include/linux/sched.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _LINUX_SCHED_H
   3 #define _LINUX_SCHED_H
   4
   5 /*
   6  * Define 'struct task_struct' and provide the main scheduler
   7  * APIs (schedule(), wakeup variants, etc.)
   8  */
   9
  10 #include <uapi/linux/sched.h>
  11
  12 #include <asm/current.h>
  13 #include <asm/processor.h>
  14 #include <linux/thread_info.h>
  15 #include <linux/preempt.h>
  16 #include <linux/cpumask_types.h>
  17
  18 #include <linux/cache.h>
  19 #include <linux/irqflags_types.h>
  20 #include <linux/smp_types.h>
  21 #include <linux/pid_types.h>
  22 #include <linux/sem_types.h>
  23 #include <linux/shm.h>
  24 #include <linux/kmsan_types.h>
  25 #include <linux/mutex_types.h>
  26 #include <linux/plist_types.h>
  27 #include <linux/hrtimer_types.h>
  28 #include <linux/timer_types.h>
  29 #include <linux/seccomp_types.h>
  30 #include <linux/nodemask_types.h>
  31 #include <linux/refcount_types.h>
  32 #include <linux/resource.h>
  33 #include <linux/latencytop.h>
  34 #include <linux/sched/prio.h>
  35 #include <linux/sched/types.h>
  36 #include <linux/signal_types.h>
  37 #include <linux/syscall_user_dispatch_types.h>
  38 #include <linux/mm_types_task.h>
  39 #include <linux/netdevice_xmit.h>
  40 #include <linux/task_io_accounting.h>
  41 #include <linux/posix-timers_types.h>
  42 #include <linux/restart_block.h>
  43 #include <uapi/linux/rseq.h>
  44 #include <linux/seqlock_types.h>
  45 #include <linux/kcsan.h>
  46 #include <linux/rv.h>
  47 #include <linux/livepatch_sched.h>
  48 #include <linux/uidgid_types.h>
  49 #include <asm/kmap_size.h>
  50
  51 /* task_struct member predeclarations (sorted alphabetically): */
  52 struct audit_context;
  53 struct bio_list;
  54 struct blk_plug;
  55 struct bpf_local_storage;
  56 struct bpf_run_ctx;
  57 struct bpf_net_context;
  58 struct capture_control;
  59 struct cfs_rq;
  60 struct fs_struct;
  61 struct futex_pi_state;
  62 struct io_context;
  63 struct io_uring_task;
  64 struct mempolicy;
  65 struct nameidata;
  66 struct nsproxy;
  67 struct perf_event_context;
  68 struct pid_namespace;
  69 struct pipe_inode_info;
  70 struct rcu_node;
  71 struct reclaim_state;
  72 struct robust_list_head;
  73 struct root_domain;
  74 struct rq;
  75 struct sched_attr;
  76 struct sched_dl_entity;
  77 struct seq_file;
  78 struct sighand_struct;
  79 struct signal_struct;
  80 struct task_delay_info;
  81 struct task_group;
  82 struct task_struct;
  83 struct user_event_mm;
  84
  85 #include <linux/sched/ext.h>
  86
  87 /*
  88  * Task state bitmask. NOTE! These bits are also
  89  * encoded in fs/proc/array.c: get_task_state().
  90  *
  91  * We have two separate sets of flags: task->__state
  92  * is about runnability, while task->exit_state are
  93  * about the task exiting. Confusing, but this way
  94  * modifying one set can't modify the other one by
  95  * mistake.
  96  */
  97
  98 /* Used in tsk->__state: */
  99 #define TASK_RUNNING                    0x00000000
 100 #define TASK_INTERRUPTIBLE              0x00000001
 101 #define TASK_UNINTERRUPTIBLE            0x00000002
 102 #define __TASK_STOPPED                  0x00000004
 103 #define __TASK_TRACED                   0x00000008
 104 /* Used in tsk->exit_state: */
 105 #define EXIT_DEAD                       0x00000010
 106 #define EXIT_ZOMBIE                     0x00000020
 107 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
 108 /* Used in tsk->__state again: */
 109 #define TASK_PARKED                     0x00000040
 110 #define TASK_DEAD                       0x00000080
 111 #define TASK_WAKEKILL                   0x00000100
 112 #define TASK_WAKING                     0x00000200
 113 #define TASK_NOLOAD                     0x00000400
 114 #define TASK_NEW                        0x00000800
 115 #define TASK_RTLOCK_WAIT                0x00001000
 116 #define TASK_FREEZABLE                  0x00002000
 117 #define __TASK_FREEZABLE_UNSAFE        (0x00004000 * IS_ENABLED(CONFIG_LOCKDEP))
 118 #define TASK_FROZEN                     0x00008000
 119 #define TASK_STATE_MAX                  0x00010000
 120
 121 #define TASK_ANY                        (TASK_STATE_MAX-1)
 122
 123 /*
 124  * DO NOT ADD ANY NEW USERS !
 125  */
 126 #define TASK_FREEZABLE_UNSAFE           (TASK_FREEZABLE | __TASK_FREEZABLE_UNSAFE)
 127
 128 /* Convenience macros for the sake of set_current_state: */
 129 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
 130 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
 131 #define TASK_TRACED                     __TASK_TRACED
 132
 133 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
 134
 135 /* Convenience macros for the sake of wake_up(): */
 136 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
 137
 138 /* get_task_state(): */
 139 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
 140                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
 141                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
 142                                          TASK_PARKED)
 143
 144 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
 145
 146 #define task_is_traced(task)            ((READ_ONCE(task->jobctl) & JOBCTL_TRACED) != 0)
 147 #define task_is_stopped(task)           ((READ_ONCE(task->jobctl) & JOBCTL_STOPPED) != 0)
 148 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->jobctl) & (JOBCTL_STOPPED | JOBCTL_TRACED)) != 0)
 149
 150 /*
 151  * Special states are those that do not use the normal wait-loop pattern. See
 152  * the comment with set_special_state().
 153  */
 154 #define is_special_task_state(state)                                    \
 155         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED |      \
 156                     TASK_DEAD | TASK_FROZEN))
 157
 158 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 159 # define debug_normal_state_change(state_value)                         \
 160         do {                                                            \
 161                 WARN_ON_ONCE(is_special_task_state(state_value));       \
 162                 current->task_state_change = _THIS_IP_;                 \
 163         } while (0)
 164
 165 # define debug_special_state_change(state_value)                        \
 166         do {                                                            \
 167                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
 168                 current->task_state_change = _THIS_IP_;                 \
 169         } while (0)
 170
 171 # define debug_rtlock_wait_set_state()                                  \
 172         do {                                                             \
 173                 current->saved_state_change = current->task_state_change;\
 174                 current->task_state_change = _THIS_IP_;                  \
 175         } while (0)
 176
 177 # define debug_rtlock_wait_restore_state()                              \
 178         do {                                                             \
 179                 current->task_state_change = current->saved_state_change;\
 180         } while (0)
 181
 182 #else
 183 # define debug_normal_state_change(cond)        do { } while (0)
 184 # define debug_special_state_change(cond)       do { } while (0)
 185 # define debug_rtlock_wait_set_state()          do { } while (0)
 186 # define debug_rtlock_wait_restore_state()      do { } while (0)
 187 #endif
 188
 189 /*
 190  * set_current_state() includes a barrier so that the write of current->__state
 191  * is correctly serialised wrt the caller's subsequent test of whether to
 192  * actually sleep:
 193  *
 194  *   for (;;) {
 195  *      set_current_state(TASK_UNINTERRUPTIBLE);
 196  *      if (CONDITION)
 197  *         break;
 198  *
 199  *      schedule();
 200  *   }
 201  *   __set_current_state(TASK_RUNNING);
 202  *
 203  * If the caller does not need such serialisation (because, for instance, the
 204  * CONDITION test and condition change and wakeup are under the same lock) then
 205  * use __set_current_state().
 206  *
 207  * The above is typically ordered against the wakeup, which does:
 208  *
 209  *   CONDITION = 1;
 210  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
 211  *
 212  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
 213  * accessing p->__state.
 214  *
 215  * Wakeup will do: if (@state & p->__state) p->__state = TASK_RUNNING, that is,
 216  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
 217  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
 218  *
 219  * However, with slightly different timing the wakeup TASK_RUNNING store can
 220  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
 221  * a problem either because that will result in one extra go around the loop
 222  * and our @cond test will save the day.
 223  *
 224  * Also see the comments of try_to_wake_up().
 225  */
 226 #define __set_current_state(state_value)                                \
 227         do {                                                            \
 228                 debug_normal_state_change((state_value));               \
 229                 WRITE_ONCE(current->__state, (state_value));            \
 230         } while (0)
 231
 232 #define set_current_state(state_value)                                  \
 233         do {                                                            \
 234                 debug_normal_state_change((state_value));               \
 235                 smp_store_mb(current->__state, (state_value));          \
 236         } while (0)
 237
 238 /*
 239  * set_special_state() should be used for those states when the blocking task
 240  * can not use the regular condition based wait-loop. In that case we must
 241  * serialize against wakeups such that any possible in-flight TASK_RUNNING
 242  * stores will not collide with our state change.
