include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13
  14 #include <linux/pid.h>
  15 #include <linux/sem.h>
  16 #include <linux/shm.h>
  17 #include <linux/mutex.h>
  18 #include <linux/plist.h>
  19 #include <linux/hrtimer.h>
  20 #include <linux/irqflags.h>
  21 #include <linux/seccomp.h>
  22 #include <linux/nodemask.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/refcount.h>
  25 #include <linux/resource.h>
  26 #include <linux/latencytop.h>
  27 #include <linux/sched/prio.h>
  28 #include <linux/sched/types.h>
  29 #include <linux/signal_types.h>
  30 #include <linux/syscall_user_dispatch.h>
  31 #include <linux/mm_types_task.h>
  32 #include <linux/task_io_accounting.h>
  33 #include <linux/posix-timers.h>
  34 #include <linux/rseq.h>
  35 #include <linux/seqlock.h>
  36 #include <linux/kcsan.h>
  37 #include <asm/kmap_size.h>
  38
  39 /* task_struct member predeclarations (sorted alphabetically): */
  40 struct audit_context;
  41 struct backing_dev_info;
  42 struct bio_list;
  43 struct blk_plug;
  44 struct bpf_local_storage;
  45 struct bpf_run_ctx;
  46 struct capture_control;
  47 struct cfs_rq;
  48 struct fs_struct;
  49 struct futex_pi_state;
  50 struct io_context;
  51 struct io_uring_task;
  52 struct mempolicy;
  53 struct nameidata;
  54 struct nsproxy;
  55 struct perf_event_context;
  56 struct pid_namespace;
  57 struct pipe_inode_info;
  58 struct rcu_node;
  59 struct reclaim_state;
  60 struct robust_list_head;
  61 struct root_domain;
  62 struct rq;
  63 struct sched_attr;
  64 struct sched_param;
  65 struct seq_file;
  66 struct sighand_struct;
  67 struct signal_struct;
  68 struct task_delay_info;
  69 struct task_group;
  70
  71 /*
  72  * Task state bitmask. NOTE! These bits are also
  73  * encoded in fs/proc/array.c: get_task_state().
  74  *
  75  * We have two separate sets of flags: task->state
  76  * is about runnability, while task->exit_state are
  77  * about the task exiting. Confusing, but this way
  78  * modifying one set can't modify the other one by
  79  * mistake.
  80  */
  81
  82 /* Used in tsk->state: */
  83 #define TASK_RUNNING                    0x0000
  84 #define TASK_INTERRUPTIBLE              0x0001
  85 #define TASK_UNINTERRUPTIBLE            0x0002
  86 #define __TASK_STOPPED                  0x0004
  87 #define __TASK_TRACED                   0x0008
  88 /* Used in tsk->exit_state: */
  89 #define EXIT_DEAD                       0x0010
  90 #define EXIT_ZOMBIE                     0x0020
  91 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
  92 /* Used in tsk->state again: */
  93 #define TASK_PARKED                     0x0040
  94 #define TASK_DEAD                       0x0080
  95 #define TASK_WAKEKILL                   0x0100
  96 #define TASK_WAKING                     0x0200
  97 #define TASK_NOLOAD                     0x0400
  98 #define TASK_NEW                        0x0800
  99 /* RT specific auxilliary flag to mark RT lock waiters */
 100 #define TASK_RTLOCK_WAIT                0x1000
 101 #define TASK_STATE_MAX                  0x2000
 102
 103 /* Convenience macros for the sake of set_current_state: */
 104 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 105 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 106 #define TASK_TRACED                     (TASK_WAKEKILL | __TASK_TRACED)
 107
 108 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 109
 110 /* Convenience macros for the sake of wake_up(): */
 111 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 112
 113 /* get_task_state(): */
 114 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 115                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 116                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 117                                          TASK_PARKED)
 118
 119 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 120
 121 #define task_is_traced(task)            ((READ_ONCE(task->__state) & __TASK_TRACED) != 0)
 122
 123 #define task_is_stopped(task)           ((READ_ONCE(task->__state) & __TASK_STOPPED) != 0)
 124
 125 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->__state) & (__TASK_STOPPED | __TASK_TRACED)) != 0)
 126
 127 /*
 128  * Special states are those that do not use the normal wait-loop pattern. See
 129  * the comment with set_special_state().
 130  */
 131 #define is_special_task_state(state)                            \
 132         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
 133
 134 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 135 # define debug_normal_state_change(state_value)                         \
 136         do {                                                            \
 137                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 138                 current->task_state_change = _THIS_IP_;                 \
 139         } while (0)
 140
 141 # define debug_special_state_change(state_value)                        \
 142         do {                                                            \
 143                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 144                 current->task_state_change = _THIS_IP_;                 \
 145         } while (0)
 146
 147 # define debug_rtlock_wait_set_state()                                  \
 148         do {                                                             \
 149                 current->saved_state_change = current->task_state_change;\
 150                 current->task_state_change = _THIS_IP_;                  \
 151         } while (0)
 152
 153 # define debug_rtlock_wait_restore_state()                              \
 154         do {                                                             \
 155                 current->task_state_change = current->saved_state_change;\
 156         } while (0)
 157
 158 #else
 159 # define debug_normal_state_change(cond)        do { } while (0)
 160 # define debug_special_state_change(cond)       do { } while (0)
 161 # define debug_rtlock_wait_set_state()          do { } while (0)
 162 # define debug_rtlock_wait_restore_state()      do { } while (0)
 163 #endif
 164
 165 /*
 166  * set_current_state() includes a barrier so that the write of current->state
 167  * is correctly serialised wrt the caller's subsequent test of whether to
 168  * actually sleep:
 169  *
 170  *   for (;;) {
 171  *      set_current_state(TASK_UNINTERRUPTIBLE);
 172  *      if (CONDITION)
 173  *         break;
 174  *
 175  *      schedule();
 176  *   }
 177  *   __set_current_state(TASK_RUNNING);
 178  *
 179  * If the caller does not need such serialisation (because, for instance, the
 180  * CONDITION test and condition change and wakeup are under the same lock) then
 181  * use __set_current_state().
 182  *
 183  * The above is typically ordered against the wakeup, which does:
 184  *
 185  *   CONDITION = 1;
 186  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 187  *
 188  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 189  * accessing p->state.
 190  *
 191  * Wakeup will do: if (@state & p->state) p->state = TASK_RUNNING, that is,
 192  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 193  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 194  *
 195  * However, with slightly different timing the wakeup TASK_RUNNING store can
 196  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 197  * a problem either because that will result in one extra go around the loop
 198  * and our @cond test will save the day.
 199  *
 200  * Also see the comments of try_to_wake_up().
 201  */
 202 #define __set_current_state(state_value)                                \
 203         do {                                                            \
 204                 debug_normal_state_change((state_value));               \
 205                 WRITE_ONCE(current->__state, (state_value));            \
 206         } while (0)
 207
 208 #define set_current_state(state_value)                                  \
 209         do {                                                            \
 210                 debug_normal_state_change((state_value));               \
 211                 smp_store_mb(current->__state, (state_value));          \
 212         } while (0)
 213
 214 /*
 215  * set_special_state() should be used for those states when the blocking task
 216  * can not use the regular condition based wait-loop. In that case we must
 217  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 218  * stores will not collide with our state change.
 219  */
 220 #define set_special_state(state_value)                                  \
 221         do {                                                            \
 222                 unsigned long flags; /* may shadow */                   \
 223                                                                         \
 224                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 225                 debug_special_state_change((state_value));              \
 226                 WRITE_ONCE(current->__state, (state_value));            \
 227                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 228         } while (0)
 229
 230 /*
 231  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 232  *
 233  * RT's spin/rwlock substitutions are state preserving. The state of the
 234  * task when blocking on the lock is saved in task_struct::saved_state and
 235  * restored after the lock has been acquired.  These operations are
 236  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 237  * lock related wakeups while the task is blocked on the lock are
 238  * redirected to operate on task_struct::saved_state to ensure that these
 239  * are not dropped. On restore task_struct::saved_state is set to
 240  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 241  *
 242  * The lock operation looks like this:
 243  *
 244  *      current_save_and_set_rtlock_wait_state();
 245  *      for (;;) {
 246  *              if (try_lock())
 247  *                      break;
 248  *              raw_spin_unlock_irq(&lock->wait_lock);
 249  *              schedule_rtlock();
 250  *              raw_spin_lock_irq(&lock->wait_lock);
 251  *              set_current_state(TASK_RTLOCK_WAIT);
 252  *      }
 253  *      current_restore_rtlock_saved_state();
 254  */
 255 #define current_save_and_set_rtlock_wait_state()                        \
 256         do {                                                            \
 257                 lockdep_assert_irqs_disabled();                         \
 258                 raw_spin_lock(&current->pi_lock);                       \
 259                 current->saved_state = current->__state;                \
 260                 debug_rtlock_wait_set_state();                          \
 261                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 262                 raw_spin_unlock(&current->pi_lock);                     \
 263         } while (0);
 264
 265 #define current_restore_rtlock_saved_state()                            \
 266         do {                                                            \
 267                 lockdep_assert_irqs_disabled();                         \
 268                 raw_spin_lock(&current->pi_lock);                       \
 269                 debug_rtlock_wait_restore_state();                      \
 270                 WRITE_ONCE(current->__state, current->saved_state);     \
 271                 current->saved_state = TASK_RUNNING;                    \
 272                 raw_spin_unlock(&current->pi_lock);                     \
 273         } while (0);
 274
 275 #define get_current_state()     READ_ONCE(current->__state)
 276
 277 /*
 278  * Define the task command name length as enum, then it can be visible to
 279  * BPF programs.
