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 * 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 [email protected]
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 <mach_kdp.h>
71
72 #include <mach/boolean.h>
73 #include <mach/machine.h>
74 #include <mach/thread_act.h>
75 #include <mach/task_special_ports.h>
76 #include <mach/vm_param.h>
77 #include <mach/exclaves.h>
78 #include <kern/assert.h>
79 #include <kern/mach_param.h>
80 #include <kern/misc_protos.h>
81 #include <kern/clock.h>
82 #include <kern/coalition.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/restartable.h>
88 #include <kern/sched_prim.h>
89 #include <kern/turnstile.h>
90 #if CONFIG_SCHED_SFI
91 #include <kern/sfi.h>
92 #endif
93 #include <kern/smr.h>
94 #include <kern/startup.h>
95 #include <kern/task.h>
96 #include <kern/thread.h>
97 #include <kern/timer.h>
98 #if CONFIG_TELEMETRY
99 #include <kern/telemetry.h>
100 #include <kern/trap_telemetry.h>
101 #endif
102 #include <kern/kpc.h>
103 #include <kern/zalloc.h>
104 #include <kern/locks.h>
105 #include <kern/debug.h>
106 #if KPERF
107 #include <kperf/kperf.h>
108 #endif /* KPERF */
109 #include <corpses/task_corpse.h>
110 #include <prng/random.h>
111 #include <console/serial_protos.h>
112 #include <vm/vm_kern_xnu.h>
113 #include <vm/vm_init_xnu.h>
114 #include <vm/vm_map.h>
115 #include <vm/vm_object_xnu.h>
116 #include <vm/vm_page.h>
117 #include <vm/vm_pageout_xnu.h>
118 #include <vm/vm_shared_region_xnu.h>
119 #include <machine/pmap.h>
120 #include <machine/commpage.h>
121 #include <machine/machine_routines.h>
122 #include <machine/static_if.h>
123 #include <libkern/version.h>
124 #include <pexpert/device_tree.h>
125 #include <sys/codesign.h>
126 #include <sys/kdebug.h>
127 #include <sys/random.h>
128 #include <sys/ktrace.h>
129 #include <sys/trust_caches.h>
130 #include <sys/code_signing.h>
131 #include <libkern/section_keywords.h>
132
133 #include <kern/waitq.h>
134 #include <ipc/ipc_voucher.h>
135 #include <mach/host_info.h>
136 #include <pthread/workqueue_internal.h>
137
138 #if SOCKETS
139 extern void mbuf_tag_init(void);
140 #endif
141
142 #if CONFIG_XNUPOST
143 #include <tests/ktest.h>
144 #include <tests/xnupost.h>
145 #endif
146
147 #if CONFIG_ATM
148 #include <atm/atm_internal.h>
149 #endif
150
151 #if ALTERNATE_DEBUGGER
152 #include <arm64/alternate_debugger.h>
153 #endif
154
155 #if MACH_KDP
156 #include <kdp/kdp.h>
157 #endif
158
159 #if CONFIG_MACF
160 #include <security/mac_mach_internal.h>
161 #if CONFIG_VNGUARD
162 extern void vnguard_policy_init(void);
163 #endif
164 #endif
165
166 #if HYPERVISOR
167 #include <kern/hv_support.h>
168 #endif
169
170 #if CONFIG_UBSAN_MINIMAL
171 #include <san/ubsan_minimal.h>
172 #endif
173
174 #include <san/kasan.h>
175
176 #include <i386/pmCPU.h>
177 static void kernel_bootstrap_thread(void);
178
179 static void load_context(
180 thread_t thread);
181
182 #if CONFIG_ECC_LOGGING
183 #include <kern/ecc.h>
184 #endif
185
186 #if (defined(__i386__) || defined(__x86_64__)) && CONFIG_VMX
187 #include <i386/vmx/vmx_cpu.h>
188 #endif
189
190 #if CONFIG_DTRACE
191 extern void dtrace_early_init(void);
192 extern void sdt_early_init(void);
193 #endif
194
195 // libkern/OSKextLib.cpp
196 extern void OSKextRemoveKextBootstrap(void);
197
198 void scale_setup(void);
199 extern void bsd_scale_setup(int);
200 extern unsigned int semaphore_max;
201 extern void stackshot_init(void);
202
203 /*
204 * Running in virtual memory, on the interrupt stack.
