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 } pset_cluster_type_t;
 151
 152 typedef bitmap_t cpumap_t;
 153
 154 struct processor_set {
 155         int                     online_processor_count;
 156         int                     load_average;
 157
 158         int                     cpu_set_low, cpu_set_hi;
 159         int                     cpu_set_count;
 160         int                     last_chosen;
 161         cpumap_t                cpu_bitmask;
 162         cpumap_t                recommended_bitmask;
 163         cpumap_t                cpu_state_map[PROCESSOR_STATE_LEN];
 164         cpumap_t                primary_map;
 165 #define SCHED_PSET_TLOCK (1)
 166 #if __SMP__
 167 #if     defined(SCHED_PSET_TLOCK)
 168 /* TODO: reorder struct for temporal cache locality */
 169         __attribute__((aligned(128))) lck_ticket_t      sched_lock;
 170 #else /* SCHED_PSET_TLOCK*/
 171         __attribute__((aligned(128))) lck_spin_t        sched_lock;     /* lock for above */
 172 #endif /* SCHED_PSET_TLOCK*/
 173 #endif
 174
 175 #if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_MULTIQ)
 176         struct run_queue        pset_runq;      /* runq for this processor set */
 177 #endif
 178         struct rt_queue         rt_runq;        /* realtime runq for this processor set */
 179 #if CONFIG_SCHED_CLUTCH
 180         struct sched_clutch_root                pset_clutch_root; /* clutch hierarchy root */
 181 #endif /* CONFIG_SCHED_CLUTCH */
 182
 183 #if defined(CONFIG_SCHED_TRADITIONAL)
 184         int                                     pset_runq_bound_count;
 185         /* # of threads in runq bound to any processor in pset */
 186 #endif
 187
 188         /* CPUs that have been sent an unacknowledged remote AST for scheduling purposes */
 189         cpumap_t                        pending_AST_URGENT_cpu_mask;
 190         cpumap_t                        pending_AST_PREEMPT_cpu_mask;
 191 #if defined(CONFIG_SCHED_DEFERRED_AST)
 192         /*
 193          * A separate mask, for ASTs that we may be able to cancel.  This is dependent on
 194          * some level of support for requesting an AST on a processor, and then quashing
 195          * that request later.
 196          *
 197          * The purpose of this field (and the associated codepaths) is to infer when we
 198          * no longer need a processor that is DISPATCHING to come up, and to prevent it
 199          * from coming out of IDLE if possible.  This should serve to decrease the number
 200          * of spurious ASTs in the system, and let processors spend longer periods in
 201          * IDLE.
 202          */
 203         cpumap_t                        pending_deferred_AST_cpu_mask;
 204 #endif
 205         cpumap_t                        pending_spill_cpu_mask;
 206
 207         struct ipc_port *       pset_self;              /* port for operations */
 208         struct ipc_port *       pset_name_self; /* port for information */
 209
 210         processor_set_t         pset_list;              /* chain of associated psets */
 211         pset_node_t             node;
 212         uint32_t                pset_cluster_id;
 213         pset_cluster_type_t     pset_cluster_type;
 214 };
 215
 216 extern struct processor_set     pset0;
 217
 218 struct pset_node {
 219         processor_set_t         psets;                  /* list of associated psets */
 220         uint32_t                pset_count;             /* count of associated psets */
 221
 222         pset_node_t                     nodes;                  /* list of associated subnodes */
 223         pset_node_t                     node_list;              /* chain of associated nodes */
 224
 225         pset_node_t                     parent;
 226 };
 227
 228 extern struct pset_node pset_node0;
 229
 230 extern queue_head_t             tasks, terminated_tasks, threads, corpse_tasks; /* Terminated tasks are ONLY for stackshot */
 231 extern int                              tasks_count, terminated_tasks_count, threads_count;
 232 decl_lck_mtx_data(extern, tasks_threads_lock);
 233 decl_lck_mtx_data(extern, tasks_corpse_lock);
 234
 235 struct processor {
 236         processor_state_t       state;                  /* See above */
 237         bool                    is_SMT;
 238         bool                    is_recommended;
 239         struct thread           *active_thread;         /* thread running on processor */
 240         struct thread           *idle_thread;           /* this processor's idle thread. */
 241         struct thread           *startup_thread;
 242
 243         processor_set_t         processor_set;  /* assigned set */
 244
 245         int                     current_pri;            /* priority of current thread */
 246         sfi_class_id_t          current_sfi_class;      /* SFI class of current thread */
 247         perfcontrol_class_t     current_perfctl_class;  /* Perfcontrol class for current thread */
 248         pset_cluster_type_t     current_recommended_pset_type;  /* Cluster type recommended for current thread */
