osfmk/i386/cpu_data.h

   1 /*
   2  * Copyright (c) 2000-2020 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         vm_offset_t             cpu_pcpu_base;
 203         thread_t                cpu_active_thread;
 204         thread_t                cpu_nthread;
 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         volatile int            cpu_signals;            /* IPI events */
 211         volatile int            cpu_prior_signals;      /* Last set of events,
 212                                                          * debugging
 213                                                          */
 214         ast_t                   cpu_pending_ast;
 215         /*
 216          * Note if rearranging fields:
 217          * We want cpu_preemption_level on a different
 218          * cache line than cpu_active_thread
 219          * for optimizing mtx_spin phase.
 220          */
 221         int                     cpu_interrupt_level;
 222         volatile int            cpu_preemption_level;
 223         volatile int            cpu_running;
 224 #if !MONOTONIC
 225         boolean_t               cpu_fixed_pmcs_enabled;
 226 #endif /* !MONOTONIC */
 227         rtclock_timer_t         rtclock_timer;
 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         struct real_descriptor  *cpu_ldtp;
 266         struct cpu_desc_table   *cpu_desc_tablep;
 267         cpu_desc_index_t        cpu_desc_index;
 268         int                     cpu_ldt;
 269
 270 #define HWINTCNT_SIZE 256
 271         uint32_t                cpu_hwIntCnt[HWINTCNT_SIZE];    /* Interrupt counts */
 272         uint64_t                cpu_hwIntpexits[HWINTCNT_SIZE];
 273         uint64_t                cpu_dr7; /* debug control register */
 274         uint64_t                cpu_int_event_time;     /* intr entry/exit time */
 275         pal_rtc_nanotime_t      *cpu_nanotime;          /* Nanotime info */
 276 #if KPC
 277         /* double-buffered performance counter data */
 278         uint64_t                *cpu_kpc_buf[2];
 279         /* PMC shadow and reload value buffers */
 280         uint64_t                *cpu_kpc_shadow;
 281         uint64_t                *cpu_kpc_reload;
 282 #endif
 283 #if MONOTONIC
 284         struct mt_cpu cpu_monotonic;
 285 #endif /* MONOTONIC */
 286         uint32_t                cpu_pmap_pcid_enabled;
 287         pcid_t                  cpu_active_pcid;
 288         pcid_t                  cpu_last_pcid;
 289         pcid_t                  cpu_kernel_pcid;
 290         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp;
 291         volatile pcid_ref_t     *cpu_pmap_pcid_coherentp_kernel;
 292         pcid_cdata_t            *cpu_pcid_data;
 293 #ifdef  PCID_STATS
 294         uint64_t                cpu_pmap_pcid_flushes;
 295         uint64_t                cpu_pmap_pcid_preserves;
 296 #endif
 297         uint64_t                cpu_aperf;
 298         uint64_t                cpu_mperf;
 299         uint64_t                cpu_c3res;
 300         uint64_t                cpu_c6res;
 301         uint64_t                cpu_c7res;
 302         uint64_t                cpu_itime_total;
 303         uint64_t                cpu_rtime_total;
 304         uint64_t                cpu_ixtime;
 305         uint64_t                cpu_idle_exits;
 306         /*
 307          * Note that the cacheline-copy mechanism uses the cpu_rtimes field in the shadow CPU
 308          * structures to temporarily stash the code cacheline that includes the instruction
 309          * pointer at the time of the fault (this field is otherwise unused in the shadow
 310          * CPU structures).
 311          */
 312         uint64_t                cpu_rtimes[CPU_RTIME_BINS];
 313         uint64_t                cpu_itimes[CPU_ITIME_BINS];
 314 #if !MONOTONIC
 315         uint64_t                cpu_cur_insns;
 316         uint64_t                cpu_cur_ucc;
 317         uint64_t                cpu_cur_urc;
 318 #endif /* !MONOTONIC */
 319         uint64_t                cpu_gpmcs[4];
 320         uint64_t                cpu_max_observed_int_latency;
 321         int                     cpu_max_observed_int_latency_vector;
 322         volatile boolean_t      cpu_NMI_acknowledged;
 323         uint64_t                debugger_entry_time;
 324         uint64_t                debugger_ipi_time;
 325         /* A separate nested interrupt stack flag, to account
 326          * for non-nested interrupts arriving while on the interrupt stack
 327          * Currently only occurs when AICPM enables interrupts on the
 328          * interrupt stack during processor offlining.