 280  */
 281 enum {
 282         TASK_COMM_LEN = 16,
 283 };
 284
 285 extern void scheduler_tick(void);
 286
 287 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 288
 289 extern long schedule_timeout(long timeout);
 290 extern long schedule_timeout_interruptible(long timeout);
 291 extern long schedule_timeout_killable(long timeout);
 292 extern long schedule_timeout_uninterruptible(long timeout);
 293 extern long schedule_timeout_idle(long timeout);
 294 asmlinkage void schedule(void);
 295 extern void schedule_preempt_disabled(void);
 296 asmlinkage void preempt_schedule_irq(void);
 297 #ifdef CONFIG_PREEMPT_RT
 298  extern void schedule_rtlock(void);
 299 #endif
 300
 301 extern int __must_check io_schedule_prepare(void);
 302 extern void io_schedule_finish(int token);
 303 extern long io_schedule_timeout(long timeout);
 304 extern void io_schedule(void);
 305
 306 /**
 307  * struct prev_cputime - snapshot of system and user cputime
 308  * @utime: time spent in user mode
 309  * @stime: time spent in system mode
 310  * @lock: protects the above two fields
 311  *
 312  * Stores previous user/system time values such that we can guarantee
 313  * monotonicity.
 314  */
 315 struct prev_cputime {
 316 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 317         u64                             utime;
 318         u64                             stime;
 319         raw_spinlock_t                  lock;
 320 #endif
 321 };
 322
 323 enum vtime_state {
 324         /* Task is sleeping or running in a CPU with VTIME inactive: */
 325         VTIME_INACTIVE = 0,
 326         /* Task is idle */
 327         VTIME_IDLE,
 328         /* Task runs in kernelspace in a CPU with VTIME active: */
 329         VTIME_SYS,
 330         /* Task runs in userspace in a CPU with VTIME active: */
 331         VTIME_USER,
 332         /* Task runs as guests in a CPU with VTIME active: */
 333         VTIME_GUEST,
 334 };
 335
 336 struct vtime {
 337         seqcount_t              seqcount;
 338         unsigned long long      starttime;
 339         enum vtime_state        state;
 340         unsigned int            cpu;
 341         u64                     utime;
 342         u64                     stime;
 343         u64                     gtime;
 344 };
 345
 346 /*
 347  * Utilization clamp constraints.
 348  * @UCLAMP_MIN: Minimum utilization
 349  * @UCLAMP_MAX: Maximum utilization
 350  * @UCLAMP_CNT: Utilization clamp constraints count
 351  */
 352 enum uclamp_id {
 353         UCLAMP_MIN = 0,
 354         UCLAMP_MAX,
 355         UCLAMP_CNT
 356 };
 357
 358 #ifdef CONFIG_SMP
 359 extern struct root_domain def_root_domain;
 360 extern struct mutex sched_domains_mutex;
 361 #endif
 362
 363 struct sched_info {
 364 #ifdef CONFIG_SCHED_INFO
 365         /* Cumulative counters: */
 366
 367         /* # of times we have run on this CPU: */
 368         unsigned long                   pcount;
 369
 370         /* Time spent waiting on a runqueue: */
 371         unsigned long long              run_delay;
 372
 373         /* Timestamps: */
 374
 375         /* When did we last run on a CPU? */
 376         unsigned long long              last_arrival;
 377
 378         /* When were we last queued to run? */
 379         unsigned long long              last_queued;
 380
 381 #endif /* CONFIG_SCHED_INFO */
 382 };
 383
 384 /*
 385  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 386  * has a few: load, load_avg, util_avg, freq, and capacity.
 387  *
 388  * We define a basic fixed point arithmetic range, and then formalize
 389  * all these metrics based on that basic range.
 390  */
 391 # define SCHED_FIXEDPOINT_SHIFT         10
 392 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 393
 394 /* Increase resolution of cpu_capacity calculations */
 395 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 396 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 397
 398 struct load_weight {
 399         unsigned long                   weight;
 400         u32                             inv_weight;
 401 };
 402
 403 /**
 404  * struct util_est - Estimation utilization of FAIR tasks
 405  * @enqueued: instantaneous estimated utilization of a task/cpu
 406  * @ewma:     the Exponential Weighted Moving Average (EWMA)
 407  *            utilization of a task
 408  *
 409  * Support data structure to track an Exponential Weighted Moving Average
 410  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
 411  * average each time a task completes an activation. Sample's weight is chosen
 412  * so that the EWMA will be relatively insensitive to transient changes to the
 413  * task's workload.
 414  *
 415  * The enqueued attribute has a slightly different meaning for tasks and cpus:
 416  * - task:   the task's util_avg at last task dequeue time
 417  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
 418  * Thus, the util_est.enqueued of a task represents the contribution on the
 419  * estimated utilization of the CPU where that task is currently enqueued.
 420  *
 421  * Only for tasks we track a moving average of the past instantaneous
 422  * estimated utilization. This allows to absorb sporadic drops in utilization
 423  * of an otherwise almost periodic task.
 424  *
 425  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 426  * updates. When a task is dequeued, its util_est should not be updated if its
 427  * util_avg has not been updated in the meantime.
 428  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
 429  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
 430  * for a task) it is safe to use MSB.
 431  */
 432 struct util_est {
 433         unsigned int                    enqueued;
 434         unsigned int                    ewma;
 435 #define UTIL_EST_WEIGHT_SHIFT           2
 436 #define UTIL_AVG_UNCHANGED              0x80000000
 437 } __attribute__((__aligned__(sizeof(u64))));
 438
 439 /*
 440  * The load/runnable/util_avg accumulates an infinite geometric series
 441  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 442  *
 443  * [load_avg definition]
 444  *
 445  *   load_avg = runnable% * scale_load_down(load)
 446  *
 447  * [runnable_avg definition]
 448  *
 449  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 450  *
 451  * [util_avg definition]
 452  *
 453  *   util_avg = running% * SCHED_CAPACITY_SCALE
 454  *
 455  * where runnable% is the time ratio that a sched_entity is runnable and
 456  * running% the time ratio that a sched_entity is running.
 457  *
 458  * For cfs_rq, they are the aggregated values of all runnable and blocked
 459  * sched_entities.
 460  *
 461  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 462  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 463  * for computing those signals (see update_rq_clock_pelt())
 464  *
 465  * N.B., the above ratios (runnable% and running%) themselves are in the
 466  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 467  * to as large a range as necessary. This is for example reflected by
 468  * util_avg's SCHED_CAPACITY_SCALE.
 469  *
 470  * [Overflow issue]
 471  *
 472  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 473  * with the highest load (=88761), always runnable on a single cfs_rq,
 474  * and should not overflow as the number already hits PID_MAX_LIMIT.
 475  *
 476  * For all other cases (including 32-bit kernels), struct load_weight's
 477  * weight will overflow first before we do, because:
 478  *
 479  *    Max(load_avg) <= Max(load.weight)
 480  *
 481  * Then it is the load_weight's responsibility to consider overflow
 482  * issues.
 483  */
 484 struct sched_avg {
 485         u64                             last_update_time;
 486         u64                             load_sum;
 487         u64                             runnable_sum;
 488         u32                             util_sum;
 489         u32                             period_contrib;
 490         unsigned long                   load_avg;
 491         unsigned long                   runnable_avg;
 492         unsigned long                   util_avg;
 493         struct util_est                 util_est;
 494 } ____cacheline_aligned;
 495
 496 struct sched_statistics {
 497 #ifdef CONFIG_SCHEDSTATS
 498         u64                             wait_start;
 499         u64                             wait_max;
 500         u64                             wait_count;
 501         u64                             wait_sum;
 502         u64                             iowait_count;
 503         u64                             iowait_sum;
 504
 505         u64                             sleep_start;
 506         u64                             sleep_max;
 507         s64                             sum_sleep_runtime;
 508
 509         u64                             block_start;
 510         u64                             block_max;
 511         s64                             sum_block_runtime;
 512
 513         u64                             exec_max;
 514         u64                             slice_max;
 515
 516         u64                             nr_migrations_cold;
 517         u64                             nr_failed_migrations_affine;
 518         u64                             nr_failed_migrations_running;
 519         u64                             nr_failed_migrations_hot;
 520         u64                             nr_forced_migrations;
 521
 522         u64                             nr_wakeups;
 523         u64                             nr_wakeups_sync;
 524         u64                             nr_wakeups_migrate;
 525         u64                             nr_wakeups_local;
 526         u64                             nr_wakeups_remote;
 527         u64                             nr_wakeups_affine;
 528         u64                             nr_wakeups_affine_attempts;
 529         u64                             nr_wakeups_passive;
 530         u64                             nr_wakeups_idle;
 531
 532 #ifdef CONFIG_SCHED_CORE
 533         u64                             core_forceidle_sum;
 534 #endif
 535 #endif /* CONFIG_SCHEDSTATS */
 536 } ____cacheline_aligned;
 537
 538 struct sched_entity {
 539         /* For load-balancing: */
 540         struct load_weight              load;
 541         struct rb_node                  run_node;
 542         struct list_head                group_node;
 543         unsigned int                    on_rq;
 544
 545         u64                             exec_start;
 546         u64                             sum_exec_runtime;
 547         u64                             vruntime;
 548         u64                             prev_sum_exec_runtime;
 549
 550         u64                             nr_migrations;
 551
 552 #ifdef CONFIG_FAIR_GROUP_SCHED
 553         int                             depth;
 554         struct sched_entity             *parent;
 555         /* rq on which this entity is (to be) queued: */
 556         struct cfs_rq                   *cfs_rq;
 557         /* rq "owned" by this entity/group: */
 558         struct cfs_rq                   *my_q;
 559         /* cached value of my_q->h_nr_running */
 560         unsigned long                   runnable_weight;
 561 #endif
 562
 563 #ifdef CONFIG_SMP
 564         /*
 565          * Per entity load average tracking.
