osfmk/i386/cpu_data.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
  33 #ifndef I386_CPU_DATA
  34 #define I386_CPU_DATA
  35
  36 #include <mach_assert.h>
  37 #include <machine/atomic.h>
  38
  39 #include <kern/assert.h>
  40 #include <kern/kern_types.h>
  41 #include <kern/mpqueue.h>
  42 #include <kern/queue.h>
  43 #include <kern/processor.h>
  44 #include <kern/pms.h>
  45 #include <pexpert/pexpert.h>
  46 #include <mach/i386/thread_status.h>
  47 #include <mach/i386/vm_param.h>
  48 #include <i386/locks.h>
  49 #include <i386/rtclock_protos.h>
  50 #include <i386/pmCPU.h>
  51 #include <i386/cpu_topology.h>
  52 #include <i386/seg.h>
  53 #include <i386/mp.h>
  54
  55 #if CONFIG_VMX
  56 #include <i386/vmx/vmx_cpu.h>
  57 #endif
  58
  59 #if MONOTONIC
  60 #include <machine/monotonic.h>
  61 #endif /* MONOTONIC */
  62
  63 #include <machine/pal_routines.h>
  64
  65 /*
  66  * Data structures referenced (anonymously) from per-cpu data:
  67  */
  68 struct cpu_cons_buffer;
  69 struct cpu_desc_table;
  70 struct mca_state;
  71 struct prngContext;
  72
  73 /*
  74  * Data structures embedded in per-cpu data:
  75  */
  76 typedef struct rtclock_timer {
  77         mpqueue_head_t          queue;
  78         uint64_t                deadline;
  79         uint64_t                when_set;
  80         boolean_t               has_expired;
  81 } rtclock_timer_t;
  82
  83 typedef struct {
  84         /* The 'u' suffixed fields store the double-mapped descriptor addresses */
  85         struct x86_64_tss       *cdi_ktssu;
  86         struct x86_64_tss       *cdi_ktssb;
  87         x86_64_desc_register_t  cdi_gdtu;
  88         x86_64_desc_register_t  cdi_gdtb;
  89         x86_64_desc_register_t  cdi_idtu;
  90         x86_64_desc_register_t  cdi_idtb;
  91         struct real_descriptor  *cdi_ldtu;
  92         struct real_descriptor  *cdi_ldtb;
  93         vm_offset_t             cdi_sstku;
  94         vm_offset_t             cdi_sstkb;
  95 } cpu_desc_index_t;
  96
  97 typedef enum {
  98         TASK_MAP_32BIT,                 /* 32-bit user, compatibility mode */
  99         TASK_MAP_64BIT,                 /* 64-bit user thread, shared space */
 100 } task_map_t;
 101
 102
 103 /*
 104  * This structure is used on entry into the (uber-)kernel on syscall from
 105  * a 64-bit user. It contains the address of the machine state save area
 106  * for the current thread and a temporary place to save the user's rsp
 107  * before loading this address into rsp.
 108  */
 109 typedef struct {
 110         addr64_t        cu_isf;         /* thread->pcb->iss.isf */
 111         uint64_t        cu_tmp;         /* temporary scratch */
 112         addr64_t        cu_user_gs_base;
 113 } cpu_uber_t;
 114
 115 typedef uint16_t        pcid_t;
 116 typedef uint8_t         pcid_ref_t;
 117
 118 #define CPU_RTIME_BINS (12)
 119 #define CPU_ITIME_BINS (CPU_RTIME_BINS)
 120
 121 #define MAX_TRACE_BTFRAMES (16)
 122 typedef struct {
 123         boolean_t pltype;
 124         int plevel;
 125         uint64_t plbt[MAX_TRACE_BTFRAMES];
 126 } plrecord_t;
 127
 128 typedef enum {
 129         IOTRACE_PHYS_READ = 1,
 130         IOTRACE_PHYS_WRITE,
 131         IOTRACE_IO_READ,
 132         IOTRACE_IO_WRITE,
 133         IOTRACE_PORTIO_READ,
 134         IOTRACE_PORTIO_WRITE
 135 } iotrace_type_e;
 136
 137 typedef struct {
 138         iotrace_type_e  iotype;
 139         int             size;
 140         uint64_t        vaddr;
 141         uint64_t        paddr;
 142         uint64_t        val;
 143         uint64_t        start_time_abs;
 144         uint64_t        duration;
 145         uint64_t        backtrace[MAX_TRACE_BTFRAMES];
 146 } iotrace_entry_t;
 147
 148 #if     DEVELOPMENT || DEBUG
 149 #define DEFAULT_IOTRACE_ENTRIES_PER_CPU (64)
 150 #define IOTRACE_MAX_ENTRIES_PER_CPU (256)
 151 extern volatile int mmiotrace_enabled;
 152 extern int iotrace_generators;
 153 extern int iotrace_entries_per_cpu;
 154 extern int *iotrace_next;
 155 extern iotrace_entry_t **iotrace_ring;
 156
 157 extern void init_iotrace_bufs(int cpucnt, int entries_per_cpu);
 158 #endif /* DEVELOPMENT || DEBUG */
 159
 160 /*
 161  * Per-cpu data.