 329          */
 330         uint32_t                cpu_nested_istack;
 331         uint32_t                cpu_nested_istack_events;
 332         x86_saved_state64_t     *cpu_fatal_trap_state;
 333         x86_saved_state64_t     *cpu_post_fatal_trap_state;
 334 #if CONFIG_VMX
 335         vmx_cpu_t               cpu_vmx;                /* wonderful world of virtualization */
 336 #endif
 337 #if CONFIG_MCA
 338         struct mca_state        *cpu_mca_state;         /* State at MC fault */
 339 #endif
 340         int                     cpu_type;
 341         int                     cpu_subtype;
 342         int                     cpu_threadtype;
 343         boolean_t               cpu_iflag;
 344         boolean_t               cpu_boot_complete;
 345         int                     cpu_hibernate;
 346 #define MAX_PREEMPTION_RECORDS (8)
 347 #if     DEVELOPMENT || DEBUG
 348         int                     cpu_plri;
 349         plrecord_t              plrecords[MAX_PREEMPTION_RECORDS];
 350 #endif
 351         void                    *cpu_console_buf;
 352         struct x86_lcpu         lcpu;
 353         int                     cpu_phys_number;        /* Physical CPU */
 354         cpu_id_t                cpu_id;                 /* Platform Expert */
 355 #if DEBUG
 356         uint64_t                cpu_entry_cr3;
 357         uint64_t                cpu_exit_cr3;
 358         uint64_t                cpu_pcid_last_cr3;
 359 #endif
 360         boolean_t               cpu_rendezvous_in_progress;
 361 #if CST_DEMOTION_DEBUG
 362         /* Count of thread wakeups issued by this processor */
 363         uint64_t                cpu_wakeups_issued_total;
 364 #endif
 365 #if DEBUG || DEVELOPMENT
 366         uint64_t                tsc_sync_delta;
 367 #endif
 368 } cpu_data_t;
 369
 370 extern cpu_data_t       *cpu_data_ptr[];
 371
 372 /*
 373  * __SEG_GS marks %gs-relative operations:
 374  *   https://clang.llvm.org/docs/LanguageExtensions.html#memory-references-to-specified-segments
 375  *   https://gcc.gnu.org/onlinedocs/gcc/Named-Address-Spaces.html#x86-Named-Address-Spaces
 376  */
 377 #if defined(__SEG_GS)
 378 // __seg_gs exists
 379 #elif defined(__clang__)
 380 #define __seg_gs __attribute__((address_space(256)))
 381 #else
 382 #error use a compiler that supports address spaces or __seg_gs
 383 #endif
 384
 385 #define CPU_DATA()            ((cpu_data_t __seg_gs *)0UL)
 386
 387 /*
 388  * Everyone within the osfmk part of the kernel can use the fast
 389  * inline versions of these routines.  Everyone outside, must call
 390  * the real thing,
 391  */
 392
 393
 394 /*
 395  * The "volatile" flavor of current_thread() is intended for use by
 396  * scheduler code which may need to update the thread pointer in the
 397  * course of a context switch.  Any call to current_thread() made
 398  * prior to the thread pointer update should be safe to optimize away
 399  * as it should be consistent with that thread's state to the extent
 400  * the compiler can reason about it.  Likewise, the context switch
 401  * path will eventually result in an arbitrary branch to the new
 402  * thread's pc, about which the compiler won't be able to reason.
 403  * Thus any compile-time optimization of current_thread() calls made
 404  * within the new thread should be safely encapsulated in its
 405  * register/stack state.  The volatile form therefore exists to cover
 406  * the window between the thread pointer update and the branch to
 407  * the new pc.