 566          *
 567          * Put into separate cache line so it does not
 568          * collide with read-mostly values above.
 569          */
 570         struct sched_avg                avg;
 571 #endif
 572 };
 573
 574 struct sched_rt_entity {
 575         struct list_head                run_list;
 576         unsigned long                   timeout;
 577         unsigned long                   watchdog_stamp;
 578         unsigned int                    time_slice;
 579         unsigned short                  on_rq;
 580         unsigned short                  on_list;
 581
 582         struct sched_rt_entity          *back;
 583 #ifdef CONFIG_RT_GROUP_SCHED
 584         struct sched_rt_entity          *parent;
 585         /* rq on which this entity is (to be) queued: */
 586         struct rt_rq                    *rt_rq;
 587         /* rq "owned" by this entity/group: */
 588         struct rt_rq                    *my_q;
 589 #endif
 590 } __randomize_layout;
 591
 592 struct sched_dl_entity {
 593         struct rb_node                  rb_node;
 594
 595         /*
 596          * Original scheduling parameters. Copied here from sched_attr
 597          * during sched_setattr(), they will remain the same until
 598          * the next sched_setattr().
 599          */
 600         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 601         u64                             dl_deadline;    /* Relative deadline of each instance   */
 602         u64                             dl_period;      /* Separation of two instances (period) */
 603         u64                             dl_bw;          /* dl_runtime / dl_period               */
 604         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 605
 606         /*
 607          * Actual scheduling parameters. Initialized with the values above,
 608          * they are continuously updated during task execution. Note that
 609          * the remaining runtime could be < 0 in case we are in overrun.
 610          */
 611         s64                             runtime;        /* Remaining runtime for this instance  */
 612         u64                             deadline;       /* Absolute deadline for this instance  */
 613         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 614
 615         /*
 616          * Some bool flags:
 617          *
 618          * @dl_throttled tells if we exhausted the runtime. If so, the
 619          * task has to wait for a replenishment to be performed at the
 620          * next firing of dl_timer.
 621          *
 622          * @dl_yielded tells if task gave up the CPU before consuming
 623          * all its available runtime during the last job.
 624          *
 625          * @dl_non_contending tells if the task is inactive while still
 626          * contributing to the active utilization. In other words, it
 627          * indicates if the inactive timer has been armed and its handler
 628          * has not been executed yet. This flag is useful to avoid race
 629          * conditions between the inactive timer handler and the wakeup
 630          * code.
 631          *
 632          * @dl_overrun tells if the task asked to be informed about runtime
 633          * overruns.
 634          */
 635         unsigned int                    dl_throttled      : 1;
 636         unsigned int                    dl_yielded        : 1;
 637         unsigned int                    dl_non_contending : 1;
 638         unsigned int                    dl_overrun        : 1;
 639
 640         /*
 641          * Bandwidth enforcement timer. Each -deadline task has its
 642          * own bandwidth to be enforced, thus we need one timer per task.
 643          */
 644         struct hrtimer                  dl_timer;
 645
 646         /*
 647          * Inactive timer, responsible for decreasing the active utilization
 648          * at the "0-lag time". When a -deadline task blocks, it contributes
 649          * to GRUB's active utilization until the "0-lag time", hence a
 650          * timer is needed to decrease the active utilization at the correct
 651          * time.
 652          */
 653         struct hrtimer inactive_timer;
 654
 655 #ifdef CONFIG_RT_MUTEXES
 656         /*
 657          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 658          * pi_se points to the donor, otherwise points to the dl_se it belongs
 659          * to (the original one/itself).
 660          */
 661         struct sched_dl_entity *pi_se;
 662 #endif
 663 };
 664
 665 #ifdef CONFIG_UCLAMP_TASK
 666 /* Number of utilization clamp buckets (shorter alias) */
 667 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 668
 669 /*
 670  * Utilization clamp for a scheduling entity
 671  * @value:              clamp value "assigned" to a se
 672  * @bucket_id:          bucket index corresponding to the "assigned" value
 673  * @active:             the se is currently refcounted in a rq's bucket
 674  * @user_defined:       the requested clamp value comes from user-space
 675  *
 676  * The bucket_id is the index of the clamp bucket matching the clamp value
 677  * which is pre-computed and stored to avoid expensive integer divisions from
 678  * the fast path.
 679  *
 680  * The active bit is set whenever a task has got an "effective" value assigned,
 681  * which can be different from the clamp value "requested" from user-space.
 682  * This allows to know a task is refcounted in the rq's bucket corresponding
 683  * to the "effective" bucket_id.
 684  *
 685  * The user_defined bit is set whenever a task has got a task-specific clamp
 686  * value requested from userspace, i.e. the system defaults apply to this task
 687  * just as a restriction. This allows to relax default clamps when a less
 688  * restrictive task-specific value has been requested, thus allowing to
 689  * implement a "nice" semantic. For example, a task running with a 20%
 690  * default boost can still drop its own boosting to 0%.
 691  */
 692 struct uclamp_se {
 693         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 694         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 695         unsigned int active             : 1;
 696         unsigned int user_defined       : 1;
 697 };
 698 #endif /* CONFIG_UCLAMP_TASK */
 699
 700 union rcu_special {
 701         struct {
 702                 u8                      blocked;
 703                 u8                      need_qs;
 704                 u8                      exp_hint; /* Hint for performance. */
 705                 u8                      need_mb; /* Readers need smp_mb(). */
 706         } b; /* Bits. */
 707         u32 s; /* Set of bits. */
 708 };
 709
 710 enum perf_event_task_context {
 711         perf_invalid_context = -1,
 712         perf_hw_context = 0,
 713         perf_sw_context,
 714         perf_nr_task_contexts,
 715 };
 716
 717 struct wake_q_node {
 718         struct wake_q_node *next;
 719 };
 720
 721 struct kmap_ctrl {
 722 #ifdef CONFIG_KMAP_LOCAL
 723         int                             idx;
 724         pte_t                           pteval[KM_MAX_IDX];
 725 #endif
 726 };
 727
 728 struct task_struct {
 729 #ifdef CONFIG_THREAD_INFO_IN_TASK
 730         /*
 731          * For reasons of header soup (see current_thread_info()), this
 732          * must be the first element of task_struct.
 733          */
 734         struct thread_info              thread_info;
 735 #endif
 736         unsigned int                    __state;
 737
 738 #ifdef CONFIG_PREEMPT_RT
 739         /* saved state for "spinlock sleepers" */
 740         unsigned int                    saved_state;
 741 #endif
 742
 743         /*
 744          * This begins the randomizable portion of task_struct. Only
 745          * scheduling-critical items should be added above here.
 746          */
 747         randomized_struct_fields_start
 748
 749         void                            *stack;
 750         refcount_t                      usage;
 751         /* Per task flags (PF_*), defined further below: */
 752         unsigned int                    flags;
 753         unsigned int                    ptrace;
 754
 755 #ifdef CONFIG_SMP
 756         int                             on_cpu;
 757         struct __call_single_node       wake_entry;
 758         unsigned int                    wakee_flips;
 759         unsigned long                   wakee_flip_decay_ts;
 760         struct task_struct              *last_wakee;
 761
 762         /*
 763          * recent_used_cpu is initially set as the last CPU used by a task
 764          * that wakes affine another task. Waker/wakee relationships can
 765          * push tasks around a CPU where each wakeup moves to the next one.
 766          * Tracking a recently used CPU allows a quick search for a recently
 767          * used CPU that may be idle.
 768          */
 769         int                             recent_used_cpu;
 770         int                             wake_cpu;
 771 #endif
 772         int                             on_rq;
 773
 774         int                             prio;
 775         int                             static_prio;
 776         int                             normal_prio;
 777         unsigned int                    rt_priority;
 778
 779         struct sched_entity             se;
 780         struct sched_rt_entity          rt;
 781         struct sched_dl_entity          dl;
 782         const struct sched_class        *sched_class;
 783
 784 #ifdef CONFIG_SCHED_CORE
 785         struct rb_node                  core_node;
 786         unsigned long                   core_cookie;
 787         unsigned int                    core_occupation;
 788 #endif
 789
 790 #ifdef CONFIG_CGROUP_SCHED
 791         struct task_group               *sched_task_group;
 792 #endif
 793
 794 #ifdef CONFIG_UCLAMP_TASK
 795         /*
 796          * Clamp values requested for a scheduling entity.
 797          * Must be updated with task_rq_lock() held.
 798          */
 799         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 800         /*
 801          * Effective clamp values used for a scheduling entity.
 802          * Must be updated with task_rq_lock() held.
