osfmk/kern/processor.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 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 /*
  60  *      processor.h:    Processor and processor-related definitions.
  61  */
  62
  63 #ifndef _KERN_PROCESSOR_H_
  64 #define _KERN_PROCESSOR_H_
  65
  66 #include <mach/boolean.h>
  67 #include <mach/kern_return.h>
  68 #include <kern/kern_types.h>
  69
  70 #include <sys/cdefs.h>
  71
  72 #ifdef  MACH_KERNEL_PRIVATE
  73
  74 #include <mach/mach_types.h>
  75 #include <kern/ast.h>
  76 #include <kern/cpu_number.h>
  77 #include <kern/smp.h>
  78 #include <kern/simple_lock.h>
  79 #include <kern/locks.h>
  80 #include <kern/percpu.h>
  81 #include <kern/queue.h>
  82 #include <kern/sched.h>
  83 #include <kern/sched_urgency.h>
  84 #include <kern/timer.h>
  85 #include <mach/sfi_class.h>
  86 #include <kern/sched_clutch.h>
  87 #include <kern/timer_call.h>
  88 #include <kern/assert.h>
  89 #include <machine/limits.h>
  90
  91 /*
  92  *      Processor state is accessed by locking the scheduling lock
  93  *      for the assigned processor set.
  94  *
  95  *           -------------------- SHUTDOWN
  96  *          /                     ^     ^
  97  *        _/                      |      \
  98  *  OFF_LINE ---> START ---> RUNNING ---> IDLE ---> DISPATCHING
  99  *         \_________________^   ^ ^______/           /
 100  *                                \__________________/
 101  *
 102  *  Most of these state transitions are externally driven as a
 103  *  a directive (for instance telling an IDLE processor to start
 104  *  coming out of the idle state to run a thread). However these
 105  *  are typically paired with a handshake by the processor itself
 106  *  to indicate that it has completed a transition of indeterminate
 107  *  length (for example, the DISPATCHING->RUNNING or START->RUNNING
 108  *  transitions must occur on the processor itself).
 109  *
 110  *  The boot processor has some special cases, and skips the START state,
 111  *  since it has already bootstrapped and is ready to context switch threads.
 112  *
 113  *  When a processor is in DISPATCHING or RUNNING state, the current_pri,
 114  *  current_thmode, and deadline fields should be set, so that other
 115  *  processors can evaluate if it is an appropriate candidate for preemption.
 116  */
 117 #if defined(CONFIG_SCHED_DEFERRED_AST)
 118 /*
 119  *           -------------------- SHUTDOWN
 120  *          /                     ^     ^
 121  *        _/                      |      \
 122  *  OFF_LINE ---> START ---> RUNNING ---> IDLE ---> DISPATCHING
 123  *         \_________________^   ^ ^______/ ^_____ /  /
 124  *                                \__________________/
 125  *
 126  *  A DISPATCHING processor may be put back into IDLE, if another
 127  *  processor determines that the target processor will have nothing to do
 128  *  upon reaching the RUNNING state.  This is racy, but if the target
 129  *  responds and becomes RUNNING, it will not break the processor state
 130  *  machine.
 131  *
 132  *  This change allows us to cancel an outstanding signal/AST on a processor
 133  *  (if such an operation is supported through hardware or software), and
 134  *  push the processor back into the IDLE state as a power optimization.
