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 #if     DEVELOPMENT || DEBUG
 129 typedef enum {
 130         IOTRACE_PHYS_READ = 1,
 131         IOTRACE_PHYS_WRITE,
 132         IOTRACE_IO_READ,
 133         IOTRACE_IO_WRITE,
 134         IOTRACE_PORTIO_READ,
 135         IOTRACE_PORTIO_WRITE
 136 } iotrace_type_e;
 137
 138 typedef struct {
 139         iotrace_type_e  iotype;
 140         int             size;
 141         uint64_t        vaddr;
 142         uint64_t        paddr;
 143         uint64_t        val;
 144         uint64_t        start_time_abs;
 145         uint64_t        duration;
 146         uint64_t        backtrace[MAX_TRACE_BTFRAMES];
 147 } iotrace_entry_t;
 148
 149 typedef struct {
 150         int             vector;                 /* Vector number of interrupt */
 151         thread_t        curthread;              /* Current thread at the time of the interrupt */
 152         uint64_t        interrupted_pc;
 153         int             curpl;                  /* Current preemption level */
 154         int             curil;                  /* Current interrupt level */
 155         uint64_t        start_time_abs;
 156         uint64_t        duration;
 157         uint64_t        backtrace[MAX_TRACE_BTFRAMES];
 158 } traptrace_entry_t;
 159
 160 #define DEFAULT_IOTRACE_ENTRIES_PER_CPU (64)
 161 #define IOTRACE_MAX_ENTRIES_PER_CPU (256)
 162 extern volatile int mmiotrace_enabled;
 163 extern int iotrace_generators;
 164 extern int iotrace_entries_per_cpu;
 165 extern int *iotrace_next;
 166 extern iotrace_entry_t **iotrace_ring;
 167
 168 #define TRAPTRACE_INVALID_INDEX (~0U)
 169 #define DEFAULT_TRAPTRACE_ENTRIES_PER_CPU (16)
 170 #define TRAPTRACE_MAX_ENTRIES_PER_CPU (256)
 171 extern volatile int traptrace_enabled;
 172 extern int traptrace_generators;
 173 extern int traptrace_entries_per_cpu;
 174 extern int *traptrace_next;
 175 extern traptrace_entry_t **traptrace_ring;
 176 #endif /* DEVELOPMENT || DEBUG */
 177
 178 /*
 179  * Per-cpu data.
 180  *
 181  * Each processor has a per-cpu data area which is dereferenced through the
 182  * current_cpu_datap() macro. For speed, the %gs segment is based here, and
 183  * using this, inlines provides single-instruction access to frequently used
 184  * members - such as get_cpu_number()/cpu_number(), and get_active_thread()/
 185  * current_thread().
 186  *
 187  * Cpu data owned by another processor can be accessed using the
 188  * cpu_datap(cpu_number) macro which uses the cpu_data_ptr[] array of per-cpu
 189  * pointers.
 190  */
 191 typedef struct {
 192         pcid_t                  cpu_pcid_free_hint;
 193 #define PMAP_PCID_MAX_PCID      (0x800)
 194         pcid_ref_t              cpu_pcid_refcounts[PMAP_PCID_MAX_PCID];
 195         pmap_t                  cpu_pcid_last_pmap_dispatched[PMAP_PCID_MAX_PCID];
 196 } pcid_cdata_t;
 197
 198 typedef struct cpu_data {
 199         struct pal_cpu_data     cpu_pal_data;           /* PAL-specific data */
 200 #define                         cpu_pd cpu_pal_data     /* convenience alias */
 201         struct cpu_data         *cpu_this;              /* pointer to myself */
 202         thread_t                cpu_active_thread;
 203         thread_t                cpu_nthread;
 204         int                     cpu_number;             /* Logical CPU */
 205         void                    *cpu_int_state;         /* interrupt state */
 206         vm_offset_t             cpu_active_stack;       /* kernel stack base */
 207         vm_offset_t             cpu_kernel_stack;       /* kernel stack top */
 208         vm_offset_t             cpu_int_stack_top;
 209         volatile int            cpu_signals;            /* IPI events */
 210         volatile int            cpu_prior_signals;      /* Last set of events,
 211                                                          * debugging
 212                                                          */
 213         ast_t                   cpu_pending_ast;
 214         /*
 215          * Note if rearranging fields:
 216          * We want cpu_preemption_level on a different
 217          * cache line than cpu_active_thread
 218          * for optimizing mtx_spin phase.
