osfmk/kern/startup.c

   1 /*
   2  * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * @OSF_COPYRIGHT@
  30  */
  31 /*
  32  * Mach Operating System
  33  * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
  34  * All Rights Reserved.
  35  *
  36  * Permission to use, copy, modify and distribute this software and its
  37  * documentation is hereby granted, provided that both the copyright
  38  * notice and this permission notice appear in all copies of the
  39  * software, derivative works or modified versions, and any portions
  40  * thereof, and that both notices appear in supporting documentation.
  41  *
  42  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
  43  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
  44  * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
  45  *
  46  * Carnegie Mellon requests users of this software to return to
  47  *
  48  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
  49  *  School of Computer Science
  50  *  Carnegie Mellon University
  51  *  Pittsburgh PA 15213-3890
  52  *
  53  * any improvements or extensions that they make and grant Carnegie Mellon
  54  * the rights to redistribute these changes.
  55  */
  56 /*
  57  * NOTICE: This file was modified by McAfee Research in 2004 to introduce
  58  * support for mandatory and extensible security protections.  This notice
  59  * is included in support of clause 2.2 (b) of the Apple Public License,
  60  * Version 2.0.
  61  */
  62 /*
  63  */
  64
  65 /*
  66  *      Mach kernel startup.
  67  */
  68
  69 #include <debug.h>
  70 #include <xpr_debug.h>
  71 #include <mach_kdp.h>
  72
  73 #include <mach/boolean.h>
  74 #include <mach/machine.h>
  75 #include <mach/thread_act.h>
  76 #include <mach/task_special_ports.h>
  77 #include <mach/vm_param.h>
  78 #include <ipc/ipc_init.h>
  79 #include <kern/assert.h>
  80 #include <kern/mach_param.h>
  81 #include <kern/misc_protos.h>
  82 #include <kern/clock.h>
  83 #include <kern/cpu_number.h>
  84 #include <kern/ledger.h>
  85 #include <kern/machine.h>
  86 #include <kern/processor.h>
  87 #include <kern/sched_prim.h>
  88 #include <kern/startup.h>
  89 #include <kern/task.h>
  90 #include <kern/thread.h>
  91 #include <kern/timer.h>
  92 #if CONFIG_TELEMETRY
  93 #include <kern/telemetry.h>
  94 #endif
  95 #include <kern/wait_queue.h>
  96 #include <kern/xpr.h>
  97 #include <kern/zalloc.h>
  98 #include <kern/locks.h>
  99 #include <console/serial_protos.h>
 100 #include <vm/vm_kern.h>
 101 #include <vm/vm_init.h>
 102 #include <vm/vm_map.h>
 103 #include <vm/vm_object.h>
 104 #include <vm/vm_page.h>
 105 #include <vm/vm_pageout.h>
 106 #include <vm/vm_shared_region.h>
 107 #include <machine/pmap.h>
 108 #include <machine/commpage.h>
 109 #include <libkern/version.h>
 110 #include <sys/codesign.h>
 111 #include <sys/kdebug.h>
 112
 113 #if MACH_KDP
 114 #include <kdp/kdp.h>
 115 #endif
 116
 117 #if CONFIG_MACF
 118 #include <security/mac_mach_internal.h>
 119 #endif
 120
 121 #if CONFIG_COUNTERS
 122 #include <pmc/pmc.h>
 123 #endif
 124
 125 #if KPC
 126 #include <kern/kpc.h>
 127 #endif
 128
 129 #if KPERF
 130 #include <kperf/kperf.h>
 131 #endif
 132
 133
 134 #include <i386/pmCPU.h>
 135 static void             kernel_bootstrap_thread(void);
 136
 137 static void             load_context(
 138                                         thread_t        thread);
 139 #if (defined(__i386__) || defined(__x86_64__)) && NCOPY_WINDOWS > 0
 140 extern void cpu_userwindow_init(int);
 141 extern void cpu_physwindow_init(int);
 142 #endif
 143
 144 // libkern/OSKextLib.cpp
 145 extern void     OSKextRemoveKextBootstrap(void);
 146
 147 void scale_setup(void);
 148 extern void bsd_scale_setup(int);
 149 extern unsigned int semaphore_max;
 150 extern void stackshot_lock_init(void);
 151
 152 /*
 153  *      Running in virtual memory, on the interrupt stack.