 135  */
 136 #endif
 137
 138 typedef enum {
 139         PROCESSOR_OFF_LINE      = 0,    /* Not available */
 140         PROCESSOR_SHUTDOWN      = 1,    /* Going off-line */
 141         PROCESSOR_START         = 2,    /* Being started */
 142         PROCESSOR_UNUSED        = 3,    /* Formerly Inactive (unavailable) */
 143         PROCESSOR_IDLE          = 4,    /* Idle (available) */
 144         PROCESSOR_DISPATCHING   = 5,    /* Dispatching (idle -> active) */
 145         PROCESSOR_RUNNING       = 6,    /* Normal execution */
 146         PROCESSOR_STATE_LEN     = (PROCESSOR_RUNNING + 1)
 147 } processor_state_t;
 148
 149 typedef enum {
 150         PSET_SMP,
 151 #if __AMP__
 152         PSET_AMP_E,
 153         PSET_AMP_P,
 154 #endif
 155 } pset_cluster_type_t;
 156
 157 #if __AMP__
 158
 159 typedef enum {
 160         SCHED_PERFCTL_POLICY_DEFAULT,           /*  static policy: set at boot */
 161         SCHED_PERFCTL_POLICY_FOLLOW_GROUP,      /* dynamic policy: perfctl_class follows thread group across amp clusters */
 162         SCHED_PERFCTL_POLICY_RESTRICT_E,        /* dynamic policy: limits perfctl_class to amp e cluster */
 163 } sched_perfctl_class_policy_t;
 164
 165 extern _Atomic sched_perfctl_class_policy_t sched_perfctl_policy_util;
 166 extern _Atomic sched_perfctl_class_policy_t sched_perfctl_policy_bg;
 167
 168 #endif /* __AMP__ */
 169
 170 typedef bitmap_t cpumap_t;
 171
 172 #if __arm64__
 173
 174 /*
 175  * pset_execution_time_t
 176  *
 177  * The pset_execution_time_t type is used to maintain the average
 178  * execution time of threads on a pset. Since the avg. execution time is
 179  * updated from contexts where the pset lock is not held, it uses a
 180  * double-wide RMW loop to update these values atomically.
 181  */
 182 typedef union {
 183         struct {
 184                 uint64_t        pset_avg_thread_execution_time;
 185                 uint64_t        pset_execution_time_last_update;
 186         };
 187         unsigned __int128       pset_execution_time_packed;
 188 } pset_execution_time_t;
 189
 190 #endif /* __arm64__ */
 191
 192 struct processor_set {
 193         int                     pset_id;
 194         int                     online_processor_count;
 195         int                     cpu_set_low, cpu_set_hi;
 196         int                     cpu_set_count;
 197         int                     last_chosen;
 198
 199         uint64_t                load_average;
 200         uint64_t                pset_load_average[TH_BUCKET_SCHED_MAX];
 201         uint64_t                pset_load_last_update;
 202         cpumap_t                cpu_bitmask;
 203         cpumap_t                recommended_bitmask;
 204         cpumap_t                cpu_state_map[PROCESSOR_STATE_LEN];
 205         cpumap_t                primary_map;
 206         cpumap_t                realtime_map;
 207         cpumap_t                cpu_running_foreign;
 208         sched_bucket_t          cpu_running_buckets[MAX_CPUS];
 209
 210 #define SCHED_PSET_TLOCK (1)
 211 #if     defined(SCHED_PSET_TLOCK)
 212 /* TODO: reorder struct for temporal cache locality */
 213         __attribute__((aligned(128))) lck_ticket_t      sched_lock;
 214 #else /* SCHED_PSET_TLOCK*/
 215         __attribute__((aligned(128))) lck_spin_t        sched_lock;     /* lock for above */
 216 #endif /* SCHED_PSET_TLOCK*/
 217
 218 #if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_MULTIQ)
 219         struct run_queue        pset_runq;      /* runq for this processor set */
 220 #endif
 221         struct rt_queue         rt_runq;        /* realtime runq for this processor set */
 222 #if CONFIG_SCHED_CLUTCH
 223         struct sched_clutch_root pset_clutch_root; /* clutch hierarchy root */
 224 #endif /* CONFIG_SCHED_CLUTCH */
 225
 226 #if defined(CONFIG_SCHED_TRADITIONAL)
 227         int                     pset_runq_bound_count;
 228         /* # of threads in runq bound to any processor in pset */
 229 #endif
 230
 231         /* CPUs that have been sent an unacknowledged remote AST for scheduling purposes */
 232         cpumap_t                pending_AST_URGENT_cpu_mask;
 233         cpumap_t                pending_AST_PREEMPT_cpu_mask;
 234 #if defined(CONFIG_SCHED_DEFERRED_AST)
 235         /*
 236          * A separate mask, for ASTs that we may be able to cancel.  This is dependent on
 237          * some level of support for requesting an AST on a processor, and then quashing
 238          * that request later.