205 */
206
207 extern struct startup_entry startup_entries[]
208 __SECTION_START_SYM(STARTUP_HOOK_SEGMENT, STARTUP_HOOK_SECTION);
209
210 extern struct startup_entry startup_entries_end[]
211 __SECTION_END_SYM(STARTUP_HOOK_SEGMENT, STARTUP_HOOK_SECTION);
212
213 static struct startup_entry *__startup_data startup_entry_cur = startup_entries;
214
215 SECURITY_READ_ONLY_LATE(startup_subsystem_id_t) startup_phase = STARTUP_SUB_NONE;
216
217 TUNABLE(startup_debug_t, startup_debug, "startup_debug", 0);
218
219 /* Indicates a server boot when set */
220 TUNABLE(int, serverperfmode, "serverperfmode", 0);
221
222 static inline void
kernel_bootstrap_log(const char * message)223 kernel_bootstrap_log(const char *message)
224 {
225 if ((startup_debug & STARTUP_DEBUG_VERBOSE) &&
226 startup_phase >= STARTUP_SUB_KPRINTF) {
227 kprintf("kernel_bootstrap: %s\n", message);
228 }
229 kernel_debug_string_early(message);
230 }
231
232 static inline void
kernel_bootstrap_thread_log(const char * message)233 kernel_bootstrap_thread_log(const char *message)
234 {
235 if ((startup_debug & STARTUP_DEBUG_VERBOSE) &&
236 startup_phase >= STARTUP_SUB_KPRINTF) {
237 kprintf("kernel_bootstrap_thread: %s\n", message);
238 }
239 kernel_debug_string_early(message);
240 }
241
242 extern void
243 qsort(void *a, size_t n, size_t es, int (*cmp)(const void *, const void *));
244
245 __startup_func
246 static int
startup_entry_cmp(const void * e1,const void * e2)247 startup_entry_cmp(const void *e1, const void *e2)
248 {
249 const struct startup_entry *a = e1;
250 const struct startup_entry *b = e2;
251 if (a->subsystem == b->subsystem) {
252 if (a->rank == b->rank) {
253 return 0;
254 }
255 return a->rank > b->rank ? 1 : -1;
256 }
257 return a->subsystem > b->subsystem ? 1 : -1;
258 }
259
260 __startup_func
261 void
kernel_startup_bootstrap(void)262 kernel_startup_bootstrap(void)
263 {
264 /*
265 * Sort the various STARTUP() entries by subsystem/rank.
266 */
267 size_t n = startup_entries_end - startup_entries;
268
269 if (n == 0) {
270 panic("Section %s,%s missing",
271 STARTUP_HOOK_SEGMENT, STARTUP_HOOK_SECTION);
272 }
273 if (((uintptr_t)startup_entries_end - (uintptr_t)startup_entries) %
274 sizeof(struct startup_entry)) {
275 panic("Section %s,%s has invalid size",
276 STARTUP_HOOK_SEGMENT, STARTUP_HOOK_SECTION);
277 }
278
279 qsort(startup_entries, n, sizeof(struct startup_entry), startup_entry_cmp);
280
281 #if !CONFIG_SPTM && !defined(__BUILDING_XNU_LIBRARY__)
282 /* static_if relies on TEXT editing and not supported in user-mode build*/
283 static_if_init(PE_boot_args());
284 #endif
285
286 /*
287 * Then initialize all tunables, timeouts, and locks
288 */
289 kernel_startup_initialize_upto(STARTUP_SUB_LOCKS);
290 }
291
292 __startup_func
293 void
kernel_startup_tunable_init(const struct startup_tunable_spec * spec)294 kernel_startup_tunable_init(const struct startup_tunable_spec *spec)
295 {
296 if (spec->var_is_str) {
297 PE_parse_boot_arg_str(spec->name, spec->var_addr, spec->var_len);
298 } else if (PE_parse_boot_argn(spec->name, spec->var_addr, spec->var_len)) {
299 if (spec->var_is_bool) {
300 /* make sure bool's are valued in {0, 1} */
301 *(bool *)spec->var_addr = *(uint8_t *)spec->var_addr;
302 }
303 }
304 }
305
306 __startup_func
307 void
kernel_startup_tunable_dt_source_init(const struct startup_tunable_dt_source_spec * spec)308 kernel_startup_tunable_dt_source_init(const struct startup_tunable_dt_source_spec *spec)
309 {
310 DTEntry base;
311
312 *spec->source_addr = STARTUP_SOURCE_DEFAULT;
313 if (SecureDTLookupEntry(NULL, spec->dt_base, &base) != kSuccess) {
314 base = NULL;
315 }
316
317 bool found_in_chosen = false;
318
319 if (spec->dt_chosen_override) {
320 DTEntry chosen, chosen_base;
321
322 if (SecureDTLookupEntry(NULL, "chosen", &chosen) != kSuccess) {