 154  */
 155
 156 extern int serverperfmode;
 157
 158 /* size of kernel trace buffer, disabled by default */
 159 unsigned int new_nkdbufs = 0;
 160 unsigned int wake_nkdbufs = 0;
 161
 162 /* mach leak logging */
 163 int log_leaks = 0;
 164 int turn_on_log_leaks = 0;
 165
 166
 167 void
 168 kernel_early_bootstrap(void)
 169 {
 170         /* serverperfmode is needed by timer setup */
 171         if (PE_parse_boot_argn("serverperfmode", &serverperfmode, sizeof (serverperfmode))) {
 172                 serverperfmode = 1;
 173         }
 174
 175         lck_mod_init();
 176
 177         /*
 178          * Initialize the timer callout world
 179          */
 180         timer_call_init();
 181 }
 182
 183
 184 void
 185 kernel_bootstrap(void)
 186 {
 187         kern_return_t   result;
 188         thread_t        thread;
 189         char            namep[16];
 190
 191         printf("%s\n", version); /* log kernel version */
 192
 193 #define kernel_bootstrap_kprintf(x...) /* kprintf("kernel_bootstrap: " x) */
 194
 195         if (PE_parse_boot_argn("-l", namep, sizeof (namep))) /* leaks logging */
 196                 turn_on_log_leaks = 1;
 197
 198         PE_parse_boot_argn("trace", &new_nkdbufs, sizeof (new_nkdbufs));
 199
 200         PE_parse_boot_argn("trace_wake", &wake_nkdbufs, sizeof (wake_nkdbufs));
 201
 202         scale_setup();
 203
 204         kernel_bootstrap_kprintf("calling vm_mem_bootstrap\n");
 205         vm_mem_bootstrap();
 206
 207         kernel_bootstrap_kprintf("calling cs_init\n");
 208         cs_init();
 209
 210         kernel_bootstrap_kprintf("calling vm_mem_init\n");
 211         vm_mem_init();
 212
 213         machine_info.memory_size = (uint32_t)mem_size;
 214         machine_info.max_mem = max_mem;
 215         machine_info.major_version = version_major;
 216         machine_info.minor_version = version_minor;
 217
 218 #if CONFIG_TELEMETRY
 219         kernel_bootstrap_kprintf("calling telemetry_init\n");
 220         telemetry_init();
 221 #endif
 222
 223         kernel_bootstrap_kprintf("calling stackshot_lock_init\n");
 224         stackshot_lock_init();
 225
 226         kernel_bootstrap_kprintf("calling sched_init\n");
 227         sched_init();
 228
 229         kernel_bootstrap_kprintf("calling wait_queue_bootstrap\n");
 230         wait_queue_bootstrap();
 231
 232         kernel_bootstrap_kprintf("calling ipc_bootstrap\n");
 233         ipc_bootstrap();
 234
 235 #if CONFIG_MACF
 236         mac_policy_init();
 237 #endif
 238         kernel_bootstrap_kprintf("calling ipc_init\n");
 239         ipc_init();
 240
 241         /*
 242          * As soon as the virtual memory system is up, we record
 243          * that this CPU is using the kernel pmap.
 244          */
 245         kernel_bootstrap_kprintf("calling PMAP_ACTIVATE_KERNEL\n");
 246         PMAP_ACTIVATE_KERNEL(master_cpu);
 247
 248         kernel_bootstrap_kprintf("calling mapping_free_prime\n");
 249         mapping_free_prime();                                           /* Load up with temporary mapping blocks */
 250
 251         kernel_bootstrap_kprintf("calling machine_init\n");
 252         machine_init();
 253
 254         kernel_bootstrap_kprintf("calling clock_init\n");
 255         clock_init();
 256
 257         ledger_init();
 258
 259         /*
 260          *      Initialize the IPC, task, and thread subsystems.
