osfmk/kern/sched.h

   1 /*
   2  * Copyright (c) 2000-2019 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * @OSF_COPYRIGHT@
  30  */
  31 /*
  32  * Mach Operating System
  33  * Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
  34  * All Rights Reserved.
  35  *
  36  * Permission to use, copy, modify and distribute this software and its
  37  * documentation is hereby granted, provided that both the copyright
  38  * notice and this permission notice appear in all copies of the
  39  * software, derivative works or modified versions, and any portions
  40  * thereof, and that both notices appear in supporting documentation.
  41  *
  42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  45  *
  46  * Carnegie Mellon requests users of this software to return to
  47  *
  48  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  49  *  School of Computer Science
  50  *  Carnegie Mellon University
  51  *  Pittsburgh PA 15213-3890
  52  *
  53  * any improvements or extensions that they make and grant Carnegie Mellon
  54  * the rights to redistribute these changes.
  55  */
  56 /*
  57  */
  58 /*
  59  *      File:   sched.h
  60  *      Author: Avadis Tevanian, Jr.
  61  *      Date:   1985
  62  *
  63  *      Header file for scheduler.
  64  *
  65  */
  66
  67 #ifndef _KERN_SCHED_H_
  68 #define _KERN_SCHED_H_
  69
  70 #include <mach/policy.h>
  71 #include <kern/kern_types.h>
  72 #include <kern/smp.h>
  73 #include <kern/circle_queue.h>
  74 #include <kern/macro_help.h>
  75 #include <kern/timer_call.h>
  76 #include <kern/ast.h>
  77 #include <kern/kalloc.h>
  78 #include <kern/bits.h>
  79
  80 #define NRQS_MAX        (128)                           /* maximum number of priority levels */
  81
  82 #define MAXPRI          (NRQS_MAX-1)
  83 #define MINPRI          0                               /* lowest legal priority schedulable */
  84 #define IDLEPRI         MINPRI                          /* idle thread priority */
  85 #define NOPRI           -1
  86
  87 /*
  88  *      High-level priority assignments
  89  *
  90  *************************************************************************
  91  * 127          Reserved (real-time)
  92  *                              A
  93  *                              +
  94  *                      (32 levels)
  95  *                              +
  96  *                              V
  97  * 96           Reserved (real-time)
  98  * 95           Kernel mode only
  99  *                              A
 100  *                              +
 101  *                      (16 levels)
 102  *                              +
 103  *                              V
 104  * 80           Kernel mode only
 105  * 79           System high priority
 106  *                              A
 107  *                              +
 108  *                      (16 levels)
 109  *                              +
 110  *                              V
 111  * 64           System high priority
 112  * 63           Elevated priorities
 113  *                              A
 114  *                              +
 115  *                      (12 levels)
 116  *                              +
 117  *                              V
 118  * 52           Elevated priorities
 119  * 51           Elevated priorities (incl. BSD +nice)
 120  *                              A
 121  *                              +
 122  *                      (20 levels)
 123  *                              +
 124  *                              V
 125  * 32           Elevated priorities (incl. BSD +nice)
 126  * 31           Default (default base for threads)
 127  * 30           Lowered priorities (incl. BSD -nice)
 128  *                              A
 129  *                              +
 130  *                      (20 levels)
 131  *                              +
 132  *                              V
 133  * 11           Lowered priorities (incl. BSD -nice)
 134  * 10           Lowered priorities (aged pri's)
 135  *                              A
 136  *                              +
 137  *                      (11 levels)
 138  *                              +
 139  *                              V
 140  * 0            Lowered priorities (aged pri's / idle)
 141  *************************************************************************
 142  */
 143
 144 #define BASEPRI_RTQUEUES        (BASEPRI_REALTIME + 1)                          /* 97 */
 145 #define BASEPRI_REALTIME        (MAXPRI - (NRQS_MAX / 4) + 1)                   /* 96 */