323 chosen = NULL;
324 }
325
326 if (chosen != NULL && SecureDTLookupEntry(chosen, spec->dt_base, &chosen_base) == kSuccess) {
327 base = chosen_base;
328 found_in_chosen = true;
329 *spec->source_addr = STARTUP_SOURCE_DEVICETREE;
330 }
331 }
332
333 uint64_t const *data;
334 unsigned int data_size = spec->var_len;
335
336 if (base != NULL && SecureDTGetProperty(base, spec->dt_name, (const void **)&data, &data_size) == kSuccess) {
337 if (data_size != spec->var_len) {
338 panic("unexpected tunable size %u in DT entry %s/%s/%s",
339 data_size, found_in_chosen ? "/chosen" : "", spec->dt_base, spec->dt_name);
340 }
341
342 /* No need to handle bools specially, they are 1 byte integers in the DT. */
343 memcpy(spec->var_addr, data, spec->var_len);
344 *spec->source_addr = STARTUP_SOURCE_DEVICETREE;
345 }
346
347 /* boot-arg overrides. */
348
349 if (spec->boot_arg_name != NULL) {
350 if (PE_parse_boot_argn(spec->boot_arg_name, spec->var_addr, spec->var_len)) {
351 if (spec->var_is_bool) {
352 *(bool *)spec->var_addr = *(uint8_t *)spec->var_addr;
353 }
354 *spec->source_addr = STARTUP_SOURCE_BOOTPARAM;
355 }
356 }
357 }
358
359 __startup_func
360 void
kernel_startup_tunable_dt_init(const struct startup_tunable_dt_spec * spec)361 kernel_startup_tunable_dt_init(const struct startup_tunable_dt_spec *spec)
362 {
363 DTEntry base;
364
365 if (SecureDTLookupEntry(NULL, spec->dt_base, &base) != kSuccess) {
366 base = NULL;
367 }
368
369 bool found_in_chosen = false;
370
371 if (spec->dt_chosen_override) {
372 DTEntry chosen, chosen_base;
373
374 if (SecureDTLookupEntry(NULL, "chosen", &chosen) != kSuccess) {
375 chosen = NULL;
376 }
377
378 if (chosen != NULL && SecureDTLookupEntry(chosen, spec->dt_base, &chosen_base) == kSuccess) {
379 base = chosen_base;
380 found_in_chosen = true;
381 }
382 }
383
384 uint64_t const *data;
385 unsigned int data_size = spec->var_len;
386
387 if (base != NULL && SecureDTGetProperty(base, spec->dt_name, (const void **)&data, &data_size) == kSuccess) {
388 if (data_size != spec->var_len) {
389 panic("unexpected tunable size %u in DT entry %s/%s/%s",
390 data_size, found_in_chosen ? "/chosen" : "", spec->dt_base, spec->dt_name);
391 }
392
393 /* No need to handle bools specially, they are 1 byte integers in the DT. */
394 memcpy(spec->var_addr, data, spec->var_len);
395 }
396
397 /* boot-arg overrides. */
398
399 if (spec->boot_arg_name != NULL) {
400 if (PE_parse_boot_argn(spec->boot_arg_name, spec->var_addr, spec->var_len)) {
401 if (spec->var_is_bool) {
402 *(bool *)spec->var_addr = *(uint8_t *)spec->var_addr;
403 }
404 }
405 }
406 }
407
408 static void
kernel_startup_log(startup_subsystem_id_t subsystem)409 kernel_startup_log(startup_subsystem_id_t subsystem)
410 {
411 static const char *names[] = {
412 [STARTUP_SUB_TUNABLES] = "tunables",
413 [STARTUP_SUB_TIMEOUTS] = "timeouts",
414 [STARTUP_SUB_LOCKS] = "locks",
415 [STARTUP_SUB_KPRINTF] = "kprintf",
416
417 [STARTUP_SUB_PMAP_STEAL] = "pmap_steal",
418 [STARTUP_SUB_KMEM] = "kmem",
419 [STARTUP_SUB_ZALLOC] = "zalloc",
420 [STARTUP_SUB_PERCPU] = "percpu",
421 [STARTUP_SUB_EVENT] = "event",
422
423 [STARTUP_SUB_CODESIGNING] = "codesigning",
424 [STARTUP_SUB_KTRACE] = "ktrace",
425 [STARTUP_SUB_OSLOG] = "oslog",
426 [STARTUP_SUB_MACH_IPC] = "mach_ipc",
427 [STARTUP_SUB_THREAD_CALL] = "thread_call",
428 [STARTUP_SUB_SYSCTL] = "sysctl",
429 [STARTUP_SUB_EARLY_BOOT] = "early_boot",
430
431 /* LOCKDOWN is special and its value won't fit here. */
432 };
433 static startup_subsystem_id_t logged = STARTUP_SUB_NONE;
434
435 if (subsystem <= logged) {
436 return;
437 }
438
439 if (subsystem < sizeof(names) / sizeof(names[0]) && names[subsystem]) {
440 kernel_bootstrap_log(names[subsystem]);
441 }
442 logged = subsystem;
443 }
444
445 __startup_func
