Merge tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost
[linux-block.git] / include / linux / sched.h
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_SCHED_H
3 #define _LINUX_SCHED_H
4
5 /*
6  * Define 'struct task_struct' and provide the main scheduler
7  * APIs (schedule(), wakeup variants, etc.)
8  */
9
10 #include <uapi/linux/sched.h>
11
12 #include <asm/current.h>
13
14 #include <linux/pid.h>
15 #include <linux/sem.h>
16 #include <linux/shm.h>
17 #include <linux/kmsan_types.h>
18 #include <linux/mutex.h>
19 #include <linux/plist.h>
20 #include <linux/hrtimer.h>
21 #include <linux/irqflags.h>
22 #include <linux/seccomp.h>
23 #include <linux/nodemask.h>
24 #include <linux/rcupdate.h>
25 #include <linux/refcount.h>
26 #include <linux/resource.h>
27 #include <linux/latencytop.h>
28 #include <linux/sched/prio.h>
29 #include <linux/sched/types.h>
30 #include <linux/signal_types.h>
31 #include <linux/syscall_user_dispatch.h>
32 #include <linux/mm_types_task.h>
33 #include <linux/task_io_accounting.h>
34 #include <linux/posix-timers.h>
35 #include <linux/rseq.h>
36 #include <linux/seqlock.h>
37 #include <linux/kcsan.h>
38 #include <linux/rv.h>
39 #include <asm/kmap_size.h>
40
41 /* task_struct member predeclarations (sorted alphabetically): */
42 struct audit_context;
43 struct backing_dev_info;
44 struct bio_list;
45 struct blk_plug;
46 struct bpf_local_storage;
47 struct bpf_run_ctx;
48 struct capture_control;
49 struct cfs_rq;
50 struct fs_struct;
51 struct futex_pi_state;
52 struct io_context;
53 struct io_uring_task;
54 struct mempolicy;
55 struct nameidata;
56 struct nsproxy;
57 struct perf_event_context;
58 struct pid_namespace;
59 struct pipe_inode_info;
60 struct rcu_node;
61 struct reclaim_state;
62 struct robust_list_head;
63 struct root_domain;
64 struct rq;
65 struct sched_attr;
66 struct sched_param;
67 struct seq_file;
68 struct sighand_struct;
69 struct signal_struct;
70 struct task_delay_info;
71 struct task_group;
72
73 /*
74  * Task state bitmask. NOTE! These bits are also
75  * encoded in fs/proc/array.c: get_task_state().
76  *
77  * We have two separate sets of flags: task->state
78  * is about runnability, while task->exit_state are
79  * about the task exiting. Confusing, but this way
80  * modifying one set can't modify the other one by
81  * mistake.
82  */
83
84 /* Used in tsk->state: */
85 #define TASK_RUNNING                    0x00000000
86 #define TASK_INTERRUPTIBLE              0x00000001
87 #define TASK_UNINTERRUPTIBLE            0x00000002
88 #define __TASK_STOPPED                  0x00000004
89 #define __TASK_TRACED                   0x00000008
90 /* Used in tsk->exit_state: */
91 #define EXIT_DEAD                       0x00000010
92 #define EXIT_ZOMBIE                     0x00000020
93 #define EXIT_TRACE                      (EXIT_ZOMBIE | EXIT_DEAD)
94 /* Used in tsk->state again: */
95 #define TASK_PARKED                     0x00000040
96 #define TASK_DEAD                       0x00000080
97 #define TASK_WAKEKILL                   0x00000100
98 #define TASK_WAKING                     0x00000200
99 #define TASK_NOLOAD                     0x00000400
100 #define TASK_NEW                        0x00000800
101 #define TASK_RTLOCK_WAIT                0x00001000
102 #define TASK_FREEZABLE                  0x00002000
103 #define __TASK_FREEZABLE_UNSAFE        (0x00004000 * IS_ENABLED(CONFIG_LOCKDEP))
104 #define TASK_FROZEN                     0x00008000
105 #define TASK_STATE_MAX                  0x00010000
106
107 #define TASK_ANY                        (TASK_STATE_MAX-1)
108
109 /*
110  * DO NOT ADD ANY NEW USERS !
111  */
112 #define TASK_FREEZABLE_UNSAFE           (TASK_FREEZABLE | __TASK_FREEZABLE_UNSAFE)
113
114 /* Convenience macros for the sake of set_current_state: */
115 #define TASK_KILLABLE                   (TASK_WAKEKILL | TASK_UNINTERRUPTIBLE)
116 #define TASK_STOPPED                    (TASK_WAKEKILL | __TASK_STOPPED)
117 #define TASK_TRACED                     __TASK_TRACED
118
119 #define TASK_IDLE                       (TASK_UNINTERRUPTIBLE | TASK_NOLOAD)
120
121 /* Convenience macros for the sake of wake_up(): */
122 #define TASK_NORMAL                     (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)
123
124 /* get_task_state(): */
125 #define TASK_REPORT                     (TASK_RUNNING | TASK_INTERRUPTIBLE | \
126                                          TASK_UNINTERRUPTIBLE | __TASK_STOPPED | \
127                                          __TASK_TRACED | EXIT_DEAD | EXIT_ZOMBIE | \
128                                          TASK_PARKED)
129
130 #define task_is_running(task)           (READ_ONCE((task)->__state) == TASK_RUNNING)
131
132 #define task_is_traced(task)            ((READ_ONCE(task->jobctl) & JOBCTL_TRACED) != 0)
133 #define task_is_stopped(task)           ((READ_ONCE(task->jobctl) & JOBCTL_STOPPED) != 0)
134 #define task_is_stopped_or_traced(task) ((READ_ONCE(task->jobctl) & (JOBCTL_STOPPED | JOBCTL_TRACED)) != 0)
135
136 /*
137  * Special states are those that do not use the normal wait-loop pattern. See
138  * the comment with set_special_state().
139  */
140 #define is_special_task_state(state)                            \
141         ((state) & (__TASK_STOPPED | __TASK_TRACED | TASK_PARKED | TASK_DEAD))
142
143 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
144 # define debug_normal_state_change(state_value)                         \
145         do {                                                            \
146                 WARN_ON_ONCE(is_special_task_state(state_value));       \
147                 current->task_state_change = _THIS_IP_;                 \
148         } while (0)
149
150 # define debug_special_state_change(state_value)                        \
151         do {                                                            \
152                 WARN_ON_ONCE(!is_special_task_state(state_value));      \
153                 current->task_state_change = _THIS_IP_;                 \
154         } while (0)
155
156 # define debug_rtlock_wait_set_state()                                  \
157         do {                                                             \
158                 current->saved_state_change = current->task_state_change;\
159                 current->task_state_change = _THIS_IP_;                  \
160         } while (0)
161
162 # define debug_rtlock_wait_restore_state()                              \
163         do {                                                             \
164                 current->task_state_change = current->saved_state_change;\
165         } while (0)
166
167 #else
168 # define debug_normal_state_change(cond)        do { } while (0)
169 # define debug_special_state_change(cond)       do { } while (0)
170 # define debug_rtlock_wait_set_state()          do { } while (0)
171 # define debug_rtlock_wait_restore_state()      do { } while (0)
172 #endif
173
174 /*
175  * set_current_state() includes a barrier so that the write of current->state
176  * is correctly serialised wrt the caller's subsequent test of whether to
177  * actually sleep:
178  *
179  *   for (;;) {
180  *      set_current_state(TASK_UNINTERRUPTIBLE);
181  *      if (CONDITION)
182  *         break;
183  *
184  *      schedule();
185  *   }
186  *   __set_current_state(TASK_RUNNING);
187  *
188  * If the caller does not need such serialisation (because, for instance, the
189  * CONDITION test and condition change and wakeup are under the same lock) then
190  * use __set_current_state().
191  *
192  * The above is typically ordered against the wakeup, which does:
193  *
194  *   CONDITION = 1;
195  *   wake_up_state(p, TASK_UNINTERRUPTIBLE);
196  *
197  * where wake_up_state()/try_to_wake_up() executes a full memory barrier before
198  * accessing p->state.
199  *
200  * Wakeup will do: if (@state & p->state) p->state = TASK_RUNNING, that is,
201  * once it observes the TASK_UNINTERRUPTIBLE store the waking CPU can issue a
202  * TASK_RUNNING store which can collide with __set_current_state(TASK_RUNNING).
203  *
204  * However, with slightly different timing the wakeup TASK_RUNNING store can
205  * also collide with the TASK_UNINTERRUPTIBLE store. Losing that store is not
206  * a problem either because that will result in one extra go around the loop
207  * and our @cond test will save the day.
208  *
209  * Also see the comments of try_to_wake_up().
210  */
211 #define __set_current_state(state_value)                                \
212         do {                                                            \
213                 debug_normal_state_change((state_value));               \
214                 WRITE_ONCE(current->__state, (state_value));            \
215         } while (0)
216
217 #define set_current_state(state_value)                                  \
218         do {                                                            \
219                 debug_normal_state_change((state_value));               \
220                 smp_store_mb(current->__state, (state_value));          \
221         } while (0)
222
223 /*
224  * set_special_state() should be used for those states when the blocking task
225  * can not use the regular condition based wait-loop. In that case we must
226  * serialize against wakeups such that any possible in-flight TASK_RUNNING
227  * stores will not collide with our state change.
228  */
229 #define set_special_state(state_value)                                  \
230         do {                                                            \
231                 unsigned long flags; /* may shadow */                   \
232                                                                         \
233                 raw_spin_lock_irqsave(&current->pi_lock, flags);        \
234                 debug_special_state_change((state_value));              \
235                 WRITE_ONCE(current->__state, (state_value));            \
236                 raw_spin_unlock_irqrestore(&current->pi_lock, flags);   \
237         } while (0)
238
239 /*
240  * PREEMPT_RT specific variants for "sleeping" spin/rwlocks
241  *
242  * RT's spin/rwlock substitutions are state preserving. The state of the
243  * task when blocking on the lock is saved in task_struct::saved_state and
244  * restored after the lock has been acquired.  These operations are
245  * serialized by task_struct::pi_lock against try_to_wake_up(). Any non RT
246  * lock related wakeups while the task is blocked on the lock are
247  * redirected to operate on task_struct::saved_state to ensure that these
248  * are not dropped. On restore task_struct::saved_state is set to
249  * TASK_RUNNING so any wakeup attempt redirected to saved_state will fail.
250  *
251  * The lock operation looks like this:
252  *
253  *      current_save_and_set_rtlock_wait_state();
254  *      for (;;) {
255  *              if (try_lock())
256  *                      break;
257  *              raw_spin_unlock_irq(&lock->wait_lock);
258  *              schedule_rtlock();
259  *              raw_spin_lock_irq(&lock->wait_lock);
260  *              set_current_state(TASK_RTLOCK_WAIT);
261  *      }
262  *      current_restore_rtlock_saved_state();
263  */
264 #define current_save_and_set_rtlock_wait_state()                        \
265         do {                                                            \
266                 lockdep_assert_irqs_disabled();                         \
267                 raw_spin_lock(&current->pi_lock);                       \
268                 current->saved_state = current->__state;                \
269                 debug_rtlock_wait_set_state();                          \
270                 WRITE_ONCE(current->__state, TASK_RTLOCK_WAIT);         \
271                 raw_spin_unlock(&current->pi_lock);                     \
272         } while (0);
273
274 #define current_restore_rtlock_saved_state()                            \
275         do {                                                            \
276                 lockdep_assert_irqs_disabled();                         \
277                 raw_spin_lock(&current->pi_lock);                       \
278                 debug_rtlock_wait_restore_state();                      \
279                 WRITE_ONCE(current->__state, current->saved_state);     \
280                 current->saved_state = TASK_RUNNING;                    \
281                 raw_spin_unlock(&current->pi_lock);                     \
282         } while (0);
283
284 #define get_current_state()     READ_ONCE(current->__state)
285
286 /*
287  * Define the task command name length as enum, then it can be visible to
288  * BPF programs.