 803          */
 804         struct uclamp_se                uclamp[UCLAMP_CNT];
 805 #endif
 806
 807         struct sched_statistics         stats;
 808
 809 #ifdef CONFIG_PREEMPT_NOTIFIERS
 810         /* List of struct preempt_notifier: */
 811         struct hlist_head               preempt_notifiers;
 812 #endif
 813
 814 #ifdef CONFIG_BLK_DEV_IO_TRACE
 815         unsigned int                    btrace_seq;
 816 #endif
 817
 818         unsigned int                    policy;
 819         int                             nr_cpus_allowed;
 820         const cpumask_t                 *cpus_ptr;
 821         cpumask_t                       *user_cpus_ptr;
 822         cpumask_t                       cpus_mask;
 823         void                            *migration_pending;
 824 #ifdef CONFIG_SMP
 825         unsigned short                  migration_disabled;
 826 #endif
 827         unsigned short                  migration_flags;
 828
 829 #ifdef CONFIG_PREEMPT_RCU
 830         int                             rcu_read_lock_nesting;
 831         union rcu_special               rcu_read_unlock_special;
 832         struct list_head                rcu_node_entry;
 833         struct rcu_node                 *rcu_blocked_node;
 834 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 835
 836 #ifdef CONFIG_TASKS_RCU
 837         unsigned long                   rcu_tasks_nvcsw;
 838         u8                              rcu_tasks_holdout;
 839         u8                              rcu_tasks_idx;
 840         int                             rcu_tasks_idle_cpu;
 841         struct list_head                rcu_tasks_holdout_list;
 842 #endif /* #ifdef CONFIG_TASKS_RCU */
 843
 844 #ifdef CONFIG_TASKS_TRACE_RCU
 845         int                             trc_reader_nesting;
 846         int                             trc_ipi_to_cpu;
 847         union rcu_special               trc_reader_special;
 848         bool                            trc_reader_checked;
 849         struct list_head                trc_holdout_list;
 850 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 851
 852         struct sched_info               sched_info;
 853
 854         struct list_head                tasks;
 855 #ifdef CONFIG_SMP
 856         struct plist_node               pushable_tasks;
 857         struct rb_node                  pushable_dl_tasks;
 858 #endif
 859
 860         struct mm_struct                *mm;
 861         struct mm_struct                *active_mm;
 862
 863         /* Per-thread vma caching: */
 864         struct vmacache                 vmacache;
 865
 866 #ifdef SPLIT_RSS_COUNTING
 867         struct task_rss_stat            rss_stat;
 868 #endif
 869         int                             exit_state;
 870         int                             exit_code;
 871         int                             exit_signal;
 872         /* The signal sent when the parent dies: */
 873         int                             pdeath_signal;
 874         /* JOBCTL_*, siglock protected: */
 875         unsigned long                   jobctl;
 876
 877         /* Used for emulating ABI behavior of previous Linux versions: */
 878         unsigned int                    personality;
 879
 880         /* Scheduler bits, serialized by scheduler locks: */
 881         unsigned                        sched_reset_on_fork:1;
 882         unsigned                        sched_contributes_to_load:1;
 883         unsigned                        sched_migrated:1;
 884 #ifdef CONFIG_PSI
 885         unsigned                        sched_psi_wake_requeue:1;
 886 #endif
 887
 888         /* Force alignment to the next boundary: */
 889         unsigned                        :0;
 890
 891         /* Unserialized, strictly 'current' */
 892
 893         /*
 894          * This field must not be in the scheduler word above due to wakelist
 895          * queueing no longer being serialized by p->on_cpu. However:
 896          *
 897          * p->XXX = X;                  ttwu()
 898          * schedule()                     if (p->on_rq && ..) // false
 899          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 900          *   deactivate_task()                ttwu_queue_wakelist())
 901          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 902          *
 903          * guarantees all stores of 'current' are visible before
 904          * ->sched_remote_wakeup gets used, so it can be in this word.
 905          */
 906         unsigned                        sched_remote_wakeup:1;
 907
 908         /* Bit to tell LSMs we're in execve(): */
 909         unsigned                        in_execve:1;
 910         unsigned                        in_iowait:1;
 911 #ifndef TIF_RESTORE_SIGMASK
 912         unsigned                        restore_sigmask:1;
 913 #endif
 914 #ifdef CONFIG_MEMCG
 915         unsigned                        in_user_fault:1;
 916 #endif
 917 #ifdef CONFIG_COMPAT_BRK
 918         unsigned                        brk_randomized:1;
 919 #endif
 920 #ifdef CONFIG_CGROUPS
 921         /* disallow userland-initiated cgroup migration */
 922         unsigned                        no_cgroup_migration:1;
 923         /* task is frozen/stopped (used by the cgroup freezer) */
 924         unsigned                        frozen:1;
 925 #endif
 926 #ifdef CONFIG_BLK_CGROUP
 927         unsigned                        use_memdelay:1;
 928 #endif
 929 #ifdef CONFIG_PSI
 930         /* Stalled due to lack of memory */
 931         unsigned                        in_memstall:1;
 932 #endif
 933 #ifdef CONFIG_PAGE_OWNER
 934         /* Used by page_owner=on to detect recursion in page tracking. */
 935         unsigned                        in_page_owner:1;
 936 #endif
 937 #ifdef CONFIG_EVENTFD
 938         /* Recursion prevention for eventfd_signal() */
 939         unsigned                        in_eventfd_signal:1;
 940 #endif
 941 #ifdef CONFIG_IOMMU_SVA
 942         unsigned                        pasid_activated:1;
 943 #endif
 944
 945         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
 946
 947         struct restart_block            restart_block;
 948
 949         pid_t                           pid;
 950         pid_t                           tgid;
 951
 952 #ifdef CONFIG_STACKPROTECTOR
 953         /* Canary value for the -fstack-protector GCC feature: */
 954         unsigned long                   stack_canary;
 955 #endif
 956         /*
 957          * Pointers to the (original) parent process, youngest child, younger sibling,
 958          * older sibling, respectively.  (p->father can be replaced with
 959          * p->real_parent->pid)
 960          */
 961
 962         /* Real parent process: */
 963         struct task_struct __rcu        *real_parent;
 964
 965         /* Recipient of SIGCHLD, wait4() reports: */
 966         struct task_struct __rcu        *parent;
 967
 968         /*
 969          * Children/sibling form the list of natural children:
 970          */
 971         struct list_head                children;
 972         struct list_head                sibling;
 973         struct task_struct              *group_leader;
 974
 975         /*
 976          * 'ptraced' is the list of tasks this task is using ptrace() on.
 977          *
 978          * This includes both natural children and PTRACE_ATTACH targets.
 979          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
 980          */
 981         struct list_head                ptraced;
 982         struct list_head                ptrace_entry;
 983
 984         /* PID/PID hash table linkage. */
 985         struct pid                      *thread_pid;
 986         struct hlist_node               pid_links[PIDTYPE_MAX];
 987         struct list_head                thread_group;
 988         struct list_head                thread_node;
 989
 990         struct completion               *vfork_done;
 991
 992         /* CLONE_CHILD_SETTID: */
 993         int __user                      *set_child_tid;
 994
 995         /* CLONE_CHILD_CLEARTID: */
 996         int __user                      *clear_child_tid;
 997
 998         /* PF_KTHREAD | PF_IO_WORKER */
 999         void                            *worker_private;
1000
1001         u64                             utime;
1002         u64                             stime;
1003 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1004         u64                             utimescaled;
1005         u64                             stimescaled;
1006 #endif
1007         u64                             gtime;
1008         struct prev_cputime             prev_cputime;
1009 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1010         struct vtime                    vtime;
1011 #endif
1012
1013 #ifdef CONFIG_NO_HZ_FULL
1014         atomic_t                        tick_dep_mask;
1015 #endif
1016         /* Context switch counts: */
1017         unsigned long                   nvcsw;
1018         unsigned long                   nivcsw;
1019
1020         /* Monotonic time in nsecs: */
1021         u64                             start_time;
1022
1023         /* Boot based time in nsecs: */
1024         u64                             start_boottime;
1025
1026         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1027         unsigned long                   min_flt;
1028         unsigned long                   maj_flt;
1029
1030         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1031         struct posix_cputimers          posix_cputimers;
1032
1033 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1034         struct posix_cputimers_work     posix_cputimers_work;
1035 #endif
1036
1037         /* Process credentials: */
1038
1039         /* Tracer's credentials at attach: */
1040         const struct cred __rcu         *ptracer_cred;
1041
1042         /* Objective and real subjective task credentials (COW): */
1043         const struct cred __rcu         *real_cred;
1044
1045         /* Effective (overridable) subjective task credentials (COW): */
1046         const struct cred __rcu         *cred;
1047
1048 #ifdef CONFIG_KEYS
1049         /* Cached requested key. */
1050         struct key                      *cached_requested_key;
1051 #endif
1052
1053         /*
1054          * executable name, excluding path.