446 void
event_register_handler(struct event_hdr * hdr)447 event_register_handler(struct event_hdr *hdr)
448 {
449 struct event_hdr *head = hdr->next;
450
451 hdr->next = head->next;
452 head->next = hdr;
453 }
454
455 __startup_func
456 void
kernel_startup_initialize_upto(startup_subsystem_id_t upto)457 kernel_startup_initialize_upto(startup_subsystem_id_t upto)
458 {
459 struct startup_entry *cur = startup_entry_cur;
460
461 assert(startup_phase < upto);
462
463 while (cur < startup_entries_end && cur->subsystem <= upto) {
464 if ((startup_debug & STARTUP_DEBUG_VERBOSE) &&
465 startup_phase >= STARTUP_SUB_KPRINTF) {
466 kprintf("%s[%d, rank %d]: %p(%p)\n", __func__,
467 cur->subsystem, cur->rank, cur->func, cur->arg);
468 }
469 startup_phase = cur->subsystem - 1;
470 kernel_startup_log(cur->subsystem);
471 cur->func(cur->arg);
472 startup_entry_cur = ++cur;
473 }
474 kernel_startup_log(upto);
475
476 if ((startup_debug & STARTUP_DEBUG_VERBOSE) &&
477 upto >= STARTUP_SUB_KPRINTF) {
478 kprintf("%s: reached phase %d\n", __func__, upto);
479 }
480 startup_phase = upto;
481 }
482
483 #ifdef __BUILDING_XNU_LIB_UNITTEST__
484 /* unit-test initialization needs to pick specific phases */
485 void
kernel_startup_initialize_only(startup_subsystem_id_t sysid)486 kernel_startup_initialize_only(startup_subsystem_id_t sysid)
487 {
488 assert(startup_phase < sysid);
489 struct startup_entry *cur = startup_entry_cur;
490 while (cur < startup_entries_end && cur->subsystem <= sysid) {
491 if (cur->subsystem == sysid) {
492 startup_phase = cur->subsystem - 1;
493 kernel_startup_log(cur->subsystem);
494 cur->func(cur->arg);
495 }
496 startup_entry_cur = ++cur;
497 }
498 startup_phase = sysid;
499 }
500 #endif
501
502 void
kernel_bootstrap(void)503 kernel_bootstrap(void)
504 {
505 kern_return_t result;
506 thread_t thread;
507 char namep[16];
508
509 code_signing_config_t cs_config;
510
511 printf("%s\n", version); /* log kernel version */
512
513 #if HAS_UPSI_FAILURE_INJECTION
514 check_for_failure_injection(XNU_STAGE_BOOTSTRAP_START);
515 #endif
516
517 scale_setup();
518
519 kernel_bootstrap_log("vm_mem_bootstrap");
520 vm_mem_bootstrap();
521
522 machine_info.memory_size = (uint32_t)mem_size;
523 #if XNU_TARGET_OS_OSX
524 machine_info.max_mem = max_mem_actual;
525 #else
526 machine_info.max_mem = max_mem;
527 #endif /* XNU_TARGET_OS_OSX */
528 machine_info.major_version = version_major;
529 machine_info.minor_version = version_minor;
530
531 #if CONFIG_ATM
532 /* Initialize the Activity Trace Resource Manager. */
533 kernel_bootstrap_log("atm_init");
534 atm_init();
535 #endif
536 kernel_startup_initialize_upto(STARTUP_SUB_OSLOG);
537
538 #if CONFIG_UBSAN_MINIMAL
539 kernel_bootstrap_log("UBSan minimal runtime init");
540 ubsan_minimal_init();
541 #endif
542
543 #if KASAN
544 kernel_bootstrap_log("kasan_late_init");
545 kasan_late_init();
546 #endif
547
548 #if CONFIG_TELEMETRY
549 kernel_bootstrap_log("trap_telemetry_init");
550 trap_telemetry_init();
551 #endif
552
553 if (PE_i_can_has_debugger(NULL)) {
554 if (PE_parse_boot_argn("-show_pointers", &namep, sizeof(namep))) {
555 doprnt_hide_pointers = FALSE;
556 }
557 if (PE_parse_boot_argn("-no_slto_panic", &namep, sizeof(namep))) {
558 extern boolean_t spinlock_timeout_panic;
559 spinlock_timeout_panic = FALSE;
560 }
561 }
562
563 kernel_bootstrap_log("console_init");
564 console_init();
565
566 kernel_bootstrap_log("stackshot_init");
567 stackshot_init();
568
569 kernel_bootstrap_log("sched_init");
570 sched_init();
571
572 #if CONFIG_MACF
573 kernel_bootstrap_log("mac_policy_init");
574 mac_policy_init();
575 #endif
576
577 kernel_startup_initialize_upto(STARTUP_SUB_MACH_IPC);
578
579 /*
580 * As soon as the virtual memory system is up, we record
581 * that this CPU is using the kernel pmap.