 261          */
 262         kernel_bootstrap_kprintf("calling task_init\n");
 263         task_init();
 264
 265         kernel_bootstrap_kprintf("calling thread_init\n");
 266         thread_init();
 267
 268         /*
 269          *      Create a kernel thread to execute the kernel bootstrap.
 270          */
 271         kernel_bootstrap_kprintf("calling kernel_thread_create\n");
 272         result = kernel_thread_create((thread_continue_t)kernel_bootstrap_thread, NULL, MAXPRI_KERNEL, &thread);
 273
 274         if (result != KERN_SUCCESS) panic("kernel_bootstrap: result = %08X\n", result);
 275
 276         thread->state = TH_RUN;
 277         thread_deallocate(thread);
 278
 279         kernel_bootstrap_kprintf("calling load_context - done\n");
 280         load_context(thread);
 281         /*NOTREACHED*/
 282 }
 283
 284 int kth_started = 0;
 285
 286 vm_offset_t vm_kernel_addrperm;
 287 vm_offset_t buf_kernel_addrperm;
 288
 289 /*
 290  * Now running in a thread.  Kick off other services,
 291  * invoke user bootstrap, enter pageout loop.
 292  */
 293 static void
 294 kernel_bootstrap_thread(void)
 295 {
 296         processor_t             processor = current_processor();
 297
 298 #define kernel_bootstrap_thread_kprintf(x...) /* kprintf("kernel_bootstrap_thread: " x) */
 299         kernel_bootstrap_thread_kprintf("calling idle_thread_create\n");
 300         /*
 301          * Create the idle processor thread.
 302          */
 303         idle_thread_create(processor);
 304
 305         /*
 306          * N.B. Do not stick anything else
 307          * before this point.
 308          *
 309          * Start up the scheduler services.
 310          */
 311         kernel_bootstrap_thread_kprintf("calling sched_startup\n");
 312         sched_startup();
 313
 314         /*
 315          * Thread lifecycle maintenance (teardown, stack allocation)
 316          */
 317         kernel_bootstrap_thread_kprintf("calling thread_daemon_init\n");
 318         thread_daemon_init();
 319
 320         /* Create kernel map entry reserve */
 321         vm_kernel_reserved_entry_init();
 322
 323         /*
 324          * Thread callout service.
 325          */
 326         kernel_bootstrap_thread_kprintf("calling thread_call_initialize\n");
 327         thread_call_initialize();
 328
 329         /*
 330          * Remain on current processor as
 331          * additional processors come online.
 332          */
 333         kernel_bootstrap_thread_kprintf("calling thread_bind\n");
 334         thread_bind(processor);
 335
 336         /*
 337          * Kick off memory mapping adjustments.
 338          */
 339         kernel_bootstrap_thread_kprintf("calling mapping_adjust\n");
 340         mapping_adjust();
 341
 342         /*
 343          *      Create the clock service.
 344          */
 345         kernel_bootstrap_thread_kprintf("calling clock_service_create\n");
 346         clock_service_create();
 347
 348         /*
 349          *      Create the device service.
 350          */
 351         device_service_create();
 352
 353         kth_started = 1;
 354
 355 #if (defined(__i386__) || defined(__x86_64__)) && NCOPY_WINDOWS > 0
 356         /*
 357          * Create and initialize the physical copy window for processor 0
 358          * This is required before starting kicking off  IOKit.
