xref: /xnu-11215/osfmk/vm/vm_resident.c (revision d4514f0b)
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) 1991,1990,1989,1988,1987 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  */
58 /*
59  *	File:	vm/vm_page.c
60  *	Author:	Avadis Tevanian, Jr., Michael Wayne Young
61  *
62  *	Resident memory management module.
63  */
64 
65 #include <debug.h>
66 #include <libkern/OSAtomic.h>
67 #include <libkern/OSDebug.h>
68 
69 #include <mach/clock_types.h>
70 #include <mach/vm_prot.h>
71 #include <mach/vm_statistics.h>
72 #include <mach/sdt.h>
73 #include <kern/counter.h>
74 #include <kern/host_statistics.h>
75 #include <kern/sched_prim.h>
76 #include <kern/policy_internal.h>
77 #include <kern/task.h>
78 #include <kern/thread.h>
79 #include <kern/kalloc.h>
80 #include <kern/zalloc_internal.h>
81 #include <kern/ledger.h>
82 #include <kern/ecc.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_init_xnu.h>
85 #include <vm/vm_map_internal.h>
86 #include <vm/vm_page_internal.h>
87 #include <vm/vm_pageout_internal.h>
88 #include <vm/vm_kern_xnu.h>                 /* kmem_alloc() */
89 #include <vm/vm_compressor_pager_internal.h>
90 #include <kern/misc_protos.h>
91 #include <mach_debug/zone_info.h>
92 #include <vm/cpm_internal.h>
93 #include <pexpert/pexpert.h>
94 #include <pexpert/device_tree.h>
95 #include <san/kasan.h>
96 #include <os/log.h>
97 
98 #include <vm/vm_protos_internal.h>
99 #include <vm/memory_object.h>
100 #include <vm/vm_purgeable_internal.h>
101 #include <vm/vm_compressor_internal.h>
102 #include <vm/vm_iokit.h>
103 #include <vm/vm_object_internal.h>
104 #if defined (__x86_64__)
105 #include <i386/misc_protos.h>
106 #endif
107 
108 #if CONFIG_SPTM
109 #include <arm64/sptm/sptm.h>
110 #endif
111 
112 #if CONFIG_PHANTOM_CACHE
113 #include <vm/vm_phantom_cache_internal.h>
114 #endif
115 
116 #if HIBERNATION
117 #include <IOKit/IOHibernatePrivate.h>
118 #include <machine/pal_hibernate.h>
119 #endif /* HIBERNATION */
120 
121 #include <sys/kdebug.h>
122 
123 #if defined(HAS_APPLE_PAC)
124 #include <ptrauth.h>
125 #endif
126 #if defined(__arm64__)
127 #include <arm/cpu_internal.h>
128 #endif /* defined(__arm64__) */
129 
130 #if MACH_ASSERT
131 
132 TUNABLE(bool, vm_check_refs_on_alloc, "vm_check_refs_on_alloc", false);
133 #define ASSERT_PMAP_FREE(mem) pmap_assert_free(VM_PAGE_GET_PHYS_PAGE(mem))
134 
135 #else /* MACH_ASSERT */
136 
137 #define ASSERT_PMAP_FREE(mem) /* nothing */
138 
139 #endif /* MACH_ASSERT */
140 
141 extern boolean_t vm_pageout_running;
142 extern thread_t  vm_pageout_scan_thread;
143 extern bool vps_dynamic_priority_enabled;
144 
145 char    vm_page_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
146 char    vm_page_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
147 char    vm_page_non_speculative_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
148 char    vm_page_active_or_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
149 
150 #if CONFIG_SECLUDED_MEMORY
151 struct vm_page_secluded_data vm_page_secluded;
152 #endif /* CONFIG_SECLUDED_MEMORY */
153 
154 #if DEVELOPMENT || DEBUG
155 extern struct memory_object_pager_ops shared_region_pager_ops;
156 unsigned int shared_region_pagers_resident_count = 0;
157 unsigned int shared_region_pagers_resident_peak = 0;
158 #endif /* DEVELOPMENT || DEBUG */
159 
160 
161 
162 int             PERCPU_DATA(start_color);
163 vm_page_t       PERCPU_DATA(free_pages);
164 boolean_t       hibernate_cleaning_in_progress = FALSE;
165 
166 uint32_t        vm_lopage_free_count = 0;
167 uint32_t        vm_lopage_free_limit = 0;
168 uint32_t        vm_lopage_lowater    = 0;
169 boolean_t       vm_lopage_refill = FALSE;
170 boolean_t       vm_lopage_needed = FALSE;
171 
172 int             speculative_age_index = 0;
173 int             speculative_steal_index = 0;
174 struct vm_speculative_age_q vm_page_queue_speculative[VM_PAGE_RESERVED_SPECULATIVE_AGE_Q + 1];
175 
176 boolean_t       hibernation_vmqueues_inspection = FALSE; /* Tracks if the hibernation code is looking at the VM queues.
177                                                           * Updated and checked behind the vm_page_queues_lock. */
178 
179 static void             vm_page_free_prepare(vm_page_t  page);
180 static vm_page_t        vm_page_grab_fictitious_common(ppnum_t, boolean_t);
181 
182 static void vm_tag_init(void);
183 
184 /* for debugging purposes */
185 SECURITY_READ_ONLY_EARLY(uint32_t) vm_packed_from_vm_pages_array_mask =
186     VM_PAGE_PACKED_FROM_ARRAY;
187 SECURITY_READ_ONLY_EARLY(vm_packing_params_t) vm_page_packing_params =
188     VM_PACKING_PARAMS(VM_PAGE_PACKED_PTR);
189 
190 /*
191  *	Associated with page of user-allocatable memory is a
192  *	page structure.
193  */
194 
195 /*
196  *	These variables record the values returned by vm_page_bootstrap,
197  *	for debugging purposes.  The implementation of pmap_steal_memory
198  *	and pmap_startup here also uses them internally.
199  */
200 
201 vm_offset_t virtual_space_start;
202 vm_offset_t virtual_space_end;
203 uint32_t        vm_page_pages;
204 
205 /*
206  *	The vm_page_lookup() routine, which provides for fast
207  *	(virtual memory object, offset) to page lookup, employs
208  *	the following hash table.  The vm_page_{insert,remove}
209  *	routines install and remove associations in the table.