582 */
583 kernel_bootstrap_log("PMAP_ACTIVATE_KERNEL");
584 PMAP_ACTIVATE_KERNEL(master_cpu);
585
586 kernel_bootstrap_log("mapping_free_prime");
587 mapping_free_prime(); /* Load up with temporary mapping blocks */
588
589 kernel_bootstrap_log("machine_init");
590 machine_init();
591
592 kernel_bootstrap_log("thread_machine_init_template");
593 thread_machine_init_template();
594
595 kernel_bootstrap_log("clock_init");
596 clock_init();
597
598 /*
599 * Initialize the IPC, task, and thread subsystems.
600 */
601 #if CONFIG_THREAD_GROUPS
602 kernel_bootstrap_log("thread_group_init");
603 thread_group_init();
604 #endif
605
606 #if CONFIG_COALITIONS
607 kernel_bootstrap_log("coalitions_init");
608 coalitions_init();
609 #endif
610
611 kernel_bootstrap_log("code_signing_init");
612 code_signing_init();
613 code_signing_configuration(NULL, &cs_config);
614 #if XNU_TARGET_OS_OSX && (DEVELOPMENT || DEBUG)
615 if (cs_config & CS_CONFIG_GET_OUT_OF_MY_WAY) {
616 AMFI_bootarg_disable_mach_hardening = true;
617 }
618 #endif /* XNU_TARGET_OS_OSX && (DEVELOPMENT || DEBUG) */
619
620 kernel_bootstrap_log("task_init");
621 task_init();
622
623 kernel_bootstrap_log("thread_init");
624 thread_init();
625
626 kernel_bootstrap_log("restartable_init");
627 restartable_init();
628
629 kernel_bootstrap_log("workq_init");
630 workq_init();
631
632 kernel_bootstrap_log("turnstiles_init");
633 turnstiles_init();
634
635 #if PAGE_SLEEP_WITH_INHERITOR
636 kernel_bootstrap_log("page_worker_init");
637 page_worker_init();
638 #endif /* PAGE_SLEEP_WITH_INHERITOR */
639
640 kernel_bootstrap_log("mach_init_activity_id");
641 mach_init_activity_id();
642
643 /* initialize host_statistics */
644 host_statistics_init();
645
646 /* initialize exceptions */
647 kernel_bootstrap_log("exception_init");
648 exception_init();
649
650 #if CONFIG_SCHED_SFI
651 kernel_bootstrap_log("sfi_init");
652 sfi_init();
653 #endif
654
655 /*
656 * Create a kernel thread to execute the kernel bootstrap.
657 */
658
659 kernel_bootstrap_log("kernel_thread_create");
660 result = kernel_thread_create((thread_continue_t)kernel_bootstrap_thread, NULL, MAXPRI_KERNEL, &thread);
661
662 if (result != KERN_SUCCESS) {
663 panic("kernel_bootstrap: result = %08X", result);
664 }
665
666 /* TODO: do a proper thread_start() (without the thread_setrun()) */
667 thread->state = TH_RUN;
668 thread->last_made_runnable_time = mach_absolute_time();
669 thread_set_thread_name(thread, "kernel_bootstrap_thread");
670
671 thread_deallocate(thread);
672
673 kernel_bootstrap_log("load_context - done");
674 load_context(thread);
675 /*NOTREACHED*/
676 }
677
678 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_addrperm;
679 SECURITY_READ_ONLY_LATE(vm_offset_t) buf_kernel_addrperm;
680 SECURITY_READ_ONLY_LATE(vm_offset_t) vm_kernel_addrperm_ext;
681 SECURITY_READ_ONLY_LATE(uint64_t) vm_kernel_addrhash_salt;
682 SECURITY_READ_ONLY_LATE(uint64_t) vm_kernel_addrhash_salt_ext;
683
684 /*
685 * Now running in a thread. Kick off other services,
686 * invoke user bootstrap, enter pageout loop.
687 */
688 static void
kernel_bootstrap_thread(void)689 kernel_bootstrap_thread(void)
690 {
691 processor_t processor = current_processor();
692
693 #if HAS_UPSI_FAILURE_INJECTION
694 check_for_failure_injection(XNU_STAGE_SCHEDULER_START);
695 #endif
696
697 kernel_bootstrap_thread_log("idle_thread_create");
698 /*
699 * Create the idle processor thread for the boot processor.
700 */
701 idle_thread_create(processor, idle_thread);
702
703 /*
704 * N.B. Do not stick anything else
705 * before this point.
706 *
707 * Start up the scheduler services.
