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/queue.h>
  81 #include <kern/sched.h>
  82 #include <kern/sched_urgency.h>
  83 #include <mach/sfi_class.h>
  84 #include <kern/processor_data.h>
  85 #include <kern/cpu_quiesce.h>
  86 #include <kern/sched_clutch.h>
  87 #include <kern/assert.h>
  88 #include <machine/limits.h>
  89
  90 /*
  91  *      Processor state is accessed by locking the scheduling lock
  92  *      for the assigned processor set.
  93  *
  94  *           -------------------- SHUTDOWN
  95  *          /                     ^     ^
  96  *        _/                      |      \
  97  *  OFF_LINE ---> START ---> RUNNING ---> IDLE ---> DISPATCHING
  98  *         \_________________^   ^ ^______/           /
  99  *                                \__________________/
 100  *
 101  *  Most of these state transitions are externally driven as a
 102  *  a directive (for instance telling an IDLE processor to start
 103  *  coming out of the idle state to run a thread). However these
 104  *  are typically paired with a handshake by the processor itself
 105  *  to indicate that it has completed a transition of indeterminate
 106  *  length (for example, the DISPATCHING->RUNNING or START->RUNNING
 107  *  transitions must occur on the processor itself).
 108  *
 109  *  The boot processor has some special cases, and skips the START state,
 110  *  since it has already bootstrapped and is ready to context switch threads.
 111  *
 112  *  When a processor is in DISPATCHING or RUNNING state, the current_pri,
 113  *  current_thmode, and deadline fields should be set, so that other
 114  *  processors can evaluate if it is an appropriate candidate for preemption.
 115  */
 116 #if defined(CONFIG_SCHED_DEFERRED_AST)
 117 /*
 118  *           -------------------- SHUTDOWN
 119  *          /                     ^     ^
 120  *        _/                      |      \
 121  *  OFF_LINE ---> START ---> RUNNING ---> IDLE ---> DISPATCHING
 122  *         \_________________^   ^ ^______/ ^_____ /  /
 123  *                                \__________________/
 124  *
 125  *  A DISPATCHING processor may be put back into IDLE, if another
 126  *  processor determines that the target processor will have nothing to do
 127  *  upon reaching the RUNNING state.  This is racy, but if the target
 128  *  responds and becomes RUNNING, it will not break the processor state
 129  *  machine.
 130  *
 131  *  This change allows us to cancel an outstanding signal/AST on a processor
 132  *  (if such an operation is supported through hardware or software), and
 133  *  push the processor back into the IDLE state as a power optimization.
 134  */
 135 #endif
 136
 137 typedef enum {
 138         PROCESSOR_OFF_LINE      = 0,    /* Not available */
 139         PROCESSOR_SHUTDOWN      = 1,    /* Going off-line */
 140         PROCESSOR_START         = 2,    /* Being started */
 141         PROCESSOR_UNUSED        = 3,    /* Formerly Inactive (unavailable) */
 142         PROCESSOR_IDLE          = 4,    /* Idle (available) */
 143         PROCESSOR_DISPATCHING   = 5,    /* Dispatching (idle -> active) */
 144         PROCESSOR_RUNNING       = 6,    /* Normal execution */
 145         PROCESSOR_STATE_LEN     = (PROCESSOR_RUNNING + 1)
 146 } processor_state_t;
 147
 148 typedef enum {
 149         PSET_SMP,
 150 #if __AMP__
 151         PSET_AMP_E,
 152         PSET_AMP_P,
 153 #endif
 154 } pset_cluster_type_t;
 155
 156 typedef bitmap_t cpumap_t;
 157
 158 struct processor_set {
 159         int                     online_processor_count;
 160         int                     load_average;
 161
 162         int                     cpu_set_low, cpu_set_hi;
 163         int                     cpu_set_count;
 164         int                     last_chosen;
 165         cpumap_t                cpu_bitmask;
 166         cpumap_t                recommended_bitmask;
 167         cpumap_t                cpu_state_map[PROCESSOR_STATE_LEN];
 168         cpumap_t                primary_map;
 169 #define SCHED_PSET_TLOCK (1)
 170 #if __SMP__
 171 #if     defined(SCHED_PSET_TLOCK)
 172 /* TODO: reorder struct for temporal cache locality */
 173         __attribute__((aligned(128))) lck_ticket_t      sched_lock;
 174 #else /* SCHED_PSET_TLOCK*/
 175         __attribute__((aligned(128))) lck_spin_t        sched_lock;     /* lock for above */
 176 #endif /* SCHED_PSET_TLOCK*/
 177 #endif
 178
 179 #if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_MULTIQ)
 180         struct run_queue        pset_runq;      /* runq for this processor set */
 181 #endif
 182         struct rt_queue         rt_runq;        /* realtime runq for this processor set */
 183 #if CONFIG_SCHED_CLUTCH
 184         struct sched_clutch_root                pset_clutch_root; /* clutch hierarchy root */
 185 #endif /* CONFIG_SCHED_CLUTCH */
 186
 187 #if defined(CONFIG_SCHED_TRADITIONAL)
 188         int                                     pset_runq_bound_count;
 189         /* # of threads in runq bound to any processor in pset */
 190 #endif
 191
 192         /* CPUs that have been sent an unacknowledged remote AST for scheduling purposes */
 193         cpumap_t                        pending_AST_URGENT_cpu_mask;
 194         cpumap_t                        pending_AST_PREEMPT_cpu_mask;
 195 #if defined(CONFIG_SCHED_DEFERRED_AST)
 196         /*
 197          * A separate mask, for ASTs that we may be able to cancel.  This is dependent on
 198          * some level of support for requesting an AST on a processor, and then quashing
 199          * that request later.