 162  *
 163  * Each processor has a per-cpu data area which is dereferenced through the
 164  * current_cpu_datap() macro. For speed, the %gs segment is based here, and
 165  * using this, inlines provides single-instruction access to frequently used
 166  * members - such as get_cpu_number()/cpu_number(), and get_active_thread()/
 167  * current_thread().
 168  *
 169  * Cpu data owned by another processor can be accessed using the
 170  * cpu_datap(cpu_number) macro which uses the cpu_data_ptr[] array of per-cpu
 171  * pointers.
 172  */
 173 typedef struct {
 174         pcid_t                  cpu_pcid_free_hint;
 175 #define PMAP_PCID_MAX_PCID      (0x800)
 176         pcid_ref_t              cpu_pcid_refcounts[PMAP_PCID_MAX_PCID];
 177         pmap_t                  cpu_pcid_last_pmap_dispatched[PMAP_PCID_MAX_PCID];
 178 } pcid_cdata_t;
 179
 180 typedef struct cpu_data {
 181         struct pal_cpu_data     cpu_pal_data;           /* PAL-specific data */
 182 #define                         cpu_pd cpu_pal_data     /* convenience alias */
 183         struct cpu_data         *cpu_this;              /* pointer to myself */
 184         thread_t                cpu_active_thread;
 185         thread_t                cpu_nthread;
 186         volatile int            cpu_preemption_level;
 187         int                     cpu_number;             /* Logical CPU */
 188         void                    *cpu_int_state;         /* interrupt state */
 189         vm_offset_t             cpu_active_stack;       /* kernel stack base */
 190         vm_offset_t             cpu_kernel_stack;       /* kernel stack top */
 191         vm_offset_t             cpu_int_stack_top;
 192         int                     cpu_interrupt_level;
 193         volatile int            cpu_signals;            /* IPI events */
 194         volatile int            cpu_prior_signals;      /* Last set of events,
 195                                                          * debugging
 196                                                          */
 197         ast_t                   cpu_pending_ast;
 198         volatile int            cpu_running;
 199 #if !MONOTONIC
 200         boolean_t               cpu_fixed_pmcs_enabled;
 201 #endif /* !MONOTONIC */
 202         rtclock_timer_t         rtclock_timer;
 203         uint64_t                quantum_timer_deadline;
 204         volatile addr64_t       cpu_active_cr3 __attribute((aligned(64)));
 205         union {
 206                 volatile uint32_t cpu_tlb_invalid;
 207                 struct {
 208                         volatile uint16_t cpu_tlb_invalid_local;
 209                         volatile uint16_t cpu_tlb_invalid_global;
 210                 };
 211         };
 212         uint64_t                cpu_ip_desc[2];
 213         volatile task_map_t     cpu_task_map;
 214         volatile addr64_t       cpu_task_cr3;
 215         addr64_t                cpu_kernel_cr3;
 216         volatile addr64_t       cpu_ucr3;
 217         volatile addr64_t       cpu_shadowtask_cr3;
 218         boolean_t               cpu_pagezero_mapped;
 219         cpu_uber_t              cpu_uber;
 220 /* Double-mapped per-CPU exception stack address */
 221         uintptr_t               cd_estack;
 222         int                     cpu_xstate;
 223         int                     cpu_curtask_has_ldt;
 224         int                     cpu_curthread_do_segchk;
 225 /* Address of shadowed, partially mirrored CPU data structures located
 226  * in the double mapped PML4
 227  */
 228         void                    *cd_shadow;
 229         union {
 230                 volatile uint32_t cpu_tlb_invalid_count;
 231                 struct {
 232                         volatile uint16_t cpu_tlb_invalid_local_count;
