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