 249         thread_urgency_t        current_urgency;        /* cached urgency of current thread */
 250         bool                    current_is_NO_SMT;         /* cached TH_SFLAG_NO_SMT of current thread */
 251         bool                    current_is_bound;       /* current thread is bound to this processor */
 252
 253         int                     starting_pri;           /* priority of current thread as it was when scheduled */
 254         int                     cpu_id;                 /* platform numeric id */
 255         cpu_quiescent_state_t   cpu_quiesce_state;
 256         uint64_t                cpu_quiesce_last_checkin;
 257
 258         timer_call_data_t       quantum_timer;  /* timer for quantum expiration */
 259         uint64_t                        quantum_end;    /* time when current quantum ends */
 260         uint64_t                        last_dispatch;  /* time of last dispatch */
 261
 262         uint64_t                        kperf_last_sample_time; /* time of last kperf sample */
 263
 264         uint64_t                        deadline;               /* current deadline */
 265         bool                    first_timeslice;        /* has the quantum expired since context switch */
 266         bool                    processor_offlined;        /* has the processor been explicitly processor_offline'ed */
 267         bool                    must_idle;              /* Needs to be forced idle as next selected thread is allowed on this processor */
 268
 269         processor_t             processor_primary;      /* pointer to primary processor for
 270                                                          * secondary SMT processors, or a pointer
 271                                                          * to ourselves for primaries or non-SMT */
 272         processor_t             processor_secondary;
 273
 274 #if defined(CONFIG_SCHED_TRADITIONAL) || defined(CONFIG_SCHED_MULTIQ)
 275         struct run_queue        runq;                   /* runq for this processor */
 276 #endif
 277
 278 #if defined(CONFIG_SCHED_TRADITIONAL)
 279         int                                     runq_bound_count; /* # of threads bound to this processor */
 280 #endif
 281 #if defined(CONFIG_SCHED_GRRR)
 282         struct grrr_run_queue   grrr_runq;      /* Group Ratio Round-Robin runq */
 283 #endif
 284         struct ipc_port *       processor_self; /* port for operations */
 285
 286         processor_t                     processor_list; /* all existing processors */
 287         processor_data_t        processor_data; /* per-processor data */
 288 };
 289
 290 extern processor_t              processor_list;
 291 decl_simple_lock_data(extern, processor_list_lock);
 292
 293 #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 */
 294 extern processor_t              processor_array[MAX_SCHED_CPUS]; /* array indexed by cpuid */
 295
 296 extern uint32_t                 processor_avail_count;
 297 extern uint32_t                 processor_avail_count_user;
 298
 299 extern processor_t              master_processor;
 300
 301 extern boolean_t                sched_stats_active;
 302
 303 extern processor_t      current_processor(void);
 304
 305 /* Lock macros, always acquired and released with interrupts disabled (splsched()) */
 306
 307 extern lck_grp_t pset_lck_grp;
 308
 309 #if __SMP__
 310 #if defined(SCHED_PSET_TLOCK)
 311 #define pset_lock_init(p)               lck_ticket_init(&(p)->sched_lock)
 312 #define pset_lock(p)                    lck_ticket_lock(&(p)->sched_lock)
 313 #define pset_unlock(p)                  lck_ticket_unlock(&(p)->sched_lock)
 314 #define pset_assert_locked(p)           lck_ticket_assert_owned(&(p)->sched_lock)
 315 #else /* SCHED_PSET_TLOCK*/
 316 #define pset_lock_init(p)               lck_spin_init(&(p)->sched_lock, &pset_lck_grp, NULL)
 317 #define pset_lock(p)                    lck_spin_lock_grp(&(p)->sched_lock, &pset_lck_grp)
 318 #define pset_unlock(p)                  lck_spin_unlock(&(p)->sched_lock)
 319 #define pset_assert_locked(p)           LCK_SPIN_ASSERT(&(p)->sched_lock, LCK_ASSERT_OWNED)
 320 #endif /*!SCHED_PSET_TLOCK*/
 321
 322 #define rt_lock_lock(p)                 simple_lock(&SCHED(rt_runq)(p)->rt_lock, &pset_lck_grp)
 323 #define rt_lock_unlock(p)               simple_unlock(&SCHED(rt_runq)(p)->rt_lock)
 324 #define rt_lock_init(p)                 simple_lock_init(&SCHED(rt_runq)(p)->rt_lock, 0)
 325 #else
 326 #define pset_lock(p)                    do { (void)p; } while(0)
 327 #define pset_unlock(p)                  do { (void)p; } while(0)
 328 #define pset_lock_init(p)               do { (void)p; } while(0)
 329 #define pset_assert_locked(p)           do { (void)p; } while(0)
 330
 331 #define rt_lock_lock(p)                 do { (void)p; } while(0)
 332 #define rt_lock_unlock(p)               do { (void)p; } while(0)
 333 #define rt_lock_init(p)                 do { (void)p; } while(0)
 334 #endif /* SMP */
 335
 336 extern void             processor_bootstrap(void);