210  *	[This table is often called the virtual-to-physical,
211  *	or VP, table.]
212  */
213 typedef struct {
214 	vm_page_packed_t page_list;
215 #if     MACH_PAGE_HASH_STATS
216 	int             cur_count;              /* current count */
217 	int             hi_count;               /* high water mark */
218 #endif /* MACH_PAGE_HASH_STATS */
219 } vm_page_bucket_t;
220 
221 
222 #define BUCKETS_PER_LOCK        16
223 
224 SECURITY_READ_ONLY_LATE(vm_page_bucket_t *) vm_page_buckets;                /* Array of buckets */
225 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_bucket_count = 0;       /* How big is array? */
226 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_hash_mask;              /* Mask for hash function */
227 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_hash_shift;             /* Shift for hash function */
228 SECURITY_READ_ONLY_LATE(uint32_t)           vm_page_bucket_hash;            /* Basic bucket hash */
229 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_bucket_lock_count = 0;  /* How big is array of locks? */
230 
231 #ifndef VM_TAG_ACTIVE_UPDATE
232 #error VM_TAG_ACTIVE_UPDATE
233 #endif
234 #ifndef VM_TAG_SIZECLASSES
235 #error VM_TAG_SIZECLASSES
236 #endif
237 
238 /* for debugging */
239 SECURITY_READ_ONLY_LATE(bool) vm_tag_active_update = VM_TAG_ACTIVE_UPDATE;
240 SECURITY_READ_ONLY_LATE(lck_spin_t *) vm_page_bucket_locks;
241 
242 vm_allocation_site_t            vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC + 1];
243 vm_allocation_site_t *          vm_allocation_sites[VM_MAX_TAG_VALUE];
244 #if VM_TAG_SIZECLASSES
245 static vm_allocation_zone_total_t **vm_allocation_zone_totals;
246 #endif /* VM_TAG_SIZECLASSES */
247 
248 vm_tag_t vm_allocation_tag_highest;
249 
250 #if VM_PAGE_BUCKETS_CHECK
251 boolean_t vm_page_buckets_check_ready = FALSE;
252 #if VM_PAGE_FAKE_BUCKETS
253 vm_page_bucket_t *vm_page_fake_buckets; /* decoy buckets */
254 vm_map_offset_t vm_page_fake_buckets_start, vm_page_fake_buckets_end;
255 #endif /* VM_PAGE_FAKE_BUCKETS */
256 #endif /* VM_PAGE_BUCKETS_CHECK */
257 
258 #if     MACH_PAGE_HASH_STATS
259 /* This routine is only for debug.  It is intended to be called by
260  * hand by a developer using a kernel debugger.  This routine prints
261  * out vm_page_hash table statistics to the kernel debug console.
262  */
263 void
hash_debug(void)264 hash_debug(void)
265 {
266 	int     i;
267 	int     numbuckets = 0;
268 	int     highsum = 0;
269 	int     maxdepth = 0;
270 
271 	for (i = 0; i < vm_page_bucket_count; i++) {
272 		if (vm_page_buckets[i].hi_count) {
273 			numbuckets++;
274 			highsum += vm_page_buckets[i].hi_count;
275 			if (vm_page_buckets[i].hi_count > maxdepth) {
276 				maxdepth = vm_page_buckets[i].hi_count;
277 			}
278 		}
279 	}
280 	printf("Total number of buckets: %d\n", vm_page_bucket_count);
281 	printf("Number used buckets:     %d = %d%%\n",
282 	    numbuckets, 100 * numbuckets / vm_page_bucket_count);
283 	printf("Number unused buckets:   %d = %d%%\n",
284 	    vm_page_bucket_count - numbuckets,
285 	    100 * (vm_page_bucket_count - numbuckets) / vm_page_bucket_count);
286 	printf("Sum of bucket max depth: %d\n", highsum);
287 	printf("Average bucket depth:    %d.%2d\n",
288 	    highsum / vm_page_bucket_count,
289 	    highsum % vm_page_bucket_count);
290 	printf("Maximum bucket depth:    %d\n", maxdepth);
291 }
292 #endif /* MACH_PAGE_HASH_STATS */
293 
294 /*
295  *	The virtual page size is currently implemented as a runtime
296  *	variable, but is constant once initialized using vm_set_page_size.
297  *	This initialization must be done in the machine-dependent
298  *	bootstrap sequence, before calling other machine-independent
299  *	initializations.
300  *
301  *	All references to the virtual page size outside this
302  *	module must use the PAGE_SIZE, PAGE_MASK and PAGE_SHIFT
303  *	constants.
304  */
305 #if defined(__arm64__)
306 vm_size_t       page_size;
307 vm_size_t       page_mask;
308 int             page_shift;
309 #else
310 vm_size_t       page_size  = PAGE_SIZE;
311 vm_size_t       page_mask  = PAGE_MASK;
312 int             page_shift = PAGE_SHIFT;
313 #endif
314 
315 SECURITY_READ_ONLY_LATE(vm_page_t) vm_pages = VM_PAGE_NULL;
316 SECURITY_READ_ONLY_LATE(vm_page_t) vm_page_array_beginning_addr;
317 vm_page_t                          vm_page_array_ending_addr;
318 
319 unsigned int    vm_pages_count = 0;
320 
321 /*
322  *	Resident pages that represent real memory
323  *	are allocated from a set of free lists,
324  *	one per color.
325  */
326 unsigned int    vm_colors;
327 unsigned int    vm_color_mask;                  /* mask is == (vm_colors-1) */
328 unsigned int    vm_cache_geometry_colors = 0;   /* set by hw dependent code during startup */
329 unsigned int    vm_free_magazine_refill_limit = 0;
330 
331 
332 struct vm_page_queue_free_head {
333 	vm_page_queue_head_t    qhead;
334 } VM_PAGE_PACKED_ALIGNED;
335 
336 struct vm_page_queue_free_head  vm_page_queue_free[MAX_COLORS];
337 
338 
339 unsigned int    vm_page_free_wanted;
340 unsigned int    vm_page_free_wanted_privileged;
341 #if CONFIG_SECLUDED_MEMORY
342 unsigned int    vm_page_free_wanted_secluded;
343 #endif /* CONFIG_SECLUDED_MEMORY */
344 unsigned int    vm_page_free_count;
345 
346 unsigned int    vm_page_realtime_count;
347 
348 /*
349  *	Occasionally, the virtual memory system uses
350  *	resident page structures that do not refer to
351  *	real pages, for example to leave a page with
352  *	important state information in the VP table.