 219          */
 220         int                     cpu_interrupt_level;
 221         volatile int            cpu_preemption_level;
 222         volatile int            cpu_running;
 223 #if !MONOTONIC
 224         boolean_t               cpu_fixed_pmcs_enabled;
 225 #endif /* !MONOTONIC */
 226         rtclock_timer_t         rtclock_timer;
 227         uint64_t                quantum_timer_deadline;
 228         volatile addr64_t       cpu_active_cr3 __attribute((aligned(64)));
 229         union {
 230                 volatile uint32_t cpu_tlb_invalid;
 231                 struct {
 232                         volatile uint16_t cpu_tlb_invalid_local;
 233                         volatile uint16_t cpu_tlb_invalid_global;
 234                 };
 235         };
 236         uint64_t                cpu_ip_desc[2];
 237         volatile task_map_t     cpu_task_map;
 238         volatile addr64_t       cpu_task_cr3;
 239         addr64_t                cpu_kernel_cr3;
 240         volatile addr64_t       cpu_ucr3;
 241         volatile addr64_t       cpu_shadowtask_cr3;
 242         boolean_t               cpu_pagezero_mapped;
 243         cpu_uber_t              cpu_uber;
 244 /* Double-mapped per-CPU exception stack address */
 245         uintptr_t               cd_estack;
 246         int                     cpu_xstate;
 247         int                     cpu_curtask_has_ldt;
 248         int                     cpu_curthread_do_segchk;
 249 /* Address of shadowed, partially mirrored CPU data structures located
 250  * in the double mapped PML4
 251  */
 252         void                    *cd_shadow;
 253         union {
 254                 volatile uint32_t cpu_tlb_invalid_count;
 255                 struct {
 256                         volatile uint16_t cpu_tlb_invalid_local_count;
 257                         volatile uint16_t cpu_tlb_invalid_global_count;
 258                 };
 259         };
 260
 261         uint16_t                cpu_tlb_gen_counts_local[MAX_CPUS];
 262         uint16_t                cpu_tlb_gen_counts_global[MAX_CPUS];
 263
 264         struct processor        *cpu_processor;
 265 #if NCOPY_WINDOWS > 0
 266         struct cpu_pmap         *cpu_pmap;
 267 #endif
 268         struct real_descriptor  *cpu_ldtp;
 269         struct cpu_desc_table   *cpu_desc_tablep;
 270         cpu_desc_index_t        cpu_desc_index;
 271         int                     cpu_ldt;
 272 #if NCOPY_WINDOWS > 0
 273         vm_offset_t             cpu_copywindow_base;
 274         uint64_t                *cpu_copywindow_pdp;
 275
 276         vm_offset_t             cpu_physwindow_base;
 277         uint64_t                *cpu_physwindow_ptep;
 278 #endif
 279
 280 #define HWINTCNT_SIZE 256
 281         uint32_t                cpu_hwIntCnt[HWINTCNT_SIZE];    /* Interrupt counts */
 282         uint64_t                cpu_hwIntpexits[HWINTCNT_SIZE];
 283         uint64_t                cpu_dr7; /* debug control register */
 284         uint64_t                cpu_int_event_time;     /* intr entry/exit time */
 285         pal_rtc_nanotime_t      *cpu_nanotime;          /* Nanotime info */
 286 #if KPC
 287         /* double-buffered performance counter data */
 288         uint64_t                *cpu_kpc_buf[2];
 289         /* PMC shadow and reload value buffers */
 290         uint64_t                *cpu_kpc_shadow;
 291         uint64_t                *cpu_kpc_reload;
 292 #endif
 293 #if MONOTONIC
 294         struct mt_cpu cpu_monotonic;
 295 #endif /* MONOTONIC */
 296         uint32_t                cpu_pmap_pcid_enabled;
 297         pcid_t                  cpu_active_pcid;
 298         pcid_t                  cpu_last_pcid;
 299         pcid_t                  cpu_kernel_pcid;
 300         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp;
 301         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp_kernel;
 302         pcid_cdata_t            *cpu_pcid_data;
 303 #ifdef  PCID_STATS
 304         uint64_t                cpu_pmap_pcid_flushes;
 305         uint64_t                cpu_pmap_pcid_preserves;
 306 #endif
 307         uint64_t                cpu_aperf;
 308         uint64_t                cpu_mperf;
 309         uint64_t                cpu_c3res;
 310         uint64_t                cpu_c6res;
 311         uint64_t                cpu_c7res;
 312         uint64_t                cpu_itime_total;
 313         uint64_t                cpu_rtime_total;
 314         uint64_t                cpu_ixtime;
 315         uint64_t                cpu_idle_exits;
 316         uint64_t                cpu_rtimes[CPU_RTIME_BINS];
 317         uint64_t                cpu_itimes[CPU_ITIME_BINS];
 318 #if !MONOTONIC
 319         uint64_t                cpu_cur_insns;