 243  */
 244 #define set_special_state(state_value)                                  \
 245         do {                                                            \
 246                 unsigned long flags; /* may shadow */                   \
 247                                                                         \
 248                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
 249                 debug_special_state_change((state_value));              \
 250                 WRITE_ONCE(current->__state, (state_value));            \
 251                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
 252         } while (0)
 253
 254 /*
 255  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
 256  *
 257  * RT's spin/rwlock substitutions are state preserving. The state of the
 258  * task when blocking on the lock is saved in task_struct::saved_state and
 259  * restored after the lock has been acquired.  These operations are
 260  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
 261  * lock related wakeups while the task is blocked on the lock are
 262  * redirected to operate on task_struct::saved_state to ensure that these
 263  * are not dropped. On restore task_struct::saved_state is set to
 264  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
 265  *
 266  * The lock operation looks like this:
 267  *
 268  *      current_save_and_set_rtlock_wait_state();
 269  *      for (;;) {
 270  *              if (try_lock())
 271  *                      break;
 272  *              raw_spin_unlock_irq(&lock->wait_lock);
 273  *              schedule_rtlock();
 274  *              raw_spin_lock_irq(&lock->wait_lock);
 275  *              set_current_state(TASK_RTLOCK_WAIT);
 276  *      }
 277  *      current_restore_rtlock_saved_state();
 278  */
 279 #define current_save_and_set_rtlock_wait_state()                        \
 280         do {                                                            \
 281                 lockdep_assert_irqs_disabled();                         \
 282                 raw_spin_lock(&current->pi_lock);                       \
 283                 current->saved_state = current->__state;                \
 284                 debug_rtlock_wait_set_state();                          \
 285                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
 286                 raw_spin_unlock(&current->pi_lock);                     \
 287         } while (0);
 288
 289 #define current_restore_rtlock_saved_state()                            \
 290         do {                                                            \
 291                 lockdep_assert_irqs_disabled();                         \
 292                 raw_spin_lock(&current->pi_lock);                       \
 293                 debug_rtlock_wait_restore_state();                      \
 294                 WRITE_ONCE(current->__state, current->saved_state);     \
 295                 current->saved_state = TASK_RUNNING;                    \
 296                 raw_spin_unlock(&current->pi_lock);                     \
 297         } while (0);
 298
 299 #define get_current_state()     READ_ONCE(current->__state)
 300
 301 /*
 302  * Define the task command name length as enum, then it can be visible to
 303  * BPF programs.
 304  */
 305 enum {
 306         TASK_COMM_LEN = 16,
 307 };
 308
 309 extern void sched_tick(void);
 310
 311 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
 312
 313 extern long schedule_timeout(long timeout);
 314 extern long schedule_timeout_interruptible(long timeout);
 315 extern long schedule_timeout_killable(long timeout);
 316 extern long schedule_timeout_uninterruptible(long timeout);
 317 extern long schedule_timeout_idle(long timeout);
 318 asmlinkage void schedule(void);
 319 extern void schedule_preempt_disabled(void);
 320 asmlinkage void preempt_schedule_irq(void);
 321 #ifdef CONFIG_PREEMPT_RT
 322  extern void schedule_rtlock(void);
 323 #endif
 324
 325 extern int __must_check io_schedule_prepare(void);
 326 extern void io_schedule_finish(int token);
 327 extern long io_schedule_timeout(long timeout);
 328 extern void io_schedule(void);
 329
 330 /**
 331  * struct prev_cputime - snapshot of system and user cputime
 332  * @utime: time spent in user mode
 333  * @stime: time spent in system mode
 334  * @lock: protects the above two fields
 335  *
 336  * Stores previous user/system time values such that we can guarantee
 337  * monotonicity.
 338  */
 339 struct prev_cputime {
 340 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 341         u64                             utime;
 342         u64                             stime;
 343         raw_spinlock_t                  lock;
 344 #endif
 345 };
 346
 347 enum vtime_state {
 348         /* Task is sleeping or running in a CPU with VTIME inactive: */
 349         VTIME_INACTIVE = 0,
 350         /* Task is idle */
 351         VTIME_IDLE,
 352         /* Task runs in kernelspace in a CPU with VTIME active: */
 353         VTIME_SYS,
 354         /* Task runs in userspace in a CPU with VTIME active: */
 355         VTIME_USER,
 356         /* Task runs as guests in a CPU with VTIME active: */
 357         VTIME_GUEST,
 358 };
 359
 360 struct vtime {
 361         seqcount_t              seqcount;
 362         unsigned long long      starttime;
 363         enum vtime_state        state;
 364         unsigned int            cpu;
 365         u64                     utime;
 366         u64                     stime;
 367         u64                     gtime;
 368 };
 369
 370 /*
 371  * Utilization clamp constraints.
 372  * @UCLAMP_MIN: Minimum utilization
 373  * @UCLAMP_MAX: Maximum utilization
 374  * @UCLAMP_CNT: Utilization clamp constraints count
 375  */
 376 enum uclamp_id {
 377         UCLAMP_MIN = 0,
 378         UCLAMP_MAX,
 379         UCLAMP_CNT
 380 };
 381
 382 #ifdef CONFIG_SMP
 383 extern struct root_domain def_root_domain;
 384 extern struct mutex sched_domains_mutex;
 385 #endif
 386
 387 struct sched_param {
 388         int sched_priority;
 389 };
 390
 391 struct sched_info {
 392 #ifdef CONFIG_SCHED_INFO
 393         /* Cumulative counters: */
 394
 395         /* # of times we have run on this CPU: */
 396         unsigned long                   pcount;
 397
 398         /* Time spent waiting on a runqueue: */
 399         unsigned long long              run_delay;
 400
 401         /* Timestamps: */
 402
 403         /* When did we last run on a CPU? */
 404         unsigned long long              last_arrival;
 405
 406         /* When were we last queued to run? */
 407         unsigned long long              last_queued;
 408
 409 #endif /* CONFIG_SCHED_INFO */
 410 };
 411
 412 /*
 413  * Integer metrics need fixed point arithmetic, e.g., sched/fair
 414  * has a few: load, load_avg, util_avg, freq, and capacity.
 415  *
 416  * We define a basic fixed point arithmetic range, and then formalize
 417  * all these metrics based on that basic range.
 418  */
 419 # define SCHED_FIXEDPOINT_SHIFT         10
 420 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
 421
 422 /* Increase resolution of cpu_capacity calculations */
 423 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
 424 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
 425
 426 struct load_weight {
 427         unsigned long                   weight;
 428         u32                             inv_weight;
 429 };
 430
 431 /*
 432  * The load/runnable/util_avg accumulates an infinite geometric series
 433  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
 434  *
 435  * [load_avg definition]
 436  *
 437  *   load_avg = runnable% * scale_load_down(load)
 438  *
 439  * [runnable_avg definition]
 440  *
 441  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
 442  *
 443  * [util_avg definition]
 444  *
 445  *   util_avg = running% * SCHED_CAPACITY_SCALE
 446  *
 447  * where runnable% is the time ratio that a sched_entity is runnable and
 448  * running% the time ratio that a sched_entity is running.
 449  *
 450  * For cfs_rq, they are the aggregated values of all runnable and blocked
 451  * sched_entities.
 452  *
 453  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
 454  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
 455  * for computing those signals (see update_rq_clock_pelt())
 456  *
 457  * N.B., the above ratios (runnable% and running%) themselves are in the
 458  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
 459  * to as large a range as necessary. This is for example reflected by
 460  * util_avg's SCHED_CAPACITY_SCALE.
 461  *
 462  * [Overflow issue]
 463  *
 464  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
 465  * with the highest load (=88761), always runnable on a single cfs_rq,
 466  * and should not overflow as the number already hits PID_MAX_LIMIT.
 467  *
 468  * For all other cases (including 32-bit kernels), struct load_weight's
 469  * weight will overflow first before we do, because:
 470  *
 471  *    Max(load_avg) <= Max(load.weight)
 472  *
 473  * Then it is the load_weight's responsibility to consider overflow
 474  * issues.
 475  */
 476 struct sched_avg {
 477         u64                             last_update_time;
 478         u64                             load_sum;
 479         u64                             runnable_sum;
 480         u32                             util_sum;
 481         u32                             period_contrib;
 482         unsigned long                   load_avg;
 483         unsigned long                   runnable_avg;
 484         unsigned long                   util_avg;
 485         unsigned int                    util_est;
 486 } ____cacheline_aligned;
 487
 488 /*
 489  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
 490  * updates. When a task is dequeued, its util_est should not be updated if its
 491  * util_avg has not been updated in the meantime.
 492  * This information is mapped into the MSB bit of util_est at dequeue time.
 493  * Since max value of util_est for a task is 1024 (PELT util_avg for a task)
 494  * it is safe to use MSB.