 337
 338 extern void             processor_init(
 339         processor_t             processor,
 340         int                             cpu_id,
 341         processor_set_t processor_set);
 342
 343 extern void             processor_set_primary(
 344         processor_t             processor,
 345         processor_t             primary);
 346
 347 extern kern_return_t    processor_shutdown(
 348         processor_t             processor);
 349
 350 extern kern_return_t    processor_start_from_user(
 351         processor_t     processor);
 352 extern kern_return_t    processor_exit_from_user(
 353         processor_t     processor);
 354
 355 kern_return_t
 356 sched_processor_enable(processor_t processor, boolean_t enable);
 357
 358 extern void             processor_queue_shutdown(
 359         processor_t             processor);
 360
 361 extern void             processor_queue_shutdown(
 362         processor_t             processor);
 363
 364 extern processor_set_t  processor_pset(
 365         processor_t             processor);
 366
 367 extern pset_node_t              pset_node_root(void);
 368
 369 extern processor_set_t  pset_create(
 370         pset_node_t             node);
 371
 372 extern void             pset_init(
 373         processor_set_t         pset,
 374         pset_node_t                     node);
 375
 376 extern processor_set_t pset_find(
 377         uint32_t cluster_id,
 378         processor_set_t default_pset);
 379
 380 extern kern_return_t    processor_info_count(
 381         processor_flavor_t              flavor,
 382         mach_msg_type_number_t  *count);
 383
 384 #define pset_deallocate(x)
 385 #define pset_reference(x)
 386
 387 extern void                             machine_run_count(
 388         uint32_t        count);
 389
 390 extern processor_t              machine_choose_processor(
 391         processor_set_t         pset,
 392         processor_t                     processor);
 393
 394 #define next_pset(p)    (((p)->pset_list != PROCESSOR_SET_NULL)? (p)->pset_list: (p)->node->psets)
 395
 396 #define PSET_THING_TASK         0
 397 #define PSET_THING_THREAD       1
 398
 399 extern kern_return_t    processor_set_things(
 400         processor_set_t pset,
 401         void **thing_list,
 402         mach_msg_type_number_t *count,
 403         int type);
 404
 405 extern pset_cluster_type_t recommended_pset_type(thread_t thread);
 406
 407 inline static bool
 408 pset_is_recommended(processor_set_t pset)
 409 {
 410         return (pset->recommended_bitmask & pset->cpu_bitmask) != 0;
 411 }
 412
 413 extern void processor_state_update_idle(processor_t processor);
 414 extern void processor_state_update_from_thread(processor_t processor, thread_t thread);
 415 extern void processor_state_update_explicit(processor_t processor, int pri,
 416     sfi_class_id_t sfi_class, pset_cluster_type_t pset_type,
 417     perfcontrol_class_t perfctl_class, thread_urgency_t urgency);
 418
 419 #define PSET_LOAD_NUMERATOR_SHIFT   16
 420 #define PSET_LOAD_FRACTIONAL_SHIFT   4
 421
 422 inline static int
 423 sched_get_pset_load_average(processor_set_t pset)
 424 {
 425         return pset->load_average >> (PSET_LOAD_NUMERATOR_SHIFT - PSET_LOAD_FRACTIONAL_SHIFT);
 426 }
 427 extern void sched_update_pset_load_average(processor_set_t pset);
 428
 429 inline static void
 430 pset_update_processor_state(processor_set_t pset, processor_t processor, uint new_state)
 431 {
 432         pset_assert_locked(pset);
 433
 434         uint old_state = processor->state;
 435         uint cpuid = processor->cpu_id;
 436
 437         assert(processor->processor_set == pset);
 438         assert(bit_test(pset->cpu_bitmask, cpuid));
 439
 440         assert(old_state < PROCESSOR_STATE_LEN);
 441         assert(new_state < PROCESSOR_STATE_LEN);
 442
 443         processor->state = new_state;
 444
 445         bit_clear(pset->cpu_state_map[old_state], cpuid);
 446         bit_set(pset->cpu_state_map[new_state], cpuid);
 447
 448         if ((old_state == PROCESSOR_RUNNING) || (new_state == PROCESSOR_RUNNING)) {
 449                 sched_update_pset_load_average(pset);
 450                 if (new_state == PROCESSOR_RUNNING) {
 451                         assert(processor == current_processor());
 452                 }
 453         }
 454 }
 455
 456 #else   /* MACH_KERNEL_PRIVATE */
 457
 458 __BEGIN_DECLS
 459
 460 extern void             pset_deallocate(
 461         processor_set_t pset);
 462
 463 extern void             pset_reference(
 464         processor_set_t pset);
 465
 466 __END_DECLS
 467
 468 #endif  /* MACH_KERNEL_PRIVATE */
 469
 470 #ifdef KERNEL_PRIVATE
 471 __BEGIN_DECLS
 472 extern unsigned int             processor_count;
 473 extern processor_t      cpu_to_processor(int cpu);
 474
 475 extern kern_return_t    enable_smt_processors(bool enable);
 476
 477 extern boolean_t        processor_in_panic_context(processor_t processor);
 478 __END_DECLS
 479
 480 #endif /* KERNEL_PRIVATE */
 481
 482 #endif  /* _KERN_PROCESSOR_H_ */