 320         uint64_t                cpu_cur_ucc;
 321         uint64_t                cpu_cur_urc;
 322 #endif /* !MONOTONIC */
 323         uint64_t                cpu_gpmcs[4];
 324         uint64_t                cpu_max_observed_int_latency;
 325         int                     cpu_max_observed_int_latency_vector;
 326         volatile boolean_t      cpu_NMI_acknowledged;
 327         uint64_t                debugger_entry_time;
 328         uint64_t                debugger_ipi_time;
 329         /* A separate nested interrupt stack flag, to account
 330          * for non-nested interrupts arriving while on the interrupt stack
 331          * Currently only occurs when AICPM enables interrupts on the
 332          * interrupt stack during processor offlining.
 333          */
 334         uint32_t                cpu_nested_istack;
 335         uint32_t                cpu_nested_istack_events;
 336         x86_saved_state64_t     *cpu_fatal_trap_state;
 337         x86_saved_state64_t     *cpu_post_fatal_trap_state;
 338 #if CONFIG_VMX
 339         vmx_cpu_t               cpu_vmx;                /* wonderful world of virtualization */
 340 #endif
 341 #if CONFIG_MCA
 342         struct mca_state        *cpu_mca_state;         /* State at MC fault */
 343 #endif
 344         int                     cpu_type;
 345         int                     cpu_subtype;
 346         int                     cpu_threadtype;
 347         boolean_t               cpu_iflag;
 348         boolean_t               cpu_boot_complete;
 349         int                     cpu_hibernate;
 350 #define MAX_PREEMPTION_RECORDS (8)
 351 #if     DEVELOPMENT || DEBUG
 352         int                     cpu_plri;
 353         plrecord_t              plrecords[MAX_PREEMPTION_RECORDS];
 354 #endif
 355         void                    *cpu_console_buf;
 356         struct x86_lcpu         lcpu;
 357         int                     cpu_phys_number;        /* Physical CPU */
 358         cpu_id_t                cpu_id;                 /* Platform Expert */
 359 #if DEBUG
 360         uint64_t                cpu_entry_cr3;
 361         uint64_t                cpu_exit_cr3;
 362         uint64_t                cpu_pcid_last_cr3;
 363 #endif
 364         boolean_t               cpu_rendezvous_in_progress;
 365 } cpu_data_t;
 366
 367 extern cpu_data_t       *cpu_data_ptr[];
 368
 369 /* Macro to generate inline bodies to retrieve per-cpu data fields. */
 370 #if defined(__clang__)
 371 #define GS_RELATIVE volatile __attribute__((address_space(256)))
 372 #ifndef offsetof
 373 #define offsetof(TYPE, MEMBER) __builtin_offsetof(TYPE,MEMBER)
 374 #endif
 375
 376 #define CPU_DATA_GET(member, type)                                                                               \
 377         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 378                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 379         type ret;                                                                                                                       \
 380         ret = cpu_data->member;                                                                                         \
 381         return ret;
 382
 383 #define CPU_DATA_GET_INDEX(member, index, type)                                                   \
 384         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 385                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 386         type ret;                                                                                                                       \
 387         ret = cpu_data->member[index];                                                                          \
 388         return ret;
 389
 390 #define CPU_DATA_SET(member, value)                                                                              \
 391         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 392                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 393         cpu_data->member = value;
 394
 395 #define CPU_DATA_XCHG(member, value, type)                                                                \
 396         cpu_data_t GS_RELATIVE *cpu_data =                                                      \
 397                 (cpu_data_t GS_RELATIVE *)0UL;                                                                  \
 398         type ret;                                                                                                                       \
 399         ret = cpu_data->member;                                                                                         \
 400         cpu_data->member = value;                                                                                       \