289  */
290 enum {
291         TASK_COMM_LEN = 16,
292 };
293
294 extern void scheduler_tick(void);
295
296 #define MAX_SCHEDULE_TIMEOUT            LONG_MAX
297
298 extern long schedule_timeout(long timeout);
299 extern long schedule_timeout_interruptible(long timeout);
300 extern long schedule_timeout_killable(long timeout);
301 extern long schedule_timeout_uninterruptible(long timeout);
302 extern long schedule_timeout_idle(long timeout);
303 asmlinkage void schedule(void);
304 extern void schedule_preempt_disabled(void);
305 asmlinkage void preempt_schedule_irq(void);
306 #ifdef CONFIG_PREEMPT_RT
307  extern void schedule_rtlock(void);
308 #endif
309
310 extern int __must_check io_schedule_prepare(void);
311 extern void io_schedule_finish(int token);
312 extern long io_schedule_timeout(long timeout);
313 extern void io_schedule(void);
314
315 /**
316  * struct prev_cputime - snapshot of system and user cputime
317  * @utime: time spent in user mode
318  * @stime: time spent in system mode
319  * @lock: protects the above two fields
320  *
321  * Stores previous user/system time values such that we can guarantee
322  * monotonicity.
323  */
324 struct prev_cputime {
325 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
326         u64                             utime;
327         u64                             stime;
328         raw_spinlock_t                  lock;
329 #endif
330 };
331
332 enum vtime_state {
333         /* Task is sleeping or running in a CPU with VTIME inactive: */
334         VTIME_INACTIVE = 0,
335         /* Task is idle */
336         VTIME_IDLE,
337         /* Task runs in kernelspace in a CPU with VTIME active: */
338         VTIME_SYS,
339         /* Task runs in userspace in a CPU with VTIME active: */
340         VTIME_USER,
341         /* Task runs as guests in a CPU with VTIME active: */
342         VTIME_GUEST,
343 };
344
345 struct vtime {
346         seqcount_t              seqcount;
347         unsigned long long      starttime;
348         enum vtime_state        state;
349         unsigned int            cpu;
350         u64                     utime;
351         u64                     stime;
352         u64                     gtime;
353 };
354
355 /*
356  * Utilization clamp constraints.
357  * @UCLAMP_MIN: Minimum utilization
358  * @UCLAMP_MAX: Maximum utilization
359  * @UCLAMP_CNT: Utilization clamp constraints count
360  */
361 enum uclamp_id {
362         UCLAMP_MIN = 0,
363         UCLAMP_MAX,
364         UCLAMP_CNT
365 };
366
367 #ifdef CONFIG_SMP
368 extern struct root_domain def_root_domain;
369 extern struct mutex sched_domains_mutex;
370 #endif
371
372 struct sched_info {
373 #ifdef CONFIG_SCHED_INFO
374         /* Cumulative counters: */
375
376         /* # of times we have run on this CPU: */
377         unsigned long                   pcount;
378
379         /* Time spent waiting on a runqueue: */
380         unsigned long long              run_delay;
381
382         /* Timestamps: */
383
384         /* When did we last run on a CPU? */
385         unsigned long long              last_arrival;
386
387         /* When were we last queued to run? */
388         unsigned long long              last_queued;
389
390 #endif /* CONFIG_SCHED_INFO */
391 };
392
393 /*
394  * Integer metrics need fixed point arithmetic, e.g., sched/fair
395  * has a few: load, load_avg, util_avg, freq, and capacity.
396  *
397  * We define a basic fixed point arithmetic range, and then formalize
398  * all these metrics based on that basic range.
399  */
400 # define SCHED_FIXEDPOINT_SHIFT         10
401 # define SCHED_FIXEDPOINT_SCALE         (1L << SCHED_FIXEDPOINT_SHIFT)
402
403 /* Increase resolution of cpu_capacity calculations */
404 # define SCHED_CAPACITY_SHIFT           SCHED_FIXEDPOINT_SHIFT
405 # define SCHED_CAPACITY_SCALE           (1L << SCHED_CAPACITY_SHIFT)
406
407 struct load_weight {
408         unsigned long                   weight;
409         u32                             inv_weight;
410 };
411
412 /**
413  * struct util_est - Estimation utilization of FAIR tasks
414  * @enqueued: instantaneous estimated utilization of a task/cpu
415  * @ewma:     the Exponential Weighted Moving Average (EWMA)
416  *            utilization of a task
417  *
418  * Support data structure to track an Exponential Weighted Moving Average
419  * (EWMA) of a FAIR task's utilization. New samples are added to the moving
420  * average each time a task completes an activation. Sample's weight is chosen
421  * so that the EWMA will be relatively insensitive to transient changes to the
422  * task's workload.
423  *
424  * The enqueued attribute has a slightly different meaning for tasks and cpus:
425  * - task:   the task's util_avg at last task dequeue time
426  * - cfs_rq: the sum of util_est.enqueued for each RUNNABLE task on that CPU
427  * Thus, the util_est.enqueued of a task represents the contribution on the
428  * estimated utilization of the CPU where that task is currently enqueued.
429  *
430  * Only for tasks we track a moving average of the past instantaneous
431  * estimated utilization. This allows to absorb sporadic drops in utilization
432  * of an otherwise almost periodic task.
433  *
434  * The UTIL_AVG_UNCHANGED flag is used to synchronize util_est with util_avg
435  * updates. When a task is dequeued, its util_est should not be updated if its
436  * util_avg has not been updated in the meantime.
437  * This information is mapped into the MSB bit of util_est.enqueued at dequeue
438  * time. Since max value of util_est.enqueued for a task is 1024 (PELT util_avg
439  * for a task) it is safe to use MSB.
440  */
441 struct util_est {
442         unsigned int                    enqueued;
443         unsigned int                    ewma;
444 #define UTIL_EST_WEIGHT_SHIFT           2
445 #define UTIL_AVG_UNCHANGED              0x80000000
446 } __attribute__((__aligned__(sizeof(u64))));
447
448 /*
449  * The load/runnable/util_avg accumulates an infinite geometric series
450  * (see __update_load_avg_cfs_rq() in kernel/sched/pelt.c).
451  *
452  * [load_avg definition]
453  *
454  *   load_avg = runnable% * scale_load_down(load)
455  *
456  * [runnable_avg definition]
457  *
458  *   runnable_avg = runnable% * SCHED_CAPACITY_SCALE
459  *
460  * [util_avg definition]
461  *
462  *   util_avg = running% * SCHED_CAPACITY_SCALE
463  *
464  * where runnable% is the time ratio that a sched_entity is runnable and
465  * running% the time ratio that a sched_entity is running.
466  *
467  * For cfs_rq, they are the aggregated values of all runnable and blocked
468  * sched_entities.
469  *
470  * The load/runnable/util_avg doesn't directly factor frequency scaling and CPU
471  * capacity scaling. The scaling is done through the rq_clock_pelt that is used
472  * for computing those signals (see update_rq_clock_pelt())
473  *
474  * N.B., the above ratios (runnable% and running%) themselves are in the
475  * range of [0, 1]. To do fixed point arithmetics, we therefore scale them
476  * to as large a range as necessary. This is for example reflected by
477  * util_avg's SCHED_CAPACITY_SCALE.
478  *
479  * [Overflow issue]
480  *
481  * The 64-bit load_sum can have 4353082796 (=2^64/47742/88761) entities
482  * with the highest load (=88761), always runnable on a single cfs_rq,
483  * and should not overflow as the number already hits PID_MAX_LIMIT.
484  *
485  * For all other cases (including 32-bit kernels), struct load_weight's
486  * weight will overflow first before we do, because:
487  *
488  *    Max(load_avg) <= Max(load.weight)
489  *
490  * Then it is the load_weight's responsibility to consider overflow
491  * issues.
492  */
493 struct sched_avg {
494         u64                             last_update_time;
495         u64                             load_sum;
496         u64                             runnable_sum;
497         u32                             util_sum;
498         u32                             period_contrib;
499         unsigned long                   load_avg;
500         unsigned long                   runnable_avg;
501         unsigned long                   util_avg;
502         struct util_est                 util_est;
503 } ____cacheline_aligned;
504
505 struct sched_statistics {
506 #ifdef CONFIG_SCHEDSTATS
507         u64                             wait_start;
508         u64                             wait_max;
509         u64                             wait_count;
510         u64                             wait_sum;
511         u64                             iowait_count;
512         u64                             iowait_sum;
513
514         u64                             sleep_start;
515         u64                             sleep_max;
516         s64                             sum_sleep_runtime;
517
518         u64                             block_start;
519         u64                             block_max;
520         s64                             sum_block_runtime;
521
522         u64                             exec_max;
523         u64                             slice_max;
524
525         u64                             nr_migrations_cold;
526         u64                             nr_failed_migrations_affine;
527         u64                             nr_failed_migrations_running;
528         u64                             nr_failed_migrations_hot;
529         u64                             nr_forced_migrations;
530
531         u64                             nr_wakeups;
532         u64                             nr_wakeups_sync;
533         u64                             nr_wakeups_migrate;
534         u64                             nr_wakeups_local;
535         u64                             nr_wakeups_remote;
536         u64                             nr_wakeups_affine;
537         u64                             nr_wakeups_affine_attempts;
538         u64                             nr_wakeups_passive;
539         u64                             nr_wakeups_idle;
540
541 #ifdef CONFIG_SCHED_CORE
542         u64                             core_forceidle_sum;
543 #endif
544 #endif /* CONFIG_SCHEDSTATS */
545 } ____cacheline_aligned;
546
547 struct sched_entity {
548         /* For load-balancing: */
549         struct load_weight              load;
550         struct rb_node                  run_node;
551         struct list_head                group_node;
552         unsigned int                    on_rq;
553
554         u64                             exec_start;
555         u64                             sum_exec_runtime;
556         u64                             vruntime;
557         u64                             prev_sum_exec_runtime;
558
559         u64                             nr_migrations;
560
561 #ifdef CONFIG_FAIR_GROUP_SCHED
562         int                             depth;
563         struct sched_entity             *parent;
564         /* rq on which this entity is (to be) queued: */
565         struct cfs_rq                   *cfs_rq;
566         /* rq "owned" by this entity/group: */
567         struct cfs_rq                   *my_q;
568         /* cached value of my_q->h_nr_running */
569         unsigned long                   runnable_weight;
570 #endif
571
572 #ifdef CONFIG_SMP
573         /*
574          * Per entity load average tracking.
575          *
576          * Put into separate cache line so it does not
577          * collide with read-mostly values above.
578          */
579         struct sched_avg                avg;
580 #endif
581 };
582
583 struct sched_rt_entity {
584         struct list_head                run_list;
585         unsigned long                   timeout;
586         unsigned long                   watchdog_stamp;
587         unsigned int                    time_slice;
588         unsigned short                  on_rq;
589         unsigned short                  on_list;
590
591         struct sched_rt_entity          *back;
592 #ifdef CONFIG_RT_GROUP_SCHED
593         struct sched_rt_entity          *parent;
594         /* rq on which this entity is (to be) queued: */
595         struct rt_rq                    *rt_rq;
596         /* rq "owned" by this entity/group: */
597         struct rt_rq                    *my_q;
598 #endif
599 } __randomize_layout;
600
601 struct sched_dl_entity {
602         struct rb_node                  rb_node;
603
604         /*
605          * Original scheduling parameters. Copied here from sched_attr
606          * during sched_setattr(), they will remain the same until
607          * the next sched_setattr().