 200          *
 201          * The purpose of this field (and the associated codepaths) is to infer when we
 202          * no longer need a processor that is DISPATCHING to come up, and to prevent it
 203          * from coming out of IDLE if possible.  This should serve to decrease the number
 204          * of spurious ASTs in the system, and let processors spend longer periods in
 205          * IDLE.
 206          */
 207         cpumap_t                        pending_deferred_AST_cpu_mask;
 208 #endif
 209         cpumap_t                        pending_spill_cpu_mask;
 210
 211         struct ipc_port *       pset_self;              /* port for operations */
 212         struct ipc_port *       pset_name_self; /* port for information */
 213
 214         processor_set_t         pset_list;              /* chain of associated psets */
 215         pset_node_t             node;
 216         uint32_t                pset_cluster_id;
 217         pset_cluster_type_t     pset_cluster_type;
 218 };
 219
 220 extern struct processor_set     pset0;
 221
 222 struct pset_node {
 223         processor_set_t         psets;                  /* list of associated psets */
 224         uint32_t                pset_count;             /* count of associated psets */
 225
 226         pset_node_t                     nodes;                  /* list of associated subnodes */
 227         pset_node_t                     node_list;              /* chain of associated nodes */
 228
 229         pset_node_t                     parent;
 230 };
 231
 232 extern struct pset_node pset_node0;
 233
 234 extern queue_head_t             tasks, terminated_tasks, threads, corpse_tasks; /* Terminated tasks are ONLY for stackshot */
 235 extern int                              tasks_count, terminated_tasks_count, threads_count;
 236 decl_lck_mtx_data(extern, tasks_threads_lock);
 237 decl_lck_mtx_data(extern, tasks_corpse_lock);
 238
 239 struct processor {
 240         processor_state_t       state;                  /* See above */
 241         bool                    is_SMT;
 242         bool                    is_recommended;
 243         struct thread           *active_thread;         /* thread running on processor */
 244         struct thread           *idle_thread;           /* this processor's idle thread. */
 245         struct thread           *startup_thread;
 246
 247         processor_set_t         processor_set;  /* assigned set */
 248
 249         int                     current_pri;            /* priority of current thread */
 250         sfi_class_id_t          current_sfi_class;      /* SFI class of current thread */
 251         perfcontrol_class_t     current_perfctl_class;  /* Perfcontrol class for current thread */
 252         pset_cluster_type_t     current_recommended_pset_type;  /* Cluster type recommended for current thread */
 253         thread_urgency_t        current_urgency;        /* cached urgency of current thread */
 254         bool                    current_is_NO_SMT;         /* cached TH_SFLAG_NO_SMT of current thread */
 255         bool                    current_is_bound;       /* current thread is bound to this processor */
 256
 257         int                     starting_pri;           /* priority of current thread as it was when scheduled */
 258         int                     cpu_id;                 /* platform numeric id */
 259         cpu_quiescent_state_t   cpu_quiesce_state;
 260         uint64_t                cpu_quiesce_last_checkin;
 261
 262         timer_call_data_t       quantum_timer;  /* timer for quantum expiration */
 263         uint64_t                        quantum_end;    /* time when current quantum ends */
 264         uint64_t                        last_dispatch;  /* time of last dispatch */
 265
 266         uint64_t                        kperf_last_sample_time; /* time of last kperf sample */
 267
 268         uint64_t                        deadline;               /* current deadline */
 269         bool                    first_timeslice;        /* has the quantum expired since context switch */
 270         bool                    processor_offlined;        /* has the processor been explicitly processor_offline'ed */
 271         bool                    must_idle;              /* Needs to be forced idle as next selected thread is allowed on this processor */