708 */
709 kernel_bootstrap_thread_log("sched_startup");
710 sched_startup();
711
712 /*
713 * Thread lifecycle maintenance (teardown, stack allocation)
714 */
715 kernel_bootstrap_thread_log("thread_daemon_init");
716 thread_daemon_init();
717
718 /*
719 * Thread callout service.
720 */
721 kernel_startup_initialize_upto(STARTUP_SUB_THREAD_CALL);
722
723 /*
724 * Remain on current processor as
725 * additional processors come online.
726 */
727 kernel_bootstrap_thread_log("thread_bind");
728 suspend_cluster_powerdown();
729 thread_bind(processor);
730
731 /*
732 * Kick off memory mapping adjustments.
733 */
734 kernel_bootstrap_thread_log("mapping_adjust");
735 mapping_adjust();
736
737 /*
738 * Create the clock service.
739 */
740 kernel_bootstrap_thread_log("clock_service_create");
741 clock_service_create();
742
743 /*
744 * Create the device service.
745 */
746 device_service_create();
747
748 phys_carveout_init();
749
750 /* Now that carveouts are allocated, start tracing (primary CPU). */
751 #if __arm64__ && (DEVELOPMENT || DEBUG)
752 pe_arm_debug_init_late();
753 PE_arm_debug_enable_trace(true);
754 #endif /* __arm64__ && (DEVELOPMENT || DEBUG) */
755
756 #if MACH_KDP
757 kernel_bootstrap_log("kdp_init");
758 kdp_init();
759 #endif
760
761 #if ALTERNATE_DEBUGGER
762 alternate_debugger_init();
763 #endif
764
765 #if HYPERVISOR
766 kernel_bootstrap_thread_log("hv_support_init");
767 hv_support_init();
768 #endif
769
770 kernel_startup_initialize_upto(STARTUP_SUB_SYSCTL);
771
772 /*
773 * Initialize the globals used for permuting kernel
774 * addresses that may be exported to userland as tokens
775 * using VM_KERNEL_ADDRPERM()/VM_KERNEL_ADDRPERM_EXTERNAL().
776 * Force the random number to be odd to avoid mapping a non-zero
777 * word-aligned address to zero via addition.
778 */
779 vm_kernel_addrperm = (vm_offset_t)(early_random() | 1);
780 buf_kernel_addrperm = (vm_offset_t)(early_random() | 1);
781 vm_kernel_addrperm_ext = (vm_offset_t)(early_random() | 1);
782 vm_kernel_addrhash_salt = early_random();
783 vm_kernel_addrhash_salt_ext = early_random();
784
785 #ifdef IOKIT
786 kernel_bootstrap_log("PE_init_iokit");
787 PE_init_iokit();
788 #endif
789
790 assert(ml_get_interrupts_enabled() == FALSE);
791
792 /*
793 * Past this point, kernel subsystems that expect to operate with
794 * interrupts or preemption enabled may begin enforcement.
795 */
796 kernel_startup_initialize_upto(STARTUP_SUB_EARLY_BOOT);
797
798 #if SCHED_HYGIENE_DEBUG
799 // Reset interrupts masked timeout before we enable interrupts
800 ml_spin_debug_clear_self();
801 #endif
802 (void) spllo(); /* Allow interruptions */
803
804 /*
805 * This will start displaying progress to the user, start as early as possible
806 */
807 initialize_screen(NULL, kPEAcquireScreen);
808
809 /*
810 * Initialize the shared region module.
811 */
812 vm_commpage_init();
813 vm_commpage_text_init();
814
815 #if CONFIG_MACF
816 kernel_bootstrap_log("mac_policy_initmach");
817 mac_policy_initmach();
818 #if CONFIG_VNGUARD
819 kernel_bootstrap_log("vnguard_policy_init");
820 vnguard_policy_init();
821 #endif
822 #endif
823
824 #if CONFIG_DTRACE
825 kernel_bootstrap_log("dtrace_early_init");
826 dtrace_early_init();
827 sdt_early_init();
828 #endif
829
830 #if CONFIG_EXCLAVES
831 exclaves_early_init();
832 #endif
833
834 #if CODE_SIGNING_MONITOR
835 /*
836 * Lockdown mode is initialized as a startup function within the early boot
837 * category, which means it has been initialized by now. Query the state and
838 * pass it to the code-signing-monitor if required.
839 */
840 kernel_bootstrap_log("code-signing-monitor lockdown mode");
841 csm_check_lockdown_mode();
842 #endif
843
844 #if CODE_SIGNING_MONITOR
845 kernel_bootstrap_log("provisioning_profile_init");
846 csm_initialize_provisioning_profiles();
847 #endif
848
849 kernel_bootstrap_log("trust_cache_init");
850
851 /* Initialize the runtime for the trust cache interface */
852 trust_cache_runtime_init();
853
854 /* Load the static and engineering trust caches */
855 load_static_trust_cache();
856
857 kernel_startup_initialize_upto(STARTUP_SUB_LOCKDOWN);
858
859 /*
860 * Get rid of segments used to bootstrap kext loading. This removes
861 * the KLD, PRELINK symtab, LINKEDIT, and symtab segments/load commands.