608          */
609         u64                             dl_runtime;     /* Maximum runtime for each instance    */
610         u64                             dl_deadline;    /* Relative deadline of each instance   */
611         u64                             dl_period;      /* Separation of two instances (period) */
612         u64                             dl_bw;          /* dl_runtime / dl_period               */
613         u64                             dl_density;     /* dl_runtime / dl_deadline             */
614
615         /*
616          * Actual scheduling parameters. Initialized with the values above,
617          * they are continuously updated during task execution. Note that
618          * the remaining runtime could be < 0 in case we are in overrun.
619          */
620         s64                             runtime;        /* Remaining runtime for this instance  */
621         u64                             deadline;       /* Absolute deadline for this instance  */
622         unsigned int                    flags;          /* Specifying the scheduler behaviour   */
623
624         /*
625          * Some bool flags:
626          *
627          * @dl_throttled tells if we exhausted the runtime. If so, the
628          * task has to wait for a replenishment to be performed at the
629          * next firing of dl_timer.
630          *
631          * @dl_yielded tells if task gave up the CPU before consuming
632          * all its available runtime during the last job.
633          *
634          * @dl_non_contending tells if the task is inactive while still
635          * contributing to the active utilization. In other words, it
636          * indicates if the inactive timer has been armed and its handler
637          * has not been executed yet. This flag is useful to avoid race
638          * conditions between the inactive timer handler and the wakeup
639          * code.
640          *
641          * @dl_overrun tells if the task asked to be informed about runtime
642          * overruns.
643          */
644         unsigned int                    dl_throttled      : 1;
645         unsigned int                    dl_yielded        : 1;
646         unsigned int                    dl_non_contending : 1;
647         unsigned int                    dl_overrun        : 1;
648
649         /*
650          * Bandwidth enforcement timer. Each -deadline task has its
651          * own bandwidth to be enforced, thus we need one timer per task.
652          */
653         struct hrtimer                  dl_timer;
654
655         /*
656          * Inactive timer, responsible for decreasing the active utilization
657          * at the "0-lag time". When a -deadline task blocks, it contributes
658          * to GRUB's active utilization until the "0-lag time", hence a
659          * timer is needed to decrease the active utilization at the correct
660          * time.
661          */
662         struct hrtimer inactive_timer;
663
664 #ifdef CONFIG_RT_MUTEXES
665         /*
666          * Priority Inheritance. When a DEADLINE scheduling entity is boosted
667          * pi_se points to the donor, otherwise points to the dl_se it belongs
668          * to (the original one/itself).
669          */
670         struct sched_dl_entity *pi_se;
671 #endif
672 };
673
674 #ifdef CONFIG_UCLAMP_TASK
675 /* Number of utilization clamp buckets (shorter alias) */
676 #define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
677
678 /*
679  * Utilization clamp for a scheduling entity
680  * @value:              clamp value "assigned" to a se
681  * @bucket_id:          bucket index corresponding to the "assigned" value
682  * @active:             the se is currently refcounted in a rq's bucket
683  * @user_defined:       the requested clamp value comes from user-space
684  *
685  * The bucket_id is the index of the clamp bucket matching the clamp value
686  * which is pre-computed and stored to avoid expensive integer divisions from
687  * the fast path.
688  *
689  * The active bit is set whenever a task has got an "effective" value assigned,
690  * which can be different from the clamp value "requested" from user-space.
691  * This allows to know a task is refcounted in the rq's bucket corresponding
692  * to the "effective" bucket_id.
693  *
694  * The user_defined bit is set whenever a task has got a task-specific clamp
695  * value requested from userspace, i.e. the system defaults apply to this task
696  * just as a restriction. This allows to relax default clamps when a less
697  * restrictive task-specific value has been requested, thus allowing to
698  * implement a "nice" semantic. For example, a task running with a 20%
699  * default boost can still drop its own boosting to 0%.
700  */
701 struct uclamp_se {
702         unsigned int value              : bits_per(SCHED_CAPACITY_SCALE);
703         unsigned int bucket_id          : bits_per(UCLAMP_BUCKETS);
704         unsigned int active             : 1;
705         unsigned int user_defined       : 1;
706 };
707 #endif /* CONFIG_UCLAMP_TASK */
708
709 union rcu_special {
710         struct {
711                 u8                      blocked;
712                 u8                      need_qs;
713                 u8                      exp_hint; /* Hint for performance. */
714                 u8                      need_mb; /* Readers need smp_mb(). */
715         } b; /* Bits. */
716         u32 s; /* Set of bits. */
717 };
718
719 enum perf_event_task_context {
720         perf_invalid_context = -1,
721         perf_hw_context = 0,
722         perf_sw_context,
723         perf_nr_task_contexts,
724 };
725
726 struct wake_q_node {
727         struct wake_q_node *next;
728 };
729
730 struct kmap_ctrl {
731 #ifdef CONFIG_KMAP_LOCAL
732         int                             idx;
733         pte_t                           pteval[KM_MAX_IDX];
734 #endif
735 };
736
737 struct task_struct {
738 #ifdef CONFIG_THREAD_INFO_IN_TASK
739         /*
740          * For reasons of header soup (see current_thread_info()), this
741          * must be the first element of task_struct.
742          */
743         struct thread_info              thread_info;
744 #endif
745         unsigned int                    __state;
746
747 #ifdef CONFIG_PREEMPT_RT
748         /* saved state for "spinlock sleepers" */
749         unsigned int                    saved_state;
750 #endif
751
752         /*
753          * This begins the randomizable portion of task_struct. Only
754          * scheduling-critical items should be added above here.
755          */
756         randomized_struct_fields_start
757
758         void                            *stack;
759         refcount_t                      usage;
760         /* Per task flags (PF_*), defined further below: */
761         unsigned int                    flags;
762         unsigned int                    ptrace;
763
764 #ifdef CONFIG_SMP
765         int                             on_cpu;
766         struct __call_single_node       wake_entry;
767         unsigned int                    wakee_flips;
768         unsigned long                   wakee_flip_decay_ts;
769         struct task_struct              *last_wakee;
770
771         /*
772          * recent_used_cpu is initially set as the last CPU used by a task
773          * that wakes affine another task. Waker/wakee relationships can
774          * push tasks around a CPU where each wakeup moves to the next one.
775          * Tracking a recently used CPU allows a quick search for a recently
776          * used CPU that may be idle.
777          */
778         int                             recent_used_cpu;
779         int                             wake_cpu;
780 #endif
781         int                             on_rq;
782
783         int                             prio;
784         int                             static_prio;
785         int                             normal_prio;
786         unsigned int                    rt_priority;
787
788         struct sched_entity             se;
789         struct sched_rt_entity          rt;
790         struct sched_dl_entity          dl;
791         const struct sched_class        *sched_class;
792
793 #ifdef CONFIG_SCHED_CORE
794         struct rb_node                  core_node;
795         unsigned long                   core_cookie;
796         unsigned int                    core_occupation;
797 #endif
798
799 #ifdef CONFIG_CGROUP_SCHED
800         struct task_group               *sched_task_group;
801 #endif
802
803 #ifdef CONFIG_UCLAMP_TASK
804         /*
805          * Clamp values requested for a scheduling entity.
806          * Must be updated with task_rq_lock() held.
807          */
808         struct uclamp_se                uclamp_req[UCLAMP_CNT];
809         /*
810          * Effective clamp values used for a scheduling entity.
811          * Must be updated with task_rq_lock() held.
812          */
813         struct uclamp_se                uclamp[UCLAMP_CNT];
814 #endif
815
816         struct sched_statistics         stats;
817
818 #ifdef CONFIG_PREEMPT_NOTIFIERS
819         /* List of struct preempt_notifier: */
820         struct hlist_head               preempt_notifiers;
821 #endif
822
823 #ifdef CONFIG_BLK_DEV_IO_TRACE
824         unsigned int                    btrace_seq;
825 #endif
826
827         unsigned int                    policy;
828         int                             nr_cpus_allowed;
829         const cpumask_t                 *cpus_ptr;
830         cpumask_t                       *user_cpus_ptr;
831         cpumask_t                       cpus_mask;
832         void                            *migration_pending;
833 #ifdef CONFIG_SMP
834         unsigned short                  migration_disabled;
835 #endif
836         unsigned short                  migration_flags;
837
838 #ifdef CONFIG_PREEMPT_RCU
839         int                             rcu_read_lock_nesting;
840         union rcu_special               rcu_read_unlock_special;
841         struct list_head                rcu_node_entry;
842         struct rcu_node                 *rcu_blocked_node;
843 #endif /* #ifdef CONFIG_PREEMPT_RCU */
844
845 #ifdef CONFIG_TASKS_RCU
846         unsigned long                   rcu_tasks_nvcsw;
847         u8                              rcu_tasks_holdout;
848         u8                              rcu_tasks_idx;
849         int                             rcu_tasks_idle_cpu;
850         struct list_head                rcu_tasks_holdout_list;
851 #endif /* #ifdef CONFIG_TASKS_RCU */
852
853 #ifdef CONFIG_TASKS_TRACE_RCU
854         int                             trc_reader_nesting;
855         int                             trc_ipi_to_cpu;
856         union rcu_special               trc_reader_special;
857         struct list_head                trc_holdout_list;
858         struct list_head                trc_blkd_node;
859         int                             trc_blkd_cpu;
860 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
861
862         struct sched_info               sched_info;
863
864         struct list_head                tasks;
865 #ifdef CONFIG_SMP
866         struct plist_node               pushable_tasks;
867         struct rb_node                  pushable_dl_tasks;
868 #endif
869
870         struct mm_struct                *mm;
871         struct mm_struct                *active_mm;
872
873         int                             exit_state;
874         int                             exit_code;
875         int                             exit_signal;
876         /* The signal sent when the parent dies: */
877         int                             pdeath_signal;
878         /* JOBCTL_*, siglock protected: */
879         unsigned long                   jobctl;
880
881         /* Used for emulating ABI behavior of previous Linux versions: */
882         unsigned int                    personality;
883
884         /* Scheduler bits, serialized by scheduler locks: */
885         unsigned                        sched_reset_on_fork:1;
886         unsigned                        sched_contributes_to_load:1;
887         unsigned                        sched_migrated:1;
888
889         /* Force alignment to the next boundary: */
890         unsigned                        :0;
891
892         /* Unserialized, strictly 'current' */
893
894         /*
895          * This field must not be in the scheduler word above due to wakelist
896          * queueing no longer being serialized by p->on_cpu. However:
897          *
898          * p->XXX = X;                  ttwu()
899          * schedule()                     if (p->on_rq && ..) // false
900          *   smp_mb__after_spinlock();    if (smp_load_acquire(&p->on_cpu) && //true
901          *   deactivate_task()                ttwu_queue_wakelist())
902          *     p->on_rq = 0;                    p->sched_remote_wakeup = Y;
903          *
904          * guarantees all stores of 'current' are visible before
905          * ->sched_remote_wakeup gets used, so it can be in this word.