 272
 273         processor_t             processor_primary;      /* pointer to primary processor for
 274                                                          * secondary SMT processors, or a pointer
 275                                                          * to ourselves for primaries or non-SMT */
 276         processor_t             processor_secondary;
 277
 278 #if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_MULTIQ)
 279         struct run_queue        runq;                   /* runq for this processor */
 280 #endif
 281
 282 #if defined(CONFIG_SCHED_TRADITIONAL)
 283         int                                     runq_bound_count; /* # of threads bound to this processor */
 284 #endif
 285 #if defined(CONFIG_SCHED_GRRR)
 286         struct grrr_run_queue   grrr_runq;      /* Group Ratio Round-Robin runq */
 287 #endif
 288         struct ipc_port *       processor_self; /* port for operations */
 289
 290         processor_t                     processor_list; /* all existing processors */
 291         processor_data_t        processor_data; /* per-processor data */
 292 };
 293
 294 extern processor_t              processor_list;
 295 decl_simple_lock_data(extern, processor_list_lock);
 296
 297 #define MAX_SCHED_CPUS          64 /* Maximum number of CPUs supported by the scheduler.  bits.h:bitmap_*() macros need to be used to support greater than 64 */
 298 extern processor_t              processor_array[MAX_SCHED_CPUS]; /* array indexed by cpuid */
 299
 300 extern uint32_t                 processor_avail_count;
 301 extern uint32_t                 processor_avail_count_user;
 302
 303 extern processor_t              master_processor;
 304
 305 extern boolean_t                sched_stats_active;
 306
 307 extern processor_t      current_processor(void);
 308
 309 /* Lock macros, always acquired and released with interrupts disabled (splsched()) */
 310
 311 extern lck_grp_t pset_lck_grp;
 312
 313 #if __SMP__
 314 #if defined(SCHED_PSET_TLOCK)
 315 #define pset_lock_init(p)               lck_ticket_init(&(p)->sched_lock)
 316 #define pset_lock(p)                    lck_ticket_lock(&(p)->sched_lock)
 317 #define pset_unlock(p)                  lck_ticket_unlock(&(p)->sched_lock)
 318 #define pset_assert_locked(p)           lck_ticket_assert_owned(&(p)->sched_lock)
 319 #else /* SCHED_PSET_TLOCK*/
 320 #define pset_lock_init(p)               lck_spin_init(&(p)->sched_lock, &pset_lck_grp, NULL)
 321 #define pset_lock(p)                    lck_spin_lock_grp(&(p)->sched_lock, &pset_lck_grp)
 322 #define pset_unlock(p)                  lck_spin_unlock(&(p)->sched_lock)
 323 #define pset_assert_locked(p)           LCK_SPIN_ASSERT(&(p)->sched_lock, LCK_ASSERT_OWNED)
 324 #endif /*!SCHED_PSET_TLOCK*/
 325
 326 #define rt_lock_lock(p)                 simple_lock(&SCHED(rt_runq)(p)->rt_lock, &pset_lck_grp)
 327 #define rt_lock_unlock(p)               simple_unlock(&SCHED(rt_runq)(p)->rt_lock)
 328 #define rt_lock_init(p)                 simple_lock_init(&SCHED(rt_runq)(p)->rt_lock, 0)
 329 #else
 330 #define pset_lock(p)                    do { (void)p; } while(0)
 331 #define pset_unlock(p)                  do { (void)p; } while(0)
 332 #define pset_lock_init(p)               do { (void)p; } while(0)
 333 #define pset_assert_locked(p)           do { (void)p; } while(0)
 334
 335 #define rt_lock_lock(p)                 do { (void)p; } while(0)
 336 #define rt_lock_unlock(p)               do { (void)p; } while(0)
 337 #define rt_lock_init(p)                 do { (void)p; } while(0)
 338 #endif /* SMP */
 339
 340 extern void             processor_bootstrap(void);
 341
 342 extern void             processor_init(
 343         processor_t             processor,
 344         int                             cpu_id,
 345         processor_set_t processor_set);
 346
 347 extern void             processor_set_primary(
 348         processor_t             processor,
 349         processor_t             primary);
 350
 351 extern kern_return_t    processor_shutdown(
 352         processor_t             processor);
 353
 354 extern kern_return_t    processor_start_from_user(
 355         processor_t     processor);
 356 extern kern_return_t    processor_exit_from_user(
 357         processor_t     processor);
 358
 359 kern_return_t