 233                         volatile uint16_t cpu_tlb_invalid_global_count;
 234                 };
 235         };
 236
 237         uint16_t                cpu_tlb_gen_counts_local[MAX_CPUS];
 238         uint16_t                cpu_tlb_gen_counts_global[MAX_CPUS];
 239
 240         struct processor        *cpu_processor;
 241 #if NCOPY_WINDOWS > 0
 242         struct cpu_pmap         *cpu_pmap;
 243 #endif
 244         struct real_descriptor  *cpu_ldtp;
 245         struct cpu_desc_table   *cpu_desc_tablep;
 246         cpu_desc_index_t        cpu_desc_index;
 247         int                     cpu_ldt;
 248 #if NCOPY_WINDOWS > 0
 249         vm_offset_t             cpu_copywindow_base;
 250         uint64_t                *cpu_copywindow_pdp;
 251
 252         vm_offset_t             cpu_physwindow_base;
 253         uint64_t                *cpu_physwindow_ptep;
 254 #endif
 255
 256 #define HWINTCNT_SIZE 256
 257         uint32_t                cpu_hwIntCnt[HWINTCNT_SIZE];    /* Interrupt counts */
 258         uint64_t                cpu_hwIntpexits[HWINTCNT_SIZE];
 259         uint64_t                cpu_dr7; /* debug control register */
 260         uint64_t                cpu_int_event_time;     /* intr entry/exit time */
 261         pal_rtc_nanotime_t      *cpu_nanotime;          /* Nanotime info */
 262 #if KPC
 263         /* double-buffered performance counter data */
 264         uint64_t                *cpu_kpc_buf[2];
 265         /* PMC shadow and reload value buffers */
 266         uint64_t                *cpu_kpc_shadow;
 267         uint64_t                *cpu_kpc_reload;
 268 #endif
 269 #if MONOTONIC
 270         struct mt_cpu cpu_monotonic;
 271 #endif /* MONOTONIC */
 272         uint32_t                cpu_pmap_pcid_enabled;
 273         pcid_t                  cpu_active_pcid;
 274         pcid_t                  cpu_last_pcid;
 275         pcid_t                  cpu_kernel_pcid;
 276         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp;
 277         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp_kernel;
 278         pcid_cdata_t            *cpu_pcid_data;
 279 #ifdef  PCID_STATS
 280         uint64_t                cpu_pmap_pcid_flushes;
 281         uint64_t                cpu_pmap_pcid_preserves;
 282 #endif
 283         uint64_t                cpu_aperf;
 284         uint64_t                cpu_mperf;
 285         uint64_t                cpu_c3res;
 286         uint64_t                cpu_c6res;
 287         uint64_t                cpu_c7res;
 288         uint64_t                cpu_itime_total;
 289         uint64_t                cpu_rtime_total;
 290         uint64_t                cpu_ixtime;
 291         uint64_t                cpu_idle_exits;
 292         uint64_t                cpu_rtimes[CPU_RTIME_BINS];
 293         uint64_t                cpu_itimes[CPU_ITIME_BINS];
 294 #if !MONOTONIC
 295         uint64_t                cpu_cur_insns;
 296         uint64_t                cpu_cur_ucc;
 297         uint64_t                cpu_cur_urc;
 298 #endif /* !MONOTONIC */
 299         uint64_t                cpu_gpmcs[4];
 300         uint64_t                cpu_max_observed_int_latency;
 301         int                     cpu_max_observed_int_latency_vector;
 302         volatile boolean_t      cpu_NMI_acknowledged;
 303         uint64_t                debugger_entry_time;
 304         uint64_t                debugger_ipi_time;
 305         /* A separate nested interrupt stack flag, to account
 306          * for non-nested interrupts arriving while on the interrupt stack
 307          * Currently only occurs when AICPM enables interrupts on the
 308          * interrupt stack during processor offlining.