 359          */
 360         cpu_physwindow_init(0);
 361 #endif
 362
 363
 364
 365 #if MACH_KDP
 366         kernel_bootstrap_kprintf("calling kdp_init\n");
 367         kdp_init();
 368 #endif
 369
 370 #if ALTERNATE_DEBUGGER
 371         alternate_debugger_init();
 372 #endif
 373
 374 #if CONFIG_COUNTERS
 375         pmc_bootstrap();
 376 #endif
 377
 378 #if KPC
 379         kpc_init();
 380 #endif
 381
 382 #if KPERF
 383         kperf_bootstrap();
 384 #endif
 385
 386 #if CONFIG_TELEMETRY
 387         kernel_bootstrap_kprintf("calling bootprofile_init\n");
 388         bootprofile_init();
 389 #endif
 390
 391 #if (defined(__i386__) || defined(__x86_64__))
 392         if (turn_on_log_leaks && !new_nkdbufs)
 393                 new_nkdbufs = 200000;
 394         start_kern_tracing(new_nkdbufs, FALSE);
 395         if (turn_on_log_leaks)
 396                 log_leaks = 1;
 397
 398 #endif
 399
 400 #ifdef  IOKIT
 401         PE_init_iokit();
 402 #endif
 403
 404         (void) spllo();         /* Allow interruptions */
 405
 406 #if (defined(__i386__) || defined(__x86_64__)) && NCOPY_WINDOWS > 0
 407         /*
 408          * Create and initialize the copy window for processor 0
 409          * This also allocates window space for all other processors.
 410          * However, this is dependent on the number of processors - so this call
 411          * must be after IOKit has been started because IOKit performs processor
 412          * discovery.
 413          */
 414         cpu_userwindow_init(0);
 415 #endif
 416
 417 #if (!defined(__i386__) && !defined(__x86_64__))
 418         if (turn_on_log_leaks && !new_nkdbufs)
 419                 new_nkdbufs = 200000;
 420         start_kern_tracing(new_nkdbufs, FALSE);
 421         if (turn_on_log_leaks)
 422                 log_leaks = 1;
 423 #endif
 424
 425         /*
 426          *      Initialize the shared region module.
 427          */
 428         vm_shared_region_init();
 429         vm_commpage_init();
 430         vm_commpage_text_init();
 431
 432
 433 #if CONFIG_MACF
 434         mac_policy_initmach();
 435 #endif
 436
 437         /*
 438          * Initialize the global used for permuting kernel
 439          * addresses that may be exported to userland as tokens
 440          * using VM_KERNEL_ADDRPERM(). Force the random number
 441          * to be odd to avoid mapping a non-zero
 442          * word-aligned address to zero via addition.
 443          */
 444         vm_kernel_addrperm = (vm_offset_t)early_random() | 1;
 445         buf_kernel_addrperm = (vm_offset_t)early_random() | 1;
 446
 447         /*
 448          *      Start the user bootstrap.
 449          */
 450 #ifdef  MACH_BSD
 451         bsd_init();
 452 #endif
 453
 454     /*
 455      * Get rid of segments used to bootstrap kext loading. This removes
 456      * the KLD, PRELINK symtab, LINKEDIT, and symtab segments/load commands.
 457      */
 458         OSKextRemoveKextBootstrap();
 459
 460         serial_keyboard_init();         /* Start serial keyboard if wanted */
 461
 462         vm_page_init_local_q();
 463
 464         thread_bind(PROCESSOR_NULL);
 465
 466         /*
 467          *      Become the pageout daemon.
 468          */
 469         vm_pageout();
 470         /*NOTREACHED*/
 471 }
 472
 473 /*
 474  *      slave_main:
 475  *
 476  *      Load the first thread to start a processor.
 477  */
 478 void
 479 slave_main(void *machine_param)
 480 {
 481         processor_t             processor = current_processor();
 482         thread_t                thread;
 483
 484         /*
 485          *      Use the idle processor thread if there
 486          *      is no dedicated start up thread.
 487          */
 488         if (processor->next_thread == THREAD_NULL) {
 489                 thread = processor->idle_thread;
 490                 thread->continuation = (thread_continue_t)processor_start_thread;
 491                 thread->parameter = machine_param;
 492         }
 493         else {
 494                 thread = processor->next_thread;
 495                 processor->next_thread = THREAD_NULL;
 496         }
 497
 498         load_context(thread);
 499         /*NOTREACHED*/
 500 }
 501
 502 /*
 503  *      processor_start_thread:
 504  *
 505  *      First thread to execute on a started processor.
 506  *
 507  *      Called at splsched.