353  *
354  *	These page structures are allocated the way
355  *	most other kernel structures are.
356  */
357 SECURITY_READ_ONLY_LATE(zone_t) vm_page_zone;
358 vm_locks_array_t vm_page_locks;
359 
360 LCK_ATTR_DECLARE(vm_page_lck_attr, 0, 0);
361 LCK_GRP_DECLARE(vm_page_lck_grp_free, "vm_page_free");
362 LCK_GRP_DECLARE(vm_page_lck_grp_queue, "vm_page_queue");
363 LCK_GRP_DECLARE(vm_page_lck_grp_local, "vm_page_queue_local");
364 LCK_GRP_DECLARE(vm_page_lck_grp_purge, "vm_page_purge");
365 LCK_GRP_DECLARE(vm_page_lck_grp_alloc, "vm_page_alloc");
366 LCK_GRP_DECLARE(vm_page_lck_grp_bucket, "vm_page_bucket");
367 LCK_SPIN_DECLARE_ATTR(vm_objects_wired_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr);
368 LCK_TICKET_DECLARE(vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
369 
370 unsigned int    vm_page_local_q_soft_limit = 250;
371 unsigned int    vm_page_local_q_hard_limit = 500;
372 struct vpl     *__zpercpu vm_page_local_q;
373 
374 /* N.B. Guard and fictitious pages must not
375  * be assigned a zero phys_page value.
376  */
377 /*
378  *	Fictitious pages don't have a physical address,
379  *	but we must initialize phys_page to something.
380  *	For debugging, this should be a strange value
381  *	that the pmap module can recognize in assertions.
382  */
383 const ppnum_t vm_page_fictitious_addr = (ppnum_t) -1;
384 
385 /*
386  *	Guard pages are not accessible so they don't
387  *      need a physical address, but we need to enter
388  *	one in the pmap.
389  *	Let's make it recognizable and make sure that
390  *	we don't use a real physical page with that
391  *	physical address.
392  */
393 const ppnum_t vm_page_guard_addr = (ppnum_t) -2;
394 
395 /*
396  *	Resident page structures are also chained on
397  *	queues that are used by the page replacement
398  *	system (pageout daemon).  These queues are
399  *	defined here, but are shared by the pageout
400  *	module.  The inactive queue is broken into
401  *	file backed and anonymous for convenience as the
402  *	pageout daemon often assignes a higher
403  *	importance to anonymous pages (less likely to pick)
404  */
405 vm_page_queue_head_t    vm_page_queue_active VM_PAGE_PACKED_ALIGNED;
406 vm_page_queue_head_t    vm_page_queue_inactive VM_PAGE_PACKED_ALIGNED;
407 #if CONFIG_SECLUDED_MEMORY
408 vm_page_queue_head_t    vm_page_queue_secluded VM_PAGE_PACKED_ALIGNED;
409 #endif /* CONFIG_SECLUDED_MEMORY */
410 vm_page_queue_head_t    vm_page_queue_anonymous VM_PAGE_PACKED_ALIGNED;  /* inactive memory queue for anonymous pages */
411 vm_page_queue_head_t    vm_page_queue_throttled VM_PAGE_PACKED_ALIGNED;
412 
413 queue_head_t    vm_objects_wired;
414 
415 vm_page_queue_head_t    vm_page_queue_donate VM_PAGE_PACKED_ALIGNED;
416 uint32_t        vm_page_donate_mode;
417 uint32_t        vm_page_donate_target, vm_page_donate_target_high, vm_page_donate_target_low;
418 uint32_t        vm_page_donate_count;
419 bool            vm_page_donate_queue_ripe;
420 
421 
422 vm_page_queue_head_t    vm_page_queue_background VM_PAGE_PACKED_ALIGNED;
423 uint32_t        vm_page_background_target;
424 uint32_t        vm_page_background_target_snapshot;
425 uint32_t        vm_page_background_count;
426 uint64_t        vm_page_background_promoted_count;
427 
428 uint32_t        vm_page_background_internal_count;
429 uint32_t        vm_page_background_external_count;
430 
431 uint32_t        vm_page_background_mode;
432 uint32_t        vm_page_background_exclude_external;
433 
434 unsigned int    vm_page_active_count;
435 unsigned int    vm_page_inactive_count;
436 unsigned int    vm_page_kernelcache_count;
437 #if CONFIG_SECLUDED_MEMORY
438 unsigned int    vm_page_secluded_count;
439 unsigned int    vm_page_secluded_count_free;
440 unsigned int    vm_page_secluded_count_inuse;
441 unsigned int    vm_page_secluded_count_over_target;
442 #endif /* CONFIG_SECLUDED_MEMORY */
443 unsigned int    vm_page_anonymous_count;
444 unsigned int    vm_page_throttled_count;
445 unsigned int    vm_page_speculative_count;
446 
447 unsigned int    vm_page_wire_count;
448 unsigned int    vm_page_wire_count_on_boot = 0;
449 unsigned int    vm_page_stolen_count = 0;
450 unsigned int    vm_page_wire_count_initial;
451 unsigned int    vm_page_gobble_count = 0;
452 unsigned int    vm_page_kern_lpage_count = 0;
453 
454 uint64_t        booter_size;  /* external so it can be found in core dumps */
455 
456 #define VM_PAGE_WIRE_COUNT_WARNING      0
457 #define VM_PAGE_GOBBLE_COUNT_WARNING    0
458 
459 unsigned int    vm_page_purgeable_count = 0; /* # of pages purgeable now */
460 unsigned int    vm_page_purgeable_wired_count = 0; /* # of purgeable pages that are wired now */
461 uint64_t        vm_page_purged_count = 0;    /* total count of purged pages */
462 
463 unsigned int    vm_page_xpmapped_external_count = 0;
464 unsigned int    vm_page_external_count = 0;
465 unsigned int    vm_page_internal_count = 0;
466 unsigned int    vm_page_pageable_external_count = 0;
467 unsigned int    vm_page_pageable_internal_count = 0;
468 
469 #if DEVELOPMENT || DEBUG
470 unsigned int    vm_page_speculative_recreated = 0;
471 unsigned int    vm_page_speculative_created = 0;
472 unsigned int    vm_page_speculative_used = 0;
473 #endif
474 
475 vm_page_queue_head_t    vm_page_queue_cleaned VM_PAGE_PACKED_ALIGNED;
476 
477 unsigned int    vm_page_cleaned_count = 0;
478 
479 uint64_t        max_valid_dma_address = 0xffffffffffffffffULL;
480 ppnum_t         max_valid_low_ppnum = PPNUM_MAX;
481 
482 
483 /*
484  *	Several page replacement parameters are also
485  *	shared with this module, so that page allocation
486  *	(done here in vm_page_alloc) can trigger the
487  *	pageout daemon.