906          */
907         unsigned                        sched_remote_wakeup:1;
908
909         /* Bit to tell LSMs we're in execve(): */
910         unsigned                        in_execve:1;
911         unsigned                        in_iowait:1;
912 #ifndef TIF_RESTORE_SIGMASK
913         unsigned                        restore_sigmask:1;
914 #endif
915 #ifdef CONFIG_MEMCG
916         unsigned                        in_user_fault:1;
917 #endif
918 #ifdef CONFIG_LRU_GEN
919         /* whether the LRU algorithm may apply to this access */
920         unsigned                        in_lru_fault:1;
921 #endif
922 #ifdef CONFIG_COMPAT_BRK
923         unsigned                        brk_randomized:1;
924 #endif
925 #ifdef CONFIG_CGROUPS
926         /* disallow userland-initiated cgroup migration */
927         unsigned                        no_cgroup_migration:1;
928         /* task is frozen/stopped (used by the cgroup freezer) */
929         unsigned                        frozen:1;
930 #endif
931 #ifdef CONFIG_BLK_CGROUP
932         unsigned                        use_memdelay:1;
933 #endif
934 #ifdef CONFIG_PSI
935         /* Stalled due to lack of memory */
936         unsigned                        in_memstall:1;
937 #endif
938 #ifdef CONFIG_PAGE_OWNER
939         /* Used by page_owner=on to detect recursion in page tracking. */
940         unsigned                        in_page_owner:1;
941 #endif
942 #ifdef CONFIG_EVENTFD
943         /* Recursion prevention for eventfd_signal() */
944         unsigned                        in_eventfd:1;
945 #endif
946 #ifdef CONFIG_IOMMU_SVA
947         unsigned                        pasid_activated:1;
948 #endif
949 #ifdef  CONFIG_CPU_SUP_INTEL
950         unsigned                        reported_split_lock:1;
951 #endif
952 #ifdef CONFIG_TASK_DELAY_ACCT
953         /* delay due to memory thrashing */
954         unsigned                        in_thrashing:1;
955 #endif
956
957         unsigned long                   atomic_flags; /* Flags requiring atomic access. */
958
959         struct restart_block            restart_block;
960
961         pid_t                           pid;
962         pid_t                           tgid;
963
964 #ifdef CONFIG_STACKPROTECTOR
965         /* Canary value for the -fstack-protector GCC feature: */
966         unsigned long                   stack_canary;
967 #endif
968         /*
969          * Pointers to the (original) parent process, youngest child, younger sibling,
970          * older sibling, respectively.  (p->father can be replaced with
971          * p->real_parent->pid)
972          */
973
974         /* Real parent process: */
975         struct task_struct __rcu        *real_parent;
976
977         /* Recipient of SIGCHLD, wait4() reports: */
978         struct task_struct __rcu        *parent;
979
980         /*
981          * Children/sibling form the list of natural children:
982          */
983         struct list_head                children;
984         struct list_head                sibling;
985         struct task_struct              *group_leader;
986
987         /*
988          * 'ptraced' is the list of tasks this task is using ptrace() on.
989          *
990          * This includes both natural children and PTRACE_ATTACH targets.
991          * 'ptrace_entry' is this task's link on the p->parent->ptraced list.
992          */
993         struct list_head                ptraced;
994         struct list_head                ptrace_entry;
995
996         /* PID/PID hash table linkage. */
997         struct pid                      *thread_pid;
998         struct hlist_node               pid_links[PIDTYPE_MAX];
999         struct list_head                thread_group;
1000         struct list_head                thread_node;
1001
1002         struct completion               *vfork_done;
1003
1004         /* CLONE_CHILD_SETTID: */
1005         int __user                      *set_child_tid;
1006
1007         /* CLONE_CHILD_CLEARTID: */
1008         int __user                      *clear_child_tid;
1009
1010         /* PF_KTHREAD | PF_IO_WORKER */
1011         void                            *worker_private;
1012
1013         u64                             utime;
1014         u64                             stime;
1015 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1016         u64                             utimescaled;
1017         u64                             stimescaled;
1018 #endif
1019         u64                             gtime;
1020         struct prev_cputime             prev_cputime;
1021 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1022         struct vtime                    vtime;
1023 #endif
1024
1025 #ifdef CONFIG_NO_HZ_FULL
1026         atomic_t                        tick_dep_mask;
1027 #endif
1028         /* Context switch counts: */
1029         unsigned long                   nvcsw;
1030         unsigned long                   nivcsw;
1031
1032         /* Monotonic time in nsecs: */
1033         u64                             start_time;
1034
1035         /* Boot based time in nsecs: */
1036         u64                             start_boottime;
1037
1038         /* MM fault and swap info: this can arguably be seen as either mm-specific or thread-specific: */
1039         unsigned long                   min_flt;
1040         unsigned long                   maj_flt;
1041
1042         /* Empty if CONFIG_POSIX_CPUTIMERS=n */
1043         struct posix_cputimers          posix_cputimers;
1044
1045 #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1046         struct posix_cputimers_work     posix_cputimers_work;
1047 #endif
1048
1049         /* Process credentials: */
1050
1051         /* Tracer's credentials at attach: */
1052         const struct cred __rcu         *ptracer_cred;
1053
1054         /* Objective and real subjective task credentials (COW): */
1055         const struct cred __rcu         *real_cred;
1056
1057         /* Effective (overridable) subjective task credentials (COW): */
1058         const struct cred __rcu         *cred;
1059
1060 #ifdef CONFIG_KEYS
1061         /* Cached requested key. */
1062         struct key                      *cached_requested_key;
1063 #endif
1064
1065         /*
1066          * executable name, excluding path.
1067          *
1068          * - normally initialized setup_new_exec()
1069          * - access it with [gs]et_task_comm()
1070          * - lock it with task_lock()
1071          */
1072         char                            comm[TASK_COMM_LEN];
1073
1074         struct nameidata                *nameidata;
1075
1076 #ifdef CONFIG_SYSVIPC
1077         struct sysv_sem                 sysvsem;
1078         struct sysv_shm                 sysvshm;
1079 #endif
1080 #ifdef CONFIG_DETECT_HUNG_TASK
1081         unsigned long                   last_switch_count;
1082         unsigned long                   last_switch_time;
1083 #endif
1084         /* Filesystem information: */
1085         struct fs_struct                *fs;
1086
1087         /* Open file information: */
1088         struct files_struct             *files;
1089
1090 #ifdef CONFIG_IO_URING
1091         struct io_uring_task            *io_uring;
1092 #endif
1093
1094         /* Namespaces: */
1095         struct nsproxy                  *nsproxy;
1096
1097         /* Signal handlers: */
1098         struct signal_struct            *signal;
1099         struct sighand_struct __rcu             *sighand;
1100         sigset_t                        blocked;
1101         sigset_t                        real_blocked;
1102         /* Restored if set_restore_sigmask() was used: */
1103         sigset_t                        saved_sigmask;
1104         struct sigpending               pending;
1105         unsigned long                   sas_ss_sp;
1106         size_t                          sas_ss_size;
1107         unsigned int                    sas_ss_flags;
1108
1109         struct callback_head            *task_works;
1110
1111 #ifdef CONFIG_AUDIT
1112 #ifdef CONFIG_AUDITSYSCALL
1113         struct audit_context            *audit_context;
1114 #endif
1115         kuid_t                          loginuid;
1116         unsigned int                    sessionid;
1117 #endif
1118         struct seccomp                  seccomp;
1119         struct syscall_user_dispatch    syscall_dispatch;
1120
1121         /* Thread group tracking: */
1122         u64                             parent_exec_id;
1123         u64                             self_exec_id;
1124
1125         /* Protection against (de-)allocation: mm, files, fs, tty, keyrings, mems_allowed, mempolicy: */
1126         spinlock_t                      alloc_lock;
1127
1128         /* Protection of the PI data structures: */
1129         raw_spinlock_t                  pi_lock;
1130
1131         struct wake_q_node              wake_q;
1132
1133 #ifdef CONFIG_RT_MUTEXES
1134         /* PI waiters blocked on a rt_mutex held by this task: */
1135         struct rb_root_cached           pi_waiters;
1136         /* Updated under owner's pi_lock and rq lock */
1137         struct task_struct              *pi_top_task;
1138         /* Deadlock detection and priority inheritance handling: */
1139         struct rt_mutex_waiter          *pi_blocked_on;
1140 #endif
1141
1142 #ifdef CONFIG_DEBUG_MUTEXES
1143         /* Mutex deadlock detection: */
1144         struct mutex_waiter             *blocked_on;
1145 #endif
1146
1147 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1148         int                             non_block_count;
1149 #endif
1150
1151 #ifdef CONFIG_TRACE_IRQFLAGS
1152         struct irqtrace_events          irqtrace;
1153         unsigned int                    hardirq_threaded;
1154         u64                             hardirq_chain_key;
1155         int                             softirqs_enabled;
1156         int                             softirq_context;
1157         int                             irq_config;
1158 #endif
1159 #ifdef CONFIG_PREEMPT_RT
1160         int                             softirq_disable_cnt;
1161 #endif
1162
1163 #ifdef CONFIG_LOCKDEP
1164 # define MAX_LOCK_DEPTH                 48UL
1165         u64                             curr_chain_key;
1166         int                             lockdep_depth;
1167         unsigned int                    lockdep_recursion;
1168         struct held_lock                held_locks[MAX_LOCK_DEPTH];
1169 #endif
1170
1171 #if defined(CONFIG_UBSAN) && !defined(CONFIG_UBSAN_TRAP)
1172         unsigned int                    in_ubsan;
1173 #endif
1174
1175         /* Journalling filesystem info: */
1176         void                            *journal_info;
1177
1178         /* Stacked block device info: */
1179         struct bio_list                 *bio_list;
1180
1181         /* Stack plugging: */
1182         struct blk_plug                 *plug;
1183
1184         /* VM state: */
1185         struct reclaim_state            *reclaim_state;
1186
1187         struct backing_dev_info         *backing_dev_info;
1188
1189         struct io_context               *io_context;
1190
1191 #ifdef CONFIG_COMPACTION
1192         struct capture_control          *capture_control;
1193 #endif
1194         /* Ptrace state: */
1195         unsigned long                   ptrace_message;
1196         kernel_siginfo_t                *last_siginfo;
1197
1198         struct task_io_accounting       ioac;
1199 #ifdef CONFIG_PSI
1200         /* Pressure stall state */
1201         unsigned int                    psi_flags;
1202 #endif
1203 #ifdef CONFIG_TASK_XACCT
1204         /* Accumulated RSS usage: */
1205         u64                             acct_rss_mem1;
1206         /* Accumulated virtual memory usage: */
1207         u64                             acct_vm_mem1;
1208         /* stime + utime since last update: */
1209         u64                             acct_timexpd;
1210 #endif
1211 #ifdef CONFIG_CPUSETS
1212         /* Protected by ->alloc_lock: */
1213         nodemask_t                      mems_allowed;
1214         /* Sequence number to catch updates: */
1215         seqcount_spinlock_t             mems_allowed_seq;
1216         int                             cpuset_mem_spread_rotor;
1217         int                             cpuset_slab_spread_rotor;
1218 #endif
1219 #ifdef CONFIG_CGROUPS
1220         /* Control Group info protected by css_set_lock: */
1221         struct css_set __rcu            *cgroups;
1222         /* cg_list protected by css_set_lock and tsk->alloc_lock: */
1223         struct list_head                cg_list;
1224 #endif
1225 #ifdef CONFIG_X86_CPU_RESCTRL
1226         u32                             closid;
1227         u32                             rmid;
1228 #endif
1229 #ifdef CONFIG_FUTEX
1230         struct robust_list_head __user  *robust_list;
1231 #ifdef CONFIG_COMPAT
1232         struct compat_robust_list_head __user *compat_robust_list;
1233 #endif
1234         struct list_head                pi_state_list;
1235         struct futex_pi_state           *pi_state_cache;
1236         struct mutex                    futex_exit_mutex;
1237         unsigned int                    futex_state;
1238 #endif
1239 #ifdef CONFIG_PERF_EVENTS
1240         struct perf_event_context       *perf_event_ctxp;
1241         struct mutex                    perf_event_mutex;
1242         struct list_head                perf_event_list;
1243 #endif
1244 #ifdef CONFIG_DEBUG_PREEMPT
1245         unsigned long                   preempt_disable_ip;
1246 #endif
1247 #ifdef CONFIG_NUMA
1248         /* Protected by alloc_lock: */
1249         struct mempolicy                *mempolicy;
1250         short                           il_prev;
1251         short                           pref_node_fork;
1252 #endif
1253 #ifdef CONFIG_NUMA_BALANCING
1254         int                             numa_scan_seq;
1255         unsigned int                    numa_scan_period;
1256         unsigned int                    numa_scan_period_max;
1257         int                             numa_preferred_nid;
1258         unsigned long                   numa_migrate_retry;
1259         /* Migration stamp: */
1260         u64                             node_stamp;
1261         u64                             last_task_numa_placement;
1262         u64                             last_sum_exec_runtime;
1263         struct callback_head            numa_work;
1264
1265         /*
1266          * This pointer is only modified for current in syscall and
1267          * pagefault context (and for tasks being destroyed), so it can be read
1268          * from any of the following contexts:
1269          *  - RCU read-side critical section
1270          *  - current->numa_group from everywhere
1271          *  - task's runqueue locked, task not running
1272          */
1273         struct numa_group __rcu         *numa_group;
1274
1275         /*
1276          * numa_faults is an array split into four regions:
1277          * faults_memory, faults_cpu, faults_memory_buffer, faults_cpu_buffer
1278          * in this precise order.