 495  */
 496 #define UTIL_EST_WEIGHT_SHIFT           2
 497 #define UTIL_AVG_UNCHANGED              0x80000000
 498
 499 struct sched_statistics {
 500 #ifdef CONFIG_SCHEDSTATS
 501         u64                             wait_start;
 502         u64                             wait_max;
 503         u64                             wait_count;
 504         u64                             wait_sum;
 505         u64                             iowait_count;
 506         u64                             iowait_sum;
 507
 508         u64                             sleep_start;
 509         u64                             sleep_max;
 510         s64                             sum_sleep_runtime;
 511
 512         u64                             block_start;
 513         u64                             block_max;
 514         s64                             sum_block_runtime;
 515
 516         s64                             exec_max;
 517         u64                             slice_max;
 518
 519         u64                             nr_migrations_cold;
 520         u64                             nr_failed_migrations_affine;
 521         u64                             nr_failed_migrations_running;
 522         u64                             nr_failed_migrations_hot;
 523         u64                             nr_forced_migrations;
 524
 525         u64                             nr_wakeups;
 526         u64                             nr_wakeups_sync;
 527         u64                             nr_wakeups_migrate;
 528         u64                             nr_wakeups_local;
 529         u64                             nr_wakeups_remote;
 530         u64                             nr_wakeups_affine;
 531         u64                             nr_wakeups_affine_attempts;
 532         u64                             nr_wakeups_passive;
 533         u64                             nr_wakeups_idle;
 534
 535 #ifdef CONFIG_SCHED_CORE
 536         u64                             core_forceidle_sum;
 537 #endif
 538 #endif /* CONFIG_SCHEDSTATS */
 539 } ____cacheline_aligned;
 540
 541 struct sched_entity {
 542         /* For load-balancing: */
 543         struct load_weight              load;
 544         struct rb_node                  run_node;
 545         u64                             deadline;
 546         u64                             min_vruntime;
 547         u64                             min_slice;
 548
 549         struct list_head                group_node;
 550         unsigned char                   on_rq;
 551         unsigned char                   sched_delayed;
 552         unsigned char                   rel_deadline;
 553         unsigned char                   custom_slice;
 554                                         /* hole */
 555
 556         u64                             exec_start;
 557         u64                             sum_exec_runtime;
 558         u64                             prev_sum_exec_runtime;
 559         u64                             vruntime;
 560         s64                             vlag;
 561         u64                             slice;
 562
 563         u64                             nr_migrations;
 564
 565 #ifdef CONFIG_FAIR_GROUP_SCHED
 566         int                             depth;
 567         struct sched_entity             *parent;
 568         /* rq on which this entity is (to be) queued: */
 569         struct cfs_rq                   *cfs_rq;
 570         /* rq "owned" by this entity/group: */
 571         struct cfs_rq                   *my_q;
 572         /* cached value of my_q->h_nr_running */
 573         unsigned long                   runnable_weight;
 574 #endif
 575
 576 #ifdef CONFIG_SMP
 577         /*
 578          * Per entity load average tracking.
 579          *
 580          * Put into separate cache line so it does not
 581          * collide with read-mostly values above.
 582          */
 583         struct sched_avg                avg;
 584 #endif
 585 };
 586
 587 struct sched_rt_entity {
 588         struct list_head                run_list;
 589         unsigned long                   timeout;
 590         unsigned long                   watchdog_stamp;
 591         unsigned int                    time_slice;
 592         unsigned short                  on_rq;
 593         unsigned short                  on_list;
 594
 595         struct sched_rt_entity          *back;
 596 #ifdef CONFIG_RT_GROUP_SCHED
 597         struct sched_rt_entity          *parent;
 598         /* rq on which this entity is (to be) queued: */
 599         struct rt_rq                    *rt_rq;
 600         /* rq "owned" by this entity/group: */
 601         struct rt_rq                    *my_q;
 602 #endif
 603 } __randomize_layout;
 604
 605 typedef bool (*dl_server_has_tasks_f)(struct sched_dl_entity *);
 606 typedef struct task_struct *(*dl_server_pick_f)(struct sched_dl_entity *);
 607
 608 struct sched_dl_entity {
 609         struct rb_node                  rb_node;
 610
 611         /*
 612          * Original scheduling parameters. Copied here from sched_attr
 613          * during sched_setattr(), they will remain the same until
 614          * the next sched_setattr().
 615          */
 616         u64                             dl_runtime;     /* Maximum runtime for each instance    */
 617         u64                             dl_deadline;    /* Relative deadline of each instance   */
 618         u64                             dl_period;      /* Separation of two instances (period) */
 619         u64                             dl_bw;          /* dl_runtime / dl_period               */
 620         u64                             dl_density;     /* dl_runtime / dl_deadline             */
 621
 622         /*
 623          * Actual scheduling parameters. Initialized with the values above,
 624          * they are continuously updated during task execution. Note that
 625          * the remaining runtime could be < 0 in case we are in overrun.
 626          */
 627         s64                             runtime;        /* Remaining runtime for this instance  */
 628         u64                             deadline;       /* Absolute deadline for this instance  */
 629         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
 630
 631         /*
 632          * Some bool flags:
 633          *
 634          * @dl_throttled tells if we exhausted the runtime. If so, the
 635          * task has to wait for a replenishment to be performed at the
 636          * next firing of dl_timer.
 637          *
 638          * @dl_yielded tells if task gave up the CPU before consuming
 639          * all its available runtime during the last job.
 640          *
 641          * @dl_non_contending tells if the task is inactive while still
 642          * contributing to the active utilization. In other words, it
 643          * indicates if the inactive timer has been armed and its handler
 644          * has not been executed yet. This flag is useful to avoid race
 645          * conditions between the inactive timer handler and the wakeup
 646          * code.
 647          *
 648          * @dl_overrun tells if the task asked to be informed about runtime
 649          * overruns.
 650          *
 651          * @dl_server tells if this is a server entity.
 652          *
 653          * @dl_defer tells if this is a deferred or regular server. For
 654          * now only defer server exists.
 655          *
 656          * @dl_defer_armed tells if the deferrable server is waiting
 657          * for the replenishment timer to activate it.
 658          *
 659          * @dl_server_active tells if the dlserver is active(started).
 660          * dlserver is started on first cfs enqueue on an idle runqueue
 661          * and is stopped when a dequeue results in 0 cfs tasks on the
 662          * runqueue. In other words, dlserver is active only when cpu's
 663          * runqueue has atleast one cfs task.
 664          *
 665          * @dl_defer_running tells if the deferrable server is actually
 666          * running, skipping the defer phase.
 667          */
 668         unsigned int                    dl_throttled      : 1;
 669         unsigned int                    dl_yielded        : 1;
 670         unsigned int                    dl_non_contending : 1;
 671         unsigned int                    dl_overrun        : 1;
 672         unsigned int                    dl_server         : 1;
 673         unsigned int                    dl_server_active  : 1;
 674         unsigned int                    dl_defer          : 1;
 675         unsigned int                    dl_defer_armed    : 1;
 676         unsigned int                    dl_defer_running  : 1;
 677
 678         /*
 679          * Bandwidth enforcement timer. Each -deadline task has its
 680          * own bandwidth to be enforced, thus we need one timer per task.
 681          */
 682         struct hrtimer                  dl_timer;
 683
 684         /*
 685          * Inactive timer, responsible for decreasing the active utilization
 686          * at the "0-lag time". When a -deadline task blocks, it contributes
 687          * to GRUB's active utilization until the "0-lag time", hence a
 688          * timer is needed to decrease the active utilization at the correct
 689          * time.
 690          */
 691         struct hrtimer                  inactive_timer;
 692
 693         /*
 694          * Bits for DL-server functionality. Also see the comment near
 695          * dl_server_update().
 696          *
 697          * @rq the runqueue this server is for
 698          *
 699          * @server_has_tasks() returns true if @server_pick return a
 700          * runnable task.
 701          */
 702         struct rq                       *rq;
 703         dl_server_has_tasks_f           server_has_tasks;
 704         dl_server_pick_f                server_pick_task;
 705
 706 #ifdef CONFIG_RT_MUTEXES
 707         /*
 708          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
 709          * pi_se points to the donor, otherwise points to the dl_se it belongs
 710          * to (the original one/itself).
 711          */
 712         struct sched_dl_entity *pi_se;
 713 #endif
 714 };
 715
 716 #ifdef CONFIG_UCLAMP_TASK
 717 /* Number of utilization clamp buckets (shorter alias) */
 718 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
 719
 720 /*
 721  * Utilization clamp for a scheduling entity
 722  * @value:              clamp value "assigned" to a se
 723  * @bucket_id:          bucket index corresponding to the "assigned" value
 724  * @active:             the se is currently refcounted in a rq's bucket
 725  * @user_defined:       the requested clamp value comes from user-space
 726  *
 727  * The bucket_id is the index of the clamp bucket matching the clamp value
 728  * which is pre-computed and stored to avoid expensive integer divisions from
 729  * the fast path.
 730  *
 731  * The active bit is set whenever a task has got an "effective" value assigned,
 732  * which can be different from the clamp value "requested" from user-space.
 733  * This allows to know a task is refcounted in the rq's bucket corresponding
 734  * to the "effective" bucket_id.
 735  *
 736  * The user_defined bit is set whenever a task has got a task-specific clamp
 737  * value requested from userspace, i.e. the system defaults apply to this task
 738  * just as a restriction. This allows to relax default clamps when a less
 739  * restrictive task-specific value has been requested, thus allowing to
 740  * implement a "nice" semantic. For example, a task running with a 20%
 741  * default boost can still drop its own boosting to 0%.
 742  */
 743 struct uclamp_se {
 744         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
 745         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
 746         unsigned int active             : 1;
 747         unsigned int user_defined       : 1;
 748 };
 749 #endif /* CONFIG_UCLAMP_TASK */
 750
 751 union rcu_special {
 752         struct {
 753                 u8                      blocked;
 754                 u8                      need_qs;
 755                 u8                      exp_hint; /* Hint for performance. */
 756                 u8                      need_mb; /* Readers need smp_mb(). */
 757         } b; /* Bits. */
 758         u32 s; /* Set of bits. */
 759 };
 760
 761 enum perf_event_task_context {
 762         perf_invalid_context = -1,
 763         perf_hw_context = 0,
 764         perf_sw_context,
 765         perf_nr_task_contexts,
 766 };
 767
 768 /*
 769  * Number of contexts where an event can trigger:
 770  *      task, softirq, hardirq, nmi.
 771  */
 772 #define PERF_NR_CONTEXTS        4
 773
 774 struct wake_q_node {
 775         struct wake_q_node *next;
 776 };
 777
 778 struct kmap_ctrl {
 779 #ifdef CONFIG_KMAP_LOCAL
 780         int                             idx;
 781         pte_t                           pteval[KM_MAX_IDX];
 782 #endif
 783 };
 784
 785 struct task_struct {
 786 #ifdef CONFIG_THREAD_INFO_IN_TASK
 787         /*
 788          * For reasons of header soup (see current_thread_info()), this
 789          * must be the first element of task_struct.