488  */
489 unsigned int    vm_page_free_target = 0;
490 unsigned int    vm_page_free_min = 0;
491 unsigned int    vm_page_throttle_limit = 0;
492 unsigned int    vm_page_inactive_target = 0;
493 #if CONFIG_SECLUDED_MEMORY
494 unsigned int    vm_page_secluded_target = 0;
495 #endif /* CONFIG_SECLUDED_MEMORY */
496 unsigned int    vm_page_anonymous_min = 0;
497 unsigned int    vm_page_free_reserved = 0;
498 
499 
500 /*
501  *	The VM system has a couple of heuristics for deciding
502  *	that pages are "uninteresting" and should be placed
503  *	on the inactive queue as likely candidates for replacement.
504  *	These variables let the heuristics be controlled at run-time
505  *	to make experimentation easier.
506  */
507 
508 boolean_t vm_page_deactivate_hint = TRUE;
509 
510 struct vm_page_stats_reusable vm_page_stats_reusable;
511 
512 /*
513  *	vm_set_page_size:
514  *
515  *	Sets the page size, perhaps based upon the memory
516  *	size.  Must be called before any use of page-size
517  *	dependent functions.
518  *
519  *	Sets page_shift and page_mask from page_size.
520  */
521 void
vm_set_page_size(void)522 vm_set_page_size(void)
523 {
524 	page_size  = PAGE_SIZE;
525 	page_mask  = PAGE_MASK;
526 	page_shift = PAGE_SHIFT;
527 
528 	if ((page_mask & page_size) != 0) {
529 		panic("vm_set_page_size: page size not a power of two");
530 	}
531 
532 	for (page_shift = 0;; page_shift++) {
533 		if ((1U << page_shift) == page_size) {
534 			break;
535 		}
536 	}
537 }
538 
539 #if defined (__x86_64__)
540 
541 #define MAX_CLUMP_SIZE      16
542 #define DEFAULT_CLUMP_SIZE  4
543 
544 unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold;
545 
546 #if DEVELOPMENT || DEBUG
547 unsigned long vm_clump_stats[MAX_CLUMP_SIZE + 1];
548 unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes;
549 
550 static inline void
vm_clump_update_stats(unsigned int c)551 vm_clump_update_stats(unsigned int c)
552 {
553 	assert(c <= vm_clump_size);
554 	if (c > 0 && c <= vm_clump_size) {
555 		vm_clump_stats[c] += c;
556 	}
557 	vm_clump_allocs += c;
558 }
559 #endif  /*  if DEVELOPMENT || DEBUG */
560 
561 /* Called once to setup the VM clump knobs */
562 static void
vm_page_setup_clump(void)563 vm_page_setup_clump( void )
564 {
565 	unsigned int override, n;
566 
567 	vm_clump_size = DEFAULT_CLUMP_SIZE;
568 	if (PE_parse_boot_argn("clump_size", &override, sizeof(override))) {
569 		vm_clump_size = override;
570 	}
571 
572 	if (vm_clump_size > MAX_CLUMP_SIZE) {
573 		panic("vm_page_setup_clump:: clump_size is too large!");
574 	}
575 	if (vm_clump_size < 1) {
576 		panic("vm_page_setup_clump:: clump_size must be >= 1");
577 	}
578 	if ((vm_clump_size & (vm_clump_size - 1)) != 0) {
579 		panic("vm_page_setup_clump:: clump_size must be a power of 2");
580 	}
581 
582 	vm_clump_promote_threshold = vm_clump_size;
583 	vm_clump_mask = vm_clump_size - 1;
584 	for (vm_clump_shift = 0, n = vm_clump_size; n > 1; n >>= 1, vm_clump_shift++) {
585 		;
586 	}
587 
588 #if DEVELOPMENT || DEBUG
589 	bzero(vm_clump_stats, sizeof(vm_clump_stats));
590 	vm_clump_allocs = vm_clump_inserts = vm_clump_inrange = vm_clump_promotes = 0;
591 #endif  /*  if DEVELOPMENT || DEBUG */
592 }
593 
594 #endif  /* #if defined (__x86_64__) */
595 
596 #define COLOR_GROUPS_TO_STEAL   4
597 
598 /* Called once during statup, once the cache geometry is known.
599  */
600 static void
vm_page_set_colors(void)601 vm_page_set_colors( void )
602 {
603 	unsigned int    n, override;
604 
605 #if defined (__x86_64__)
606 	/* adjust #colors because we need to color outside the clump boundary */
607 	vm_cache_geometry_colors >>= vm_clump_shift;
608 #endif
609 	if (PE_parse_boot_argn("colors", &override, sizeof(override))) {                /* colors specified as a boot-arg? */
610 		n = override;
611 	} else if (vm_cache_geometry_colors) {                  /* do we know what the cache geometry is? */
612 		n = vm_cache_geometry_colors;
613 	} else {
614 		n = DEFAULT_COLORS;                             /* use default if all else fails */
615 	}
616 	if (n == 0) {
617 		n = 1;
618 	}
619 	if (n > MAX_COLORS) {
620 		n = MAX_COLORS;
621 	}
622 
623 	/* the count must be a power of 2  */
624 	if ((n & (n - 1)) != 0) {
625 		n = DEFAULT_COLORS;                             /* use default if all else fails */
626 	}
627 	vm_colors = n;
628 	vm_color_mask = n - 1;
629 
630 	vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL;
631 
632 #if defined (__x86_64__)
633 	/* adjust for reduction in colors due to clumping and multiple cores */
634 	if (real_ncpus) {
635 		vm_free_magazine_refill_limit *= (vm_clump_size * real_ncpus);
636 	}
637 #endif
638 }
639 
640 /*
641  * During single threaded early boot we don't initialize all pages.