 508  */
 509 void
 510 processor_start_thread(void *machine_param)
 511 {
 512         processor_t             processor = current_processor();
 513         thread_t                self = current_thread();
 514
 515         slave_machine_init(machine_param);
 516
 517         /*
 518          *      If running the idle processor thread,
 519          *      reenter the idle loop, else terminate.
 520          */
 521         if (self == processor->idle_thread)
 522                 thread_block((thread_continue_t)idle_thread);
 523
 524         thread_terminate(self);
 525         /*NOTREACHED*/
 526 }
 527
 528 /*
 529  *      load_context:
 530  *
 531  *      Start the first thread on a processor.
 532  */
 533 static void
 534 load_context(
 535         thread_t                thread)
 536 {
 537         processor_t             processor = current_processor();
 538
 539
 540 #define load_context_kprintf(x...) /* kprintf("load_context: " x) */
 541
 542         load_context_kprintf("calling machine_set_current_thread\n");
 543         machine_set_current_thread(thread);
 544
 545         load_context_kprintf("calling processor_up\n");
 546         processor_up(processor);
 547
 548         PMAP_ACTIVATE_KERNEL(processor->cpu_id);
 549
 550         /*
 551          * Acquire a stack if none attached.  The panic
 552          * should never occur since the thread is expected
 553          * to have reserved stack.
 554          */
 555         load_context_kprintf("thread %p, stack %lx, stackptr %lx\n", thread,
 556                              thread->kernel_stack, thread->machine.kstackptr);
 557         if (!thread->kernel_stack) {
 558                 load_context_kprintf("calling stack_alloc_try\n");
 559                 if (!stack_alloc_try(thread))
 560                         panic("load_context");
 561         }
 562
 563         /*
 564          * The idle processor threads are not counted as
 565          * running for load calculations.
 566          */
 567         if (!(thread->state & TH_IDLE))
 568                 sched_run_incr();
 569
 570         processor->active_thread = thread;
 571         processor->current_pri = thread->sched_pri;
 572         processor->current_thmode = thread->sched_mode;
 573         processor->deadline = UINT64_MAX;
 574         thread->last_processor = processor;
 575
 576         processor->last_dispatch = mach_absolute_time();
 577         timer_start(&thread->system_timer, processor->last_dispatch);
 578         PROCESSOR_DATA(processor, thread_timer) = PROCESSOR_DATA(processor, kernel_timer) = &thread->system_timer;
 579
 580         timer_start(&PROCESSOR_DATA(processor, system_state), processor->last_dispatch);
 581         PROCESSOR_DATA(processor, current_state) = &PROCESSOR_DATA(processor, system_state);
 582
 583         PMAP_ACTIVATE_USER(thread, processor->cpu_id);
 584
 585         load_context_kprintf("calling machine_load_context\n");
 586         machine_load_context(thread);
 587         /*NOTREACHED*/
 588 }
 589
 590 void
 591 scale_setup()
 592 {
 593         int scale = 0;
 594 #if defined(__LP64__)
 595         typeof(task_max) task_max_base = task_max;
 596
 597         /* Raise limits for servers with >= 16G */
 598         if ((serverperfmode != 0) && ((uint64_t)sane_size >= (uint64_t)(16 * 1024 * 1024 *1024ULL))) {
 599                 scale = (int)((uint64_t)sane_size / (uint64_t)(8 * 1024 * 1024 *1024ULL));
 600                 /* limit to 128 G */
 601                 if (scale > 16)
 602                         scale = 16;
 603                 task_max_base = 2500;
 604         } else if ((uint64_t)sane_size >= (uint64_t)(3 * 1024 * 1024 *1024ULL))
 605                 scale = 2;
 606
 607         task_max = MAX(task_max, task_max_base * scale);
 608
 609         if (scale != 0) {
 610                 task_threadmax = task_max;
 611                 thread_max = task_max * 5;
 612         }
 613
 614 #endif
 615
 616         bsd_scale_setup(scale);
 617
 618         ipc_space_max = SPACE_MAX;
 619         ipc_port_max = PORT_MAX;
 620         ipc_pset_max = SET_MAX;
 621         semaphore_max = SEMAPHORE_MAX;
 622 }
 623