 790          */
 791         struct thread_info              thread_info;
 792 #endif
 793         unsigned int                    __state;
 794
 795         /* saved state for "spinlock sleepers" */
 796         unsigned int                    saved_state;
 797
 798         /*
 799          * This begins the randomizable portion of task_struct. Only
 800          * scheduling-critical items should be added above here.
 801          */
 802         randomized_struct_fields_start
 803
 804         void                            *stack;
 805         refcount_t                      usage;
 806         /* Per task flags (PF_*), defined further below: */
 807         unsigned int                    flags;
 808         unsigned int                    ptrace;
 809
 810 #ifdef CONFIG_MEM_ALLOC_PROFILING
 811         struct alloc_tag                *alloc_tag;
 812 #endif
 813
 814 #ifdef CONFIG_SMP
 815         int                             on_cpu;
 816         struct __call_single_node       wake_entry;
 817         unsigned int                    wakee_flips;
 818         unsigned long                   wakee_flip_decay_ts;
 819         struct task_struct              *last_wakee;
 820
 821         /*
 822          * recent_used_cpu is initially set as the last CPU used by a task
 823          * that wakes affine another task. Waker/wakee relationships can
 824          * push tasks around a CPU where each wakeup moves to the next one.
 825          * Tracking a recently used CPU allows a quick search for a recently
 826          * used CPU that may be idle.
 827          */
 828         int                             recent_used_cpu;
 829         int                             wake_cpu;
 830 #endif
 831         int                             on_rq;
 832
 833         int                             prio;
 834         int                             static_prio;
 835         int                             normal_prio;
 836         unsigned int                    rt_priority;
 837
 838         struct sched_entity             se;
 839         struct sched_rt_entity          rt;
 840         struct sched_dl_entity          dl;
 841         struct sched_dl_entity          *dl_server;
 842 #ifdef CONFIG_SCHED_CLASS_EXT
 843         struct sched_ext_entity         scx;
 844 #endif
 845         const struct sched_class        *sched_class;
 846
 847 #ifdef CONFIG_SCHED_CORE
 848         struct rb_node                  core_node;
 849         unsigned long                   core_cookie;
 850         unsigned int                    core_occupation;
 851 #endif
 852
 853 #ifdef CONFIG_CGROUP_SCHED
 854         struct task_group               *sched_task_group;
 855 #endif
 856
 857
 858 #ifdef CONFIG_UCLAMP_TASK
 859         /*
 860          * Clamp values requested for a scheduling entity.
 861          * Must be updated with task_rq_lock() held.
 862          */
 863         struct uclamp_se                uclamp_req[UCLAMP_CNT];
 864         /*
 865          * Effective clamp values used for a scheduling entity.
 866          * Must be updated with task_rq_lock() held.
 867          */
 868         struct uclamp_se                uclamp[UCLAMP_CNT];
 869 #endif
 870
 871         struct sched_statistics         stats;
 872
 873 #ifdef CONFIG_PREEMPT_NOTIFIERS
 874         /* List of struct preempt_notifier: */
 875         struct hlist_head               preempt_notifiers;
 876 #endif
 877
 878 #ifdef CONFIG_BLK_DEV_IO_TRACE
 879         unsigned int                    btrace_seq;
 880 #endif
 881
 882         unsigned int                    policy;
 883         unsigned long                   max_allowed_capacity;
 884         int                             nr_cpus_allowed;
 885         const cpumask_t                 *cpus_ptr;
 886         cpumask_t                       *user_cpus_ptr;
 887         cpumask_t                       cpus_mask;
 888         void                            *migration_pending;
 889 #ifdef CONFIG_SMP
 890         unsigned short                  migration_disabled;
 891 #endif
 892         unsigned short                  migration_flags;
 893
 894 #ifdef CONFIG_PREEMPT_RCU
 895         int                             rcu_read_lock_nesting;
 896         union rcu_special               rcu_read_unlock_special;
 897         struct list_head                rcu_node_entry;
 898         struct rcu_node                 *rcu_blocked_node;
 899 #endif /* #ifdef CONFIG_PREEMPT_RCU */
 900
 901 #ifdef CONFIG_TASKS_RCU
 902         unsigned long                   rcu_tasks_nvcsw;
 903         u8                              rcu_tasks_holdout;
 904         u8                              rcu_tasks_idx;
 905         int                             rcu_tasks_idle_cpu;
 906         struct list_head                rcu_tasks_holdout_list;
 907         int                             rcu_tasks_exit_cpu;
 908         struct list_head                rcu_tasks_exit_list;
 909 #endif /* #ifdef CONFIG_TASKS_RCU */
 910
 911 #ifdef CONFIG_TASKS_TRACE_RCU
 912         int                             trc_reader_nesting;
 913         int                             trc_ipi_to_cpu;
 914         union rcu_special               trc_reader_special;
 915         struct list_head                trc_holdout_list;
 916         struct list_head                trc_blkd_node;
 917         int                             trc_blkd_cpu;
 918 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
 919
 920         struct sched_info               sched_info;
 921
 922         struct list_head                tasks;
 923 #ifdef CONFIG_SMP
 924         struct plist_node               pushable_tasks;
 925         struct rb_node                  pushable_dl_tasks;
 926 #endif
 927
 928         struct mm_struct                *mm;
 929         struct mm_struct                *active_mm;
 930         struct address_space            *faults_disabled_mapping;
 931
 932         int                             exit_state;
 933         int                             exit_code;
 934         int                             exit_signal;
 935         /* The signal sent when the parent dies: */
 936         int                             pdeath_signal;
 937         /* JOBCTL_*, siglock protected: */
 938         unsigned long                   jobctl;
 939
 940         /* Used for emulating ABI behavior of previous Linux versions: */
 941         unsigned int                    personality;
 942
 943         /* Scheduler bits, serialized by scheduler locks: */
 944         unsigned                        sched_reset_on_fork:1;
 945         unsigned                        sched_contributes_to_load:1;
 946         unsigned                        sched_migrated:1;
 947
 948         /* Force alignment to the next boundary: */
 949         unsigned                        :0;
 950
 951         /* Unserialized, strictly 'current' */
 952
 953         /*
 954          * This field must not be in the scheduler word above due to wakelist
 955          * queueing no longer being serialized by p->on_cpu. However:
 956          *
 957          * p->XXX = X;                  ttwu()
 958          * schedule()                     if (p->on_rq && ..) // false
 959          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
 960          *   deactivate_task()                ttwu_queue_wakelist())
 961          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
 962          *
 963          * guarantees all stores of 'current' are visible before
 964          * ->sched_remote_wakeup gets used, so it can be in this word.
 965          */
 966         unsigned                        sched_remote_wakeup:1;
 967 #ifdef CONFIG_RT_MUTEXES
 968         unsigned                        sched_rt_mutex:1;
 969 #endif
 970
 971         /* Bit to tell TOMOYO we're in execve(): */
 972         unsigned                        in_execve:1;
 973         unsigned                        in_iowait:1;
 974 #ifndef TIF_RESTORE_SIGMASK
 975         unsigned                        restore_sigmask:1;
 976 #endif
 977 #ifdef CONFIG_MEMCG_V1
 978         unsigned                        in_user_fault:1;
 979 #endif
 980 #ifdef CONFIG_LRU_GEN
 981         /* whether the LRU algorithm may apply to this access */
 982         unsigned                        in_lru_fault:1;
 983 #endif
 984 #ifdef CONFIG_COMPAT_BRK
 985         unsigned                        brk_randomized:1;
 986 #endif
 987 #ifdef CONFIG_CGROUPS
 988         /* disallow userland-initiated cgroup migration */
 989         unsigned                        no_cgroup_migration:1;
 990         /* task is frozen/stopped (used by the cgroup freezer) */
 991         unsigned                        frozen:1;
 992 #endif
 993 #ifdef CONFIG_BLK_CGROUP
 994         unsigned                        use_memdelay:1;
 995 #endif
 996 #ifdef CONFIG_PSI
 997         /* Stalled due to lack of memory */
 998         unsigned                        in_memstall:1;
 999 #endif
1000 #ifdef CONFIG_PAGE_OWNER
1001         /* Used by page_owner=on to detect recursion in page tracking. */
1002         unsigned                        in_page_owner:1;
1003 #endif
1004 #ifdef CONFIG_EVENTFD
1005         /* Recursion prevention for eventfd_signal() */
1006         unsigned                        in_eventfd:1;
1007 #endif
1008 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1009         unsigned                        pasid_activated:1;
1010 #endif
1011 #ifdef CONFIG_X86_BUS_LOCK_DETECT
1012         unsigned                        reported_split_lock:1;
1013 #endif
1014 #ifdef CONFIG_TASK_DELAY_ACCT
1015         /* delay due to memory thrashing */
1016         unsigned                        in_thrashing:1;
1017 #endif
1018 #ifdef CONFIG_PREEMPT_RT
1019         struct netdev_xmit              net_xmit;
1020 #endif
1021         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
1022
1023         struct restart_block            restart_block;
1024
1025         pid_t                           pid;
1026         pid_t                           tgid;
1027
1028 #ifdef CONFIG_STACKPROTECTOR
1029         /* Canary value for the -fstack-protector GCC feature: */
1030         unsigned long                   stack_canary;
1031 #endif
1032         /*
1033          * Pointers to the (original) parent process, youngest child, younger sibling,
1034          * older sibling, respectively.  (p->father can be replaced with
1035          * p->real_parent->pid)
1036          */
1037
1038         /* Real parent process: */
1039         struct task_struct __rcu        *real_parent;
1040
1041         /* Recipient of SIGCHLD, wait4() reports: */
1042         struct task_struct __rcu        *parent;
1043
1044         /*
1045          * Children/sibling form the list of natural children:
1046          */
1047         struct list_head                children;
1048         struct list_head                sibling;
1049         struct task_struct              *group_leader;
1050
1051         /*
1052          * 'ptraced' is the list of tasks this task is using ptrace() on.