642  * This avoids some delay during boot. They'll be initialized and
643  * added to the free list as needed or after we are multithreaded by
644  * what becomes the pageout thread.
645  */
646 static boolean_t fill = FALSE;
647 static unsigned int fillval;
648 uint_t vm_delayed_count = 0;    /* when non-zero, indicates we may have more pages to init */
649 ppnum_t delay_above_pnum = PPNUM_MAX;
650 
651 /*
652  * For x86 first 8 Gig initializes quickly and gives us lots of lowmem + mem above to start off with.
653  * If ARM ever uses delayed page initialization, this value may need to be quite different.
654  */
655 #define DEFAULT_DELAY_ABOVE_PHYS_GB (8)
656 
657 /*
658  * When we have to dip into more delayed pages due to low memory, free up
659  * a large chunk to get things back to normal. This avoids contention on the
660  * delayed code allocating page by page.
661  */
662 #define VM_DELAY_PAGE_CHUNK ((1024 * 1024 * 1024) / PAGE_SIZE)
663 
664 /*
665  * Get and initialize the next delayed page.
666  */
667 static vm_page_t
vm_get_delayed_page(int grab_options)668 vm_get_delayed_page(int grab_options)
669 {
670 	vm_page_t p;
671 	ppnum_t   pnum;
672 
673 	/*
674 	 * Get a new page if we have one.
675 	 */
676 	vm_free_page_lock();
677 	if (vm_delayed_count == 0) {
678 		vm_free_page_unlock();
679 		return NULL;
680 	}
681 
682 	if (!pmap_next_page(&pnum)) {
683 		vm_delayed_count = 0;
684 		vm_free_page_unlock();
685 		return NULL;
686 	}
687 
688 
689 	assert(vm_delayed_count > 0);
690 	--vm_delayed_count;
691 
692 #if defined(__x86_64__)
693 	/* x86 cluster code requires increasing phys_page in vm_pages[] */
694 	if (vm_pages_count > 0) {
695 		assert(pnum > vm_pages[vm_pages_count - 1].vmp_phys_page);
696 	}
697 #endif
698 	p = &vm_pages[vm_pages_count];
699 	assert(p < vm_page_array_ending_addr);
700 	vm_page_init(p, pnum, FALSE);
701 	++vm_pages_count;
702 	++vm_page_pages;
703 	vm_free_page_unlock();
704 
705 	/*
706 	 * These pages were initially counted as wired, undo that now.
707 	 */
708 	if (grab_options & VM_PAGE_GRAB_Q_LOCK_HELD) {
709 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
710 	} else {
711 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
712 		vm_page_lockspin_queues();
713 	}
714 	--vm_page_wire_count;
715 	--vm_page_wire_count_initial;
716 	if (vm_page_wire_count_on_boot != 0) {
717 		--vm_page_wire_count_on_boot;
718 	}
719 	if (!(grab_options & VM_PAGE_GRAB_Q_LOCK_HELD)) {
720 		vm_page_unlock_queues();
721 	}
722 
723 
724 	if (fill) {
725 		fillPage(pnum, fillval);
726 	}
727 	return p;
728 }
729 
730 static void vm_page_module_init_delayed(void);
731 
732 /*
733  * Free all remaining delayed pages to the free lists.
734  */
735 void
vm_free_delayed_pages(void)736 vm_free_delayed_pages(void)
737 {
738 	vm_page_t   p;
739 	vm_page_t   list = NULL;
740 	uint_t      cnt = 0;
741 	vm_offset_t start_free_va;
742 	int64_t     free_size;
743 
744 	while ((p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE)) != NULL) {
745 		if (vm_himemory_mode) {
746 			vm_page_release(p, FALSE);
747 		} else {
748 			p->vmp_snext = list;
749 			list = p;
750 		}
751 		++cnt;
752 	}
753 
754 	/*
755 	 * Free the pages in reverse order if not himemory mode.
756 	 * Hence the low memory pages will be first on free lists. (LIFO)
757 	 */
758 	while (list != NULL) {
759 		p = list;
760 		list = p->vmp_snext;
761 		p->vmp_snext = NULL;
762 		vm_page_release(p, FALSE);
763 	}
764 #if DEVELOPMENT || DEBUG
765 	kprintf("vm_free_delayed_pages: initialized %d free pages\n", cnt);
766 #endif
767 
768 	/*
769 	 * Free up any unused full pages at the end of the vm_pages[] array
770 	 */
771 	start_free_va = round_page((vm_offset_t)&vm_pages[vm_pages_count]);
772 
773 #if defined(__x86_64__)
774 	/*
775 	 * Since x86 might have used large pages for vm_pages[], we can't
776 	 * free starting in the middle of a partially used large page.
777 	 */
778 	if (pmap_query_pagesize(kernel_pmap, start_free_va) == I386_LPGBYTES) {
779 		start_free_va = ((start_free_va + I386_LPGMASK) & ~I386_LPGMASK);
780 	}
781 #endif
782 	if (start_free_va < (vm_offset_t)vm_page_array_ending_addr) {
783 		free_size = trunc_page((vm_offset_t)vm_page_array_ending_addr - start_free_va);
784 		if (free_size > 0) {
785 			ml_static_mfree(start_free_va, (vm_offset_t)free_size);
786 			vm_page_array_ending_addr = (void *)start_free_va;
787 
788 			/*
789 			 * Note there's no locking here, as only this thread will ever change this value.
790 			 * The reader, vm_page_diagnose, doesn't grab any locks for the counts it looks at.
791 			 */
792 			vm_page_stolen_count -= (free_size >> PAGE_SHIFT);
793 
794 #if DEVELOPMENT || DEBUG
795 			kprintf("Freeing final unused %ld bytes from vm_pages[] at 0x%lx\n",
796 			    (long)free_size, (long)start_free_va);
797 #endif
798 		}
799 	}
800 
801 
802 	/*
803 	 * now we can create the VM page array zone
804 	 */
805 	vm_page_module_init_delayed();
806 }
807 
808 /*
809  * Try and free up enough delayed pages to match a contig memory allocation.