1055          *
1056          * - normally initialized setup_new_exec()
1057          * - access it with [gs]et_task_comm()
1058          * - lock it with task_lock()
1059          */
1060         char                            comm[TASK_COMM_LEN];
1061
1062         struct nameidata                *nameidata;
1063
1064 #ifdef CONFIG_SYSVIPC
1065         struct sysv_sem                 sysvsem;
1066         struct sysv_shm                 sysvshm;
1067 #endif
1068 #ifdef CONFIG_DETECT_HUNG_TASK
1069         unsigned long                   last_switch_count;
1070         unsigned long                   last_switch_time;
1071 #endif
1072         /* Filesystem information: */
1073         struct fs_struct                *fs;
1074
1075         /* Open file information: */
1076         struct files_struct             *files;
1077
1078 #ifdef CONFIG_IO_URING
1079         struct io_uring_task            *io_uring;
1080 #endif
1081
1082         /* Namespaces: */
1083         struct nsproxy                  *nsproxy;
1084
1085         /* Signal handlers: */
1086         struct signal_struct            *signal;
1087         struct sighand_struct __rcu             *sighand;
1088         sigset_t                        blocked;
1089         sigset_t                        real_blocked;
1090         /* Restored if set_restore_sigmask() was used: */
1091         sigset_t                        saved_sigmask;
1092         struct sigpending               pending;
1093         unsigned long                   sas_ss_sp;
1094         size_t                          sas_ss_size;
1095         unsigned int                    sas_ss_flags;
1096
1097         struct callback_head            *task_works;
1098
1099 #ifdef CONFIG_AUDIT
1100 #ifdef CONFIG_AUDITSYSCALL
1101         struct audit_context            *audit_context;
1102 #endif
1103         kuid_t                          loginuid;
1104         unsigned int                    sessionid;
1105 #endif
1106         struct seccomp                  seccomp;
1107         struct syscall_user_dispatch    syscall_dispatch;
1108
1109         /* Thread group tracking: */
1110         u64                             parent_exec_id;
1111         u64                             self_exec_id;
1112
1113         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1114         spinlock_t                      alloc_lock;
1115
1116         /* Protection of the PI data structures: */
1117         raw_spinlock_t                  pi_lock;
1118
1119         struct wake_q_node              wake_q;
1120
1121 #ifdef CONFIG_RT_MUTEXES
1122         /* PI waiters blocked on a rt_mutex held by this task: */
1123         struct rb_root_cached           pi_waiters;
1124         /* Updated under owner's pi_lock and rq lock */
1125         struct task_struct              *pi_top_task;
1126         /* Deadlock detection and priority inheritance handling: */
1127         struct rt_mutex_waiter          *pi_blocked_on;
1128 #endif
1129
1130 #ifdef CONFIG_DEBUG_MUTEXES
1131         /* Mutex deadlock detection: */
1132         struct mutex_waiter             *blocked_on;
1133 #endif
1134
1135 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1136         int                             non_block_count;
1137 #endif
1138
1139 #ifdef CONFIG_TRACE_IRQFLAGS
1140         struct irqtrace_events          irqtrace;
1141         unsigned int                    hardirq_threaded;
1142         u64                             hardirq_chain_key;
1143         int                             softirqs_enabled;
1144         int                             softirq_context;
1145         int                             irq_config;
1146 #endif
1147 #ifdef CONFIG_PREEMPT_RT
1148         int                             softirq_disable_cnt;
1149 #endif
1150
1151 #ifdef CONFIG_LOCKDEP
1152 # define MAX_LOCK_DEPTH                 48UL
1153         u64                             curr_chain_key;
1154         int                             lockdep_depth;
1155         unsigned int                    lockdep_recursion;
1156         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1157 #endif
1158
1159 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1160         unsigned int                    in_ubsan;
1161 #endif
1162
1163         /* Journalling filesystem info: */
1164         void                            *journal_info;
1165
1166         /* Stacked block device info: */
1167         struct bio_list                 *bio_list;
1168
1169         /* Stack plugging: */
1170         struct blk_plug                 *plug;
1171
1172         /* VM state: */
1173         struct reclaim_state            *reclaim_state;
1174
1175         struct backing_dev_info         *backing_dev_info;
1176
1177         struct io_context               *io_context;
1178
1179 #ifdef CONFIG_COMPACTION
1180         struct capture_control          *capture_control;
1181 #endif
1182         /* Ptrace state: */
1183         unsigned long                   ptrace_message;
1184         kernel_siginfo_t                *last_siginfo;
1185
1186         struct task_io_accounting       ioac;
1187 #ifdef CONFIG_PSI
1188         /* Pressure stall state */
1189         unsigned int                    psi_flags;
1190 #endif
1191 #ifdef CONFIG_TASK_XACCT
1192         /* Accumulated RSS usage: */
1193         u64                             acct_rss_mem1;
1194         /* Accumulated virtual memory usage: */
1195         u64                             acct_vm_mem1;
1196         /* stime + utime since last update: */
1197         u64                             acct_timexpd;
1198 #endif
1199 #ifdef CONFIG_CPUSETS
1200         /* Protected by ->alloc_lock: */
1201         nodemask_t                      mems_allowed;
1202         /* Sequence number to catch updates: */
1203         seqcount_spinlock_t             mems_allowed_seq;
1204         int                             cpuset_mem_spread_rotor;
1205         int                             cpuset_slab_spread_rotor;
1206 #endif
1207 #ifdef CONFIG_CGROUPS
1208         /* Control Group info protected by css_set_lock: */
1209         struct css_set __rcu            *cgroups;
1210         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1211         struct list_head                cg_list;
1212 #endif
1213 #ifdef CONFIG_X86_CPU_RESCTRL
1214         u32                             closid;
1215         u32                             rmid;
1216 #endif
1217 #ifdef CONFIG_FUTEX
1218         struct robust_list_head __user  *robust_list;
1219 #ifdef CONFIG_COMPAT
1220         struct compat_robust_list_head __user *compat_robust_list;
1221 #endif
1222         struct list_head                pi_state_list;
1223         struct futex_pi_state           *pi_state_cache;
1224         struct mutex                    futex_exit_mutex;
1225         unsigned int                    futex_state;
1226 #endif
1227 #ifdef CONFIG_PERF_EVENTS
1228         struct perf_event_context       *perf_event_ctxp[perf_nr_task_contexts];
1229         struct mutex                    perf_event_mutex;
1230         struct list_head                perf_event_list;
1231 #endif
1232 #ifdef CONFIG_DEBUG_PREEMPT
1233         unsigned long                   preempt_disable_ip;
1234 #endif
1235 #ifdef CONFIG_NUMA
1236         /* Protected by alloc_lock: */
1237         struct mempolicy                *mempolicy;
1238         short                           il_prev;
1239         short                           pref_node_fork;
1240 #endif
1241 #ifdef CONFIG_NUMA_BALANCING
1242         int                             numa_scan_seq;
1243         unsigned int                    numa_scan_period;
1244         unsigned int                    numa_scan_period_max;
1245         int                             numa_preferred_nid;
1246         unsigned long                   numa_migrate_retry;
1247         /* Migration stamp: */
1248         u64                             node_stamp;
1249         u64                             last_task_numa_placement;
1250         u64                             last_sum_exec_runtime;
1251         struct callback_head            numa_work;
1252
1253         /*
1254          * This pointer is only modified for current in syscall and
1255          * pagefault context (and for tasks being destroyed), so it can be read
1256          * from any of the following contexts:
1257          *  - RCU read-side critical section
1258          *  - current->numa_group from everywhere
1259          *  - task's runqueue locked, task not running
1260          */
1261         struct numa_group __rcu         *numa_group;
1262
1263         /*
1264          * numa_faults is an array split into four regions:
1265          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1266          * in this precise order.
1267          *
1268          * faults_memory: Exponential decaying average of faults on a per-node
1269          * basis. Scheduling placement decisions are made based on these
1270          * counts. The values remain static for the duration of a PTE scan.
1271          * faults_cpu: Track the nodes the process was running on when a NUMA
1272          * hinting fault was incurred.
1273          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1274          * during the current scan window. When the scan completes, the counts
1275          * in faults_memory and faults_cpu decay and these values are copied.
1276          */
1277         unsigned long                   *numa_faults;
1278         unsigned long                   total_numa_faults;
1279
1280         /*
1281          * numa_faults_locality tracks if faults recorded during the last
1282          * scan window were remote/local or failed to migrate. The task scan
1283          * period is adapted based on the locality of the faults with different
1284          * weights depending on whether they were shared or private faults
1285          */
1286         unsigned long                   numa_faults_locality[3];
1287
1288         unsigned long                   numa_pages_migrated;
1289 #endif /* CONFIG_NUMA_BALANCING */
1290
1291 #ifdef CONFIG_RSEQ
1292         struct rseq __user *rseq;
1293         u32 rseq_sig;
1294         /*
1295          * RmW on rseq_event_mask must be performed atomically
1296          * with respect to preemption.
1297          */
1298         unsigned long rseq_event_mask;
1299 #endif
1300
1301         struct tlbflush_unmap_batch     tlb_ubc;
1302
1303         union {
1304                 refcount_t              rcu_users;
1305                 struct rcu_head         rcu;
1306         };
1307
1308         /* Cache last used pipe for splice(): */
1309         struct pipe_inode_info          *splice_pipe;
1310
1311         struct page_frag                task_frag;
1312
1313 #ifdef CONFIG_TASK_DELAY_ACCT
1314         struct task_delay_info          *delays;
1315 #endif
1316
1317 #ifdef CONFIG_FAULT_INJECTION
1318         int                             make_it_fail;
1319         unsigned int                    fail_nth;
1320 #endif
1321         /*
1322          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1323          * balance_dirty_pages() for a dirty throttling pause:
1324          */
1325         int                             nr_dirtied;
1326         int                             nr_dirtied_pause;
1327         /* Start of a write-and-pause period: */
1328         unsigned long                   dirty_paused_when;
1329
1330 #ifdef CONFIG_LATENCYTOP
1331         int                             latency_record_count;
1332         struct latency_record           latency_record[LT_SAVECOUNT];
1333 #endif
1334         /*
1335          * Time slack values; these are used to round up poll() and
1336          * select() etc timeout values. These are in nanoseconds.
