include/linux/sched.h

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