 401         return ret;
 402
 403 #else /* !defined(__clang__) */
 404
 405 #ifndef offsetof
 406 #define offsetof(TYPE, MEMBER) ((size_t) &((TYPE *)0)->MEMBER)
 407 #endif /* offsetof */
 408 #define CPU_DATA_GET(member, type)                                       \
 409         type ret;                                                       \
 410         __asm__ volatile ("mov %%gs:%P1,%0"                             \
 411                 : "=r" (ret)                                            \
 412                 : "i" (offsetof(cpu_data_t,member)));                   \
 413         return ret;
 414
 415 #define CPU_DATA_GET_INDEX(member, index, type)   \
 416         type ret;                                                       \
 417         __asm__ volatile ("mov %%gs:(%1),%0"                            \
 418                 : "=r" (ret)                                            \
 419                 : "r" (offsetof(cpu_data_t,member[index])));                    \
 420         return ret;
 421
 422 #define CPU_DATA_SET(member, value)                                      \
 423         __asm__ volatile ("mov %0,%%gs:%P1"                             \
 424                 :                                                       \
 425                 : "r" (value), "i" (offsetof(cpu_data_t,member)));
 426
 427 #define CPU_DATA_XCHG(member, value, type)                                \
 428         type ret;                                                       \
 429         __asm__ volatile ("xchg %0,%%gs:%P1"                            \
 430                 : "=r" (ret)                                            \
 431                 : "i" (offsetof(cpu_data_t,member)), "0" (value));      \
 432         return ret;
 433
 434 #endif /* !defined(__clang__) */
 435
 436 /*
 437  * Everyone within the osfmk part of the kernel can use the fast
 438  * inline versions of these routines.  Everyone outside, must call
 439  * the real thing,
 440  */
 441
 442
 443 /*
 444  * The "volatile" flavor of current_thread() is intended for use by
 445  * scheduler code which may need to update the thread pointer in the
 446  * course of a context switch.  Any call to current_thread() made
 447  * prior to the thread pointer update should be safe to optimize away
 448  * as it should be consistent with that thread's state to the extent
 449  * the compiler can reason about it.  Likewise, the context switch
 450  * path will eventually result in an arbitrary branch to the new
 451  * thread's pc, about which the compiler won't be able to reason.
 452  * Thus any compile-time optimization of current_thread() calls made
 453  * within the new thread should be safely encapsulated in its
 454  * register/stack state.  The volatile form therefore exists to cover
 455  * the window between the thread pointer update and the branch to
 456  * the new pc.
 457  */
 458 static inline thread_t
 459 get_active_thread_volatile(void)
 460 {
 461         CPU_DATA_GET(cpu_active_thread, thread_t)
 462 }
 463
 464 static inline __attribute__((const)) thread_t
 465 get_active_thread(void)
 466 {
 467         CPU_DATA_GET(cpu_active_thread, thread_t)
 468 }
 469
 470 #define current_thread_fast()           get_active_thread()
 471 #define current_thread_volatile()       get_active_thread_volatile()
 472 #define current_thread()                current_thread_fast()
 473
 474 #define cpu_mode_is64bit()              TRUE
 475
 476 static inline int
 477 get_preemption_level(void)
 478 {
 479         CPU_DATA_GET(cpu_preemption_level, int)
 480 }
 481 static inline int
 482 get_interrupt_level(void)
 483 {
 484         CPU_DATA_GET(cpu_interrupt_level, int)
 485 }
 486 static inline int
 487 get_cpu_number(void)
 488 {
 489         CPU_DATA_GET(cpu_number, int)
 490 }
 491 static inline int
 492 get_cpu_phys_number(void)
 493 {
 494         CPU_DATA_GET(cpu_phys_number, int)
 495 }
 496
 497 static inline cpu_data_t *
 498 current_cpu_datap(void)
 499 {
 500         CPU_DATA_GET(cpu_this, cpu_data_t *);
 501 }
 502
 503 /*
 504  * Facility to diagnose preemption-level imbalances, which are otherwise
 505  * challenging to debug. On each operation that enables or disables preemption,
 506  * we record a backtrace into a per-CPU ring buffer, along with the current
 507  * preemption level and operation type. Thus, if an imbalance is observed,
 508  * one can examine these per-CPU records to determine which codepath failed
 509  * to re-enable preemption, enabled premption without a corresponding
 510  * disablement etc. The backtracer determines which stack is currently active,
 511  * and uses that to perform bounds checks on unterminated stacks.