 360 sched_processor_enable(processor_t processor, boolean_t enable);
 361
 362 extern void             processor_queue_shutdown(
 363         processor_t             processor);
 364
 365 extern void             processor_queue_shutdown(
 366         processor_t             processor);
 367
 368 extern processor_set_t  processor_pset(
 369         processor_t             processor);
 370
 371 extern pset_node_t              pset_node_root(void);
 372
 373 extern processor_set_t  pset_create(
 374         pset_node_t             node);
 375
 376 extern void             pset_init(
 377         processor_set_t         pset,
 378         pset_node_t                     node);
 379
 380 extern processor_set_t pset_find(
 381         uint32_t cluster_id,
 382         processor_set_t default_pset);
 383
 384 extern kern_return_t    processor_info_count(
 385         processor_flavor_t              flavor,
 386         mach_msg_type_number_t  *count);
 387
 388 #define pset_deallocate(x)
 389 #define pset_reference(x)
 390
 391 extern void                             machine_run_count(
 392         uint32_t        count);
 393
 394 extern processor_t              machine_choose_processor(
 395         processor_set_t         pset,
 396         processor_t                     processor);
 397
 398 #define next_pset(p)    (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
 399
 400 #define PSET_THING_TASK         0
 401 #define PSET_THING_THREAD       1
 402
 403 extern kern_return_t    processor_set_things(
 404         processor_set_t pset,
 405         void **thing_list,
 406         mach_msg_type_number_t *count,
 407         int type);
 408
 409 extern pset_cluster_type_t recommended_pset_type(thread_t thread);
 410
 411 inline static bool
 412 pset_is_recommended(processor_set_t pset)
 413 {
 414         return (pset->recommended_bitmask & pset->cpu_bitmask) != 0;
 415 }
 416
 417 extern void processor_state_update_idle(processor_t processor);
 418 extern void processor_state_update_from_thread(processor_t processor, thread_t thread);
 419 extern void processor_state_update_explicit(processor_t processor, int pri,
 420     sfi_class_id_t sfi_class, pset_cluster_type_t pset_type,
 421     perfcontrol_class_t perfctl_class, thread_urgency_t urgency);
 422
 423 #define PSET_LOAD_NUMERATOR_SHIFT   16
 424 #define PSET_LOAD_FRACTIONAL_SHIFT   4
 425
 426 inline static int
 427 sched_get_pset_load_average(processor_set_t pset)
 428 {
 429         return pset->load_average >> (PSET_LOAD_NUMERATOR_SHIFT - PSET_LOAD_FRACTIONAL_SHIFT);
 430 }
 431 extern void sched_update_pset_load_average(processor_set_t pset);
 432
 433 inline static void
 434 pset_update_processor_state(processor_set_t pset, processor_t processor, uint new_state)
 435 {
 436         pset_assert_locked(pset);
 437
 438         uint old_state = processor->state;
 439         uint cpuid = processor->cpu_id;
 440
 441         assert(processor->processor_set == pset);
 442         assert(bit_test(pset->cpu_bitmask, cpuid));
 443
 444         assert(old_state < PROCESSOR_STATE_LEN);
 445         assert(new_state < PROCESSOR_STATE_LEN);
 446
 447         processor->state = new_state;
 448
 449         bit_clear(pset->cpu_state_map[old_state], cpuid);
 450         bit_set(pset->cpu_state_map[new_state], cpuid);
 451
 452         if ((old_state == PROCESSOR_RUNNING) || (new_state == PROCESSOR_RUNNING)) {
 453                 sched_update_pset_load_average(pset);
 454                 if (new_state == PROCESSOR_RUNNING) {
 455                         assert(processor == current_processor());
 456                 }
 457         }
 458 }
 459
 460 #else   /* MACH_KERNEL_PRIVATE */
 461
 462 __BEGIN_DECLS
 463
 464 extern void             pset_deallocate(
 465         processor_set_t pset);
 466
 467 extern void             pset_reference(
 468         processor_set_t pset);
 469
 470 __END_DECLS
 471
 472 #endif  /* MACH_KERNEL_PRIVATE */
 473
 474 #ifdef KERNEL_PRIVATE
 475 __BEGIN_DECLS
 476 extern unsigned int             processor_count;
 477 extern processor_t      cpu_to_processor(int cpu);
 478
 479 extern kern_return_t    enable_smt_processors(bool enable);
 480
 481 extern boolean_t        processor_in_panic_context(processor_t processor);
 482 __END_DECLS
 483
 484 #endif /* KERNEL_PRIVATE */
 485
 486 #endif  /* _KERN_PROCESSOR_H_ */