 239          *
 240          * The purpose of this field (and the associated codepaths) is to infer when we
 241          * no longer need a processor that is DISPATCHING to come up, and to prevent it
 242          * from coming out of IDLE if possible.  This should serve to decrease the number
 243          * of spurious ASTs in the system, and let processors spend longer periods in
 244          * IDLE.
 245          */
 246         cpumap_t                pending_deferred_AST_cpu_mask;
 247 #endif
 248         cpumap_t                pending_spill_cpu_mask;
 249
 250         struct ipc_port *       pset_self;              /* port for operations */
 251         struct ipc_port *       pset_name_self; /* port for information */
 252
 253         processor_set_t         pset_list;              /* chain of associated psets */
 254         pset_node_t             node;
 255         uint32_t                pset_cluster_id;
 256
 257         /*
 258          * Currently the scheduler uses a mix of pset_cluster_type_t & cluster_type_t
 259          * for recommendations etc. It might be useful to unify these as a single type.
 260          */
 261         pset_cluster_type_t     pset_cluster_type;
 262         cluster_type_t          pset_type;
 263
 264 #if CONFIG_SCHED_EDGE
 265         bitmap_t                foreign_psets[BITMAP_LEN(MAX_PSETS)];
 266         sched_clutch_edge       sched_edges[MAX_PSETS];
 267         pset_execution_time_t   pset_execution_time[TH_BUCKET_SCHED_MAX];
 268 #endif /* CONFIG_SCHED_EDGE */
 269         bool                    is_SMT;                 /* pset contains SMT processors */
 270 };
 271
 272 extern struct processor_set     pset0;
 273
 274 typedef bitmap_t pset_map_t;
 275
 276 struct pset_node {
 277         processor_set_t         psets;                  /* list of associated psets */
 278
 279         pset_node_t             nodes;                  /* list of associated subnodes */
 280         pset_node_t             node_list;              /* chain of associated nodes */
 281
 282         pset_node_t             parent;
 283
 284         pset_map_t              pset_map;               /* map of associated psets */
 285         _Atomic pset_map_t      pset_idle_map;          /* psets with at least one IDLE CPU */
 286         _Atomic pset_map_t      pset_idle_primary_map;  /* psets with at least one IDLE primary CPU */
 287         _Atomic pset_map_t      pset_non_rt_map;        /* psets with at least one available CPU not running a realtime thread */
 288         _Atomic pset_map_t      pset_non_rt_primary_map;/* psets with at least one available primary CPU not running a realtime thread */
 289 };
 290
 291 extern struct pset_node pset_node0;
 292
 293 extern queue_head_t tasks, threads, corpse_tasks;
 294 extern int tasks_count, terminated_tasks_count, threads_count;
 295 decl_lck_mtx_data(extern, tasks_threads_lock);
 296 decl_lck_mtx_data(extern, tasks_corpse_lock);
 297
 298 /*
 299  * The terminated tasks queue should only be inspected elsewhere by stackshot.
 300  */
 301 extern queue_head_t terminated_tasks;
 302
 303 struct processor {
 304         processor_state_t       state;                  /* See above */
 305         bool                    is_SMT;
 306         bool                    is_recommended;
 307         bool                    current_is_NO_SMT;      /* cached TH_SFLAG_NO_SMT of current thread */
 308         bool                    current_is_bound;       /* current thread is bound to this processor */
 309         bool                    current_is_eagerpreempt;/* current thread is TH_SFLAG_EAGERPREEMPT */
 310         struct thread          *active_thread;          /* thread running on processor */
 311         struct thread          *idle_thread;            /* this processor's idle thread. */
 312         struct thread          *startup_thread;
 313
 314         processor_set_t         processor_set;  /* assigned set */
 315
 316         /*
 317          * XXX All current_* fields should be grouped together, as they're
 318          * updated at the same time.