862 * Must be done prior to lockdown so that we can free (and possibly relocate)
863 * the static KVA mappings used for the jettisoned bootstrap segments.
864 */
865 kernel_bootstrap_log("OSKextRemoveKextBootstrap");
866 OSKextRemoveKextBootstrap();
867
868 #if SOCKETS
869 /*
870 * Initialize callback table before machine lockdown
871 */
872 mbuf_tag_init();
873 #endif
874
875 /* No changes to kernel text and rodata beyond this point. */
876 kernel_bootstrap_log("machine_lockdown");
877 machine_lockdown();
878
879 #ifdef CONFIG_XNUPOST
880 kern_return_t result = kernel_list_tests();
881 result = kernel_do_post();
882 if (result != KERN_SUCCESS) {
883 panic("kernel_do_post: Tests failed with result = 0x%08x", result);
884 }
885 kernel_bootstrap_log("kernel_do_post - done");
886 #endif /* CONFIG_XNUPOST */
887
888 #ifdef IOKIT
889 kernel_bootstrap_log("PE_lockdown_iokit");
890 PE_lockdown_iokit();
891 #endif
892 /*
893 * max_cpus must be nailed down by the time PE_lockdown_iokit() finishes,
894 * at the latest
895 */
896 vm_set_restrictions(machine_info.max_cpus);
897
898
899 #if KPERF
900 kperf_init_early();
901 #endif
902
903 /*
904 * Start the user bootstrap.
905 */
906 #ifdef MACH_BSD
907 bsd_init();
908 #endif
909
910
911 /*
912 * Get rid of pages used for early boot tracing.
913 */
914 kdebug_free_early_buf();
915
916 serial_keyboard_init(); /* Start serial keyboard if wanted */
917
918 vm_page_init_local_q(machine_info.max_cpus);
919
920 thread_bind(PROCESSOR_NULL);
921 resume_cluster_powerdown();
922
923 #if XNU_VM_HAS_DELAYED_PAGES
924 /*
925 * Now that all CPUs are available to run threads, this is essentially
926 * a background thread. Take this opportunity to initialize and free
927 * any remaining vm_pages that were delayed earlier by pmap_startup().
928 */
929 vm_free_delayed_pages();
930 #endif /* XNU_VM_HAS_DELAYED_PAGES */
931
932 vm_pages_array_finalize();
933
934 /*
935 * Become the pageout daemon.
936 */
937 vm_pageout();
938 /*NOTREACHED*/
939 }
940
941 /*
942 * secondary_cpu_main:
943 *
944 * Load the first thread to start a processor, or
945 * load the previous thread context when restarting a processor
946 * from shutdown.
947 * This path will also be used by the master processor
948 * after being offlined.
949 */
950 void
secondary_cpu_main(void * machine_param)951 secondary_cpu_main(void *machine_param)
952 {
953 processor_t processor = current_processor();
954 thread_t thread = processor->idle_thread;
955
956 thread->parameter = machine_param;
957
958 load_context(thread);
959 /*NOTREACHED*/
960 }
961
962 /*
963 * processor_start_thread:
964 *
965 * First thread to execute on a started processor.
966 *
967 * Called at splsched.
968 */
969 void
processor_start_thread(void * machine_param,__unused wait_result_t result)970 processor_start_thread(void *machine_param,
971 __unused wait_result_t result)
972 {
973 assert(ml_get_interrupts_enabled() == FALSE);
974 assert(current_thread() == current_processor()->idle_thread);
975
976 #if CONFIG_KCOV
977 kcov_start_cpu(current_processor()->cpu_id);
978 #endif
979
980 #if USE_APPLEARMSMP
981 /*
982 * On AppleARMSMP platforms, the cpu_boot_thread registers the AIC and
983 * FastIPI interrupt handlers before the secondary CPU is booted, so we
984 * can expect the self-IPI to deliver immediately.
985 */
986 bool wait_for_cpu_signal = true;
987 #else /* USE_APPLEARMSMP */
988 /*
989 * On AppleARMCPU platforms, the AIC and AppleARMCPU threads must be
990 * scheduled after the secondary CPUs boot in order to register the IPI
991 * interrupt handlers, so we can not be guaranteed when the self-IPI
992 * will deliver. The threads may even need to run on this CPU, so we
993 * can't spin against the self-IPI being delivered.
994 * See rdar://125383535.