 408  */
 409 static inline thread_t
 410 get_active_thread_volatile(void)
 411 {
 412         return CPU_DATA()->cpu_active_thread;
 413 }
 414
 415 static inline __attribute__((const)) thread_t
 416 get_active_thread(void)
 417 {
 418         return CPU_DATA()->cpu_active_thread;
 419 }
 420
 421 #define current_thread_fast()           get_active_thread()
 422 #define current_thread_volatile()       get_active_thread_volatile()
 423 #define current_thread()                current_thread_fast()
 424
 425 #define cpu_mode_is64bit()              TRUE
 426
 427 static inline int
 428 get_preemption_level(void)
 429 {
 430         return CPU_DATA()->cpu_preemption_level;
 431 }
 432 static inline int
 433 get_interrupt_level(void)
 434 {
 435         return CPU_DATA()->cpu_interrupt_level;
 436 }
 437 static inline int
 438 get_cpu_number(void)
 439 {
 440         return CPU_DATA()->cpu_number;
 441 }
 442 static inline vm_offset_t
 443 get_current_percpu_base(void)
 444 {
 445         return CPU_DATA()->cpu_pcpu_base;
 446 }
 447 static inline int
 448 get_cpu_phys_number(void)
 449 {
 450         return CPU_DATA()->cpu_phys_number;
 451 }
 452
 453 static inline cpu_data_t *
 454 current_cpu_datap(void)
 455 {
 456         return CPU_DATA()->cpu_this;
 457 }
 458
 459 /*
 460  * Facility to diagnose preemption-level imbalances, which are otherwise
 461  * challenging to debug. On each operation that enables or disables preemption,
 462  * we record a backtrace into a per-CPU ring buffer, along with the current
 463  * preemption level and operation type. Thus, if an imbalance is observed,
 464  * one can examine these per-CPU records to determine which codepath failed
 465  * to re-enable preemption, enabled premption without a corresponding
 466  * disablement etc. The backtracer determines which stack is currently active,
 467  * and uses that to perform bounds checks on unterminated stacks.
 468  * To enable, sysctl -w machdep.pltrace=1 on DEVELOPMENT or DEBUG kernels (DRK '15)
 469  * The bounds check currently doesn't account for non-default thread stack sizes.
 470  */
 471 #if DEVELOPMENT || DEBUG
 472 static inline void
 473 rbtrace_bt(uint64_t *rets, int maxframes, cpu_data_t *cdata, uint64_t frameptr, bool use_cursp)
 474 {
 475         extern uint32_t         low_intstack[];         /* bottom */
 476         extern uint32_t         low_eintstack[];        /* top */
 477         extern char             mp_slave_stack[];
 478         int                     btidx = 0;
 479
 480         uint64_t kstackb, kstackt;
 481
 482         /* Obtain the 'current' program counter, initial backtrace
 483          * element. This will also indicate if we were unable to
 484          * trace further up the stack for some reason
 485          */
 486         if (use_cursp) {
 487                 __asm__ volatile ("leaq 1f(%%rip), %%rax; mov %%rax, %0\n1:"
 488                      : "=m" (rets[btidx++])
 489                      :
 490                      : "rax");
 491         }
 492
 493         thread_t cplthread = cdata->cpu_active_thread;
 494         if (cplthread) {
 495                 uintptr_t csp;
 496                 if (use_cursp == true) {
 497                         __asm__ __volatile__ ("movq %%rsp, %0": "=r" (csp):);
 498                 } else {
 499                         csp = frameptr;
 500                 }
 501                 /* Determine which stack we're on to populate stack bounds.
 502                  * We don't need to trace across stack boundaries for this
 503                  * routine.
 504                  */
 505                 kstackb = cdata->cpu_active_stack;
 506                 kstackt = kstackb + KERNEL_STACK_SIZE;
 507                 if (csp < kstackb || csp > kstackt) {
 508                         kstackt = cdata->cpu_kernel_stack;
 509                         kstackb = kstackt - KERNEL_STACK_SIZE;
 510                         if (csp < kstackb || csp > kstackt) {
 511                                 kstackt = cdata->cpu_int_stack_top;
 512                                 kstackb = kstackt - INTSTACK_SIZE;
 513                                 if (csp < kstackb || csp > kstackt) {