 512  * To enable, sysctl -w machdep.pltrace=1 on DEVELOPMENT or DEBUG kernels (DRK '15)
 513  * The bounds check currently doesn't account for non-default thread stack sizes.
 514  */
 515 #if DEVELOPMENT || DEBUG
 516 static inline void
 517 rbtrace_bt(uint64_t *rets, int maxframes, cpu_data_t *cdata, uint64_t frameptr, bool use_cursp)
 518 {
 519         extern uint32_t         low_intstack[];         /* bottom */
 520         extern uint32_t         low_eintstack[];        /* top */
 521         extern char             mp_slave_stack[];
 522         int                     btidx = 0;
 523
 524         uint64_t kstackb, kstackt;
 525
 526         /* Obtain the 'current' program counter, initial backtrace
 527          * element. This will also indicate if we were unable to
 528          * trace further up the stack for some reason
 529          */
 530         if (use_cursp) {
 531                 __asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
 532                      : "=m" (rets[btidx++])
 533                      :
 534                      : "rax");
 535         }
 536
 537         thread_t cplthread = cdata->cpu_active_thread;
 538         if (cplthread) {
 539                 uintptr_t csp;
 540                 if (use_cursp == true) {
 541                         __asm__ __volatile__ ("movq %%rsp, %0": "=r" (csp):);
 542                 } else {
 543                         csp = frameptr;
 544                 }
 545                 /* Determine which stack we're on to populate stack bounds.
 546                  * We don't need to trace across stack boundaries for this
 547                  * routine.
 548                  */
 549                 kstackb = cdata->cpu_active_stack;
 550                 kstackt = kstackb + KERNEL_STACK_SIZE;
 551                 if (csp < kstackb || csp > kstackt) {
 552                         kstackt = cdata->cpu_kernel_stack;
 553                         kstackb = kstackt - KERNEL_STACK_SIZE;
 554                         if (csp < kstackb || csp > kstackt) {
 555                                 kstackt = cdata->cpu_int_stack_top;
 556                                 kstackb = kstackt - INTSTACK_SIZE;
 557                                 if (csp < kstackb || csp > kstackt) {
 558                                         kstackt = (uintptr_t)low_eintstack;
 559                                         kstackb = kstackt - INTSTACK_SIZE;
 560                                         if (csp < kstackb || csp > kstackt) {
 561                                                 kstackb = (uintptr_t) mp_slave_stack;
 562                                                 kstackt = kstackb + PAGE_SIZE;
 563                                         } else {
 564                                                 kstackb = 0;
 565                                                 kstackt = 0;
 566                                         }
 567                                 }
 568                         }
 569                 }
 570
 571                 if (__probable(kstackb && kstackt)) {
 572                         uint64_t *cfp = (uint64_t *) frameptr;
 573                         int rbbtf;
 574
 575                         for (rbbtf = btidx; rbbtf < maxframes; rbbtf++) {
 576                                 if (((uint64_t)cfp == 0) || (((uint64_t)cfp < kstackb) || ((uint64_t)cfp > kstackt))) {
 577                                         rets[rbbtf] = 0;
 578                                         continue;
 579                                 }
 580                                 rets[rbbtf] = *(cfp + 1);
 581                                 cfp = (uint64_t *) (*cfp);
 582                         }
 583                 }
 584         }
 585 }
 586
 587 static inline void
 588 pltrace_internal(boolean_t enable)
 589 {
 590         cpu_data_t *cdata = current_cpu_datap();
 591         int cpli = cdata->cpu_preemption_level;