1279          *
1280          * faults_memory: Exponential decaying average of faults on a per-node
1281          * basis. Scheduling placement decisions are made based on these
1282          * counts. The values remain static for the duration of a PTE scan.
1283          * faults_cpu: Track the nodes the process was running on when a NUMA
1284          * hinting fault was incurred.
1285          * faults_memory_buffer and faults_cpu_buffer: Record faults per node
1286          * during the current scan window. When the scan completes, the counts
1287          * in faults_memory and faults_cpu decay and these values are copied.
1288          */
1289         unsigned long                   *numa_faults;
1290         unsigned long                   total_numa_faults;
1291
1292         /*
1293          * numa_faults_locality tracks if faults recorded during the last
1294          * scan window were remote/local or failed to migrate. The task scan
1295          * period is adapted based on the locality of the faults with different
1296          * weights depending on whether they were shared or private faults
1297          */
1298         unsigned long                   numa_faults_locality[3];
1299
1300         unsigned long                   numa_pages_migrated;
1301 #endif /* CONFIG_NUMA_BALANCING */
1302
1303 #ifdef CONFIG_RSEQ
1304         struct rseq __user *rseq;
1305         u32 rseq_len;
1306         u32 rseq_sig;
1307         /*
1308          * RmW on rseq_event_mask must be performed atomically
1309          * with respect to preemption.
1310          */
1311         unsigned long rseq_event_mask;
1312 #endif
1313
1314 #ifdef CONFIG_SCHED_MM_CID
1315         int                             mm_cid;         /* Current cid in mm */
1316         int                             mm_cid_active;  /* Whether cid bitmap is active */
1317 #endif
1318
1319         struct tlbflush_unmap_batch     tlb_ubc;
1320
1321         union {
1322                 refcount_t              rcu_users;
1323                 struct rcu_head         rcu;
1324         };
1325
1326         /* Cache last used pipe for splice(): */
1327         struct pipe_inode_info          *splice_pipe;
1328
1329         struct page_frag                task_frag;
1330
1331 #ifdef CONFIG_TASK_DELAY_ACCT
1332         struct task_delay_info          *delays;
1333 #endif
1334
1335 #ifdef CONFIG_FAULT_INJECTION
1336         int                             make_it_fail;
1337         unsigned int                    fail_nth;
1338 #endif
1339         /*
1340          * When (nr_dirtied >= nr_dirtied_pause), it's time to call
1341          * balance_dirty_pages() for a dirty throttling pause:
1342          */
1343         int                             nr_dirtied;
1344         int                             nr_dirtied_pause;
1345         /* Start of a write-and-pause period: */
1346         unsigned long                   dirty_paused_when;
1347
1348 #ifdef CONFIG_LATENCYTOP
1349         int                             latency_record_count;
1350         struct latency_record           latency_record[LT_SAVECOUNT];
1351 #endif
1352         /*
1353          * Time slack values; these are used to round up poll() and
1354          * select() etc timeout values. These are in nanoseconds.
1355          */
1356         u64                             timer_slack_ns;
1357         u64                             default_timer_slack_ns;
1358
1359 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
1360         unsigned int                    kasan_depth;
1361 #endif
1362
1363 #ifdef CONFIG_KCSAN
1364         struct kcsan_ctx                kcsan_ctx;
1365 #ifdef CONFIG_TRACE_IRQFLAGS
1366         struct irqtrace_events          kcsan_save_irqtrace;
1367 #endif
1368 #ifdef CONFIG_KCSAN_WEAK_MEMORY
1369         int                             kcsan_stack_depth;
1370 #endif
1371 #endif
1372
1373 #ifdef CONFIG_KMSAN
1374         struct kmsan_ctx                kmsan_ctx;
1375 #endif
1376
1377 #if IS_ENABLED(CONFIG_KUNIT)
1378         struct kunit                    *kunit_test;
1379 #endif
1380
1381 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1382         /* Index of current stored address in ret_stack: */
1383         int                             curr_ret_stack;
1384         int                             curr_ret_depth;
1385
1386         /* Stack of return addresses for return function tracing: */
1387         struct ftrace_ret_stack         *ret_stack;
1388
1389         /* Timestamp for last schedule: */
1390         unsigned long long              ftrace_timestamp;
1391
1392         /*
1393          * Number of functions that haven't been traced
1394          * because of depth overrun:
1395          */
1396         atomic_t                        trace_overrun;
1397
1398         /* Pause tracing: */
1399         atomic_t                        tracing_graph_pause;
1400 #endif
1401
1402 #ifdef CONFIG_TRACING
1403         /* Bitmask and counter of trace recursion: */
1404         unsigned long                   trace_recursion;
1405 #endif /* CONFIG_TRACING */
1406
1407 #ifdef CONFIG_KCOV
1408         /* See kernel/kcov.c for more details. */
1409
1410         /* Coverage collection mode enabled for this task (0 if disabled): */
1411         unsigned int                    kcov_mode;
1412
1413         /* Size of the kcov_area: */
1414         unsigned int                    kcov_size;
1415
1416         /* Buffer for coverage collection: */
1417         void                            *kcov_area;
1418
1419         /* KCOV descriptor wired with this task or NULL: */
1420         struct kcov                     *kcov;
1421
1422         /* KCOV common handle for remote coverage collection: */
1423         u64                             kcov_handle;
1424
1425         /* KCOV sequence number: */
1426         int                             kcov_sequence;
1427
1428         /* Collect coverage from softirq context: */
1429         unsigned int                    kcov_softirq;
1430 #endif
1431
1432 #ifdef CONFIG_MEMCG
1433         struct mem_cgroup               *memcg_in_oom;
1434         gfp_t                           memcg_oom_gfp_mask;
1435         int                             memcg_oom_order;
1436
1437         /* Number of pages to reclaim on returning to userland: */
1438         unsigned int                    memcg_nr_pages_over_high;
1439
1440         /* Used by memcontrol for targeted memcg charge: */
1441         struct mem_cgroup               *active_memcg;
1442 #endif
1443
1444 #ifdef CONFIG_BLK_CGROUP
1445         struct gendisk                  *throttle_disk;
1446 #endif
1447
1448 #ifdef CONFIG_UPROBES
1449         struct uprobe_task              *utask;
1450 #endif
1451 #if defined(CONFIG_BCACHE) || defined(CONFIG_BCACHE_MODULE)
1452         unsigned int                    sequential_io;
1453         unsigned int                    sequential_io_avg;
1454 #endif
1455         struct kmap_ctrl                kmap_ctrl;
1456 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
1457         unsigned long                   task_state_change;
1458 # ifdef CONFIG_PREEMPT_RT
1459         unsigned long                   saved_state_change;
1460 # endif
1461 #endif
1462         int                             pagefault_disabled;
1463 #ifdef CONFIG_MMU
1464         struct task_struct              *oom_reaper_list;
1465         struct timer_list               oom_reaper_timer;
1466 #endif
1467 #ifdef CONFIG_VMAP_STACK
1468         struct vm_struct                *stack_vm_area;
1469 #endif
1470 #ifdef CONFIG_THREAD_INFO_IN_TASK
1471         /* A live task holds one reference: */
1472         refcount_t                      stack_refcount;
1473 #endif
1474 #ifdef CONFIG_LIVEPATCH
1475         int patch_state;
1476 #endif
1477 #ifdef CONFIG_SECURITY
1478         /* Used by LSM modules for access restriction: */
1479         void                            *security;
1480 #endif
1481 #ifdef CONFIG_BPF_SYSCALL
1482         /* Used by BPF task local storage */
1483         struct bpf_local_storage __rcu  *bpf_storage;
1484         /* Used for BPF run context */
1485         struct bpf_run_ctx              *bpf_ctx;
1486 #endif
1487
1488 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
1489         unsigned long                   lowest_stack;
1490         unsigned long                   prev_lowest_stack;
1491 #endif
1492
1493 #ifdef CONFIG_X86_MCE
1494         void __user                     *mce_vaddr;
1495         __u64                           mce_kflags;
1496         u64                             mce_addr;
1497         __u64                           mce_ripv : 1,
1498                                         mce_whole_page : 1,
1499                                         __mce_reserved : 62;
1500         struct callback_head            mce_kill_me;
1501         int                             mce_count;
1502 #endif
1503
1504 #ifdef CONFIG_KRETPROBES
1505         struct llist_head               kretprobe_instances;
1506 #endif
1507 #ifdef CONFIG_RETHOOK
1508         struct llist_head               rethooks;
1509 #endif
1510
1511 #ifdef CONFIG_ARCH_HAS_PARANOID_L1D_FLUSH
1512         /*
1513          * If L1D flush is supported on mm context switch
1514          * then we use this callback head to queue kill work
1515          * to kill tasks that are not running on SMT disabled
1516          * cores
1517          */
1518         struct callback_head            l1d_flush_kill;
1519 #endif
1520
1521 #ifdef CONFIG_RV
1522         /*
1523          * Per-task RV monitor. Nowadays fixed in RV_PER_TASK_MONITORS.
1524          * If we find justification for more monitors, we can think
1525          * about adding more or developing a dynamic method. So far,
1526          * none of these are justified.
1527          */
1528         union rv_task_monitor           rv[RV_PER_TASK_MONITORS];
1529 #endif
1530
1531         /*
1532          * New fields for task_struct should be added above here, so that
1533          * they are included in the randomized portion of task_struct.
1534          */
1535         randomized_struct_fields_end
1536
1537         /* CPU-specific state of this task: */
1538         struct thread_struct            thread;
1539
1540         /*
1541          * WARNING: on x86, 'thread_struct' contains a variable-sized
1542          * structure.  It *MUST* be at the end of 'task_struct'.
1543          *
1544          * Do not put anything below here!
1545          */
1546 };
1547
1548 static inline struct pid *task_pid(struct task_struct *task)
1549 {
1550         return task->thread_pid;
1551 }
1552
1553 /*
1554  * the helpers to get the task's different pids as they are seen
1555  * from various namespaces
1556  *
1557  * task_xid_nr()     : global id, i.e. the id seen from the init namespace;
1558  * task_xid_vnr()    : virtual id, i.e. the id seen from the pid namespace of
1559  *                     current.