1053          *
1054          * This includes both natural children and PTRACE_ATTACH targets.
1055          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
1056          */
1057         struct list_head                ptraced;
1058         struct list_head                ptrace_entry;
1059
1060         /* PID/PID hash table linkage. */
1061         struct pid                      *thread_pid;
1062         struct hlist_node               pid_links[PIDTYPE_MAX];
1063         struct list_head                thread_node;
1064
1065         struct completion               *vfork_done;
1066
1067         /* CLONE_CHILD_SETTID: */
1068         int __user                      *set_child_tid;
1069
1070         /* CLONE_CHILD_CLEARTID: */
1071         int __user                      *clear_child_tid;
1072
1073         /* PF_KTHREAD | PF_IO_WORKER */
1074         void                            *worker_private;
1075
1076         u64                             utime;
1077         u64                             stime;
1078 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1079         u64                             utimescaled;
1080         u64                             stimescaled;
1081 #endif
1082         u64                             gtime;
1083         struct prev_cputime             prev_cputime;
1084 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1085         struct vtime                    vtime;
1086 #endif
1087
1088 #ifdef CONFIG_NO_HZ_FULL
1089         atomic_t                        tick_dep_mask;
1090 #endif
1091         /* Context switch counts: */
1092         unsigned long                   nvcsw;
1093         unsigned long                   nivcsw;
1094
1095         /* Monotonic time in nsecs: */
1096         u64                             start_time;
1097
1098         /* Boot based time in nsecs: */
1099         u64                             start_boottime;
1100
1101         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1102         unsigned long                   min_flt;
1103         unsigned long                   maj_flt;
1104
1105         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1106         struct posix_cputimers          posix_cputimers;
1107
1108 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1109         struct posix_cputimers_work     posix_cputimers_work;
1110 #endif
1111
1112         /* Process credentials: */
1113
1114         /* Tracer's credentials at attach: */
1115         const struct cred __rcu         *ptracer_cred;
1116
1117         /* Objective and real subjective task credentials (COW): */
1118         const struct cred __rcu         *real_cred;
1119
1120         /* Effective (overridable) subjective task credentials (COW): */
1121         const struct cred __rcu         *cred;
1122
1123 #ifdef CONFIG_KEYS
1124         /* Cached requested key. */
1125         struct key                      *cached_requested_key;
1126 #endif
1127
1128         /*
1129          * executable name, excluding path.
1130          *
1131          * - normally initialized begin_new_exec()
1132          * - set it with set_task_comm()
1133          *   - strscpy_pad() to ensure it is always NUL-terminated and
1134          *     zero-padded
1135          *   - task_lock() to ensure the operation is atomic and the name is
1136          *     fully updated.
1137          */
1138         char                            comm[TASK_COMM_LEN];
1139
1140         struct nameidata                *nameidata;
1141
1142 #ifdef CONFIG_SYSVIPC
1143         struct sysv_sem                 sysvsem;
1144         struct sysv_shm                 sysvshm;
1145 #endif
1146 #ifdef CONFIG_DETECT_HUNG_TASK
1147         unsigned long                   last_switch_count;
1148         unsigned long                   last_switch_time;
1149 #endif
1150         /* Filesystem information: */
1151         struct fs_struct                *fs;
1152
1153         /* Open file information: */
1154         struct files_struct             *files;
1155
1156 #ifdef CONFIG_IO_URING
1157         struct io_uring_task            *io_uring;
1158 #endif
1159
1160         /* Namespaces: */
1161         struct nsproxy                  *nsproxy;
1162
1163         /* Signal handlers: */
1164         struct signal_struct            *signal;
1165         struct sighand_struct __rcu             *sighand;
1166         sigset_t                        blocked;
1167         sigset_t                        real_blocked;
1168         /* Restored if set_restore_sigmask() was used: */
1169         sigset_t                        saved_sigmask;
1170         struct sigpending               pending;
1171         unsigned long                   sas_ss_sp;
1172         size_t                          sas_ss_size;
1173         unsigned int                    sas_ss_flags;
1174
1175         struct callback_head            *task_works;
1176
1177 #ifdef CONFIG_AUDIT
1178 #ifdef CONFIG_AUDITSYSCALL
1179         struct audit_context            *audit_context;
1180 #endif
1181         kuid_t                          loginuid;
1182         unsigned int                    sessionid;
1183 #endif
1184         struct seccomp                  seccomp;
1185         struct syscall_user_dispatch    syscall_dispatch;
1186
1187         /* Thread group tracking: */
1188         u64                             parent_exec_id;
1189         u64                             self_exec_id;
1190
1191         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1192         spinlock_t                      alloc_lock;
1193
1194         /* Protection of the PI data structures: */
1195         raw_spinlock_t                  pi_lock;
1196
1197         struct wake_q_node              wake_q;
1198
1199 #ifdef CONFIG_RT_MUTEXES
1200         /* PI waiters blocked on a rt_mutex held by this task: */
1201         struct rb_root_cached           pi_waiters;
1202         /* Updated under owner's pi_lock and rq lock */
1203         struct task_struct              *pi_top_task;
1204         /* Deadlock detection and priority inheritance handling: */
1205         struct rt_mutex_waiter          *pi_blocked_on;
1206 #endif
1207
1208 #ifdef CONFIG_DEBUG_MUTEXES
1209         /* Mutex deadlock detection: */
1210         struct mutex_waiter             *blocked_on;
1211 #endif
1212
1213 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1214         int                             non_block_count;
1215 #endif
1216
1217 #ifdef CONFIG_TRACE_IRQFLAGS
1218         struct irqtrace_events          irqtrace;
1219         unsigned int                    hardirq_threaded;
1220         u64                             hardirq_chain_key;
1221         int                             softirqs_enabled;
1222         int                             softirq_context;
1223         int                             irq_config;
1224 #endif
1225 #ifdef CONFIG_PREEMPT_RT
1226         int                             softirq_disable_cnt;
1227 #endif
1228
1229 #ifdef CONFIG_LOCKDEP
1230 # define MAX_LOCK_DEPTH                 48UL
1231         u64                             curr_chain_key;
1232         int                             lockdep_depth;
1233         unsigned int                    lockdep_recursion;
1234         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1235 #endif
1236
1237 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1238         unsigned int                    in_ubsan;
1239 #endif
1240
1241         /* Journalling filesystem info: */
1242         void                            *journal_info;
1243
1244         /* Stacked block device info: */
1245         struct bio_list                 *bio_list;
1246
1247         /* Stack plugging: */
1248         struct blk_plug                 *plug;
1249
1250         /* VM state: */
1251         struct reclaim_state            *reclaim_state;
1252
1253         struct io_context               *io_context;
1254
1255 #ifdef CONFIG_COMPACTION
1256         struct capture_control          *capture_control;
1257 #endif
1258         /* Ptrace state: */
1259         unsigned long                   ptrace_message;
1260         kernel_siginfo_t                *last_siginfo;
1261
1262         struct task_io_accounting       ioac;
1263 #ifdef CONFIG_PSI
1264         /* Pressure stall state */
1265         unsigned int                    psi_flags;
1266 #endif
1267 #ifdef CONFIG_TASK_XACCT
1268         /* Accumulated RSS usage: */
1269         u64                             acct_rss_mem1;
1270         /* Accumulated virtual memory usage: */
1271         u64                             acct_vm_mem1;
1272         /* stime + utime since last update: */
1273         u64                             acct_timexpd;
1274 #endif
1275 #ifdef CONFIG_CPUSETS
1276         /* Protected by ->alloc_lock: */
1277         nodemask_t                      mems_allowed;
1278         /* Sequence number to catch updates: */
1279         seqcount_spinlock_t             mems_allowed_seq;
1280         int                             cpuset_mem_spread_rotor;
1281 #endif
1282 #ifdef CONFIG_CGROUPS
1283         /* Control Group info protected by css_set_lock: */
1284         struct css_set __rcu            *cgroups;
1285         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1286         struct list_head                cg_list;
1287 #endif
1288 #ifdef CONFIG_X86_CPU_RESCTRL
1289         u32                             closid;
1290         u32                             rmid;
1291 #endif
1292 #ifdef CONFIG_FUTEX
1293         struct robust_list_head __user  *robust_list;
1294 #ifdef CONFIG_COMPAT
1295         struct compat_robust_list_head __user *compat_robust_list;
1296 #endif
1297         struct list_head                pi_state_list;
1298         struct futex_pi_state           *pi_state_cache;
1299         struct mutex                    futex_exit_mutex;
1300         unsigned int                    futex_state;
1301 #endif
1302 #ifdef CONFIG_PERF_EVENTS
1303         u8                              perf_recursion[PERF_NR_CONTEXTS];
1304         struct perf_event_context       *perf_event_ctxp;
1305         struct mutex                    perf_event_mutex;
1306         struct list_head                perf_event_list;
1307 #endif
1308 #ifdef CONFIG_DEBUG_PREEMPT
1309         unsigned long                   preempt_disable_ip;
1310 #endif
1311 #ifdef CONFIG_NUMA
1312         /* Protected by alloc_lock: */
1313         struct mempolicy                *mempolicy;
1314         short                           il_prev;
1315         u8                              il_weight;
1316         short                           pref_node_fork;
1317 #endif
1318 #ifdef CONFIG_NUMA_BALANCING
1319         int                             numa_scan_seq;
1320         unsigned int                    numa_scan_period;
1321         unsigned int                    numa_scan_period_max;
1322         int                             numa_preferred_nid;
1323         unsigned long                   numa_migrate_retry;
1324         /* Migration stamp: */
1325         u64                             node_stamp;
1326         u64                             last_task_numa_placement;
1327         u64                             last_sum_exec_runtime;
1328         struct callback_head            numa_work;
1329
1330         /*
1331          * This pointer is only modified for current in syscall and
1332          * pagefault context (and for tasks being destroyed), so it can be read
1333          * from any of the following contexts:
1334          *  - RCU read-side critical section
1335          *  - current->numa_group from everywhere
1336          *  - task's runqueue locked, task not running
1337          */
1338         struct numa_group __rcu         *numa_group;
1339
1340         /*
1341          * numa_faults is an array split into four regions:
1342          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1343          * in this precise order.