1337          */
1338         u64                             timer_slack_ns;
1339         u64                             default_timer_slack_ns;
1340
1341 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1342         unsigned int                    kasan_depth;
1343 #endif
1344
1345 #ifdef CONFIG_KCSAN
1346         struct kcsan_ctx                kcsan_ctx;
1347 #ifdef CONFIG_TRACE_IRQFLAGS
1348         struct irqtrace_events          kcsan_save_irqtrace;
1349 #endif
1350 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1351         int                             kcsan_stack_depth;
1352 #endif
1353 #endif
1354
1355 #if IS_ENABLED(CONFIG_KUNIT)
1356         struct kunit                    *kunit_test;
1357 #endif
1358
1359 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1360         /* Index of current stored address in ret_stack: */
1361         int                             curr_ret_stack;
1362         int                             curr_ret_depth;
1363
1364         /* Stack of return addresses for return function tracing: */
1365         struct ftrace_ret_stack         *ret_stack;
1366
1367         /* Timestamp for last schedule: */
1368         unsigned long long              ftrace_timestamp;
1369
1370         /*
1371          * Number of functions that haven't been traced
1372          * because of depth overrun:
1373          */
1374         atomic_t                        trace_overrun;
1375
1376         /* Pause tracing: */
1377         atomic_t                        tracing_graph_pause;
1378 #endif
1379
1380 #ifdef CONFIG_TRACING
1381         /* State flags for use by tracers: */
1382         unsigned long                   trace;
1383
1384         /* Bitmask and counter of trace recursion: */
1385         unsigned long                   trace_recursion;
1386 #endif /* CONFIG_TRACING */
1387
1388 #ifdef CONFIG_KCOV
1389         /* See kernel/kcov.c for more details. */
1390
1391         /* Coverage collection mode enabled for this task (0 if disabled): */
1392         unsigned int                    kcov_mode;
1393
1394         /* Size of the kcov_area: */
1395         unsigned int                    kcov_size;
1396
1397         /* Buffer for coverage collection: */
1398         void                            *kcov_area;
1399
1400         /* KCOV descriptor wired with this task or NULL: */
1401         struct kcov                     *kcov;
1402
1403         /* KCOV common handle for remote coverage collection: */
1404         u64                             kcov_handle;
1405
1406         /* KCOV sequence number: */
1407         int                             kcov_sequence;
1408
1409         /* Collect coverage from softirq context: */
1410         unsigned int                    kcov_softirq;
1411 #endif
1412
1413 #ifdef CONFIG_MEMCG
1414         struct mem_cgroup               *memcg_in_oom;
1415         gfp_t                           memcg_oom_gfp_mask;
1416         int                             memcg_oom_order;
1417
1418         /* Number of pages to reclaim on returning to userland: */
1419         unsigned int                    memcg_nr_pages_over_high;
1420
1421         /* Used by memcontrol for targeted memcg charge: */
1422         struct mem_cgroup               *active_memcg;
1423 #endif
1424
1425 #ifdef CONFIG_BLK_CGROUP
1426         struct request_queue            *throttle_queue;
1427 #endif
1428
1429 #ifdef CONFIG_UPROBES
1430         struct uprobe_task              *utask;
1431 #endif
1432 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1433         unsigned int                    sequential_io;
1434         unsigned int                    sequential_io_avg;
1435 #endif
1436         struct kmap_ctrl                kmap_ctrl;
1437 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1438         unsigned long                   task_state_change;
1439 # ifdef CONFIG_PREEMPT_RT
1440         unsigned long                   saved_state_change;
1441 # endif
1442 #endif
1443         int                             pagefault_disabled;
1444 #ifdef CONFIG_MMU
1445         struct task_struct              *oom_reaper_list;
1446 #endif
1447 #ifdef CONFIG_VMAP_STACK
1448         struct vm_struct                *stack_vm_area;
1449 #endif
1450 #ifdef CONFIG_THREAD_INFO_IN_TASK
1451         /* A live task holds one reference: */
1452         refcount_t                      stack_refcount;
1453 #endif
1454 #ifdef CONFIG_LIVEPATCH
1455         int patch_state;
1456 #endif
1457 #ifdef CONFIG_SECURITY
1458         /* Used by LSM modules for access restriction: */
1459         void                            *security;
1460 #endif
1461 #ifdef CONFIG_BPF_SYSCALL
1462         /* Used by BPF task local storage */
1463         struct bpf_local_storage __rcu  *bpf_storage;
1464         /* Used for BPF run context */
1465         struct bpf_run_ctx              *bpf_ctx;
1466 #endif
1467
1468 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1469         unsigned long                   lowest_stack;
1470         unsigned long                   prev_lowest_stack;
1471 #endif
1472
1473 #ifdef CONFIG_X86_MCE
1474         void __user                     *mce_vaddr;
1475         __u64                           mce_kflags;
1476         u64                             mce_addr;
1477         __u64                           mce_ripv : 1,
1478                                         mce_whole_page : 1,
1479                                         __mce_reserved : 62;
1480         struct callback_head            mce_kill_me;
1481         int                             mce_count;
1482 #endif
1483
1484 #ifdef CONFIG_KRETPROBES
1485         struct llist_head               kretprobe_instances;
1486 #endif
1487 #ifdef CONFIG_RETHOOK
1488         struct llist_head               rethooks;
1489 #endif
1490
1491 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1492         /*
1493          * If L1D flush is supported on mm context switch
1494          * then we use this callback head to queue kill work
1495          * to kill tasks that are not running on SMT disabled
1496          * cores
1497          */
1498         struct callback_head            l1d_flush_kill;
1499 #endif
1500
1501         /*
1502          * New fields for task_struct should be added above here, so that
1503          * they are included in the randomized portion of task_struct.
1504          */
1505         randomized_struct_fields_end
1506
1507         /* CPU-specific state of this task: */
1508         struct thread_struct            thread;
1509
1510         /*
1511          * WARNING: on x86, 'thread_struct' contains a variable-sized
1512          * structure.  It *MUST* be at the end of 'task_struct'.
1513          *
1514          * Do not put anything below here!
1515          */
1516 };
1517
1518 static inline struct pid *task_pid(struct task_struct *task)
1519 {
1520         return task->thread_pid;
1521 }
1522
1523 /*
1524  * the helpers to get the task's different pids as they are seen
1525  * from various namespaces
1526  *
1527  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1528  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1529  *                     current.
1530  * task_xid_nr_ns()  : id seen from the ns specified;
1531  *
1532  * see also pid_nr() etc in include/linux/pid.h
1533  */
1534 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1535
1536 static inline pid_t task_pid_nr(struct task_struct *tsk)
1537 {
1538         return tsk->pid;
1539 }
1540
1541 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1542 {
1543         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1544 }
1545
1546 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1547 {
1548         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1549 }
1550
1551
1552 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1553 {
1554         return tsk->tgid;
1555 }
1556
1557 /**
1558  * pid_alive - check that a task structure is not stale
1559  * @p: Task structure to be checked.
1560  *
1561  * Test if a process is not yet dead (at most zombie state)
1562  * If pid_alive fails, then pointers within the task structure
1563  * can be stale and must not be dereferenced.
1564  *
1565  * Return: 1 if the process is alive. 0 otherwise.
1566  */
1567 static inline int pid_alive(const struct task_struct *p)
1568 {
1569         return p->thread_pid != NULL;
1570 }
1571
1572 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1573 {
1574         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1575 }
1576
1577 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1578 {
1579         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1580 }
1581
1582
1583 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1584 {
1585         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1586 }
1587
1588 static inline pid_t task_session_vnr(struct task_struct *tsk)
1589 {
1590         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1591 }
1592
1593 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1594 {
1595         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1596 }
1597
1598 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1599 {
1600         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1601 }
1602
1603 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1604 {
1605         pid_t pid = 0;
1606
1607         rcu_read_lock();
1608         if (pid_alive(tsk))
1609                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1610         rcu_read_unlock();
1611
1612         return pid;
1613 }
1614
1615 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1616 {
1617         return task_ppid_nr_ns(tsk, &init_pid_ns);
1618 }
1619
1620 /* Obsolete, do not use: */
1621 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1622 {
1623         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1624 }
1625
1626 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1627 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1628
1629 static inline unsigned int __task_state_index(unsigned int tsk_state,
1630                                               unsigned int tsk_exit_state)
1631 {
1632         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1633
1634         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1635
1636         if (tsk_state == TASK_IDLE)
1637                 state = TASK_REPORT_IDLE;
1638
1639         /*
1640          * We're lying here, but rather than expose a completely new task state
1641          * to userspace, we can make this appear as if the task has gone through
1642          * a regular rt_mutex_lock() call.
1643          */
1644         if (tsk_state == TASK_RTLOCK_WAIT)
1645                 state = TASK_UNINTERRUPTIBLE;
1646
1647         return fls(state);
1648 }
1649
1650 static inline unsigned int task_state_index(struct task_struct *tsk)
1651 {
1652         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1653 }
1654
1655 static inline char task_index_to_char(unsigned int state)
1656 {
1657         static const char state_char[] = "RSDTtXZPI";
1658
1659         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1660
1661         return state_char[state];
1662 }
1663
1664 static inline char task_state_to_char(struct task_struct *tsk)
1665 {
1666         return task_index_to_char(task_state_index(tsk));
1667 }
1668
1669 /**
1670  * is_global_init - check if a task structure is init. Since init
1671  * is free to have sub-threads we need to check tgid.
1672  * @tsk: Task structure to be checked.
1673  *
1674  * Check if a task structure is the first user space task the kernel created.
1675  *
1676  * Return: 1 if the task structure is init. 0 otherwise.
1677  */
1678 static inline int is_global_init(struct task_struct *tsk)
1679 {
1680         return task_tgid_nr(tsk) == 1;
1681 }
1682
1683 extern struct pid *cad_pid;
1684
1685 /*
1686  * Per process flags
1687  */
1688 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1689 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1690 #define PF_EXITING              0x00000004      /* Getting shut down */
1691 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1692 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1693 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1694 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1695 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1696 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1697 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1698 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1699 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1700 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1701 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1702 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1703 #define PF_FROZEN               0x00010000      /* Frozen for system suspend */
1704 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1705 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1706 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1707 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1708                                                  * I am cleaning dirty pages from some other bdi. */
1709 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1710 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1711 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1712 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1713 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1714 #define PF_FREEZER_SKIP         0x40000000      /* Freezer should not count it as freezable */
1715 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1716
1717 /*
1718  * Only the _current_ task can read/write to tsk->flags, but other
1719  * tasks can access tsk->flags in readonly mode for example
1720  * with tsk_used_math (like during threaded core dumping).