1560  * task_xid_nr_ns()  : id seen from the ns specified;
1561  *
1562  * see also pid_nr() etc in include/linux/pid.h
1563  */
1564 pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type, struct pid_namespace *ns);
1565
1566 static inline pid_t task_pid_nr(struct task_struct *tsk)
1567 {
1568         return tsk->pid;
1569 }
1570
1571 static inline pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1572 {
1573         return __task_pid_nr_ns(tsk, PIDTYPE_PID, ns);
1574 }
1575
1576 static inline pid_t task_pid_vnr(struct task_struct *tsk)
1577 {
1578         return __task_pid_nr_ns(tsk, PIDTYPE_PID, NULL);
1579 }
1580
1581
1582 static inline pid_t task_tgid_nr(struct task_struct *tsk)
1583 {
1584         return tsk->tgid;
1585 }
1586
1587 /**
1588  * pid_alive - check that a task structure is not stale
1589  * @p: Task structure to be checked.
1590  *
1591  * Test if a process is not yet dead (at most zombie state)
1592  * If pid_alive fails, then pointers within the task structure
1593  * can be stale and must not be dereferenced.
1594  *
1595  * Return: 1 if the process is alive. 0 otherwise.
1596  */
1597 static inline int pid_alive(const struct task_struct *p)
1598 {
1599         return p->thread_pid != NULL;
1600 }
1601
1602 static inline pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1603 {
1604         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, ns);
1605 }
1606
1607 static inline pid_t task_pgrp_vnr(struct task_struct *tsk)
1608 {
1609         return __task_pid_nr_ns(tsk, PIDTYPE_PGID, NULL);
1610 }
1611
1612
1613 static inline pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1614 {
1615         return __task_pid_nr_ns(tsk, PIDTYPE_SID, ns);
1616 }
1617
1618 static inline pid_t task_session_vnr(struct task_struct *tsk)
1619 {
1620         return __task_pid_nr_ns(tsk, PIDTYPE_SID, NULL);
1621 }
1622
1623 static inline pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
1624 {
1625         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, ns);
1626 }
1627
1628 static inline pid_t task_tgid_vnr(struct task_struct *tsk)
1629 {
1630         return __task_pid_nr_ns(tsk, PIDTYPE_TGID, NULL);
1631 }
1632
1633 static inline pid_t task_ppid_nr_ns(const struct task_struct *tsk, struct pid_namespace *ns)
1634 {
1635         pid_t pid = 0;
1636
1637         rcu_read_lock();
1638         if (pid_alive(tsk))
1639                 pid = task_tgid_nr_ns(rcu_dereference(tsk->real_parent), ns);
1640         rcu_read_unlock();
1641
1642         return pid;
1643 }
1644
1645 static inline pid_t task_ppid_nr(const struct task_struct *tsk)
1646 {
1647         return task_ppid_nr_ns(tsk, &init_pid_ns);
1648 }
1649
1650 /* Obsolete, do not use: */
1651 static inline pid_t task_pgrp_nr(struct task_struct *tsk)
1652 {
1653         return task_pgrp_nr_ns(tsk, &init_pid_ns);
1654 }
1655
1656 #define TASK_REPORT_IDLE        (TASK_REPORT + 1)
1657 #define TASK_REPORT_MAX         (TASK_REPORT_IDLE << 1)
1658
1659 static inline unsigned int __task_state_index(unsigned int tsk_state,
1660                                               unsigned int tsk_exit_state)
1661 {
1662         unsigned int state = (tsk_state | tsk_exit_state) & TASK_REPORT;
1663
1664         BUILD_BUG_ON_NOT_POWER_OF_2(TASK_REPORT_MAX);
1665
1666         if (tsk_state == TASK_IDLE)
1667                 state = TASK_REPORT_IDLE;
1668
1669         /*
1670          * We're lying here, but rather than expose a completely new task state
1671          * to userspace, we can make this appear as if the task has gone through
1672          * a regular rt_mutex_lock() call.
1673          */
1674         if (tsk_state == TASK_RTLOCK_WAIT)
1675                 state = TASK_UNINTERRUPTIBLE;
1676
1677         return fls(state);
1678 }
1679
1680 static inline unsigned int task_state_index(struct task_struct *tsk)
1681 {
1682         return __task_state_index(READ_ONCE(tsk->__state), tsk->exit_state);
1683 }
1684
1685 static inline char task_index_to_char(unsigned int state)
1686 {
1687         static const char state_char[] = "RSDTtXZPI";
1688
1689         BUILD_BUG_ON(1 + ilog2(TASK_REPORT_MAX) != sizeof(state_char) - 1);
1690
1691         return state_char[state];
1692 }
1693
1694 static inline char task_state_to_char(struct task_struct *tsk)
1695 {
1696         return task_index_to_char(task_state_index(tsk));
1697 }
1698
1699 /**
1700  * is_global_init - check if a task structure is init. Since init
1701  * is free to have sub-threads we need to check tgid.
1702  * @tsk: Task structure to be checked.
1703  *
1704  * Check if a task structure is the first user space task the kernel created.
1705  *
1706  * Return: 1 if the task structure is init. 0 otherwise.
1707  */
1708 static inline int is_global_init(struct task_struct *tsk)
1709 {
1710         return task_tgid_nr(tsk) == 1;
1711 }
1712
1713 extern struct pid *cad_pid;
1714
1715 /*
1716  * Per process flags
1717  */
1718 #define PF_VCPU                 0x00000001      /* I'm a virtual CPU */
1719 #define PF_IDLE                 0x00000002      /* I am an IDLE thread */
1720 #define PF_EXITING              0x00000004      /* Getting shut down */
1721 #define PF_POSTCOREDUMP         0x00000008      /* Coredumps should ignore this task */
1722 #define PF_IO_WORKER            0x00000010      /* Task is an IO worker */
1723 #define PF_WQ_WORKER            0x00000020      /* I'm a workqueue worker */
1724 #define PF_FORKNOEXEC           0x00000040      /* Forked but didn't exec */
1725 #define PF_MCE_PROCESS          0x00000080      /* Process policy on mce errors */
1726 #define PF_SUPERPRIV            0x00000100      /* Used super-user privileges */
1727 #define PF_DUMPCORE             0x00000200      /* Dumped core */
1728 #define PF_SIGNALED             0x00000400      /* Killed by a signal */
1729 #define PF_MEMALLOC             0x00000800      /* Allocating memory */
1730 #define PF_NPROC_EXCEEDED       0x00001000      /* set_user() noticed that RLIMIT_NPROC was exceeded */
1731 #define PF_USED_MATH            0x00002000      /* If unset the fpu must be initialized before use */
1732 #define PF__HOLE__00004000      0x00004000
1733 #define PF_NOFREEZE             0x00008000      /* This thread should not be frozen */
1734 #define PF__HOLE__00010000      0x00010000
1735 #define PF_KSWAPD               0x00020000      /* I am kswapd */
1736 #define PF_MEMALLOC_NOFS        0x00040000      /* All allocation requests will inherit GFP_NOFS */
1737 #define PF_MEMALLOC_NOIO        0x00080000      /* All allocation requests will inherit GFP_NOIO */
1738 #define PF_LOCAL_THROTTLE       0x00100000      /* Throttle writes only against the bdi I write to,
1739                                                  * I am cleaning dirty pages from some other bdi. */
1740 #define PF_KTHREAD              0x00200000      /* I am a kernel thread */
1741 #define PF_RANDOMIZE            0x00400000      /* Randomize virtual address space */
1742 #define PF__HOLE__00800000      0x00800000
1743 #define PF__HOLE__01000000      0x01000000
1744 #define PF__HOLE__02000000      0x02000000
1745 #define PF_NO_SETAFFINITY       0x04000000      /* Userland is not allowed to meddle with cpus_mask */
1746 #define PF_MCE_EARLY            0x08000000      /* Early kill for mce process policy */
1747 #define PF_MEMALLOC_PIN         0x10000000      /* Allocation context constrained to zones which allow long term pinning. */
1748 #define PF__HOLE__20000000      0x20000000
1749 #define PF__HOLE__40000000      0x40000000
1750 #define PF_SUSPEND_TASK         0x80000000      /* This thread called freeze_processes() and should not be frozen */
1751
1752 /*
1753  * Only the _current_ task can read/write to tsk->flags, but other
1754  * tasks can access tsk->flags in readonly mode for example
1755  * with tsk_used_math (like during threaded core dumping).
1756  * There is however an exception to this rule during ptrace
1757  * or during fork: the ptracer task is allowed to write to the
1758  * child->flags of its traced child (same goes for fork, the parent
1759  * can write to the child->flags), because we're guaranteed the
1760  * child is not running and in turn not changing child->flags
1761  * at the same time the parent does it.
1762  */
1763 #define clear_stopped_child_used_math(child)    do { (child)->flags &= ~PF_USED_MATH; } while (0)
1764 #define set_stopped_child_used_math(child)      do { (child)->flags |= PF_USED_MATH; } while (0)
1765 #define clear_used_math()                       clear_stopped_child_used_math(current)
1766 #define set_used_math()                         set_stopped_child_used_math(current)
1767
1768 #define conditional_stopped_child_used_math(condition, child) \
1769         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= (condition) ? PF_USED_MATH : 0; } while (0)
1770
1771 #define conditional_used_math(condition)        conditional_stopped_child_used_math(condition, current)
1772
1773 #define copy_to_stopped_child_used_math(child) \
1774         do { (child)->flags &= ~PF_USED_MATH, (child)->flags |= current->flags & PF_USED_MATH; } while (0)
1775
1776 /* NOTE: this will return 0 or PF_USED_MATH, it will never return 1 */
1777 #define tsk_used_math(p)                        ((p)->flags & PF_USED_MATH)
1778 #define used_math()                             tsk_used_math(current)
1779
1780 static __always_inline bool is_percpu_thread(void)
1781 {
1782 #ifdef CONFIG_SMP
1783         return (current->flags & PF_NO_SETAFFINITY) &&
1784                 (current->nr_cpus_allowed  == 1);
1785 #else
1786         return true;
1787 #endif
1788 }
1789
1790 /* Per-process atomic flags. */
1791 #define PFA_NO_NEW_PRIVS                0       /* May not gain new privileges. */
1792 #define PFA_SPREAD_PAGE                 1       /* Spread page cache over cpuset */
1793 #define PFA_SPREAD_SLAB                 2       /* Spread some slab caches over cpuset */
1794 #define PFA_SPEC_SSB_DISABLE            3       /* Speculative Store Bypass disabled */
1795 #define PFA_SPEC_SSB_FORCE_DISABLE      4       /* Speculative Store Bypass force disabled*/
1796 #define PFA_SPEC_IB_DISABLE             5       /* Indirect branch speculation restricted */
1797 #define PFA_SPEC_IB_FORCE_DISABLE       6       /* Indirect branch speculation permanently restricted */
1798 #define PFA_SPEC_SSB_NOEXEC             7       /* Speculative Store Bypass clear on execve() */
1799
1800 #define TASK_PFA_TEST(name, func)                                       \
1801         static inline bool task_##func(struct task_struct *p)           \
1802         { return test_bit(PFA_##name, &p->atomic_flags); }
1803
1804 #define TASK_PFA_SET(name, func)                                        \
1805         static inline void task_set_##func(struct task_struct *p)       \
1806         { set_bit(PFA_##name, &p->atomic_flags); }
1807
1808 #define TASK_PFA_CLEAR(name, func)                                      \
1809         static inline void task_clear_##func(struct task_struct *p)     \
1810         { clear_bit(PFA_##name, &p->atomic_flags); }
1811
1812 TASK_PFA_TEST(NO_NEW_PRIVS, no_new_privs)
1813 TASK_PFA_SET(NO_NEW_PRIVS, no_new_privs)
1814
1815 TASK_PFA_TEST(SPREAD_PAGE, spread_page)
1816 TASK_PFA_SET(SPREAD_PAGE, spread_page)
1817 TASK_PFA_CLEAR(SPREAD_PAGE, spread_page)
1818
1819 TASK_PFA_TEST(SPREAD_SLAB, spread_slab)
1820 TASK_PFA_SET(SPREAD_SLAB, spread_slab)
1821 TASK_PFA_CLEAR(SPREAD_SLAB, spread_slab)
1822
1823 TASK_PFA_TEST(SPEC_SSB_DISABLE, spec_ssb_disable)
1824 TASK_PFA_SET(SPEC_SSB_DISABLE, spec_ssb_disable)
1825 TASK_PFA_CLEAR(SPEC_SSB_DISABLE, spec_ssb_disable)
1826
1827 TASK_PFA_TEST(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1828 TASK_PFA_SET(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1829 TASK_PFA_CLEAR(SPEC_SSB_NOEXEC, spec_ssb_noexec)
1830
1831 TASK_PFA_TEST(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1832 TASK_PFA_SET(SPEC_SSB_FORCE_DISABLE, spec_ssb_force_disable)
1833
1834 TASK_PFA_TEST(SPEC_IB_DISABLE, spec_ib_disable)
1835 TASK_PFA_SET(SPEC_IB_DISABLE, spec_ib_disable)
1836 TASK_PFA_CLEAR(SPEC_IB_DISABLE, spec_ib_disable)
1837
1838 TASK_PFA_TEST(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1839 TASK_PFA_SET(SPEC_IB_FORCE_DISABLE, spec_ib_force_disable)
1840
1841 static inline void
1842 current_restore_flags(unsigned long orig_flags, unsigned long flags)
1843 {
1844         current->flags &= ~flags;
1845         current->flags |= orig_flags & flags;
1846 }
1847
1848 extern int cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial);
1849 extern int task_can_attach(struct task_struct *p, const struct cpumask *cs_effective_cpus);
1850 #ifdef CONFIG_SMP
1851 extern void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask);
1852 extern int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask);
1853 extern int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node);
1854 extern void release_user_cpus_ptr(struct task_struct *p);
1855 extern int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask);
1856 extern void force_compatible_cpus_allowed_ptr(struct task_struct *p);
1857 extern void relax_compatible_cpus_allowed_ptr(struct task_struct *p);
1858 #else
1859 static inline void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
1860 {
1861 }
1862 static inline int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
1863 {
1864         if (!cpumask_test_cpu(0, new_mask))
1865                 return -EINVAL;
1866         return 0;
1867 }
1868 static inline int dup_user_cpus_ptr(struct task_struct *dst, struct task_struct *src, int node)
1869 {
1870         if (src->user_cpus_ptr)
1871                 return -EINVAL;
1872         return 0;
1873 }
1874 static inline void release_user_cpus_ptr(struct task_struct *p)
1875 {
1876         WARN_ON(p->user_cpus_ptr);
1877 }
1878
1879 static inline int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask)
1880 {
1881         return 0;
1882 }
1883 #endif
1884
1885 extern int yield_to(struct task_struct *p, bool preempt);
1886 extern void set_user_nice(struct task_struct *p, long nice);
1887 extern int task_prio(const struct task_struct *p);
1888
1889 /**
1890  * task_nice - return the nice value of a given task.
