq_event_mask = 0;
2343 }
2344
2345 #else
2346
2347 static inline void rseq_set_notify_resume(struct task_struct *t)
2348 {
2349 }
2350 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2351                                              struct pt_regs *regs)
2352 {
2353 }
2354 static inline void rseq_signal_deliver(struct ksignal *ksig,
2355                                        struct pt_regs *regs)
2356 {
2357 }
2358 static inline void rseq_preempt(struct task_struct *t)
2359 {
2360 }
2361 static inline void rseq_migrate(struct task_struct *t)
2362 {
2363 }
2364 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2365 {
2366 }
2367 static inline void rseq_execve(struct task_struct *t)
2368 {
2369 }
2370
2371 #endif
2372
2373 #ifdef CONFIG_DEBUG_RSEQ
2374
2375 void rseq_syscall(struct pt_regs *regs);
2376
2377 #else
2378
2379 static inline void rseq_syscall(struct pt_regs *regs)
2380 {
2381 }
2382
2383 #endif
2384
2385 #ifdef CONFIG_SCHED_CORE
2386 extern void sched_core_free(struct task_struct *tsk);
2387 extern void sched_core_fork(struct task_struct *p);
2388 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2389                                 unsigned long uaddr);
2390 #else
2391 static inline void sched_core_free(struct task_struct *tsk) { }
2392 static inline void sched_core_fork(struct task_struct *p) { }
2393 #endif
2394
2395 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2396
2397 #endif