810  */
811 static void
vm_free_delayed_pages_contig(uint_t npages,ppnum_t max_pnum,ppnum_t pnum_mask)812 vm_free_delayed_pages_contig(
813 	uint_t    npages,
814 	ppnum_t   max_pnum,
815 	ppnum_t   pnum_mask)
816 {
817 	vm_page_t p;
818 	ppnum_t   pnum;
819 	uint_t    cnt = 0;
820 
821 	/*
822 	 * Treat 0 as the absolute max page number.
823 	 */
824 	if (max_pnum == 0) {
825 		max_pnum = PPNUM_MAX;
826 	}
827 
828 	/*
829 	 * Free till we get a properly aligned start page
830 	 */
831 	for (;;) {
832 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
833 		if (p == NULL) {
834 			return;
835 		}
836 		pnum = VM_PAGE_GET_PHYS_PAGE(p);
837 		vm_page_release(p, FALSE);
838 		if (pnum >= max_pnum) {
839 			return;
840 		}
841 		if ((pnum & pnum_mask) == 0) {
842 			break;
843 		}
844 	}
845 
846 	/*
847 	 * Having a healthy pool of free pages will help performance. We don't
848 	 * want to fall back to the delayed code for every page allocation.
849 	 */
850 	if (vm_page_free_count < VM_DELAY_PAGE_CHUNK) {
851 		npages += VM_DELAY_PAGE_CHUNK;
852 	}
853 
854 	/*
855 	 * Now free up the pages
856 	 */
857 	for (cnt = 1; cnt < npages; ++cnt) {
858 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
859 		if (p == NULL) {
860 			return;
861 		}
862 		vm_page_release(p, FALSE);
863 	}
864 }
865 
866 #define ROUNDUP_NEXTP2(X) (1U << (32 - __builtin_clz((X) - 1)))
867 
868 void
vm_page_init_local_q(unsigned int num_cpus)869 vm_page_init_local_q(unsigned int num_cpus)
870 {
871 	struct vpl *t_local_q;
872 
873 	/*
874 	 * no point in this for a uni-processor system
875 	 */
876 	if (num_cpus >= 2) {
877 		ml_cpu_info_t cpu_info;
878 
879 		/*
880 		 * Force the allocation alignment to a cacheline,
881 		 * because the `vpl` struct has a lock and will be taken
882 		 * cross CPU so we want to isolate the rest of the per-CPU
883 		 * data to avoid false sharing due to this lock being taken.
884 		 */
885 
886 		ml_cpu_get_info(&cpu_info);
887 
888 		t_local_q = zalloc_percpu_permanent(sizeof(struct vpl),
889 		    cpu_info.cache_line_size - 1);
890 
891 		zpercpu_foreach(lq, t_local_q) {
892 			VPL_LOCK_INIT(lq, &vm_page_lck_grp_local, &vm_page_lck_attr);
893 			vm_page_queue_init(&lq->vpl_queue);
894 		}
895 
896 		/* make the initialization visible to all cores */
897 		os_atomic_store(&vm_page_local_q, t_local_q, release);
898 	}
899 }
900 
901 /*
902  * vm_init_before_launchd
903  *
904  * This should be called right before launchd is loaded.
905  */
906 void
vm_init_before_launchd()907 vm_init_before_launchd()
908 {
909 	vm_page_lockspin_queues();
910 	vm_page_wire_count_on_boot = vm_page_wire_count;
911 	vm_page_unlock_queues();
912 }
913 
914 
915 /*
916  *	vm_page_bootstrap:
917  *
918  *	Initializes the resident memory module.
919  *
920  *	Allocates memory for the page cells, and
921  *	for the object/offset-to-page hash table headers.
922  *	Each page cell is initialized and placed on the free list.
923  *	Returns the range of available kernel virtual memory.
924  */
925 __startup_func
926 void
vm_page_bootstrap(vm_offset_t * startp,vm_offset_t * endp)927 vm_page_bootstrap(
928 	vm_offset_t             *startp,
929 	vm_offset_t             *endp)
930 {
931 	unsigned int            i;
932 	unsigned int            log1;
933 	unsigned int            log2;
934 	unsigned int            size;
935 
936 	/*
937 	 *	Initialize the page queues.
938 	 */
939 
940 	lck_mtx_init(&vm_page_queue_free_lock, &vm_page_lck_grp_free, &vm_page_lck_attr);
941 	lck_mtx_init(&vm_page_queue_lock, &vm_page_lck_grp_queue, &vm_page_lck_attr);
942 	lck_mtx_init(&vm_purgeable_queue_lock, &vm_page_lck_grp_purge, &vm_page_lck_attr);
943 
944 	for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) {
945 		int group;
946 
947 		purgeable_queues[i].token_q_head = 0;
948 		purgeable_queues[i].token_q_tail = 0;
949 		for (group = 0; group < NUM_VOLATILE_GROUPS; group++) {
950 			queue_init(&purgeable_queues[i].objq[group]);
951 		}
952 
953 		purgeable_queues[i].type = i;
954 		purgeable_queues[i].new_pages = 0;
955 #if MACH_ASSERT
956 		purgeable_queues[i].debug_count_tokens = 0;
957 		purgeable_queues[i].debug_count_objects = 0;
958 #endif
959 	}
960 	;
961 	purgeable_nonvolatile_count = 0;
962 	queue_init(&purgeable_nonvolatile_queue);
963 
964 	for (i = 0; i < MAX_COLORS; i++) {
965 		vm_page_queue_init(&vm_page_queue_free[i].qhead);
966 	}
967 
968 	vm_page_queue_init(&vm_lopage_queue_free);
969 	vm_page_queue_init(&vm_page_queue_active);
970 	vm_page_queue_init(&vm_page_queue_inactive);
971 #if CONFIG_SECLUDED_MEMORY
972 	vm_page_queue_init(&vm_page_queue_secluded);
973 #endif /* CONFIG_SECLUDED_MEMORY */
974 	vm_page_queue_init(&vm_page_queue_cleaned);
975 	vm_page_queue_init(&vm_page_queue_throttled);
976 	vm_page_queue_init(&vm_page_queue_anonymous);
977 	queue_init(&vm_objects_wired);
978 
979 	for (i = 0; i <= vm_page_max_speculative_age_q; i++) {
980 		vm_page_queue_init(&vm_page_queue_speculative[i].age_q);
981 
982 		vm_page_queue_speculative[i].age_ts.tv_sec = 0;
983 		vm_page_queue_speculative[i].age_ts.tv_nsec = 0;
984 	}
985 