 319          */
 320         int                     current_pri;            /* priority of current thread */
 321         sfi_class_id_t          current_sfi_class;      /* SFI class of current thread */
 322         perfcontrol_class_t     current_perfctl_class;  /* Perfcontrol class for current thread */
 323         /*
 324          * The cluster type recommended for the current thread.
 325          */
 326         pset_cluster_type_t     current_recommended_pset_type;
 327         thread_urgency_t        current_urgency;        /* cached urgency of current thread */
 328
 329 #if CONFIG_SCHED_TRADITIONAL
 330         int                     runq_bound_count;       /* # of threads bound to this processor */
 331 #endif /* CONFIG_SCHED_TRADITIONAL */
 332
 333 #if CONFIG_THREAD_GROUPS
 334         struct thread_group    *current_thread_group;   /* thread_group of current thread */
 335 #endif
 336         int                     starting_pri;           /* priority of current thread as it was when scheduled */
 337         int                     cpu_id;                 /* platform numeric id */
 338
 339         uint64_t                quantum_end;            /* time when current quantum ends */
 340         uint64_t                last_dispatch;          /* time of last dispatch */
 341
 342 #if KPERF
 343         uint64_t                kperf_last_sample_time; /* time of last kperf sample */
 344 #endif /* KPERF */
 345
 346         uint64_t                deadline;               /* for next realtime thread */
 347         bool                    first_timeslice;        /* has the quantum expired since context switch */
 348
 349         bool                    processor_offlined;     /* has the processor been explicitly processor_offline'ed */
 350         bool                    must_idle;              /* Needs to be forced idle as next selected thread is allowed on this processor */
 351
 352         bool                    running_timers_active;  /* whether the running timers should fire */
 353         struct timer_call       running_timers[RUNNING_TIMER_MAX];
 354
 355 #if CONFIG_SCHED_TRADITIONAL || CONFIG_SCHED_MULTIQ
 356         struct run_queue        runq;                   /* runq for this processor */
 357 #endif /* CONFIG_SCHED_TRADITIONAL || CONFIG_SCHED_MULTIQ */
 358
 359 #if CONFIG_SCHED_GRRR
 360         struct grrr_run_queue   grrr_runq;              /* Group Ratio Round-Robin runq */
 361 #endif /* CONFIG_SCHED_GRRR */
 362
 363         /*
 364          * Pointer to primary processor for secondary SMT processors, or a
 365          * pointer to ourselves for primaries or non-SMT.
 366          */
 367         processor_t             processor_primary;
 368         processor_t             processor_secondary;
 369         struct ipc_port        *processor_self;         /* port for operations */
 370
 371         processor_t             processor_list;         /* all existing processors */
 372
 373         /* Processor state statistics */
 374         timer_data_t            idle_state;
 375         timer_data_t            system_state;
 376         timer_data_t            user_state;
 377
 378         timer_t                 current_state;          /* points to processor's idle, system, or user state timer */
 379
 380         /* Thread execution timers */
 381         timer_t                 thread_timer;           /* points to current thread's user or system timer */
 382         timer_t                 kernel_timer;           /* points to current thread's system_timer */
 383
 384         uint64_t                timer_call_ttd;         /* current timer call time-to-deadline */
 385 };
 386
 387 extern processor_t processor_list;
 388 decl_simple_lock_data(extern, processor_list_lock);
 389
 390 /*
 391  * Maximum number of CPUs supported by the scheduler.  bits.h bitmap macros
 392  * need to be used to support greater than 64.