1721  * There is however an exception to this rule during ptrace
1722  * or during fork: the ptracer task is allowed to write to the
1723  * child->flags of its traced child (same goes for fork, the parent
1724  * can write to the child->flags), because we're guaranteed the
1725  * child is not running and in turn not changing child->flags
1726  * at the same time the parent does it.
1727  */
1728 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1729 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1730 #define clear_used_math()                       clear_stopped_child_used_math(current)
1731 #define set_used_math()                         set_stopped_child_used_math(current)
1732
1733 #define conditional_stopped_child_used_math(condition, child) \
1734         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1735
1736 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1737
1738 #define copy_to_stopped_child_used_math(child) \
1739         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1740
1741 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1742 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1743 #define used_math()                             tsk_used_math(current)
1744
1745 static __always_inline bool is_percpu_thread(void)
1746 {
1747 #ifdef CONFIG_SMP
1748         return (current->flags & PF_NO_SETAFFINITY) &&
1749                 (current->nr_cpus_allowed  == 1);
1750 #else
1751         return true;
1752 #endif
1753 }
1754
1755 /* Per-process atomic flags. */
1756 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1757 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1758 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1759 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1760 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1761 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1762 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1763 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1764
1765 #define TASK_PFA_TEST(name, func)                                       \
1766         static inline bool task_##func(struct task_struct *p)           \
1767         { return test_bit(PFA_##name, &p->atomic_flags); }
1768
1769 #define TASK_PFA_SET(name, func)                                        \
1770         static inline void task_set_##func(struct task_struct *p)       \
1771         { set_bit(PFA_##name, &p->atomic_flags); }
1772
1773 #define TASK_PFA_CLEAR(name, func)                                      \
1774         static inline void task_clear_##func(struct task_struct *p)     \
1775         { clear_bit(PFA_##name, &p->atomic_flags); }
1776
1777 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1778 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1779
1780 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1781 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1782 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1783
1784 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1785 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1786 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1787
1788 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1789 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1790 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1791
1792 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1793 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1794 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1795
1796 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1797 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1798
1799 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1800 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1801 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1802
1803 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1804 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1805
1806 static inline void
1807 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1808 {
1809         current->flags &= ~flags;
1810         current->flags |= orig_flags & flags;
1811 }
1812
1813 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1814 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed);
1815 #ifdef CONFIG_SMP
1816 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1817 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1818 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1819 extern void release_user_cpus_ptr(struct task_struct *p);
1820 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1821 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1822 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1823 #else
1824 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1825 {
1826 }
1827 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1828 {
1829         if (!cpumask_test_cpu(0, new_mask))
1830                 return -EINVAL;
1831         return 0;
1832 }
1833 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1834 {
1835         if (src->user_cpus_ptr)
1836                 return -EINVAL;
1837         return 0;
1838 }
1839 static inline void release_user_cpus_ptr(struct task_struct *p)
1840 {
1841         WARN_ON(p->user_cpus_ptr);
1842 }
1843
1844 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1845 {
1846         return 0;
1847 }
1848 #endif
1849
1850 extern int yield_to(struct task_struct *p, bool preempt);
1851 extern void set_user_nice(struct task_struct *p, long nice);
1852 extern int task_prio(const struct task_struct *p);
1853
1854 /**
1855  * task_nice - return the nice value of a given task.
1856  * @p: the task in question.
1857  *
1858  * Return: The nice value [ -20 ... 0 ... 19 ].
1859  */
1860 static inline int task_nice(const struct task_struct *p)
1861 {
1862         return PRIO_TO_NICE((p)->static_prio);
1863 }
1864
1865 extern int can_nice(const struct task_struct *p, const int nice);
1866 extern int task_curr(const struct task_struct *p);
1867 extern int idle_cpu(int cpu);
1868 extern int available_idle_cpu(int cpu);
1869 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1870 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1871 extern void sched_set_fifo(struct task_struct *p);
1872 extern void sched_set_fifo_low(struct task_struct *p);
1873 extern void sched_set_normal(struct task_struct *p, int nice);
1874 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1875 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1876 extern struct task_struct *idle_task(int cpu);
1877
1878 /**
1879  * is_idle_task - is the specified task an idle task?
1880  * @p: the task in question.
1881  *
1882  * Return: 1 if @p is an idle task. 0 otherwise.
1883  */
1884 static __always_inline bool is_idle_task(const struct task_struct *p)
1885 {
1886         return !!(p->flags & PF_IDLE);
1887 }
1888
1889 extern struct task_struct *curr_task(int cpu);
1890 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1891
1892 void yield(void);
1893
1894 union thread_union {
1895 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1896         struct task_struct task;
1897 #endif
1898 #ifndef CONFIG_THREAD_INFO_IN_TASK
1899         struct thread_info thread_info;
1900 #endif
1901         unsigned long stack[THREAD_SIZE/sizeof(long)];
1902 };
1903
1904 #ifndef CONFIG_THREAD_INFO_IN_TASK
1905 extern struct thread_info init_thread_info;
1906 #endif
1907
1908 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1909
1910 #ifdef CONFIG_THREAD_INFO_IN_TASK
1911 # define task_thread_info(task) (&(task)->thread_info)
1912 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1913 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1914 #endif
1915
1916 /*
1917  * find a task by one of its numerical ids
1918  *
1919  * find_task_by_pid_ns():
1920  *      finds a task by its pid in the specified namespace
1921  * find_task_by_vpid():
1922  *      finds a task by its virtual pid
1923  *
1924  * see also find_vpid() etc in include/linux/pid.h
1925  */
1926
1927 extern struct task_struct *find_task_by_vpid(pid_t nr);
1928 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1929
1930 /*
1931  * find a task by its virtual pid and get the task struct
1932  */
1933 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1934
1935 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1936 extern int wake_up_process(struct task_struct *tsk);
1937 extern void wake_up_new_task(struct task_struct *tsk);
1938
1939 #ifdef CONFIG_SMP
1940 extern void kick_process(struct task_struct *tsk);
1941 #else
1942 static inline void kick_process(struct task_struct *tsk) { }
1943 #endif
1944
1945 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1946
1947 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1948 {
1949         __set_task_comm(tsk, from, false);
1950 }
1951
1952 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1953 #define get_task_comm(buf, tsk) ({                      \
1954         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1955         __get_task_comm(buf, sizeof(buf), tsk);         \
1956 })
1957
1958 #ifdef CONFIG_SMP
1959 static __always_inline void scheduler_ipi(void)
1960 {
1961         /*
1962          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1963          * TIF_NEED_RESCHED remotely (for the first time) will also send
1964          * this IPI.
1965          */
1966         preempt_fold_need_resched();
1967 }
1968 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1969 #else
1970 static inline void scheduler_ipi(void) { }
1971 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
1972 {
1973         return 1;
1974 }
1975 #endif
1976
1977 /*
1978  * Set thread flags in other task's structures.
1979  * See asm/thread_info.h for TIF_xxxx flags available:
1980  */
1981 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1982 {
1983         set_ti_thread_flag(task_thread_info(tsk), flag);
1984 }
1985
1986 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1987 {
1988         clear_ti_thread_flag(task_thread_info(tsk), flag);
1989 }
1990
1991 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
1992                                           bool value)
1993 {
1994         update_ti_thread_flag(task_thread_info(tsk), flag, value);
1995 }
1996
1997 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
1998 {
1999         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
2000 }
2001
2002 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
2003 {
2004         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
2005 }
2006
2007 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
2008 {
2009         return test_ti_thread_flag(task_thread_info(tsk), flag);
2010 }
2011
2012 static inline void set_tsk_need_resched(struct task_struct *tsk)
2013 {
2014         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2015 }
2016
2017 static inline void clear_tsk_need_resched(struct task_struct *tsk)
2018 {
2019         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2020 }
2021
2022 static inline int test_tsk_need_resched(struct task_struct *tsk)
2023 {
2024         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2025 }
2026
2027 /*
2028  * cond_resched() and cond_resched_lock(): latency reduction via
2029  * explicit rescheduling in places that are safe. The return
2030  * value indicates whether a reschedule was done in fact.
2031  * cond_resched_lock() will drop the spinlock before scheduling,
2032  */
2033 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2034 extern int __cond_resched(void);
2035
2036 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
2037
2038 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2039
2040 static __always_inline int _cond_resched(void)
2041 {
2042         return static_call_mod(cond_resched)();
2043 }
2044
2045 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
2046 extern int dynamic_cond_resched(void);
2047
2048 static __always_inline int _cond_resched(void)
2049 {
2050         return dynamic_cond_resched();
2051 }
2052
2053 #else
2054
2055 static inline int _cond_resched(void)
2056 {
2057         return __cond_resched();
2058 }
2059
2060 #endif /* CONFIG_PREEMPT_DYNAMIC */
2061
2062 #else
2063
2064 static inline int _cond_resched(void) { return 0; }
2065
2066 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2067
2068 #define cond_resched() ({                       \
2069         __might_resched(__FILE__, __LINE__, 0); \
2070         _cond_resched();                        \
2071 })
2072
2073 extern int __cond_resched_lock(spinlock_t *lock);
2074 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2075 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2076
2077 #define MIGHT_RESCHED_RCU_SHIFT         8
2078 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2079
2080 #ifndef CONFIG_PREEMPT_RT
2081 /*
2082  * Non RT kernels have an elevated preempt count due to the held lock,
2083  * but are not allowed to be inside a RCU read side critical section
2084  */
2085 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2086 #else
2087 /*
2088  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2089  * cond_resched*lock() has to take that into account because it checks for
2090  * preempt_count() and rcu_preempt_depth().