986 	vm_page_queue_init(&vm_page_queue_donate);
987 	vm_page_queue_init(&vm_page_queue_background);
988 
989 	vm_page_background_count = 0;
990 	vm_page_background_internal_count = 0;
991 	vm_page_background_external_count = 0;
992 	vm_page_background_promoted_count = 0;
993 
994 	vm_page_background_target = (unsigned int)(atop_64(max_mem) / 25);
995 
996 	if (vm_page_background_target > VM_PAGE_BACKGROUND_TARGET_MAX) {
997 		vm_page_background_target = VM_PAGE_BACKGROUND_TARGET_MAX;
998 	}
999 
1000 #if    defined(__LP64__)
1001 	vm_page_background_mode = VM_PAGE_BG_ENABLED;
1002 	vm_page_donate_mode = VM_PAGE_DONATE_ENABLED;
1003 #else
1004 	vm_page_background_mode = VM_PAGE_BG_DISABLED;
1005 	vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
1006 #endif
1007 	vm_page_background_exclude_external = 0;
1008 
1009 	PE_parse_boot_argn("vm_page_bg_mode", &vm_page_background_mode, sizeof(vm_page_background_mode));
1010 	PE_parse_boot_argn("vm_page_bg_exclude_external", &vm_page_background_exclude_external, sizeof(vm_page_background_exclude_external));
1011 	PE_parse_boot_argn("vm_page_bg_target", &vm_page_background_target, sizeof(vm_page_background_target));
1012 
1013 	if (vm_page_background_mode != VM_PAGE_BG_DISABLED && vm_page_background_mode != VM_PAGE_BG_ENABLED) {
1014 		vm_page_background_mode = VM_PAGE_BG_DISABLED;
1015 	}
1016 
1017 	PE_parse_boot_argn("vm_page_donate_mode", &vm_page_donate_mode, sizeof(vm_page_donate_mode));
1018 	if (vm_page_donate_mode != VM_PAGE_DONATE_DISABLED && vm_page_donate_mode != VM_PAGE_DONATE_ENABLED) {
1019 		vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
1020 	}
1021 
1022 	vm_page_donate_target_high = VM_PAGE_DONATE_TARGET_HIGHWATER;
1023 	vm_page_donate_target_low = VM_PAGE_DONATE_TARGET_LOWWATER;
1024 	vm_page_donate_target = vm_page_donate_target_high;
1025 	vm_page_donate_count = 0;
1026 
1027 	vm_page_free_wanted = 0;
1028 	vm_page_free_wanted_privileged = 0;
1029 #if CONFIG_SECLUDED_MEMORY
1030 	vm_page_free_wanted_secluded = 0;
1031 #endif /* CONFIG_SECLUDED_MEMORY */
1032 
1033 #if defined (__x86_64__)
1034 	/* this must be called before vm_page_set_colors() */
1035 	vm_page_setup_clump();
1036 #endif
1037 
1038 	vm_page_set_colors();
1039 
1040 	bzero(vm_page_inactive_states, sizeof(vm_page_inactive_states));
1041 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1042 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1043 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1044 
1045 	bzero(vm_page_pageable_states, sizeof(vm_page_pageable_states));
1046 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1047 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1048 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1049 	vm_page_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1050 	vm_page_pageable_states[VM_PAGE_ON_SPECULATIVE_Q] = 1;
1051 	vm_page_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1052 #if CONFIG_SECLUDED_MEMORY
1053 	vm_page_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1054 #endif /* CONFIG_SECLUDED_MEMORY */
1055 
1056 	bzero(vm_page_non_speculative_pageable_states, sizeof(vm_page_non_speculative_pageable_states));
1057 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1058 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1059 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1060 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1061 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1062 #if CONFIG_SECLUDED_MEMORY
1063 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1064 #endif /* CONFIG_SECLUDED_MEMORY */
1065 
1066 	bzero(vm_page_active_or_inactive_states, sizeof(vm_page_active_or_inactive_states));
1067 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1068 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1069 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1070 	vm_page_active_or_inactive_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1071 #if CONFIG_SECLUDED_MEMORY
1072 	vm_page_active_or_inactive_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1073 #endif /* CONFIG_SECLUDED_MEMORY */
1074 
1075 	for (vm_tag_t t = 0; t < VM_KERN_MEMORY_FIRST_DYNAMIC; t++) {
1076 		vm_allocation_sites_static[t].refcount = 2;
1077 		vm_allocation_sites_static[t].tag = t;
1078 		vm_allocation_sites[t] = &vm_allocation_sites_static[t];
1079 	}
1080 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].refcount = 2;
1081 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].tag = VM_KERN_MEMORY_ANY;
1082 	vm_allocation_sites[VM_KERN_MEMORY_ANY] = &vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC];
1083 
1084 	/*
1085 	 *	Steal memory for the map and zone subsystems.
1086 	 */
1087 	kernel_startup_initialize_upto(STARTUP_SUB_PMAP_STEAL);
1088 
1089 	/*
1090 	 *	Allocate (and initialize) the virtual-to-physical
1091 	 *	table hash buckets.
1092 	 *
1093 	 *	The number of buckets should be a power of two to
1094 	 *	get a good hash function.  The following computation
1095 	 *	chooses the first power of two that is greater
1096 	 *	than the number of physical pages in the system.