 309          */
 310         uint32_t                cpu_nested_istack;
 311         uint32_t                cpu_nested_istack_events;
 312         x86_saved_state64_t     *cpu_fatal_trap_state;
 313         x86_saved_state64_t     *cpu_post_fatal_trap_state;
 314 #if CONFIG_VMX
 315         vmx_cpu_t               cpu_vmx;                /* wonderful world of virtualization */
 316 #endif
 317 #if CONFIG_MCA
 318         struct mca_state        *cpu_mca_state;         /* State at MC fault */
 319 #endif
 320         int                     cpu_type;
 321         int                     cpu_subtype;
 322         int                     cpu_threadtype;
 323         boolean_t               cpu_iflag;
 324         boolean_t               cpu_boot_complete;
 325         int                     cpu_hibernate;
 326 #define MAX_PREEMPTION_RECORDS (8)
 327 #if     DEVELOPMENT || DEBUG
 328         int                     cpu_plri;
 329         plrecord_t              plrecords[MAX_PREEMPTION_RECORDS];
 330 #endif
 331         void                    *cpu_console_buf;
 332         struct x86_lcpu         lcpu;
 333         int                     cpu_phys_number;        /* Physical CPU */
 334         cpu_id_t                cpu_id;                 /* Platform Expert */
 335 #if DEBUG
 336         uint64_t                cpu_entry_cr3;
 337         uint64_t                cpu_exit_cr3;
 338         uint64_t                cpu_pcid_last_cr3;
 339 #endif
 340         boolean_t               cpu_rendezvous_in_progress;
 341 } cpu_data_t;
 342
 343 extern cpu_data_t       *cpu_data_ptr[];
 344
 345 /* Macro to generate inline bodies to retrieve per-cpu data fields. */
 346 #if defined(__clang__)
 347 #define GS_RELATIVE volatile __attribute__((address_space(256)))
 348 #ifndef offsetof
 349 #define offsetof(TYPE, MEMBER) __builtin_offsetof(TYPE,MEMBER)
 350 #endif
 351
 352 #define CPU_DATA_GET(member, type)                                                                               \
 353         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 354                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 355         type ret;                                                                                                                       \
 356         ret = cpu_data->member;                                                                                         \
 357         return ret;
 358
 359 #define CPU_DATA_GET_INDEX(member, index, type)                                                   \
 360         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 361                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 362         type ret;                                                                                                                       \
 363         ret = cpu_data->member[index];                                                                          \
 364         return ret;
 365
 366 #define CPU_DATA_SET(member, value)                                                                              \
 367         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 368                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 369         cpu_data->member = value;
 370
 371 #define CPU_DATA_XCHG(member, value, type)                                                                \
 372         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 373                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 374         type ret;                                                                                                                       \
 375         ret = cpu_data->member;                                                                                         \
 376         cpu_data->member = value;                                                                                       \
 377         return ret;
 378
 379 #else /* !defined(__clang__) */
 380
 381 #ifndef offsetof
 382 #define offsetof(TYPE, MEMBER) ((size_t) &((TYPE *)0)->MEMBER)
 383 #endif /* offsetof */
 384 #define CPU_DATA_GET(member, type)                                       \
 385         type ret;                                                       \
 386         __asm__ volatile ("mov %%gs:%P1,%0"                             \
 387                 : "=r" (ret)                                            \
 388                 : "i" (offsetof(cpu_data_t,member)));                   \
 389         return ret;
 390
 391 #define CPU_DATA_GET_INDEX(member, index, type)   \
 392         type ret;                                                       \
 393         __asm__ volatile ("mov %%gs:(%1),%0"                            \
 394                 : "=r" (ret)                                            \
 395                 : "r" (offsetof(cpu_data_t,member[index])));                    \
 396         return ret;
 397
 398 #define CPU_DATA_SET(member, value)                                      \
 399         __asm__ volatile ("mov %0,%%gs:%P1"                             \
 400                 :                                                       \
 401                 : "r" (value), "i" (offsetof(cpu_data_t,member)));
 402
 403 #define CPU_DATA_XCHG(member, value, type)                                \
 404         type ret;                                                       \
 405         __asm__ volatile ("xchg %0,%%gs:%P1"                            \
 406                 : "=r" (ret)                                            \
 407                 : "i" (offsetof(cpu_data_t,member)), "0" (value));      \
 408         return ret;
 409
 410 #endif /* !defined(__clang__) */
 411
 412 /*
 413  * Everyone within the osfmk part of the kernel can use the fast
 414  * inline versions of these routines.  Everyone outside, must call
 415  * the real thing,
 416  */
 417
 418
 419 /*
 420  * The "volatile" flavor of current_thread() is intended for use by
 421  * scheduler code which may need to update the thread pointer in the
 422  * course of a context switch.  Any call to current_thread() made
 423  * prior to the thread pointer update should be safe to optimize away
 424  * as it should be consistent with that thread's state to the extent
 425  * the compiler can reason about it.  Likewise, the context switch
 426  * path will eventually result in an arbitrary branch to the new
 427  * thread's pc, about which the compiler won't be able to reason.