1344          *
1345          * faults_memory: Exponential decaying average of faults on a per-node
1346          * basis. Scheduling placement decisions are made based on these
1347          * counts. The values remain static for the duration of a PTE scan.
1348          * faults_cpu: Track the nodes the process was running on when a NUMA
1349          * hinting fault was incurred.
1350          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1351          * during the current scan window. When the scan completes, the counts
1352          * in faults_memory and faults_cpu decay and these values are copied.
1353          */
1354         unsigned long                   *numa_faults;
1355         unsigned long                   total_numa_faults;
1356
1357         /*
1358          * numa_faults_locality tracks if faults recorded during the last
1359          * scan window were remote/local or failed to migrate. The task scan
1360          * period is adapted based on the locality of the faults with different
1361          * weights depending on whether they were shared or private faults
1362          */
1363         unsigned long                   numa_faults_locality[3];
1364
1365         unsigned long                   numa_pages_migrated;
1366 #endif /* CONFIG_NUMA_BALANCING */
1367
1368 #ifdef CONFIG_RSEQ
1369         struct rseq __user *rseq;
1370         u32 rseq_len;
1371         u32 rseq_sig;
1372         /*
1373          * RmW on rseq_event_mask must be performed atomically
1374          * with respect to preemption.
1375          */
1376         unsigned long rseq_event_mask;
1377 #endif
1378
1379 #ifdef CONFIG_SCHED_MM_CID
1380         int                             mm_cid;         /* Current cid in mm */
1381         int                             last_mm_cid;    /* Most recent cid in mm */
1382         int                             migrate_from_cpu;
1383         int                             mm_cid_active;  /* Whether cid bitmap is active */
1384         struct callback_head            cid_work;
1385 #endif
1386
1387         struct tlbflush_unmap_batch     tlb_ubc;
1388
1389         /* Cache last used pipe for splice(): */
1390         struct pipe_inode_info          *splice_pipe;
1391
1392         struct page_frag                task_frag;
1393
1394 #ifdef CONFIG_TASK_DELAY_ACCT
1395         struct task_delay_info          *delays;
1396 #endif
1397
1398 #ifdef CONFIG_FAULT_INJECTION
1399         int                             make_it_fail;
1400         unsigned int                    fail_nth;
1401 #endif
1402         /*
1403          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1404          * balance_dirty_pages() for a dirty throttling pause:
1405          */
1406         int                             nr_dirtied;
1407         int                             nr_dirtied_pause;
1408         /* Start of a write-and-pause period: */
1409         unsigned long                   dirty_paused_when;
1410
1411 #ifdef CONFIG_LATENCYTOP
1412         int                             latency_record_count;
1413         struct latency_record           latency_record[LT_SAVECOUNT];
1414 #endif
1415         /*
1416          * Time slack values; these are used to round up poll() and
1417          * select() etc timeout values. These are in nanoseconds.
1418          */
1419         u64                             timer_slack_ns;
1420         u64                             default_timer_slack_ns;
1421
1422 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1423         unsigned int                    kasan_depth;
1424 #endif
1425
1426 #ifdef CONFIG_KCSAN
1427         struct kcsan_ctx                kcsan_ctx;
1428 #ifdef CONFIG_TRACE_IRQFLAGS
1429         struct irqtrace_events          kcsan_save_irqtrace;
1430 #endif
1431 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1432         int                             kcsan_stack_depth;
1433 #endif
1434 #endif
1435
1436 #ifdef CONFIG_KMSAN
1437         struct kmsan_ctx                kmsan_ctx;
1438 #endif
1439
1440 #if IS_ENABLED(CONFIG_KUNIT)
1441         struct kunit                    *kunit_test;
1442 #endif
1443
1444 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1445         /* Index of current stored address in ret_stack: */
1446         int                             curr_ret_stack;
1447         int                             curr_ret_depth;
1448
1449         /* Stack of return addresses for return function tracing: */
1450         unsigned long                   *ret_stack;
1451
1452         /* Timestamp for last schedule: */
1453         unsigned long long              ftrace_timestamp;
1454         unsigned long long              ftrace_sleeptime;
1455
1456         /*
1457          * Number of functions that haven't been traced
1458          * because of depth overrun:
1459          */
1460         atomic_t                        trace_overrun;
1461
1462         /* Pause tracing: */
1463         atomic_t                        tracing_graph_pause;
1464 #endif
1465
1466 #ifdef CONFIG_TRACING
1467         /* Bitmask and counter of trace recursion: */
1468         unsigned long                   trace_recursion;
1469 #endif /* CONFIG_TRACING */
1470
1471 #ifdef CONFIG_KCOV
1472         /* See kernel/kcov.c for more details. */
1473
1474         /* Coverage collection mode enabled for this task (0 if disabled): */
1475         unsigned int                    kcov_mode;
1476
1477         /* Size of the kcov_area: */
1478         unsigned int                    kcov_size;
1479
1480         /* Buffer for coverage collection: */
1481         void                            *kcov_area;
1482
1483         /* KCOV descriptor wired with this task or NULL: */
1484         struct kcov                     *kcov;
1485
1486         /* KCOV common handle for remote coverage collection: */
1487         u64                             kcov_handle;
1488
1489         /* KCOV sequence number: */
1490         int                             kcov_sequence;
1491
1492         /* Collect coverage from softirq context: */
1493         unsigned int                    kcov_softirq;
1494 #endif
1495
1496 #ifdef CONFIG_MEMCG_V1
1497         struct mem_cgroup               *memcg_in_oom;
1498 #endif
1499
1500 #ifdef CONFIG_MEMCG
1501         /* Number of pages to reclaim on returning to userland: */
1502         unsigned int                    memcg_nr_pages_over_high;
1503
1504         /* Used by memcontrol for targeted memcg charge: */
1505         struct mem_cgroup               *active_memcg;
1506
1507         /* Cache for current->cgroups->memcg->objcg lookups: */
1508         struct obj_cgroup               *objcg;
1509 #endif
1510
1511 #ifdef CONFIG_BLK_CGROUP
1512         struct gendisk                  *throttle_disk;
1513 #endif
1514
1515 #ifdef CONFIG_UPROBES
1516         struct uprobe_task              *utask;
1517 #endif
1518 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1519         unsigned int                    sequential_io;
1520         unsigned int                    sequential_io_avg;
1521 #endif
1522         struct kmap_ctrl                kmap_ctrl;
1523 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1524         unsigned long                   task_state_change;
1525 # ifdef CONFIG_PREEMPT_RT
1526         unsigned long                   saved_state_change;
1527 # endif
1528 #endif
1529         struct rcu_head                 rcu;
1530         refcount_t                      rcu_users;
1531         int                             pagefault_disabled;
1532 #ifdef CONFIG_MMU
1533         struct task_struct              *oom_reaper_list;
1534         struct timer_list               oom_reaper_timer;
1535 #endif
1536 #ifdef CONFIG_VMAP_STACK
1537         struct vm_struct                *stack_vm_area;
1538 #endif
1539 #ifdef CONFIG_THREAD_INFO_IN_TASK
1540         /* A live task holds one reference: */
1541         refcount_t                      stack_refcount;
1542 #endif
1543 #ifdef CONFIG_LIVEPATCH
1544         int patch_state;
1545 #endif
1546 #ifdef CONFIG_SECURITY
1547         /* Used by LSM modules for access restriction: */
1548         void                            *security;
1549 #endif
1550 #ifdef CONFIG_BPF_SYSCALL
1551         /* Used by BPF task local storage */
1552         struct bpf_local_storage __rcu  *bpf_storage;
1553         /* Used for BPF run context */
1554         struct bpf_run_ctx              *bpf_ctx;
1555 #endif
1556         /* Used by BPF for per-TASK xdp storage */
1557         struct bpf_net_context          *bpf_net_context;
1558
1559 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1560         unsigned long                   lowest_stack;
1561         unsigned long                   prev_lowest_stack;
1562 #endif
1563
1564 #ifdef CONFIG_X86_MCE
1565         void __user                     *mce_vaddr;
1566         __u64                           mce_kflags;
1567         u64                             mce_addr;
1568         __u64                           mce_ripv : 1,
1569                                         mce_whole_page : 1,
1570                                         __mce_reserved : 62;
1571         struct callback_head            mce_kill_me;
1572         int                             mce_count;
1573 #endif
1574
1575 #ifdef CONFIG_KRETPROBES
1576         struct llist_head               kretprobe_instances;
1577 #endif
1578 #ifdef CONFIG_RETHOOK
1579         struct llist_head               rethooks;
1580 #endif
1581
1582 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1583         /*
1584          * If L1D flush is supported on mm context switch
1585          * then we use this callback head to queue kill work
1586          * to kill tasks that are not running on SMT disabled
1587          * cores
1588          */
1589         struct callback_head            l1d_flush_kill;
1590 #endif
1591
1592 #ifdef CONFIG_RV
1593         /*
1594          * Per-task RV monitor. Nowadays fixed in RV_PER_TASK_MONITORS.