1891  * @p: the task in question.
1892  *
1893  * Return: The nice value [ -20 ... 0 ... 19 ].
1894  */
1895 static inline int task_nice(const struct task_struct *p)
1896 {
1897         return PRIO_TO_NICE((p)->static_prio);
1898 }
1899
1900 extern int can_nice(const struct task_struct *p, const int nice);
1901 extern int task_curr(const struct task_struct *p);
1902 extern int idle_cpu(int cpu);
1903 extern int available_idle_cpu(int cpu);
1904 extern int sched_setscheduler(struct task_struct *, int, const struct sched_param *);
1905 extern int sched_setscheduler_nocheck(struct task_struct *, int, const struct sched_param *);
1906 extern void sched_set_fifo(struct task_struct *p);
1907 extern void sched_set_fifo_low(struct task_struct *p);
1908 extern void sched_set_normal(struct task_struct *p, int nice);
1909 extern int sched_setattr(struct task_struct *, const struct sched_attr *);
1910 extern int sched_setattr_nocheck(struct task_struct *, const struct sched_attr *);
1911 extern struct task_struct *idle_task(int cpu);
1912
1913 /**
1914  * is_idle_task - is the specified task an idle task?
1915  * @p: the task in question.
1916  *
1917  * Return: 1 if @p is an idle task. 0 otherwise.
1918  */
1919 static __always_inline bool is_idle_task(const struct task_struct *p)
1920 {
1921         return !!(p->flags & PF_IDLE);
1922 }
1923
1924 extern struct task_struct *curr_task(int cpu);
1925 extern void ia64_set_curr_task(int cpu, struct task_struct *p);
1926
1927 void yield(void);
1928
1929 union thread_union {
1930 #ifndef CONFIG_ARCH_TASK_STRUCT_ON_STACK
1931         struct task_struct task;
1932 #endif
1933 #ifndef CONFIG_THREAD_INFO_IN_TASK
1934         struct thread_info thread_info;
1935 #endif
1936         unsigned long stack[THREAD_SIZE/sizeof(long)];
1937 };
1938
1939 #ifndef CONFIG_THREAD_INFO_IN_TASK
1940 extern struct thread_info init_thread_info;
1941 #endif
1942
1943 extern unsigned long init_stack[THREAD_SIZE / sizeof(unsigned long)];
1944
1945 #ifdef CONFIG_THREAD_INFO_IN_TASK
1946 # define task_thread_info(task) (&(task)->thread_info)
1947 #elif !defined(__HAVE_THREAD_FUNCTIONS)
1948 # define task_thread_info(task) ((struct thread_info *)(task)->stack)
1949 #endif
1950
1951 /*
1952  * find a task by one of its numerical ids
1953  *
1954  * find_task_by_pid_ns():
1955  *      finds a task by its pid in the specified namespace
1956  * find_task_by_vpid():
1957  *      finds a task by its virtual pid
1958  *
1959  * see also find_vpid() etc in include/linux/pid.h
1960  */
1961
1962 extern struct task_struct *find_task_by_vpid(pid_t nr);
1963 extern struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns);
1964
1965 /*
1966  * find a task by its virtual pid and get the task struct
1967  */
1968 extern struct task_struct *find_get_task_by_vpid(pid_t nr);
1969
1970 extern int wake_up_state(struct task_struct *tsk, unsigned int state);
1971 extern int wake_up_process(struct task_struct *tsk);
1972 extern void wake_up_new_task(struct task_struct *tsk);
1973
1974 #ifdef CONFIG_SMP
1975 extern void kick_process(struct task_struct *tsk);
1976 #else
1977 static inline void kick_process(struct task_struct *tsk) { }
1978 #endif
1979
1980 extern void __set_task_comm(struct task_struct *tsk, const char *from, bool exec);
1981
1982 static inline void set_task_comm(struct task_struct *tsk, const char *from)
1983 {
1984         __set_task_comm(tsk, from, false);
1985 }
1986
1987 extern char *__get_task_comm(char *to, size_t len, struct task_struct *tsk);
1988 #define get_task_comm(buf, tsk) ({                      \
1989         BUILD_BUG_ON(sizeof(buf) != TASK_COMM_LEN);     \
1990         __get_task_comm(buf, sizeof(buf), tsk);         \
1991 })
1992
1993 #ifdef CONFIG_SMP
1994 static __always_inline void scheduler_ipi(void)
1995 {
1996         /*
1997          * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
1998          * TIF_NEED_RESCHED remotely (for the first time) will also send
1999          * this IPI.
2000          */
2001         preempt_fold_need_resched();
2002 }
2003 extern unsigned long wait_task_inactive(struct task_struct *, unsigned int match_state);
2004 #else
2005 static inline void scheduler_ipi(void) { }
2006 static inline unsigned long wait_task_inactive(struct task_struct *p, unsigned int match_state)
2007 {
2008         return 1;
2009 }
2010 #endif
2011
2012 /*
2013  * Set thread flags in other task's structures.
2014  * See asm/thread_info.h for TIF_xxxx flags available:
2015  */
2016 static inline void set_tsk_thread_flag(struct task_struct *tsk, int flag)
2017 {
2018         set_ti_thread_flag(task_thread_info(tsk), flag);
2019 }
2020
2021 static inline void clear_tsk_thread_flag(struct task_struct *tsk, int flag)
2022 {
2023         clear_ti_thread_flag(task_thread_info(tsk), flag);
2024 }
2025
2026 static inline void update_tsk_thread_flag(struct task_struct *tsk, int flag,
2027                                           bool value)
2028 {
2029         update_ti_thread_flag(task_thread_info(tsk), flag, value);
2030 }
2031
2032 static inline int test_and_set_tsk_thread_flag(struct task_struct *tsk, int flag)
2033 {
2034         return test_and_set_ti_thread_flag(task_thread_info(tsk), flag);
2035 }
2036
2037 static inline int test_and_clear_tsk_thread_flag(struct task_struct *tsk, int flag)
2038 {
2039         return test_and_clear_ti_thread_flag(task_thread_info(tsk), flag);
2040 }
2041
2042 static inline int test_tsk_thread_flag(struct task_struct *tsk, int flag)
2043 {
2044         return test_ti_thread_flag(task_thread_info(tsk), flag);
2045 }
2046
2047 static inline void set_tsk_need_resched(struct task_struct *tsk)
2048 {
2049         set_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2050 }
2051
2052 static inline void clear_tsk_need_resched(struct task_struct *tsk)
2053 {
2054         clear_tsk_thread_flag(tsk,TIF_NEED_RESCHED);
2055 }
2056
2057 static inline int test_tsk_need_resched(struct task_struct *tsk)
2058 {
2059         return unlikely(test_tsk_thread_flag(tsk,TIF_NEED_RESCHED));
2060 }
2061
2062 /*
2063  * cond_resched() and cond_resched_lock(): latency reduction via
2064  * explicit rescheduling in places that are safe. The return
2065  * value indicates whether a reschedule was done in fact.
2066  * cond_resched_lock() will drop the spinlock before scheduling,
2067  */
2068 #if !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC)
2069 extern int __cond_resched(void);
2070
2071 #if defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
2072
2073 DECLARE_STATIC_CALL(cond_resched, __cond_resched);
2074
2075 static __always_inline int _cond_resched(void)
2076 {
2077         return static_call_mod(cond_resched)();
2078 }
2079
2080 #elif defined(CONFIG_PREEMPT_DYNAMIC) && defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
2081 extern int dynamic_cond_resched(void);
2082
2083 static __always_inline int _cond_resched(void)
2084 {
2085         return dynamic_cond_resched();
2086 }
2087
2088 #else
2089
2090 static inline int _cond_resched(void)
2091 {
2092         return __cond_resched();
2093 }
2094
2095 #endif /* CONFIG_PREEMPT_DYNAMIC */
2096
2097 #else
2098
2099 static inline int _cond_resched(void) { return 0; }
2100
2101 #endif /* !defined(CONFIG_PREEMPTION) || defined(CONFIG_PREEMPT_DYNAMIC) */
2102
2103 #define cond_resched() ({                       \
2104         __might_resched(__FILE__, __LINE__, 0); \
2105         _cond_resched();                        \
2106 })
2107
2108 extern int __cond_resched_lock(spinlock_t *lock);
2109 extern int __cond_resched_rwlock_read(rwlock_t *lock);
2110 extern int __cond_resched_rwlock_write(rwlock_t *lock);
2111
2112 #define MIGHT_RESCHED_RCU_SHIFT         8
2113 #define MIGHT_RESCHED_PREEMPT_MASK      ((1U << MIGHT_RESCHED_RCU_SHIFT) - 1)
2114
2115 #ifndef CONFIG_PREEMPT_RT
2116 /*
2117  * Non RT kernels have an elevated preempt count due to the held lock,
2118  * but are not allowed to be inside a RCU read side critical section
2119  */
2120 # define PREEMPT_LOCK_RESCHED_OFFSETS   PREEMPT_LOCK_OFFSET
2121 #else
2122 /*
2123  * spin/rw_lock() on RT implies rcu_read_lock(). The might_sleep() check in
2124  * cond_resched*lock() has to take that into account because it checks for
2125  * preempt_count() and rcu_preempt_depth().