 428  * Thus any compile-time optimization of current_thread() calls made
 429  * within the new thread should be safely encapsulated in its
 430  * register/stack state.  The volatile form therefore exists to cover
 431  * the window between the thread pointer update and the branch to
 432  * the new pc.
 433  */
 434 static inline thread_t
 435 get_active_thread_volatile(void)
 436 {
 437         CPU_DATA_GET(cpu_active_thread, thread_t)
 438 }
 439
 440 static inline __attribute__((const)) thread_t
 441 get_active_thread(void)
 442 {
 443         CPU_DATA_GET(cpu_active_thread, thread_t)
 444 }
 445
 446 #define current_thread_fast()           get_active_thread()
 447 #define current_thread_volatile()       get_active_thread_volatile()
 448 #define current_thread()                current_thread_fast()
 449
 450 #define cpu_mode_is64bit()              TRUE
 451
 452 static inline int
 453 get_preemption_level(void)
 454 {
 455         CPU_DATA_GET(cpu_preemption_level, int)
 456 }
 457 static inline int
 458 get_interrupt_level(void)
 459 {
 460         CPU_DATA_GET(cpu_interrupt_level, int)
 461 }
 462 static inline int
 463 get_cpu_number(void)
 464 {
 465         CPU_DATA_GET(cpu_number, int)
 466 }
 467 static inline int
 468 get_cpu_phys_number(void)
 469 {
 470         CPU_DATA_GET(cpu_phys_number, int)
 471 }
 472
 473 static inline cpu_data_t *
 474 current_cpu_datap(void)
 475 {
 476         CPU_DATA_GET(cpu_this, cpu_data_t *);
 477 }
 478
 479 /*
 480  * Facility to diagnose preemption-level imbalances, which are otherwise
 481  * challenging to debug. On each operation that enables or disables preemption,
 482  * we record a backtrace into a per-CPU ring buffer, along with the current
 483  * preemption level and operation type. Thus, if an imbalance is observed,
 484  * one can examine these per-CPU records to determine which codepath failed
 485  * to re-enable preemption, enabled premption without a corresponding
 486  * disablement etc. The backtracer determines which stack is currently active,
 487  * and uses that to perform bounds checks on unterminated stacks.
 488  * To enable, sysctl -w machdep.pltrace=1 on DEVELOPMENT or DEBUG kernels (DRK '15)
 489  * The bounds check currently doesn't account for non-default thread stack sizes.
 490  */
 491 #if DEVELOPMENT || DEBUG
 492 static inline void
 493 rbtrace_bt(uint64_t *rets, int maxframes, cpu_data_t *cdata)
 494 {
 495         extern uint32_t         low_intstack[];         /* bottom */
 496         extern uint32_t         low_eintstack[];        /* top */
 497         extern char             mp_slave_stack[];
 498
 499         uint64_t kstackb, kstackt;
 500
 501         /* Obtain the 'current' program counter, initial backtrace
 502          * element. This will also indicate if we were unable to
 503          * trace further up the stack for some reason
 504          */
 505         __asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
 506              : "=m" (rets[0])
 507              :
 508              : "rax");
 509
 510
 511         thread_t cplthread = cdata->cpu_active_thread;
 512         if (cplthread) {
 513                 uintptr_t csp;
 514                 __asm__ __volatile__ ("movq %%rsp, %0": "=r" (csp):);
 515                 /* Determine which stack we're on to populate stack bounds.
 516                  * We don't need to trace across stack boundaries for this
 517                  * routine.