 146
 147 #define MAXPRI_KERNEL           (BASEPRI_REALTIME - 1)                          /* 95 */
 148 #define BASEPRI_PREEMPT_HIGH    (BASEPRI_PREEMPT + 1)                           /* 93 */
 149 #define BASEPRI_PREEMPT         (MAXPRI_KERNEL - 3)                             /* 92 */
 150 #define BASEPRI_VM              (BASEPRI_PREEMPT - 1)                           /* 91 */
 151
 152 #define BASEPRI_KERNEL          (MINPRI_KERNEL + 1)                             /* 81 */
 153 #define MINPRI_KERNEL           (MAXPRI_KERNEL - (NRQS_MAX / 8) + 1)            /* 80 */
 154
 155 #define MAXPRI_RESERVED         (MINPRI_KERNEL - 1)                             /* 79 */
 156 #define BASEPRI_GRAPHICS        (MAXPRI_RESERVED - 3)                           /* 76 */
 157 #define MINPRI_RESERVED         (MAXPRI_RESERVED - (NRQS_MAX / 8) + 1)          /* 64 */
 158
 159 #define MAXPRI_USER             (MINPRI_RESERVED - 1)                           /* 63 */
 160 #define BASEPRI_CONTROL         (BASEPRI_DEFAULT + 17)                          /* 48 */
 161 #define BASEPRI_FOREGROUND      (BASEPRI_DEFAULT + 16)                          /* 47 */
 162 #define BASEPRI_BACKGROUND      (BASEPRI_DEFAULT + 15)                          /* 46 */
 163 #define BASEPRI_USER_INITIATED  (BASEPRI_DEFAULT +  6)                          /* 37 */
 164 #define BASEPRI_DEFAULT         (MAXPRI_USER - (NRQS_MAX / 4))                  /* 31 */
 165 #define MAXPRI_SUPPRESSED       (BASEPRI_DEFAULT - 3)                           /* 28 */
 166 #define BASEPRI_UTILITY         (BASEPRI_DEFAULT - 11)                          /* 20 */
 167 #define MAXPRI_THROTTLE         (MINPRI + 4)                                    /*  4 */
 168 #define MINPRI_USER             MINPRI                                          /*  0 */
 169
 170 #define DEPRESSPRI              (MINPRI)                /* depress priority */
 171
 172 #define MAXPRI_PROMOTE          (MAXPRI_KERNEL)         /* ceiling for mutex promotion */
 173 #define MINPRI_RWLOCK           (BASEPRI_BACKGROUND)    /* floor when holding rwlock count */
 174 #define MINPRI_EXEC             (BASEPRI_DEFAULT)       /* floor when in exec state */
 175 #define MINPRI_WAITQ            (BASEPRI_DEFAULT)       /* floor when in waitq handover state */
 176
 177 #define NRQS                    (BASEPRI_REALTIME)      /* Non-realtime levels for runqs */
 178
 179 /* Ensure that NRQS is large enough to represent all non-realtime threads; even promoted ones */
 180 _Static_assert((NRQS == (MAXPRI_PROMOTE + 1)), "Runqueues are too small to hold all non-realtime threads");
 181
 182 /* Type used for thread->sched_mode and saved_mode */
 183 typedef enum {
 184         TH_MODE_NONE = 0,                                       /* unassigned, usually for saved_mode only */
 185         TH_MODE_REALTIME,                                       /* time constraints supplied */
 186         TH_MODE_FIXED,                                          /* use fixed priorities, no decay */
 187         TH_MODE_TIMESHARE,                                      /* use timesharing algorithm */
 188 } sched_mode_t;
 189
 190 /*
 191  * Since the clutch scheduler organizes threads based on the thread group
 192  * and the scheduling bucket, its important to not mix threads from multiple
 193  * priority bands into the same bucket. To achieve that, in the clutch bucket
 194  * world, there is a scheduling bucket per QoS effectively.
 195  */
 196
 197 /* Buckets used for load calculation */
 198 typedef enum {
 199         TH_BUCKET_FIXPRI = 0,                   /* Fixed-priority */
 200         TH_BUCKET_SHARE_FG,                     /* Timeshare thread above BASEPRI_DEFAULT */
 201 #if CONFIG_SCHED_CLUTCH
 202         TH_BUCKET_SHARE_IN,                     /* Timeshare thread between BASEPRI_USER_INITIATED and BASEPRI_DEFAULT */
 203 #endif /* CONFIG_SCHED_CLUTCH */
 204         TH_BUCKET_SHARE_DF,                     /* Timeshare thread between BASEPRI_DEFAULT and BASEPRI_UTILITY */
 205         TH_BUCKET_SHARE_UT,                     /* Timeshare thread between BASEPRI_UTILITY and MAXPRI_THROTTLE */
 206         TH_BUCKET_SHARE_BG,                     /* Timeshare thread between MAXPRI_THROTTLE and MINPRI */
 207         TH_BUCKET_RUN,                          /* All runnable threads */
 208         TH_BUCKET_SCHED_MAX = TH_BUCKET_RUN,    /* Maximum schedulable buckets */
 209         TH_BUCKET_MAX,
 210 } sched_bucket_t;
 211
 212 /*
 213  *      Macro to check for invalid priorities.