2126  */
2127 # define PREEMPT_LOCK_RESCHED_OFFSETS   \
2128         (PREEMPT_LOCK_OFFSET + (1U << MIGHT_RESCHED_RCU_SHIFT))
2129 #endif
2130
2131 #define cond_resched_lock(lock) ({                                              \
2132         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2133         __cond_resched_lock(lock);                                              \
2134 })
2135
2136 #define cond_resched_rwlock_read(lock) ({                                       \
2137         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2138         __cond_resched_rwlock_read(lock);                                       \
2139 })
2140
2141 #define cond_resched_rwlock_write(lock) ({                                      \
2142         __might_resched(__FILE__, __LINE__, PREEMPT_LOCK_RESCHED_OFFSETS);      \
2143         __cond_resched_rwlock_write(lock);                                      \
2144 })
2145
2146 static inline void cond_resched_rcu(void)
2147 {
2148 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP) || !defined(CONFIG_PREEMPT_RCU)
2149         rcu_read_unlock();
2150         cond_resched();
2151         rcu_read_lock();
2152 #endif
2153 }
2154
2155 #ifdef CONFIG_PREEMPT_DYNAMIC
2156
2157 extern bool preempt_model_none(void);
2158 extern bool preempt_model_voluntary(void);
2159 extern bool preempt_model_full(void);
2160
2161 #else
2162
2163 static inline bool preempt_model_none(void)
2164 {
2165         return IS_ENABLED(CONFIG_PREEMPT_NONE);
2166 }
2167 static inline bool preempt_model_voluntary(void)
2168 {
2169         return IS_ENABLED(CONFIG_PREEMPT_VOLUNTARY);
2170 }
2171 static inline bool preempt_model_full(void)
2172 {
2173         return IS_ENABLED(CONFIG_PREEMPT);
2174 }
2175
2176 #endif
2177
2178 static inline bool preempt_model_rt(void)
2179 {
2180         return IS_ENABLED(CONFIG_PREEMPT_RT);
2181 }
2182
2183 /*
2184  * Does the preemption model allow non-cooperative preemption?
2185  *
2186  * For !CONFIG_PREEMPT_DYNAMIC kernels this is an exact match with
2187  * CONFIG_PREEMPTION; for CONFIG_PREEMPT_DYNAMIC this doesn't work as the
2188  * kernel is *built* with CONFIG_PREEMPTION=y but may run with e.g. the
2189  * PREEMPT_NONE model.
2190  */
2191 static inline bool preempt_model_preemptible(void)
2192 {
2193         return preempt_model_full() || preempt_model_rt();
2194 }
2195
2196 /*
2197  * Does a critical section need to be broken due to another
2198  * task waiting?: (technically does not depend on CONFIG_PREEMPTION,
2199  * but a general need for low latency)
2200  */
2201 static inline int spin_needbreak(spinlock_t *lock)
2202 {
2203 #ifdef CONFIG_PREEMPTION
2204         return spin_is_contended(lock);
2205 #else
2206         return 0;
2207 #endif
2208 }
2209
2210 /*
2211  * Check if a rwlock is contended.
2212  * Returns non-zero if there is another task waiting on the rwlock.
2213  * Returns zero if the lock is not contended or the system / underlying
2214  * rwlock implementation does not support contention detection.
2215  * Technically does not depend on CONFIG_PREEMPTION, but a general need
2216  * for low latency.
2217  */
2218 static inline int rwlock_needbreak(rwlock_t *lock)
2219 {
2220 #ifdef CONFIG_PREEMPTION
2221         return rwlock_is_contended(lock);
2222 #else
2223         return 0;
2224 #endif
2225 }
2226
2227 static __always_inline bool need_resched(void)
2228 {
2229         return unlikely(tif_need_resched());
2230 }
2231
2232 /*
2233  * Wrappers for p->thread_info->cpu access. No-op on UP.
2234  */
2235 #ifdef CONFIG_SMP
2236
2237 static inline unsigned int task_cpu(const struct task_struct *p)
2238 {
2239         return READ_ONCE(task_thread_info(p)->cpu);
2240 }
2241
2242 extern void set_task_cpu(struct task_struct *p, unsigned int cpu);
2243
2244 #else
2245
2246 static inline unsigned int task_cpu(const struct task_struct *p)
2247 {
2248         return 0;
2249 }
2250
2251 static inline void set_task_cpu(struct task_struct *p, unsigned int cpu)
2252 {
2253 }
2254
2255 #endif /* CONFIG_SMP */
2256
2257 extern bool sched_task_on_rq(struct task_struct *p);
2258 extern unsigned long get_wchan(struct task_struct *p);
2259 extern struct task_struct *cpu_curr_snapshot(int cpu);
2260
2261 /*
2262  * In order to reduce various lock holder preemption latencies provide an
2263  * interface to see if a vCPU is currently running or not.
2264  *
2265  * This allows us to terminate optimistic spin loops and block, analogous to
2266  * the native optimistic spin heuristic of testing if the lock owner task is
2267  * running or not.
2268  */
2269 #ifndef vcpu_is_preempted
2270 static inline bool vcpu_is_preempted(int cpu)
2271 {
2272         return false;
2273 }
2274 #endif
2275
2276 extern long sched_setaffinity(pid_t pid, const struct cpumask *new_mask);
2277 extern long sched_getaffinity(pid_t pid, struct cpumask *mask);
2278
2279 #ifndef TASK_SIZE_OF
2280 #define TASK_SIZE_OF(tsk)       TASK_SIZE
2281 #endif
2282
2283 #ifdef CONFIG_SMP
2284 static inline bool owner_on_cpu(struct task_struct *owner)
2285 {
2286         /*
2287          * As lock holder preemption issue, we both skip spinning if
2288          * task is not on cpu or its cpu is preempted
2289          */
2290         return READ_ONCE(owner->on_cpu) && !vcpu_is_preempted(task_cpu(owner));
2291 }
2292
2293 /* Returns effective CPU energy utilization, as seen by the scheduler */
2294 unsigned long sched_cpu_util(int cpu);
2295 #endif /* CONFIG_SMP */
2296
2297 #ifdef CONFIG_RSEQ
2298
2299 /*
2300  * Map the event mask on the user-space ABI enum rseq_cs_flags
2301  * for direct mask checks.
2302  */
2303 enum rseq_event_mask_bits {
2304         RSEQ_EVENT_PREEMPT_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_PREEMPT_BIT,
2305         RSEQ_EVENT_SIGNAL_BIT   = RSEQ_CS_FLAG_NO_RESTART_ON_SIGNAL_BIT,
2306         RSEQ_EVENT_MIGRATE_BIT  = RSEQ_CS_FLAG_NO_RESTART_ON_MIGRATE_BIT,
2307 };
2308
2309 enum rseq_event_mask {
2310         RSEQ_EVENT_PREEMPT      = (1U << RSEQ_EVENT_PREEMPT_BIT),
2311         RSEQ_EVENT_SIGNAL       = (1U << RSEQ_EVENT_SIGNAL_BIT),
2312         RSEQ_EVENT_MIGRATE      = (1U << RSEQ_EVENT_MIGRATE_BIT),
2313 };
2314
2315 static inline void rseq_set_notify_resume(struct task_struct *t)
2316 {
2317         if (t->rseq)
2318                 set_tsk_thread_flag(t, TIF_NOTIFY_RESUME);
2319 }
2320
2321 void __rseq_handle_notify_resume(struct ksignal *sig, struct pt_regs *regs);
2322
2323 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2324                                              struct pt_regs *regs)
2325 {
2326         if (current->rseq)
2327                 __rseq_handle_notify_resume(ksig, regs);
2328 }
2329
2330 static inline void rseq_signal_deliver(struct ksignal *ksig,
2331                                        struct pt_regs *regs)
2332 {
2333         preempt_disable();
2334         __set_bit(RSEQ_EVENT_SIGNAL_BIT, &current->rseq_event_mask);
2335         preempt_enable();
2336         rseq_handle_notify_resume(ksig, regs);
2337 }
2338
2339 /* rseq_preempt() requires preemption to be disabled. */
2340 static inline void rseq_preempt(struct task_struct *t)
2341 {
2342         __set_bit(RSEQ_EVENT_PREEMPT_BIT, &t->rseq_event_mask);
2343         rseq_set_notify_resume(t);
2344 }
2345
2346 /* rseq_migrate() requires preemption to be disabled. */
2347 static inline void rseq_migrate(struct task_struct *t)
2348 {
2349         __set_bit(RSEQ_EVENT_MIGRATE_BIT, &t->rseq_event_mask);
2350         rseq_set_notify_resume(t);
2351 }
2352
2353 /*
2354  * If parent process has a registered restartable sequences area, the
2355  * child inherits. Unregister rseq for a clone with CLONE_VM set.
2356  */
2357 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2358 {
2359         if (clone_flags & CLONE_VM) {
2360                 t->rseq = NULL;
2361                 t->rseq_len = 0;
2362                 t->rseq_sig = 0;
2363                 t->rseq_event_mask = 0;
2364         } else {
2365                 t->rseq = current->rseq;
2366                 t->rseq_len = current->rseq_len;
2367                 t->rseq_sig = current->rseq_sig;
2368                 t->rseq_event_mask = current->rseq_event_mask;
2369         }
2370 }
2371
2372 static inline void rseq_execve(struct task_struct *t)
2373 {
2374         t->rseq = NULL;
2375         t->rseq_len = 0;
2376         t->rseq_sig = 0;
2377         t->rseq_event_mask = 0;
2378 }
2379
2380 #else
2381
2382 static inline void rseq_set_notify_resume(struct task_struct *t)
2383 {
2384 }
2385 static inline void rseq_handle_notify_resume(struct ksignal *ksig,
2386                                              struct pt_regs *regs)
2387 {
2388 }
2389 static inline void rseq_signal_deliver(struct ksignal *ksig,
2390                                        struct pt_regs *regs)
2391 {
2392 }
2393 static inline void rseq_preempt(struct task_struct *t)
2394 {
2395 }
2396 static inline void rseq_migrate(struct task_struct *t)
2397 {
2398 }
2399 static inline void rseq_fork(struct task_struct *t, unsigned long clone_flags)
2400 {
2401 }
2402 static inline void rseq_execve(struct task_struct *t)
2403 {
2404 }
2405
2406 #endif
2407
2408 #ifdef CONFIG_DEBUG_RSEQ
2409
2410 void rseq_syscall(struct pt_regs *regs);
2411
2412 #else
2413
2414 static inline void rseq_syscall(struct pt_regs *regs)
2415 {
2416 }
2417
2418 #endif
2419
2420 #ifdef CONFIG_SCHED_CORE
2421 extern void sched_core_free(struct task_struct *tsk);
2422 extern void sched_core_fork(struct task_struct *p);
2423 extern int sched_core_share_pid(unsigned int cmd, pid_t pid, enum pid_type type,
2424                                 unsigned long uaddr);
2425 #else
2426 static inline void sched_core_free(struct task_struct *tsk) { }
2427 static inline void sched_core_fork(struct task_struct *p) { }
2428 #endif
2429
2430 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
2431
2432 #endif