 518                  */
 519                 kstackb = cdata->cpu_active_stack;
 520                 kstackt = kstackb + KERNEL_STACK_SIZE;
 521                 if (csp < kstackb || csp > kstackt) {
 522                         kstackt = cdata->cpu_kernel_stack;
 523                         kstackb = kstackt - KERNEL_STACK_SIZE;
 524                         if (csp < kstackb || csp > kstackt) {
 525                                 kstackt = cdata->cpu_int_stack_top;
 526                                 kstackb = kstackt - INTSTACK_SIZE;
 527                                 if (csp < kstackb || csp > kstackt) {
 528                                         kstackt = (uintptr_t)low_eintstack;
 529                                         kstackb = kstackt - INTSTACK_SIZE;
 530                                         if (csp < kstackb || csp > kstackt) {
 531                                                 kstackb = (uintptr_t) mp_slave_stack;
 532                                                 kstackt = kstackb + PAGE_SIZE;
 533                                         } else {
 534                                                 kstackb = 0;
 535                                                 kstackt = 0;
 536                                         }
 537                                 }
 538                         }
 539                 }
 540
 541                 if (__probable(kstackb && kstackt)) {
 542                         uint64_t *cfp = (uint64_t *) __builtin_frame_address(0);
 543                         int rbbtf;
 544
 545                         for (rbbtf = 1; rbbtf < maxframes; rbbtf++) {
 546                                 if (((uint64_t)cfp == 0) || (((uint64_t)cfp < kstackb) || ((uint64_t)cfp > kstackt))) {
 547                                         rets[rbbtf] = 0;
 548                                         continue;
 549                                 }
 550                                 rets[rbbtf] = *(cfp + 1);
 551                                 cfp = (uint64_t *) (*cfp);
 552                         }
 553                 }
 554         }
 555 }
 556
 557 static inline void
 558 pltrace_internal(boolean_t enable)
 559 {
 560         cpu_data_t *cdata = current_cpu_datap();
 561         int cpli = cdata->cpu_preemption_level;
 562         int cplrecord = cdata->cpu_plri;
 563         uint64_t *plbts;
 564
 565         assert(cpli >= 0);
 566
 567         cdata->plrecords[cplrecord].pltype = enable;
 568         cdata->plrecords[cplrecord].plevel = cpli;
 569
 570         plbts = &cdata->plrecords[cplrecord].plbt[0];
 571
 572         cplrecord++;
 573
 574         if (cplrecord >= MAX_PREEMPTION_RECORDS) {
 575                 cplrecord = 0;
 576         }
 577
 578         cdata->cpu_plri = cplrecord;
 579
 580         rbtrace_bt(plbts, MAX_TRACE_BTFRAMES - 1, cdata);
 581 }
 582
 583 extern int plctrace_enabled;
 584
 585 static inline void
 586 iotrace(iotrace_type_e type, uint64_t vaddr, uint64_t paddr, int size, uint64_t val,
 587     uint64_t sabs, uint64_t duration)
 588 {
 589         cpu_data_t *cdata;
 590         int cpu_num, nextidx;
 591         iotrace_entry_t *cur_iotrace_ring;
 592
 593         if (__improbable(mmiotrace_enabled == 0 || iotrace_generators == 0)) {
 594                 return;
 595         }
 596
 597         cdata = current_cpu_datap();
 598         cpu_num = cdata->cpu_number;
 599         nextidx = iotrace_next[cpu_num];
 600         cur_iotrace_ring = iotrace_ring[cpu_num];
 601
 602         cur_iotrace_ring[nextidx].iotype = type;
 603         cur_iotrace_ring[nextidx].vaddr = vaddr;
 604         cur_iotrace_ring[nextidx].paddr = paddr;
 605         cur_iotrace_ring[nextidx].size = size;
 606         cur_iotrace_ring[nextidx].val = val;
 607         cur_iotrace_ring[nextidx].start_time_abs = sabs;
 608         cur_iotrace_ring[nextidx].duration = duration;
 609
 610         iotrace_next[cpu_num] = ((nextidx + 1) >= iotrace_entries_per_cpu) ? 0 : (nextidx + 1);
 611
 612         rbtrace_bt(&cur_iotrace_ring[nextidx].backtrace[0],
 613             MAX_TRACE_BTFRAMES - 1, cdata);