2091  */
2092 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2093         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2094 #endif
2095
2096 #define cond_resched_lock(lock) ({                                              \
2097         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2098         __cond_resched_lock(lock);                                              \
2099 })
2100
2101 #define cond_resched_rwlock_read(lock) ({                                       \
2102         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2103         __cond_resched_rwlock_read(lock);                                       \
2104 })
2105
2106 #define cond_resched_rwlock_write(lock) ({                                      \
2107         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2108         __cond_resched_rwlock_write(lock);                                      \
2109 })
2110
2111 static inline void cond_resched_rcu(void)
2112 {
2113 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2114         rcu_read_unlock();
2115         cond_resched();
2116         rcu_read_lock();
2117 #endif
2118 }
2119
2120 #ifdef CONFIG_PREEMPT_DYNAMIC
2121
2122 extern bool preempt_model_none(void);
2123 extern bool preempt_model_voluntary(void);
2124 extern bool preempt_model_full(void);
2125
2126 #else
2127
2128 static inline bool preempt_model_none(void)
2129 {
2130         return IS_ENABLED(CONFIG_PREEMPT_NONE);
2131 }
2132 static inline bool preempt_model_voluntary(void)
2133 {
2134         return IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY);
2135 }
2136 static inline bool preempt_model_full(void)
2137 {
2138         return IS_ENABLED(CONFIG_PREEMPT);
2139 }
2140
2141 #endif
2142
2143 static inline bool preempt_model_rt(void)
2144 {
2145         return IS_ENABLED(CONFIG_PREEMPT_RT);
2146 }
2147
2148 /*
2149  * Does the preemption model allow non-cooperative preemption?
2150  *
2151  * For !CONFIG_PREEMPT_DYNAMIC kernels this is an exact match with
2152  * CONFIG_PREEMPTION; for CONFIG_PREEMPT_DYNAMIC this doesn't work as the
2153  * kernel is *built* with CONFIG_PREEMPTION=y but may run with e.g. the
2154  * PREEMPT_NONE model.
2155  */
2156 static inline bool preempt_model_preemptible(void)
2157 {
2158         return preempt_model_full() || preempt_model_rt();
2159 }
2160
2161 /*
2162  * Does a critical section need to be broken due to another
2163  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2164  * but a general need for low latency)
2165  */
2166 static inline int spin_needbreak(spinlock_t *lock)
2167 {
2168 #ifdef CONFIG_PREEMPTION
2169         return spin_is_contended(lock);
2170 #else
2171         return 0;
2172 #endif
2173 }
2174
2175 /*
2176  * Check if a rwlock is contended.
2177  * Returns non-zero if there is another task waiting on the rwlock.
2178  * Returns zero if the lock is not contended or the system / underlying
2179  * rwlock implementation does not support contention detection.
2180  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2181  * for low latency.
2182  */
2183 static inline int rwlock_needbreak(rwlock_t *lock)
2184 {
2185 #ifdef CONFIG_PREEMPTION
2186         return rwlock_is_contended(lock);
2187 #else
2188         return 0;
2189 #endif
2190 }
2191
2192 static __always_inline bool need_resched(void)
2193 {
2194         return unlikely(tif_need_resched());
2195 }
2196
2197 /*
2198  * Wrappers for p->thread_info->cpu access. No-op on UP.
2199  */
2200 #ifdef CONFIG_SMP
2201
2202 static inline unsigned int task_cpu(const struct task_struct *p)
2203 {
2204         return READ_ONCE(task_thread_info(p)->cpu);
2205 }
2206
2207 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2208
2209 #else
2210
2211 static inline unsigned int task_cpu(const struct task_struct *p)
2212 {
2213         return 0;
2214 }
2215
2216 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2217 {
2218 }
2219
2220 #endif /* CONFIG_SMP */
2221
2222 extern bool sched_task_on_rq(struct task_struct *p);
2223 extern unsigned long get_wchan(struct task_struct *p);
2224
2225 /*
2226  * In order to reduce various lock holder preemption latencies provide an
2227  * interface to see if a vCPU is currently running or not.
2228  *
2229  * This allows us to terminate optimistic spin loops and block, analogous to
2230  * the native optimistic spin heuristic of testing if the lock owner task is
2231  * running or not.
2232  */
2233 #ifndef vcpu_is_preempted
2234 static inline bool vcpu_is_preempted(int cpu)
2235 {
2236         return false;
2237 }
2238 #endif
2239
2240 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2241 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2242
2243 #ifndef TASK_SIZE_OF
2244 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2245 #endif
2246
2247 #ifdef CONFIG_SMP
2248 static inline bool owner_on_cpu(struct task_struct *owner)
2249 {
2250         /*
2251          * As lock holder preemption issue, we both skip spinning if
2252          * task is not on cpu or its cpu is preempted
2253          */
2254         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2255 }
2256
2257 /* Returns effective CPU energy utilization, as seen by the scheduler */
2258 unsigned long sched_cpu_util(int cpu, unsigned long max);
2259 #endif /* CONFIG_SMP */
2260
2261 #ifdef CONFIG_RSEQ
2262
2263 /*
2264  * Map the event mask on the user-space ABI enum rseq_cs_flags
2265  * for direct mask checks.
2266  */
2267 enum rseq_event_mask_bits {
2268         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2269         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2270         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2271 };
2272
2273 enum rseq_event_mask {
2274         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2275         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2276         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2277 };
2278
2279 static inline void rseq_set_notify_resume(struct task_struct *t)
2280 {
2281         if (t->rseq)
2282                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2283 }
2284
2285 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2286
2287 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2288                                              struct pt_regs *regs)
2289 {
2290         if (current->rseq)
2291                 __rseq_handle_notify_resume(ksig, regs);
2292 }
2293
2294 static inline void rseq_signal_deliver(struct ksignal *ksig,
2295                                        struct pt_regs *regs)
2296 {
2297         preempt_disable();
2298         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2299         preempt_enable();
2300         rseq_handle_notify_resume(ksig, regs);
2301 }
2302
2303 /* rseq_preempt() requires preemption to be disabled. */
2304 static inline void rseq_preempt(struct task_struct *t)
2305 {
2306         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2307         rseq_set_notify_resume(t);
2308 }
2309
2310 /* rseq_migrate() requires preemption to be disabled. */
2311 static inline void rseq_migrate(struct task_struct *t)
2312 {
2313         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2314         rseq_set_notify_resume(t);
2315 }
2316
2317 /*
2318  * If parent process has a registered restartable sequences area, the
2319  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2320  */
2321 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2322 {
2323         if (clone_flags & CLONE_VM) {
2324                 t->rseq = NULL;
2325                 t->rseq_sig = 0;
2326                 t->rseq_event_mask = 0;
2327         } else {
2328                 t->rseq = current->rseq;
2329                 t->rseq_sig = current->rseq_sig;
2330                 t->rseq_event_mask = current->rseq_event_mask;
2331         }
2332 }
2333
2334 static inline void rseq_execve(struct task_struct *t)
2335 {
2336         t->rseq = NULL;
2337         t->rseq_sig = 0;
2338         t->rseq_event_mask = 0;
2339 }
2340
2341 #else
2342
2343 static inline void rseq_set_notify_resume(struct task_struct *t)
2344 {
2345 }
2346 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2347                                              struct pt_regs *regs)
2348 {
2349 }
2350 static inline void rseq_signal_deliver(struct ksignal *ksig,
2351                                        struct pt_regs *regs)
2352 {
2353 }
2354 static inline void rseq_preempt(struct task_struct *t)
2355 {
2356 }
2357 static inline void rseq_migrate(struct task_struct *t)
2358 {
2359 }
2360 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2361 {
2362 }
2363 static inline void rseq_execve(struct task_struct *t)
2364 {
2365 }
2366
2367 #endif
2368
2369 #ifdef CONFIG_DEBUG_RSEQ
2370
2371 void rseq_syscall(struct pt_regs *regs);
2372
2373 #else
2374
2375 static inline void rseq_syscall(struct pt_regs *regs)
2376 {
2377 }
2378
2379 #endif
2380
2381 const struct sched_avg *sched_trace_cfs_rq_avg(struct cfs_rq *cfs_rq);
2382 char *sched_trace_cfs_rq_path(struct cfs_rq *cfs_rq, char *str, int len);
2383 int sched_trace_cfs_rq_cpu(struct cfs_rq *cfs_rq);
2384
2385 const struct sched_avg *sched_trace_rq_avg_rt(struct rq *rq);
2386 const struct sched_avg *sched_trace_rq_avg_dl(struct rq *rq);
2387 const struct sched_avg *sched_trace_rq_avg_irq(struct rq *rq);
2388
2389 int sched_trace_rq_cpu(struct rq *rq);
2390 int sched_trace_rq_cpu_capacity(struct rq *rq);
2391 int sched_trace_rq_nr_running(struct rq *rq);
2392
2393 const struct cpumask *sched_trace_rd_span(struct root_domain *rd);
2394
2395 #ifdef CONFIG_SCHED_CORE
2396 extern void sched_core_free(struct task_struct *tsk);
2397 extern void sched_core_fork(struct task_struct *p);
2398 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2399                                 unsigned long uaddr);
2400 #else
2401 static inline void sched_core_free(struct task_struct *tsk) { }
2402 static inline void sched_core_fork(struct task_struct *p) { }
2403 #endif
2404
2405 #endif