 393  */
 394 #define MAX_SCHED_CPUS          64
 395 extern processor_t              processor_array[MAX_SCHED_CPUS]; /* array indexed by cpuid */
 396 extern processor_set_t          pset_array[MAX_PSETS];           /* array indexed by pset_id */
 397
 398 extern uint32_t                 processor_avail_count;
 399 extern uint32_t                 processor_avail_count_user;
 400 extern uint32_t                 primary_processor_avail_count;
 401 extern uint32_t                 primary_processor_avail_count_user;
 402
 403 #define master_processor PERCPU_GET_MASTER(processor)
 404 PERCPU_DECL(struct processor, processor);
 405
 406 extern processor_t      current_processor(void);
 407
 408 /* Lock macros, always acquired and released with interrupts disabled (splsched()) */
 409
 410 extern lck_grp_t pset_lck_grp;
 411
 412 #if defined(SCHED_PSET_TLOCK)
 413 #define pset_lock_init(p)               lck_ticket_init(&(p)->sched_lock, &pset_lck_grp)
 414 #define pset_lock(p)                    lck_ticket_lock(&(p)->sched_lock, &pset_lck_grp)
 415 #define pset_unlock(p)                  lck_ticket_unlock(&(p)->sched_lock)
 416 #define pset_assert_locked(p)           lck_ticket_assert_owned(&(p)->sched_lock)
 417 #else /* SCHED_PSET_TLOCK*/
 418 #define pset_lock_init(p)               lck_spin_init(&(p)->sched_lock, &pset_lck_grp, NULL)
 419 #define pset_lock(p)                    lck_spin_lock_grp(&(p)->sched_lock, &pset_lck_grp)
 420 #define pset_unlock(p)                  lck_spin_unlock(&(p)->sched_lock)
 421 #define pset_assert_locked(p)           LCK_SPIN_ASSERT(&(p)->sched_lock, LCK_ASSERT_OWNED)
 422 #endif /*!SCHED_PSET_TLOCK*/
 423
 424 extern void             processor_bootstrap(void);
 425
 426 extern void             processor_init(
 427         processor_t             processor,
 428         int                     cpu_id,
 429         processor_set_t         processor_set);
 430
 431 extern void             processor_set_primary(
 432         processor_t             processor,
 433         processor_t             primary);
 434
 435 extern kern_return_t    processor_shutdown(
 436         processor_t             processor);
 437
 438 extern kern_return_t    processor_start_from_user(
 439         processor_t             processor);
 440 extern kern_return_t    processor_exit_from_user(
 441         processor_t             processor);
 442
 443 extern kern_return_t    sched_processor_enable(
 444         processor_t             processor,
 445         boolean_t               enable);
 446
 447 extern void             processor_queue_shutdown(
 448         processor_t             processor);
 449
 450 extern void             processor_queue_shutdown(
 451         processor_t             processor);
 452
 453 extern processor_set_t  processor_pset(
 454         processor_t             processor);
 455
 456 extern pset_node_t      pset_node_root(void);
 457
 458 extern processor_set_t  pset_create(
 459         pset_node_t             node);
 460
 461 extern void             pset_init(
 462         processor_set_t         pset,
 463         pset_node_t             node);
 464
 465 extern processor_set_t  pset_find(
 466         uint32_t                cluster_id,
 467         processor_set_t         default_pset);
 468
 469 #if !defined(RC_HIDE_XNU_FIRESTORM) && (MAX_CPU_CLUSTERS > 2)
 470
 471 /*
 472  * Find the first processor_set for the given pset_cluster_type.
 473  * Should be removed with rdar://57340304, as it's only
 474  * useful for the workaround described in rdar://57306691.