 514                                         kstackt = (uintptr_t)low_eintstack;
 515                                         kstackb = kstackt - INTSTACK_SIZE;
 516                                         if (csp < kstackb || csp > kstackt) {
 517                                                 kstackb = (uintptr_t) mp_slave_stack;
 518                                                 kstackt = kstackb + PAGE_SIZE;
 519                                         } else {
 520                                                 kstackb = 0;
 521                                                 kstackt = 0;
 522                                         }
 523                                 }
 524                         }
 525                 }
 526
 527                 if (__probable(kstackb && kstackt)) {
 528                         uint64_t *cfp = (uint64_t *) frameptr;
 529                         int rbbtf;
 530
 531                         for (rbbtf = btidx; rbbtf < maxframes; rbbtf++) {
 532                                 if (((uint64_t)cfp == 0) || (((uint64_t)cfp < kstackb) || ((uint64_t)cfp > kstackt))) {
 533                                         rets[rbbtf] = 0;
 534                                         continue;
 535                                 }
 536                                 rets[rbbtf] = *(cfp + 1);
 537                                 cfp = (uint64_t *) (*cfp);
 538                         }
 539                 }
 540         }
 541 }
 542
 543 __attribute__((noinline))
 544 static inline void
 545 pltrace_internal(boolean_t enable)
 546 {
 547         cpu_data_t *cdata = current_cpu_datap();
 548         int cpli = cdata->cpu_preemption_level;
 549         int cplrecord = cdata->cpu_plri;
 550         uint64_t *plbts;
 551
 552         assert(cpli >= 0);
 553
 554         cdata->plrecords[cplrecord].pltype = enable;
 555         cdata->plrecords[cplrecord].plevel = cpli;
 556
 557         plbts = &cdata->plrecords[cplrecord].plbt[0];
 558
 559         cplrecord++;
 560
 561         if (cplrecord >= MAX_PREEMPTION_RECORDS) {
 562                 cplrecord = 0;
 563         }
 564
 565         cdata->cpu_plri = cplrecord;
 566
 567         rbtrace_bt(plbts, MAX_TRACE_BTFRAMES - 1, cdata, (uint64_t)__builtin_frame_address(0), false);
 568 }
 569
 570 extern int plctrace_enabled;
 571
 572 static inline void
 573 iotrace(iotrace_type_e type, uint64_t vaddr, uint64_t paddr, int size, uint64_t val,
 574     uint64_t sabs, uint64_t duration)
 575 {
 576         cpu_data_t *cdata;
 577         int cpu_num, nextidx;
 578         iotrace_entry_t *cur_iotrace_ring;
 579
 580         if (__improbable(mmiotrace_enabled == 0 || iotrace_generators == 0)) {
 581                 return;
 582         }
 583
 584         cdata = current_cpu_datap();
 585         cpu_num = cdata->cpu_number;
 586         nextidx = iotrace_next[cpu_num];
 587         cur_iotrace_ring = iotrace_ring[cpu_num];
 588
 589         cur_iotrace_ring[nextidx].iotype = type;
 590         cur_iotrace_ring[nextidx].vaddr = vaddr;
 591         cur_iotrace_ring[nextidx].paddr = paddr;
 592         cur_iotrace_ring[nextidx].size = size;
 593         cur_iotrace_ring[nextidx].val = val;
 594         cur_iotrace_ring[nextidx].start_time_abs = sabs;
 595         cur_iotrace_ring[nextidx].duration = duration;
 596
 597         iotrace_next[cpu_num] = ((nextidx + 1) >= iotrace_entries_per_cpu) ? 0 : (nextidx + 1);
 598
 599         rbtrace_bt(&cur_iotrace_ring[nextidx].backtrace[0],
 600             MAX_TRACE_BTFRAMES - 1, cdata, (uint64_t)__builtin_frame_address(0), true);
 601 }
 602
 603 static inline uint32_t
 604 traptrace_start(int vecnum, uint64_t ipc, uint64_t sabs, uint64_t frameptr)
 605 {
 606         cpu_data_t *cdata;
 607         unsigned int cpu_num, nextidx;
 608         traptrace_entry_t *cur_traptrace_ring;
 609
 610         if (__improbable(traptrace_enabled == 0 || traptrace_generators == 0)) {
 611                 return TRAPTRACE_INVALID_INDEX;
 612         }
 613
 614         assert(ml_get_interrupts_enabled() == FALSE);
 615         cdata = current_cpu_datap();
 616         cpu_num = (unsigned int)cdata->cpu_number;
 617         nextidx = (unsigned int)traptrace_next[cpu_num];