1595          * If we find justification for more monitors, we can think
1596          * about adding more or developing a dynamic method. So far,
1597          * none of these are justified.
1598          */
1599         union rv_task_monitor           rv[RV_PER_TASK_MONITORS];
1600 #endif
1601
1602 #ifdef CONFIG_USER_EVENTS
1603         struct user_event_mm            *user_event_mm;
1604 #endif
1605
1606         /*
1607          * New fields for task_struct should be added above here, so that
1608          * they are included in the randomized portion of task_struct.
1609          */
1610         randomized_struct_fields_end
1611
1612         /* CPU-specific state of this task: */
1613         struct thread_struct            thread;
1614
1615         /*
1616          * WARNING: on x86, 'thread_struct' contains a variable-sized
1617          * structure.  It *MUST* be at the end of 'task_struct'.
1618          *
1619          * Do not put anything below here!
1620          */
1621 };
1622
1623 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1624 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1625
1626 static inline unsigned int __task_state_index(unsigned int tsk_state,
1627                                               unsigned int tsk_exit_state)
1628 {
1629         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1630
1631         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1632
1633         if ((tsk_state & TASK_IDLE) == TASK_IDLE)
1634                 state = TASK_REPORT_IDLE;
1635
1636         /*
1637          * We're lying here, but rather than expose a completely new task state
1638          * to userspace, we can make this appear as if the task has gone through
1639          * a regular rt_mutex_lock() call.
1640          * Report frozen tasks as uninterruptible.
1641          */
1642         if ((tsk_state & TASK_RTLOCK_WAIT) || (tsk_state & TASK_FROZEN))
1643                 state = TASK_UNINTERRUPTIBLE;
1644
1645         return fls(state);
1646 }
1647
1648 static inline unsigned int task_state_index(struct task_struct *tsk)
1649 {
1650         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1651 }
1652
1653 static inline char task_index_to_char(unsigned int state)
1654 {
1655         static const char state_char[] = "RSDTtXZPI";
1656
1657         BUILD_BUG_ON(TASK_REPORT_MAX * 2 != 1 << (sizeof(state_char) - 1));
1658
1659         return state_char[state];
1660 }
1661
1662 static inline char task_state_to_char(struct task_struct *tsk)
1663 {
1664         return task_index_to_char(task_state_index(tsk));
1665 }
1666
1667 extern struct pid *cad_pid;
1668
1669 /*
1670  * Per process flags
1671  */
1672 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1673 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1674 #define PF_EXITING              0x00000004      /* Getting shut down */
1675 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1676 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1677 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1678 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1679 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1680 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1681 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1682 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1683 #define PF_MEMALLOC             0x00000800      /* Allocating memory to free memory. See memalloc_noreclaim_save() */
1684 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1685 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1686 #define PF_USER_WORKER          0x00004000      /* Kernel thread cloned from userspace thread */
1687 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1688 #define PF__HOLE__00010000      0x00010000
1689 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1690 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocations inherit GFP_NOFS. See memalloc_nfs_save() */
1691 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocations inherit GFP_NOIO. See memalloc_noio_save() */
1692 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1693                                                  * I am cleaning dirty pages from some other bdi. */
1694 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1695 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1696 #define PF__HOLE__00800000      0x00800000
1697 #define PF__HOLE__01000000      0x01000000
1698 #define PF__HOLE__02000000      0x02000000
1699 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1700 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1701 #define PF_MEMALLOC_PIN         0x10000000      /* Allocations constrained to zones which allow long term pinning.
1702                                                  * See memalloc_pin_save() */
1703 #define PF_BLOCK_TS             0x20000000      /* plug has ts that needs updating */
1704 #define PF__HOLE__40000000      0x40000000
1705 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1706
1707 /*
1708  * Only the _current_ task can read/write to tsk->flags, but other
1709  * tasks can access tsk->flags in readonly mode for example
1710  * with tsk_used_math (like during threaded core dumping).
1711  * There is however an exception to this rule during ptrace
1712  * or during fork: the ptracer task is allowed to write to the
1713  * child->flags of its traced child (same goes for fork, the parent
1714  * can write to the child->flags), because we're guaranteed the
1715  * child is not running and in turn not changing child->flags
1716  * at the same time the parent does it.
1717  */
1718 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1719 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1720 #define clear_used_math()                       clear_stopped_child_used_math(current)
1721 #define set_used_math()                         set_stopped_child_used_math(current)
1722
1723 #define conditional_stopped_child_used_math(condition, child) \
1724         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1725
1726 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1727
1728 #define copy_to_stopped_child_used_math(child) \
1729         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1730
1731 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1732 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1733 #define used_math()                             tsk_used_math(current)
1734
1735 static __always_inline bool is_percpu_thread(void)
1736 {
1737 #ifdef CONFIG_SMP
1738         return (current->flags & PF_NO_SETAFFINITY) &&
1739                 (current->nr_cpus_allowed  == 1);
1740 #else
1741         return true;
1742 #endif
1743 }
1744
1745 /* Per-process atomic flags. */
1746 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1747 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1748 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1749 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1750 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1751 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1752 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1753 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1754
1755 #define TASK_PFA_TEST(name, func)                                       \
1756         static inline bool task_##func(struct task_struct *p)           \
1757         { return test_bit(PFA_##name, &p->atomic_flags); }
1758
1759 #define TASK_PFA_SET(name, func)                                        \
1760         static inline void task_set_##func(struct task_struct *p)       \
1761         { set_bit(PFA_##name, &p->atomic_flags); }
1762
1763 #define TASK_PFA_CLEAR(name, func)                                      \
1764         static inline void task_clear_##func(struct task_struct *p)     \
1765         { clear_bit(PFA_##name, &p->atomic_flags); }
1766
1767 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1768 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1769
1770 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1771 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1772 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1773
1774 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1775 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1776 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1777
1778 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1779 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1780 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1781
1782 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1783 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1784 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1785
1786 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1787 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1788
1789 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1790 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1791 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1792
1793 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1794 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1795
1796 static inline void
1797 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1798 {
1799         current->flags &= ~flags;
1800         current->flags |= orig_flags & flags;
1801 }
1802
1803 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1804 extern int task_can_attach(struct task_struct *p);
1805 extern int dl_bw_alloc(int cpu, u64 dl_bw);
1806 extern void dl_bw_free(int cpu, u64 dl_bw);
1807 #ifdef CONFIG_SMP
1808
1809 /* do_set_cpus_allowed() - consider using set_cpus_allowed_ptr() instead */
1810 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1811
1812 /**
1813  * set_cpus_allowed_ptr - set CPU affinity mask of a task
1814  * @p: the task
1815  * @new_mask: CPU affinity mask
1816  *
1817  * Return: zero if successful, or a negative error code
1818  */
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         /* Opencoded cpumask_test_cpu(0, new_mask) to avoid dependency on cpumask.h */
1832         if ((*cpumask_bits(new_mask) & 1) == 0)
1833                 return -EINVAL;
1834         return 0;
1835 }
1836 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1837 {
1838         if (src->user_cpus_ptr)
1839                 return -EINVAL;
1840         return 0;
1841 }
1842 static inline void release_user_cpus_ptr(struct task_struct *p)
1843 {
1844         WARN_ON(p->user_cpus_ptr);
1845 }
1846
1847 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1848 {
1849         return 0;
1850 }
1851 #endif
1852
1853 extern int yield_to(struct task_struct *p, bool preempt);
1854 extern void set_user_nice(struct task_struct *p, long nice);
1855 extern int task_prio(const struct task_struct *p);
1856
1857 /**
1858  * task_nice - return the nice value of a given task.
1859  * @p: the task in question.
1860  *
1861  * Return: The nice value [ -20 ... 0 ... 19 ].
1862  */
1863 static inline int task_nice(const struct task_struct *p)
1864 {
1865         return PRIO_TO_NICE((p)->static_prio);
1866 }
1867
1868 extern int can_nice(const struct task_struct *p, const int nice);
1869 extern int task_curr(const struct task_struct *p);
1870 extern int idle_cpu(int cpu);
1871 extern int available_idle_cpu(int cpu);
1872 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1873 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1874 extern void sched_set_fifo(struct task_struct *p);
1875 extern void sched_set_fifo_low(struct task_struct *p);
1876 extern void sched_set_normal(struct task_struct *p, int nice);
1877 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1878 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1879 extern struct task_struct *idle_task(int cpu);
1880
1881 /**
1882  * is_idle_task - is the specified task an idle task?
1883  * @p: the task in question.
1884  *
1885  * Return: 1 if @p is an idle task. 0 otherwise.
1886  */
1887 static __always_inline bool is_idle_task(const struct task_struct *p)
1888 {
1889         return !!(p->flags & PF_IDLE);
1890 }
1891
1892 extern struct task_struct *curr_task(int cpu);
1893 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1894
1895 void yield(void);
1896
1897 union thread_union {
1898         struct task_struct task;
1899 #ifndef CONFIG_THREAD_INFO_IN_TASK
1900         struct thread_info thread_info;
1901 #endif
1902         unsigned long stack[THREAD_SIZE/sizeof(long)];
1903 };
1904
1905 #ifndef CONFIG_THREAD_INFO_IN_TASK
1906 extern struct thread_info init_thread_info;
1907 #endif
1908
1909 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1910
1911 #ifdef CONFIG_THREAD_INFO_IN_TASK
1912 # define task_thread_info(task) (&(task)->thread_info)
1913 #else
1914 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1915 #endif
1916
1917 /*
1918  * find a task by one of its numerical ids
1919  *
1920  * find_task_by_pid_ns():
1921  *      finds a task by its pid in the specified namespace
1922  * find_task_by_vpid():
1923  *      finds a task by its virtual pid
1924  *
1925  * see also find_vpid() etc in include/linux/pid.h
1926  */
1927
1928 extern struct task_struct *find_task_by_vpid(pid_t nr);
1929 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1930
1931 /*
1932  * find a task by its virtual pid and get the task struct
1933  */
1934 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1935
1936 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1937 extern int wake_up_process(struct task_struct *tsk);
1938 extern void wake_up_new_task(struct task_struct *tsk);
1939
1940 #ifdef CONFIG_SMP
1941 extern void kick_process(struct task_struct *tsk);
1942 #else
1943 static inline void kick_process(struct task_struct *tsk) { }
1944 #endif
1945
1946 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1947
1948 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1949 {
1950         __set_task_comm(tsk, from, false);
1951 }
1952
1953 /*
1954  * - Why not use task_lock()?