 614 }
 615 #endif /* DEVELOPMENT || DEBUG */
 616
 617 static inline void
 618 pltrace(boolean_t plenable)
 619 {
 620 #if DEVELOPMENT || DEBUG
 621         if (__improbable(plctrace_enabled != 0)) {
 622                 pltrace_internal(plenable);
 623         }
 624 #else
 625         (void)plenable;
 626 #endif
 627 }
 628
 629 static inline void
 630 disable_preemption_internal(void)
 631 {
 632         assert(get_preemption_level() >= 0);
 633
 634         os_compiler_barrier(release);
 635 #if defined(__clang__)
 636         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 637         cpu_data->cpu_preemption_level++;
 638 #else
 639         __asm__ volatile ("incl %%gs:%P0"
 640             :
 641             : "i" (offsetof(cpu_data_t, cpu_preemption_level)));
 642 #endif
 643         os_compiler_barrier(acquire);
 644         pltrace(FALSE);
 645 }
 646
 647 static inline void
 648 enable_preemption_internal(void)
 649 {
 650         assert(get_preemption_level() > 0);
 651         pltrace(TRUE);
 652         os_compiler_barrier(release);
 653 #if defined(__clang__)
 654         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 655         if (0 == --cpu_data->cpu_preemption_level) {
 656                 kernel_preempt_check();
 657         }
 658 #else
 659         __asm__ volatile ("decl %%gs:%P0                \n\t"
 660                           "jne 1f                       \n\t"
 661                           "call _kernel_preempt_check   \n\t"
 662                           "1:"
 663                         : /* no outputs */
 664                         : "i" (offsetof(cpu_data_t, cpu_preemption_level))
 665                         : "eax", "ecx", "edx", "cc", "memory");
 666 #endif
 667         os_compiler_barrier(acquire);
 668 }
 669
 670 static inline void
 671 enable_preemption_no_check(void)
 672 {
 673         assert(get_preemption_level() > 0);
 674
 675         pltrace(TRUE);
 676         os_compiler_barrier(release);
 677 #if defined(__clang__)
 678         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 679         cpu_data->cpu_preemption_level--;
 680 #else
 681         __asm__ volatile ("decl %%gs:%P0"
 682                         : /* no outputs */
 683                         : "i" (offsetof(cpu_data_t, cpu_preemption_level))
 684                         : "cc", "memory");
 685 #endif
 686         os_compiler_barrier(acquire);
 687 }
 688
 689 static inline void
 690 _enable_preemption_no_check(void)
 691 {
 692         enable_preemption_no_check();
 693 }
 694
 695 static inline void
 696 mp_disable_preemption(void)
 697 {
 698         disable_preemption_internal();
 699 }
 700
 701 static inline void
 702 _mp_disable_preemption(void)
 703 {
 704         disable_preemption_internal();
 705 }
 706
 707 static inline void
 708 mp_enable_preemption(void)
 709 {
 710         enable_preemption_internal();
 711 }
 712
 713 static inline void
 714 _mp_enable_preemption(void)
 715 {
 716         enable_preemption_internal();
 717 }
 718
 719 static inline void
 720 mp_enable_preemption_no_check(void)
 721 {
 722         enable_preemption_no_check();
 723 }
 724
 725 static inline void
 726 _mp_enable_preemption_no_check(void)
 727 {
 728         enable_preemption_no_check();
 729 }
 730
 731 #ifdef XNU_KERNEL_PRIVATE
 732 #define disable_preemption() disable_preemption_internal()
 733 #define enable_preemption() enable_preemption_internal()
 734 #define MACHINE_PREEMPTION_MACROS (1)
 735 #endif
 736
 737 static inline cpu_data_t *
 738 cpu_datap(int cpu)
 739 {
 740         return cpu_data_ptr[cpu];
 741 }
 742
 743 static inline int
 744 cpu_is_running(int cpu)
 745 {
 746         return (cpu_datap(cpu) != NULL) && (cpu_datap(cpu)->cpu_running);
 747 }
 748
 749 #ifdef MACH_KERNEL_PRIVATE
 750 static inline cpu_data_t *
 751 cpu_shadowp(int cpu)
 752 {
 753         return cpu_data_ptr[cpu]->cd_shadow;
 754 }
 755
 756 #endif
 757 extern cpu_data_t *cpu_data_alloc(boolean_t is_boot_cpu);
 758 extern void cpu_data_realloc(void);
 759
 760 #endif  /* I386_CPU_DATA */