 475  */
 476
 477 extern processor_set_t  pset_find_first_by_cluster_type(
 478         pset_cluster_type_t     pset_cluster_type);
 479
 480 #endif /* !defined(RC_HIDE_XNU_FIRESTORM) && (MAX_CPU_CLUSTERS > 2) */
 481
 482 extern kern_return_t    processor_info_count(
 483         processor_flavor_t      flavor,
 484         mach_msg_type_number_t  *count);
 485
 486 #define pset_deallocate(x)
 487 #define pset_reference(x)
 488
 489 extern void             machine_run_count(
 490         uint32_t                count);
 491
 492 extern processor_t      machine_choose_processor(
 493         processor_set_t         pset,
 494         processor_t             processor);
 495
 496 #define next_pset(p)    (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
 497
 498 #define PSET_THING_TASK         0
 499 #define PSET_THING_THREAD       1
 500
 501 extern pset_cluster_type_t recommended_pset_type(
 502         thread_t                thread);
 503 #if CONFIG_THREAD_GROUPS
 504 extern pset_cluster_type_t thread_group_pset_recommendation(
 505         struct thread_group     *tg,
 506         cluster_type_t          recommendation);
 507 #endif /* CONFIG_THREAD_GROUPS */
 508
 509 inline static bool
 510 pset_is_recommended(processor_set_t pset)
 511 {
 512         return (pset->recommended_bitmask & pset->cpu_bitmask) != 0;
 513 }
 514
 515 extern void             processor_state_update_idle(
 516         processor_t             processor);
 517
 518 extern void             processor_state_update_from_thread(
 519         processor_t             processor,
 520         thread_t                thread);
 521
 522 extern void             processor_state_update_explicit(
 523         processor_t             processor,
 524         int                     pri,
 525         sfi_class_id_t          sfi_class,
 526         pset_cluster_type_t     pset_type,
 527         perfcontrol_class_t     perfctl_class,
 528         thread_urgency_t        urgency,
 529         sched_bucket_t          bucket);
 530
 531 #define PSET_LOAD_NUMERATOR_SHIFT   16
 532 #define PSET_LOAD_FRACTIONAL_SHIFT   4
 533
 534 #if CONFIG_SCHED_EDGE
 535
 536 extern cluster_type_t pset_type_for_id(uint32_t cluster_id);
 537
 538 /*
 539  * The Edge scheduler uses average scheduling latency as the metric for making
 540  * thread migration decisions. One component of avg scheduling latency is the load
 541  * average on the cluster.
 542  *
 543  * Load Average Fixed Point Arithmetic
 544  *
 545  * The load average is maintained as a 24.8 fixed point arithmetic value for precision.
 546  * When multiplied by the average execution time, it needs to be rounded up (based on
 547  * the most significant bit of the fractional part) for better accuracy. After rounding
 548  * up, the whole number part of the value is used as the actual load value for
 549  * migrate/steal decisions.
 550  */
 551 #define SCHED_PSET_LOAD_EWMA_FRACTION_BITS 8
 552 #define SCHED_PSET_LOAD_EWMA_ROUND_BIT     (1 << (SCHED_PSET_LOAD_EWMA_FRACTION_BITS - 1))
 553 #define SCHED_PSET_LOAD_EWMA_FRACTION_MASK ((1 << SCHED_PSET_LOAD_EWMA_FRACTION_BITS) - 1)
 554
 555 inline static int
 556 sched_get_pset_load_average(processor_set_t pset, sched_bucket_t sched_bucket)
 557 {
 558         return (int)(((pset->pset_load_average[sched_bucket] + SCHED_PSET_LOAD_EWMA_ROUND_BIT) >> SCHED_PSET_LOAD_EWMA_FRACTION_BITS) *
 559                pset->pset_execution_time[sched_bucket].pset_avg_thread_execution_time);
 560 }
 561
 562 #else /* CONFIG_SCHED_EDGE */
 563 inline static int
 564 sched_get_pset_load_average(processor_set_t pset, __unused sched_bucket_t sched_bucket)
 565 {
 566         return (int)pset->load_average >> (PSET_LOAD_NUMERATOR_SHIFT - PSET_LOAD_FRACTIONAL_SHIFT);
 567 }
 568 #endif /* CONFIG_SCHED_EDGE */
 569
 570 extern void sched_update_pset_load_average(processor_set_t pset, uint64_t curtime);
 571 extern void sched_update_pset_avg_execution_time(processor_set_t pset, uint64_t delta, uint64_t curtime, sched_bucket_t sched_bucket);
 572
 573 inline static void
 574 pset_update_processor_state(processor_set_t pset, processor_t processor, uint new_state)
 575 {
 576         pset_assert_locked(pset);
 577
 578         uint old_state = processor->state;
 579         uint cpuid = (uint)processor->cpu_id;
 580