 592         int cplrecord = cdata->cpu_plri;
 593         uint64_t *plbts;
 594
 595         assert(cpli >= 0);
 596
 597         cdata->plrecords[cplrecord].pltype = enable;
 598         cdata->plrecords[cplrecord].plevel = cpli;
 599
 600         plbts = &cdata->plrecords[cplrecord].plbt[0];
 601
 602         cplrecord++;
 603
 604         if (cplrecord >= MAX_PREEMPTION_RECORDS) {
 605                 cplrecord = 0;
 606         }
 607
 608         cdata->cpu_plri = cplrecord;
 609
 610         rbtrace_bt(plbts, MAX_TRACE_BTFRAMES - 1, cdata, (uint64_t)__builtin_frame_address(0), true);
 611 }
 612
 613 extern int plctrace_enabled;
 614
 615 static inline void
 616 iotrace(iotrace_type_e type, uint64_t vaddr, uint64_t paddr, int size, uint64_t val,
 617     uint64_t sabs, uint64_t duration)
 618 {
 619         cpu_data_t *cdata;
 620         int cpu_num, nextidx;
 621         iotrace_entry_t *cur_iotrace_ring;
 622
 623         if (__improbable(mmiotrace_enabled == 0 || iotrace_generators == 0)) {
 624                 return;
 625         }
 626
 627         cdata = current_cpu_datap();
 628         cpu_num = cdata->cpu_number;
 629         nextidx = iotrace_next[cpu_num];
 630         cur_iotrace_ring = iotrace_ring[cpu_num];
 631
 632         cur_iotrace_ring[nextidx].iotype = type;
 633         cur_iotrace_ring[nextidx].vaddr = vaddr;
 634         cur_iotrace_ring[nextidx].paddr = paddr;
 635         cur_iotrace_ring[nextidx].size = size;
 636         cur_iotrace_ring[nextidx].val = val;
 637         cur_iotrace_ring[nextidx].start_time_abs = sabs;
 638         cur_iotrace_ring[nextidx].duration = duration;
 639
 640         iotrace_next[cpu_num] = ((nextidx + 1) >= iotrace_entries_per_cpu) ? 0 : (nextidx + 1);
 641
 642         rbtrace_bt(&cur_iotrace_ring[nextidx].backtrace[0],
 643             MAX_TRACE_BTFRAMES - 1, cdata, (uint64_t)__builtin_frame_address(0), true);
 644 }
 645
 646 static inline uint32_t
 647 traptrace_start(int vecnum, uint64_t ipc, uint64_t sabs, uint64_t frameptr)
 648 {
 649         cpu_data_t *cdata;
 650         int cpu_num, nextidx;
 651         traptrace_entry_t *cur_traptrace_ring;
 652
 653         if (__improbable(traptrace_enabled == 0 || traptrace_generators == 0)) {
 654                 return TRAPTRACE_INVALID_INDEX;
 655         }
 656
 657         assert(ml_get_interrupts_enabled() == FALSE);
 658         cdata = current_cpu_datap();
 659         cpu_num = cdata->cpu_number;
 660         nextidx = traptrace_next[cpu_num];
 661         /* prevent nested interrupts from clobbering this record */
 662         traptrace_next[cpu_num] = ((nextidx + 1) >= traptrace_entries_per_cpu) ? 0 : (nextidx + 1);
 663
 664         cur_traptrace_ring = traptrace_ring[cpu_num];
 665
 666         cur_traptrace_ring[nextidx].vector = vecnum;
 667         cur_traptrace_ring[nextidx].curthread = current_thread();
 668         cur_traptrace_ring[nextidx].interrupted_pc = ipc;
 669         cur_traptrace_ring[nextidx].curpl = cdata->cpu_preemption_level;
 670         cur_traptrace_ring[nextidx].curil = cdata->cpu_interrupt_level;
 671         cur_traptrace_ring[nextidx].start_time_abs = sabs;
 672         cur_traptrace_ring[nextidx].duration = ~0ULL;
 673
 674         rbtrace_bt(&cur_traptrace_ring[nextidx].backtrace[0],
 675             MAX_TRACE_BTFRAMES - 1, cdata, frameptr, false);
 676
 677         assert(nextidx <= 0xFFFF);
 678
 679         return ((unsigned)cpu_num << 16) | nextidx;
 680 }
 681
 682 static inline void
 683 traptrace_end(uint32_t index, uint64_t eabs)
 684 {
 685         if (index != TRAPTRACE_INVALID_INDEX) {
 686                 traptrace_entry_t *ttentp = &traptrace_ring[index >> 16][index & 0xFFFF];
 687
 688                 ttentp->duration = eabs - ttentp->start_time_abs;
 689         }
 690 }
 691
 692 #endif /* DEVELOPMENT || DEBUG */
 693
 694 static inline void
 695 pltrace(boolean_t plenable)
 696 {
 697 #if DEVELOPMENT || DEBUG