1955  *   User space can randomly change their names anyway, so locking for readers
1956  *   doesn't make sense. For writers, locking is probably necessary, as a race
1957  *   condition could lead to long-term mixed results.
1958  *   The strscpy_pad() in __set_task_comm() can ensure that the task comm is
1959  *   always NUL-terminated and zero-padded. Therefore the race condition between
1960  *   reader and writer is not an issue.
1961  *
1962  * - BUILD_BUG_ON() can help prevent the buf from being truncated.
1963  *   Since the callers don't perform any return value checks, this safeguard is
1964  *   necessary.
1965  */
1966 #define get_task_comm(buf, tsk) ({                      \
1967         BUILD_BUG_ON(sizeof(buf) < TASK_COMM_LEN);      \
1968         strscpy_pad(buf, (tsk)->comm);                  \
1969         buf;                                            \
1970 })
1971
1972 #ifdef CONFIG_SMP
1973 static __always_inline void scheduler_ipi(void)
1974 {
1975         /*
1976          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1977          * TIF_NEED_RESCHED remotely (for the first time) will also send
1978          * this IPI.
1979          */
1980         preempt_fold_need_resched();
1981 }
1982 #else
1983 static inline void scheduler_ipi(void) { }
1984 #endif
1985
1986 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
1987
1988 /*
1989  * Set thread flags in other task's structures.
1990  * See asm/thread_info.h for TIF_xxxx flags available:
1991  */
1992 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
1993 {
1994         set_ti_thread_flag(task_thread_info(tsk), flag);
1995 }
1996
1997 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
1998 {
1999         clear_ti_thread_flag(task_thread_info(tsk), flag);
2000 }
2001
2002 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
2003                                           bool value)
2004 {
2005         update_ti_thread_flag(task_thread_info(tsk), flag, value);
2006 }
2007
2008 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
2009 {
2010         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
2011 }
2012
2013 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
2014 {
2015         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
2016 }
2017
2018 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
2019 {
2020         return test_ti_thread_flag(task_thread_info(tsk), flag);
2021 }
2022
2023 static inline void set_tsk_need_resched(struct task_struct *tsk)
2024 {
2025         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2026 }
2027
2028 static inline void clear_tsk_need_resched(struct task_struct *tsk)
2029 {
2030         atomic_long_andnot(_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY,
2031                            (atomic_long_t *)&task_thread_info(tsk)->flags);
2032 }
2033
2034 static inline int test_tsk_need_resched(struct task_struct *tsk)
2035 {
2036         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2037 }
2038
2039 /*
2040  * cond_resched() and cond_resched_lock(): latency reduction via
2041  * explicit rescheduling in places that are safe. The return
2042  * value indicates whether a reschedule was done in fact.
2043  * cond_resched_lock() will drop the spinlock before scheduling,
2044  */
2045 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2046 extern int __cond_resched(void);
2047
2048 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
2049
2050 void sched_dynamic_klp_enable(void);
2051 void sched_dynamic_klp_disable(void);
2052
2053 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2054
2055 static __always_inline int _cond_resched(void)
2056 {
2057         return static_call_mod(cond_resched)();
2058 }
2059
2060 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
2061
2062 extern int dynamic_cond_resched(void);
2063
2064 static __always_inline int _cond_resched(void)
2065 {
2066         return dynamic_cond_resched();
2067 }
2068
2069 #else /* !CONFIG_PREEMPTION */
2070
2071 static inline int _cond_resched(void)
2072 {
2073         klp_sched_try_switch();
2074         return __cond_resched();
2075 }
2076
2077 #endif /* PREEMPT_DYNAMIC && CONFIG_HAVE_PREEMPT_DYNAMIC_CALL */
2078
2079 #else /* CONFIG_PREEMPTION && !CONFIG_PREEMPT_DYNAMIC */
2080
2081 static inline int _cond_resched(void)
2082 {
2083         klp_sched_try_switch();
2084         return 0;
2085 }
2086
2087 #endif /* !CONFIG_PREEMPTION || CONFIG_PREEMPT_DYNAMIC */
2088
2089 #define cond_resched() ({                       \
2090         __might_resched(__FILE__, __LINE__, 0); \
2091         _cond_resched();                        \
2092 })
2093
2094 extern int __cond_resched_lock(spinlock_t *lock);
2095 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2096 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2097
2098 #define MIGHT_RESCHED_RCU_SHIFT         8
2099 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2100
2101 #ifndef CONFIG_PREEMPT_RT
2102 /*
2103  * Non RT kernels have an elevated preempt count due to the held lock,
2104  * but are not allowed to be inside a RCU read side critical section
2105  */
2106 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2107 #else
2108 /*
2109  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2110  * cond_resched*lock() has to take that into account because it checks for
2111  * preempt_count() and rcu_preempt_depth().
2112  */
2113 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2114         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2115 #endif
2116
2117 #define cond_resched_lock(lock) ({                                              \
2118         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2119         __cond_resched_lock(lock);                                              \
2120 })
2121
2122 #define cond_resched_rwlock_read(lock) ({                                       \
2123         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2124         __cond_resched_rwlock_read(lock);                                       \
2125 })
2126
2127 #define cond_resched_rwlock_write(lock) ({                                      \
2128         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2129         __cond_resched_rwlock_write(lock);                                      \
2130 })
2131
2132 static __always_inline bool need_resched(void)
2133 {
2134         return unlikely(tif_need_resched());
2135 }
2136
2137 /*
2138  * Wrappers for p->thread_info->cpu access. No-op on UP.
2139  */
2140 #ifdef CONFIG_SMP
2141
2142 static inline unsigned int task_cpu(const struct task_struct *p)
2143 {
2144         return READ_ONCE(task_thread_info(p)->cpu);
2145 }
2146
2147 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2148
2149 #else
2150
2151 static inline unsigned int task_cpu(const struct task_struct *p)
2152 {
2153         return 0;
2154 }
2155
2156 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2157 {
2158 }
2159
2160 #endif /* CONFIG_SMP */
2161
2162 static inline bool task_is_runnable(struct task_struct *p)
2163 {
2164         return p->on_rq && !p->se.sched_delayed;
2165 }
2166
2167 extern bool sched_task_on_rq(struct task_struct *p);
2168 extern unsigned long get_wchan(struct task_struct *p);
2169 extern struct task_struct *cpu_curr_snapshot(int cpu);
2170
2171 #include <linux/spinlock.h>
2172
2173 /*
2174  * In order to reduce various lock holder preemption latencies provide an
2175  * interface to see if a vCPU is currently running or not.
2176  *
2177  * This allows us to terminate optimistic spin loops and block, analogous to
2178  * the native optimistic spin heuristic of testing if the lock owner task is
2179  * running or not.
2180  */
2181 #ifndef vcpu_is_preempted
2182 static inline bool vcpu_is_preempted(int cpu)
2183 {
2184         return false;
2185 }
2186 #endif
2187
2188 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2189 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2190
2191 #ifndef TASK_SIZE_OF
2192 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2193 #endif
2194
2195 #ifdef CONFIG_SMP
2196 static inline bool owner_on_cpu(struct task_struct *owner)
2197 {
2198         /*
2199          * As lock holder preemption issue, we both skip spinning if
2200          * task is not on cpu or its cpu is preempted
2201          */
2202         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2203 }
2204
2205 /* Returns effective CPU energy utilization, as seen by the scheduler */
2206 unsigned long sched_cpu_util(int cpu);
2207 #endif /* CONFIG_SMP */
2208
2209 #ifdef CONFIG_SCHED_CORE
2210 extern void sched_core_free(struct task_struct *tsk);
2211 extern void sched_core_fork(struct task_struct *p);
2212 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2213                                 unsigned long uaddr);
2214 extern int sched_core_idle_cpu(int cpu);
2215 #else
2216 static inline void sched_core_free(struct task_struct *tsk) { }
2217 static inline void sched_core_fork(struct task_struct *p) { }
2218 static inline int sched_core_idle_cpu(int cpu) { return idle_cpu(cpu); }
2219 #endif
2220
2221 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2222
2223 #ifdef CONFIG_MEM_ALLOC_PROFILING
2224 static __always_inline struct alloc_tag *alloc_tag_save(struct alloc_tag *tag)
2225 {
2226         swap(current->alloc_tag, tag);
2227         return tag;
2228 }
2229
2230 static __always_inline void alloc_tag_restore(struct alloc_tag *tag, struct alloc_tag *old)
2231 {
2232 #ifdef CONFIG_MEM_ALLOC_PROFILING_DEBUG
2233         WARN(current->alloc_tag != tag, "current->alloc_tag was changed:\n");
2234 #endif
2235         current->alloc_tag = old;
2236 }
2237 #else
2238 #define alloc_tag_save(_tag)                    NULL
2239 #define alloc_tag_restore(_tag, _old)           do {} while (0)
2240 #endif
2241
2242 #endif