 618         /* prevent nested interrupts from clobbering this record */
 619         traptrace_next[cpu_num] = (int)(((nextidx + 1) >= (unsigned int)traptrace_entries_per_cpu) ? 0 : (nextidx + 1));
 620
 621         cur_traptrace_ring = traptrace_ring[cpu_num];
 622
 623         cur_traptrace_ring[nextidx].vector = vecnum;
 624         cur_traptrace_ring[nextidx].curthread = current_thread();
 625         cur_traptrace_ring[nextidx].interrupted_pc = ipc;
 626         cur_traptrace_ring[nextidx].curpl = cdata->cpu_preemption_level;
 627         cur_traptrace_ring[nextidx].curil = cdata->cpu_interrupt_level;
 628         cur_traptrace_ring[nextidx].start_time_abs = sabs;
 629         cur_traptrace_ring[nextidx].duration = ~0ULL;
 630
 631         rbtrace_bt(&cur_traptrace_ring[nextidx].backtrace[0],
 632             MAX_TRACE_BTFRAMES - 1, cdata, frameptr, false);
 633
 634         assert(nextidx <= 0xFFFF);
 635
 636         return (uint32_t)((cpu_num << 16) | nextidx);
 637 }
 638
 639 static inline void
 640 traptrace_end(uint32_t index, uint64_t eabs)
 641 {
 642         if (index != TRAPTRACE_INVALID_INDEX) {
 643                 traptrace_entry_t *ttentp = &traptrace_ring[index >> 16][index & 0xFFFF];
 644
 645                 ttentp->duration = eabs - ttentp->start_time_abs;
 646         }
 647 }
 648
 649 #endif /* DEVELOPMENT || DEBUG */
 650
 651 __header_always_inline void
 652 pltrace(boolean_t plenable)
 653 {
 654 #if DEVELOPMENT || DEBUG
 655         if (__improbable(plctrace_enabled != 0)) {
 656                 pltrace_internal(plenable);
 657         }
 658 #else
 659         (void)plenable;
 660 #endif
 661 }
 662
 663 static inline void
 664 disable_preemption_internal(void)
 665 {
 666         assert(get_preemption_level() >= 0);
 667
 668         os_compiler_barrier();
 669         CPU_DATA()->cpu_preemption_level++;
 670         os_compiler_barrier();
 671         pltrace(FALSE);
 672 }
 673
 674 static inline void
 675 enable_preemption_internal(void)
 676 {
 677         assert(get_preemption_level() > 0);
 678         pltrace(TRUE);
 679         os_compiler_barrier();
 680         if (0 == --CPU_DATA()->cpu_preemption_level) {
 681                 kernel_preempt_check();
 682         }
 683         os_compiler_barrier();
 684 }
 685
 686 static inline void
 687 enable_preemption_no_check(void)
 688 {
 689         assert(get_preemption_level() > 0);
 690
 691         pltrace(TRUE);
 692         os_compiler_barrier();
 693         CPU_DATA()->cpu_preemption_level--;
 694         os_compiler_barrier();
 695 }
 696
 697 static inline void
 698 _enable_preemption_no_check(void)
 699 {
 700         enable_preemption_no_check();
 701 }
 702
 703 static inline void
 704 mp_disable_preemption(void)
 705 {
 706         disable_preemption_internal();
 707 }
 708
 709 static inline void
 710 _mp_disable_preemption(void)
 711 {
 712         disable_preemption_internal();
 713 }
 714
 715 static inline void
 716 mp_enable_preemption(void)
 717 {
 718         enable_preemption_internal();
 719 }
 720
 721 static inline void
 722 _mp_enable_preemption(void)
 723 {
 724         enable_preemption_internal();
 725 }
 726
 727 static inline void
 728 mp_enable_preemption_no_check(void)
 729 {
 730         enable_preemption_no_check();
 731 }
 732
 733 static inline void
 734 _mp_enable_preemption_no_check(void)
 735 {
 736         enable_preemption_no_check();
 737 }
 738
 739 #ifdef XNU_KERNEL_PRIVATE
 740 #define disable_preemption() disable_preemption_internal()
 741 #define enable_preemption() enable_preemption_internal()
 742 #define MACHINE_PREEMPTION_MACROS (1)
 743 #endif
 744
 745 static inline cpu_data_t *
 746 cpu_datap(int cpu)
 747 {
 748         return cpu_data_ptr[cpu];
 749 }
 750
 751 static inline int
 752 cpu_is_running(int cpu)
 753 {
 754         return (cpu_datap(cpu) != NULL) && (cpu_datap(cpu)->cpu_running);
 755 }
 756
 757 #ifdef MACH_KERNEL_PRIVATE
 758 static inline cpu_data_t *
 759 cpu_shadowp(int cpu)
 760 {
 761         return cpu_data_ptr[cpu]->cd_shadow;
 762 }
 763
 764 #endif
 765 extern cpu_data_t *cpu_data_alloc(boolean_t is_boot_cpu);
 766 extern void cpu_data_realloc(void);
 767
 768 #endif  /* I386_CPU_DATA */