 214  */
 215 #define invalid_pri(pri) ((pri) < MINPRI || (pri) > MAXPRI)
 216
 217 struct runq_stats {
 218         uint64_t                count_sum;
 219         uint64_t                last_change_timestamp;
 220 };
 221
 222 #if defined(CONFIG_SCHED_TIMESHARE_CORE) || defined(CONFIG_SCHED_PROTO)
 223
 224 struct run_queue {
 225         int                     highq;                          /* highest runnable queue */
 226         bitmap_t                bitmap[BITMAP_LEN(NRQS)];       /* run queue bitmap array */
 227         int                     count;                          /* # of threads total */
 228         int                     urgency;                        /* level of preemption urgency */
 229         circle_queue_head_t     queues[NRQS];           /* one for each priority */
 230
 231         struct runq_stats       runq_stats;
 232 };
 233
 234 inline static void
 235 rq_bitmap_set(bitmap_t *map, u_int n)
 236 {
 237         assert(n < NRQS);
 238         bitmap_set(map, n);
 239 }
 240
 241 inline static void
 242 rq_bitmap_clear(bitmap_t *map, u_int n)
 243 {
 244         assert(n < NRQS);
 245         bitmap_clear(map, n);
 246 }
 247
 248 #endif /* defined(CONFIG_SCHED_TIMESHARE_CORE) || defined(CONFIG_SCHED_PROTO) */
 249
 250 struct rt_queue {
 251         _Atomic int             count;                          /* # of threads total */
 252         queue_head_t            queue;                          /* all runnable RT threads */
 253 #if __SMP__
 254         decl_simple_lock_data(, rt_lock);
 255 #endif
 256         struct runq_stats       runq_stats;
 257 };
 258 typedef struct rt_queue *rt_queue_t;
 259
 260 #if defined(CONFIG_SCHED_GRRR_CORE)
 261
 262 /*
 263  * We map standard Mach priorities to an abstract scale that more properly
 264  * indicates how we want processor time allocated under contention.
 265  */
 266 typedef uint8_t grrr_proportional_priority_t;
 267 typedef uint8_t grrr_group_index_t;
 268
 269 #define NUM_GRRR_PROPORTIONAL_PRIORITIES        256
 270 #define MAX_GRRR_PROPORTIONAL_PRIORITY ((grrr_proportional_priority_t)255)
 271
 272 #if 0
 273 #define NUM_GRRR_GROUPS 8                                       /* log(256) */
 274 #endif
 275
 276 #define NUM_GRRR_GROUPS 64                                      /* 256/4 */
 277
 278 struct grrr_group {
 279         queue_chain_t                   priority_order;                         /* next greatest weight group */
 280         grrr_proportional_priority_t            minpriority;
 281         grrr_group_index_t              index;
 282
 283         queue_head_t                    clients;
 284         int                                             count;
 285         uint32_t                                weight;
 286 #if 0
 287         uint32_t                                deferred_removal_weight;
 288 #endif
 289         uint32_t                                work;
 290         thread_t                                current_client;
 291 };
 292
 293 struct grrr_run_queue {
 294         int                                     count;
 295         uint32_t                        last_rescale_tick;
 296         struct grrr_group       groups[NUM_GRRR_GROUPS];
 297         queue_head_t            sorted_group_list;
 298         uint32_t                        weight;
 299         grrr_group_t            current_group;
 300
 301         struct runq_stats   runq_stats;
 302 };
 303
 304 #endif /* defined(CONFIG_SCHED_GRRR_CORE) */
 305
 306 extern int rt_runq_count(processor_set_t);
 307 extern void rt_runq_count_incr(processor_set_t);
 308 extern void rt_runq_count_decr(processor_set_t);
 309
 310 #if defined(CONFIG_SCHED_MULTIQ)
 311 sched_group_t   sched_group_create(void);
 312 void            sched_group_destroy(sched_group_t sched_group);
 313 #endif /* defined(CONFIG_SCHED_MULTIQ) */
 314
 315
 316
 317 /*
 318  *      Scheduler routines.