 698         if (__improbable(plctrace_enabled != 0)) {
 699                 pltrace_internal(plenable);
 700         }
 701 #else
 702         (void)plenable;
 703 #endif
 704 }
 705
 706 static inline void
 707 disable_preemption_internal(void)
 708 {
 709         assert(get_preemption_level() >= 0);
 710
 711         os_compiler_barrier(release);
 712 #if defined(__clang__)
 713         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 714         cpu_data->cpu_preemption_level++;
 715 #else
 716         __asm__ volatile ("incl %%gs:%P0"
 717             :
 718             : "i" (offsetof(cpu_data_t, cpu_preemption_level)));
 719 #endif
 720         os_compiler_barrier(acquire);
 721         pltrace(FALSE);
 722 }
 723
 724 static inline void
 725 enable_preemption_internal(void)
 726 {
 727         assert(get_preemption_level() > 0);
 728         pltrace(TRUE);
 729         os_compiler_barrier(release);
 730 #if defined(__clang__)
 731         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 732         if (0 == --cpu_data->cpu_preemption_level) {
 733                 kernel_preempt_check();
 734         }
 735 #else
 736         __asm__ volatile ("decl %%gs:%P0                \n\t"
 737                           "jne 1f                       \n\t"
 738                           "call _kernel_preempt_check   \n\t"
 739                           "1:"
 740                         : /* no outputs */
 741                         : "i" (offsetof(cpu_data_t, cpu_preemption_level))
 742                         : "eax", "ecx", "edx", "cc", "memory");
 743 #endif
 744         os_compiler_barrier(acquire);
 745 }
 746
 747 static inline void
 748 enable_preemption_no_check(void)
 749 {
 750         assert(get_preemption_level() > 0);
 751
 752         pltrace(TRUE);
 753         os_compiler_barrier(release);
 754 #if defined(__clang__)
 755         cpu_data_t GS_RELATIVE *cpu_data = (cpu_data_t GS_RELATIVE *)0UL;
 756         cpu_data->cpu_preemption_level--;
 757 #else
 758         __asm__ volatile ("decl %%gs:%P0"
 759                         : /* no outputs */
 760                         : "i" (offsetof(cpu_data_t, cpu_preemption_level))
 761                         : "cc", "memory");
 762 #endif
 763         os_compiler_barrier(acquire);
 764 }
 765
 766 static inline void
 767 _enable_preemption_no_check(void)
 768 {
 769         enable_preemption_no_check();
 770 }
 771
 772 static inline void
 773 mp_disable_preemption(void)
 774 {
 775         disable_preemption_internal();
 776 }
 777
 778 static inline void
 779 _mp_disable_preemption(void)
 780 {
 781         disable_preemption_internal();
 782 }
 783
 784 static inline void
 785 mp_enable_preemption(void)
 786 {
 787         enable_preemption_internal();
 788 }
 789
 790 static inline void
 791 _mp_enable_preemption(void)
 792 {
 793         enable_preemption_internal();
 794 }
 795
 796 static inline void
 797 mp_enable_preemption_no_check(void)
 798 {
 799         enable_preemption_no_check();
 800 }
 801
 802 static inline void
 803 _mp_enable_preemption_no_check(void)
 804 {
 805         enable_preemption_no_check();
 806 }
 807
 808 #ifdef XNU_KERNEL_PRIVATE
 809 #define disable_preemption() disable_preemption_internal()
 810 #define enable_preemption() enable_preemption_internal()
 811 #define MACHINE_PREEMPTION_MACROS (1)
 812 #endif
 813
 814 static inline cpu_data_t *
 815 cpu_datap(int cpu)
 816 {
 817         return cpu_data_ptr[cpu];
 818 }
 819
 820 static inline int
 821 cpu_is_running(int cpu)
 822 {
 823         return (cpu_datap(cpu) != NULL) && (cpu_datap(cpu)->cpu_running);
 824 }
 825
 826 #ifdef MACH_KERNEL_PRIVATE
 827 static inline cpu_data_t *
 828 cpu_shadowp(int cpu)
 829 {
 830         return cpu_data_ptr[cpu]->cd_shadow;
 831 }
 832
 833 #endif
 834 extern cpu_data_t *cpu_data_alloc(boolean_t is_boot_cpu);
 835 extern void cpu_data_realloc(void);
 836
 837 #endif  /* I386_CPU_DATA */