995 */
996 bool wait_for_cpu_signal = false;
997 #endif /* USE_APPLEARMSMP */
998
999 processor_cpu_reinit(machine_param, wait_for_cpu_signal, false);
1000
1001 thread_block(idle_thread);
1002 /*NOTREACHED*/
1003 }
1004
1005 /*
1006 * load_context:
1007 *
1008 * Start the first thread on a processor.
1009 * This may be the first thread ever run on a processor, or
1010 * it could be a processor that was previously offlined.
1011 */
1012 static void __attribute__((noreturn))
load_context(thread_t thread)1013 load_context(
1014 thread_t thread)
1015 {
1016 processor_t processor = current_processor();
1017
1018
1019 #define load_context_kprintf(x...) /* kprintf("load_context: " x) */
1020
1021 load_context_kprintf("machine_set_current_thread\n");
1022 machine_set_current_thread(thread);
1023
1024 load_context_kprintf("processor_up\n");
1025
1026 PMAP_ACTIVATE_KERNEL(processor->cpu_id);
1027
1028 /*
1029 * Acquire a stack if none attached. The panic
1030 * should never occur since the thread is expected
1031 * to have reserved stack.
1032 */
1033 load_context_kprintf("thread %p, stack %lx, stackptr %lx\n", thread,
1034 thread->kernel_stack, thread->machine.kstackptr);
1035 if (!thread->kernel_stack) {
1036 load_context_kprintf("stack_alloc_try\n");
1037 if (!stack_alloc_try(thread)) {
1038 panic("load_context");
1039 }
1040 }
1041
1042 /*
1043 * The idle processor threads are not counted as
1044 * running for load calculations.
1045 */
1046 if (!(thread->state & TH_IDLE)) {
1047 SCHED(run_count_incr)(thread);
1048 }
1049
1050 processor->active_thread = thread;
1051 processor_state_update_from_thread(processor, thread, false);
1052 processor->starting_pri = thread->sched_pri;
1053 processor->deadline = UINT64_MAX;
1054 thread->last_processor = processor;
1055 processor_up(processor);
1056 struct recount_snap snap = { 0 };
1057 recount_snapshot(&snap);
1058 processor->last_dispatch = snap.rsn_time_mach;
1059 recount_processor_online(processor, &snap);
1060
1061 smr_cpu_join(processor, processor->last_dispatch);
1062
1063 PMAP_ACTIVATE_USER(thread, processor->cpu_id);
1064
1065 load_context_kprintf("machine_load_context\n");
1066
1067 #if KASAN_TBI
1068 __asan_handle_no_return();
1069 #endif /* KASAN_TBI */
1070
1071 machine_load_context(thread);
1072 /*NOTREACHED*/
1073 }
1074
1075 extern unsigned int kern_feature_overrides;
1076
1077 void
scale_setup(void)1078 scale_setup(void)
1079 {
1080 boolean_t pe_serverperfmode = FALSE;
1081 int scale = 0;
1082
1083 /*
1084 * kern_feature_override_init() will update kern_feature_override
1085 * based on the serverperfmode=1 boot-arg being present,
1086 * but doesn't take the device-tree setting into account on purpose.
1087 */
1088
1089 pe_serverperfmode = PE_get_default("kern.serverperfmode",
1090 &pe_serverperfmode, sizeof(pe_serverperfmode));
1091 if (pe_serverperfmode) {
1092 serverperfmode = (pe_serverperfmode != 0);
1093 }
1094 #if defined(__LP64__)
1095 typeof(task_max) task_max_base = task_max;
1096
1097
1098 /* Raise limits for servers with >= 16G */
1099 if (serverperfmode && ((uint64_t)max_mem_actual >= (uint64_t)(16 * 1024 * 1024 * 1024ULL))) {
1100 scale = (int)((uint64_t)sane_size / (uint64_t)(8 * 1024 * 1024 * 1024ULL));
1101 /* limit to 128 G */
1102 if (scale > 16) {
1103 scale = 16;
1104 }
1105 task_max_base = 2500;
1106 /* Raise limits for machines with >= 3GB */
1107 } else if ((uint64_t)max_mem_actual >= (uint64_t)(3 * 1024 * 1024 * 1024ULL)) {
1108 if ((uint64_t)max_mem_actual < (uint64_t)(8 * 1024 * 1024 * 1024ULL)) {
1109 scale = 2;
1110 } else {
1111 /* limit to 64GB */
1112 scale = MIN(16, (int)((uint64_t)max_mem_actual / (uint64_t)(4 * 1024 * 1024 * 1024ULL)));
1113 }
1114 }
1115
1116 task_max = MAX(task_max, task_max_base * scale);
1117
1118 if (scale != 0) {
1119 task_threadmax = task_max;
1120 thread_max = task_max * 5;
1121 }
1122
1123 #endif
1124
1125 bsd_scale_setup(scale);
1126 }
1127