 581         assert(processor->processor_set == pset);
 582         assert(bit_test(pset->cpu_bitmask, cpuid));
 583
 584         assert(old_state < PROCESSOR_STATE_LEN);
 585         assert(new_state < PROCESSOR_STATE_LEN);
 586
 587         processor->state = new_state;
 588
 589         bit_clear(pset->cpu_state_map[old_state], cpuid);
 590         bit_set(pset->cpu_state_map[new_state], cpuid);
 591
 592         if ((old_state == PROCESSOR_RUNNING) || (new_state == PROCESSOR_RUNNING)) {
 593                 sched_update_pset_load_average(pset, 0);
 594                 if (new_state == PROCESSOR_RUNNING) {
 595                         assert(processor == current_processor());
 596                 }
 597         }
 598         if ((old_state == PROCESSOR_IDLE) || (new_state == PROCESSOR_IDLE)) {
 599                 if (new_state == PROCESSOR_IDLE) {
 600                         bit_clear(pset->realtime_map, cpuid);
 601                 }
 602
 603                 pset_node_t node = pset->node;
 604
 605                 if (bit_count(node->pset_map) == 1) {
 606                         /* Node has only a single pset, so skip node pset map updates */
 607                         return;
 608                 }
 609
 610                 if (new_state == PROCESSOR_IDLE) {
 611                         if (processor->processor_primary == processor) {
 612                                 if (!bit_test(atomic_load(&node->pset_non_rt_primary_map), pset->pset_id)) {
 613                                         atomic_bit_set(&node->pset_non_rt_primary_map, pset->pset_id, memory_order_relaxed);
 614                                 }
 615                                 if (!bit_test(atomic_load(&node->pset_idle_primary_map), pset->pset_id)) {
 616                                         atomic_bit_set(&node->pset_idle_primary_map, pset->pset_id, memory_order_relaxed);
 617                                 }
 618                         }
 619                         if (!bit_test(atomic_load(&node->pset_non_rt_map), pset->pset_id)) {
 620                                 atomic_bit_set(&node->pset_non_rt_map, pset->pset_id, memory_order_relaxed);
 621                         }
 622                         if (!bit_test(atomic_load(&node->pset_idle_map), pset->pset_id)) {
 623                                 atomic_bit_set(&node->pset_idle_map, pset->pset_id, memory_order_relaxed);
 624                         }
 625                 } else {
 626                         cpumap_t idle_map = pset->cpu_state_map[PROCESSOR_IDLE];
 627                         if (idle_map == 0) {
 628                                 /* No more IDLE CPUs */
 629                                 if (bit_test(atomic_load(&node->pset_idle_map), pset->pset_id)) {
 630                                         atomic_bit_clear(&node->pset_idle_map, pset->pset_id, memory_order_relaxed);
 631                                 }
 632                         }
 633                         if (processor->processor_primary == processor) {
 634                                 idle_map &= pset->primary_map;
 635                                 if (idle_map == 0) {
 636                                         /* No more IDLE primary CPUs */
 637                                         if (bit_test(atomic_load(&node->pset_idle_primary_map), pset->pset_id)) {
 638                                                 atomic_bit_clear(&node->pset_idle_primary_map, pset->pset_id, memory_order_relaxed);
 639                                         }
 640                                 }
 641                         }
 642                 }
 643         }
 644 }
 645
 646 #else   /* MACH_KERNEL_PRIVATE */
 647
 648 __BEGIN_DECLS
 649
 650 extern void             pset_deallocate(
 651         processor_set_t         pset);
 652
 653 extern void             pset_reference(
 654         processor_set_t         pset);
 655
 656 __END_DECLS
 657
 658 #endif  /* MACH_KERNEL_PRIVATE */
 659
 660 #ifdef KERNEL_PRIVATE
 661 __BEGIN_DECLS
 662 extern unsigned int             processor_count;
 663 extern processor_t      cpu_to_processor(int cpu);
 664
 665 extern kern_return_t    enable_smt_processors(bool enable);
 666
 667 __END_DECLS
 668
 669 #endif /* KERNEL_PRIVATE */
 670
 671 #endif  /* _KERN_PROCESSOR_H_ */