 319  */
 320
 321 /* Handle quantum expiration for an executing thread */
 322 extern void             thread_quantum_expire(
 323         timer_call_param_t      processor,
 324         timer_call_param_t      thread);
 325
 326 /* Context switch check for current processor */
 327 extern ast_t    csw_check(
 328         thread_t      thread,
 329         processor_t   processor,
 330         ast_t         check_reason);
 331
 332 /* Check for pending ASTs */
 333 extern void ast_check(processor_t processor);
 334
 335 extern void sched_update_generation_count(void);
 336
 337 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
 338 extern uint32_t std_quantum, min_std_quantum;
 339 extern uint32_t std_quantum_us;
 340 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 341
 342 extern uint32_t thread_depress_time;
 343 extern uint32_t default_timeshare_computation;
 344 extern uint32_t default_timeshare_constraint;
 345
 346 extern uint32_t max_rt_quantum, min_rt_quantum;
 347
 348 extern int default_preemption_rate;
 349 extern int default_bg_preemption_rate;
 350
 351 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
 352
 353 /*
 354  *      Age usage  at approximately (1 << SCHED_TICK_SHIFT) times per second
 355  *      Aging may be deferred during periods where all processors are idle
 356  *      and cumulatively applied during periods of activity.
 357  */
 358 #define SCHED_TICK_SHIFT        3
 359 #define SCHED_TICK_MAX_DELTA    (8)
 360
 361 extern unsigned         sched_tick;
 362 extern uint32_t         sched_tick_interval;
 363
 364 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 365
 366 extern uint64_t         sched_one_second_interval;
 367
 368 /* Periodic computation of various averages */
 369 extern void            compute_sched_load(void);
 370
 371 extern void             compute_averages(uint64_t);
 372
 373 extern void             compute_averunnable(
 374         void                    *nrun);
 375
 376 extern void             compute_stack_target(
 377         void                    *arg);
 378
 379 extern void             compute_pageout_gc_throttle(
 380         void                    *arg);
 381
 382 extern void             compute_pmap_gc_throttle(
 383         void                    *arg);
 384
 385 /*
 386  *      Conversion factor from usage
 387  *      to priority.
 388  */
 389 #if defined(CONFIG_SCHED_TIMESHARE_CORE)
 390
 391 #define MAX_LOAD (NRQS - 1)
 392 extern uint32_t         sched_pri_shifts[TH_BUCKET_MAX];
 393 extern uint32_t         sched_fixed_shift;
 394 extern int8_t           sched_load_shifts[NRQS];
 395 extern uint32_t         sched_decay_usage_age_factor;
 396 void sched_timeshare_consider_maintenance(uint64_t ctime);
 397 #endif /* CONFIG_SCHED_TIMESHARE_CORE */
 398
 399 void sched_consider_recommended_cores(uint64_t ctime, thread_t thread);
 400
 401 extern int32_t          sched_poll_yield_shift;
 402 extern uint64_t         sched_safe_duration;
 403
 404 extern uint32_t         sched_load_average, sched_mach_factor;
 405
 406 extern uint32_t         avenrun[3], mach_factor[3];
 407
 408 extern uint64_t         max_unsafe_computation;
 409 extern uint64_t         max_poll_computation;
 410
 411 extern uint32_t         sched_run_buckets[TH_BUCKET_MAX];
 412
 413 extern uint32_t sched_run_incr(thread_t thread);
 414 extern uint32_t sched_run_decr(thread_t thread);
 415 extern void sched_update_thread_bucket(thread_t thread);
 416
 417 #define SCHED_DECAY_TICKS       32
 418 struct shift_data {
 419         int     shift1;
 420         int     shift2;
 421 };
 422
 423 /*
 424  *      thread_timer_delta macro takes care of both thread timers.
 425  */
 426 #define thread_timer_delta(thread, delta)                                       \
 427 MACRO_BEGIN                                                                                                     \
 428         (delta) = (typeof(delta))timer_delta(&(thread)->system_timer,                   \
 429                                                         &(thread)->system_timer_save);  \
 430         (delta) += (typeof(delta))timer_delta(&(thread)->user_timer,                    \
 431                                                         &(thread)->user_timer_save);    \
 432 MACRO_END
 433
 434 #endif  /* _KERN_SCHED_H_ */