1097 	 */
1098 
1099 	if (vm_page_bucket_count == 0) {
1100 		unsigned int npages = pmap_free_pages();
1101 
1102 		vm_page_bucket_count = 1;
1103 		while (vm_page_bucket_count < npages) {
1104 			vm_page_bucket_count <<= 1;
1105 		}
1106 	}
1107 	vm_page_bucket_lock_count = (vm_page_bucket_count + BUCKETS_PER_LOCK - 1) / BUCKETS_PER_LOCK;
1108 
1109 	vm_page_hash_mask = vm_page_bucket_count - 1;
1110 
1111 	/*
1112 	 *	Calculate object shift value for hashing algorithm:
1113 	 *		O = log2(sizeof(struct vm_object))
1114 	 *		B = log2(vm_page_bucket_count)
1115 	 *	        hash shifts the object left by
1116 	 *		B/2 - O
1117 	 */
1118 	size = vm_page_bucket_count;
1119 	for (log1 = 0; size > 1; log1++) {
1120 		size /= 2;
1121 	}
1122 	size = sizeof(struct vm_object);
1123 	for (log2 = 0; size > 1; log2++) {
1124 		size /= 2;
1125 	}
1126 	vm_page_hash_shift = log1 / 2 - log2 + 1;
1127 
1128 	vm_page_bucket_hash = 1 << ((log1 + 1) >> 1);           /* Get (ceiling of sqrt of table size) */
1129 	vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2);          /* Get (ceiling of quadroot of table size) */
1130 	vm_page_bucket_hash |= 1;                                                       /* Set bit and add 1 - always must be 1 to insure unique series */
1131 
1132 	if (vm_page_hash_mask & vm_page_bucket_count) {
1133 		printf("vm_page_bootstrap: WARNING -- strange page hash\n");
1134 	}
1135 
1136 #if VM_PAGE_BUCKETS_CHECK
1137 #if VM_PAGE_FAKE_BUCKETS
1138 	/*
1139 	 * Allocate a decoy set of page buckets, to detect
1140 	 * any stomping there.
1141 	 */
1142 	vm_page_fake_buckets = (vm_page_bucket_t *)
1143 	    pmap_steal_memory(vm_page_bucket_count *
1144 	    sizeof(vm_page_bucket_t), 0);
1145 	vm_page_fake_buckets_start = (vm_map_offset_t) vm_page_fake_buckets;
1146 	vm_page_fake_buckets_end =
1147 	    vm_map_round_page((vm_page_fake_buckets_start +
1148 	    (vm_page_bucket_count *
1149 	    sizeof(vm_page_bucket_t))),
1150 	    PAGE_MASK);
1151 	char *cp;
1152 	for (cp = (char *)vm_page_fake_buckets_start;
1153 	    cp < (char *)vm_page_fake_buckets_end;
1154 	    cp++) {
1155 		*cp = 0x5a;
1156 	}
1157 #endif /* VM_PAGE_FAKE_BUCKETS */
1158 #endif /* VM_PAGE_BUCKETS_CHECK */
1159 
1160 	kernel_debug_string_early("vm_page_buckets");
1161 	vm_page_buckets = (vm_page_bucket_t *)
1162 	    pmap_steal_memory(vm_page_bucket_count *
1163 	    sizeof(vm_page_bucket_t), 0);
1164 
1165 	kernel_debug_string_early("vm_page_bucket_locks");
1166 	vm_page_bucket_locks = (lck_spin_t *)
1167 	    pmap_steal_memory(vm_page_bucket_lock_count *
1168 	    sizeof(lck_spin_t), 0);
1169 
1170 	for (i = 0; i < vm_page_bucket_count; i++) {
1171 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
1172 
1173 		bucket->page_list = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
1174 #if     MACH_PAGE_HASH_STATS
1175 		bucket->cur_count = 0;
1176 		bucket->hi_count = 0;
1177 #endif /* MACH_PAGE_HASH_STATS */
1178 	}
1179 
1180 	for (i = 0; i < vm_page_bucket_lock_count; i++) {
1181 		lck_spin_init(&vm_page_bucket_locks[i], &vm_page_lck_grp_bucket, &vm_page_lck_attr);
1182 	}
1183 
1184 	vm_tag_init();
1185 
1186 #if VM_PAGE_BUCKETS_CHECK
1187 	vm_page_buckets_check_ready = TRUE;
1188 #endif /* VM_PAGE_BUCKETS_CHECK */
1189 
1190 	/*
1191 	 *	Machine-dependent code allocates the resident page table.
1192 	 *	It uses vm_page_init to initialize the page frames.
1193 	 *	The code also returns to us the virtual space available
1194 	 *	to the kernel.  We don't trust the pmap module
1195 	 *	to get the alignment right.
1196 	 */
1197 
1198 	kernel_debug_string_early("pmap_startup");
1199 	pmap_startup(&virtual_space_start, &virtual_space_end);
1200 	virtual_space_start = round_page(virtual_space_start);
1201 	virtual_space_end = trunc_page(virtual_space_end);
1202 
1203 	*startp = virtual_space_start;
1204 	*endp = virtual_space_end;
1205 
1206 	/*
1207 	 *	Compute the initial "wire" count.
1208 	 *	Up until now, the pages which have been set aside are not under
1209 	 *	the VM system's control, so although they aren't explicitly
1210 	 *	wired, they nonetheless can't be moved. At this moment,
1211 	 *	all VM managed pages are "free", courtesy of pmap_startup.
1212 	 */
1213 	assert((unsigned int) atop_64(max_mem) == atop_64(max_mem));
1214 	vm_page_wire_count = ((unsigned int) atop_64(max_mem)) -
1215 	    vm_page_free_count - vm_lopage_free_count;
1216 #if CONFIG_SECLUDED_MEMORY
1217 	vm_page_wire_count -= vm_page_secluded_count;
1218 #endif
1219 	vm_page_wire_count_initial = vm_page_wire_count;
1220 
1221 	/* capture this for later use */
1222 	booter_size = ml_get_booter_memory_size();
1223 
1224 	printf("vm_page_bootstrap: %d free pages, %d wired pages, (up to %d of which are delayed free)\n",
1225 	    vm_page_free_count, vm_page_wire_count, vm_delayed_count);
1226 
1227 	kernel_debug_string_early("vm_page_bootstrap complete");
1228 }
1229 
1230 #ifndef MACHINE_PAGES
1231 /*
1232  * This is the early boot time allocator for data structures needed to bootstrap the VM system.
1233  * On x86 it will allocate large pages if size is sufficiently large. We don't need to do this
1234  * on ARM yet, due to the combination of a large base page size and smaller RAM devices.
1235  */
1236 static void *
pmap_steal_memory_internal(vm_size_t size,vm_size_t alignment,boolean_t might_free,unsigned int flags,pmap_mapping_type_t mapping_type)1237 pmap_steal_memory_internal(
1238 	vm_size_t size,
1239 	vm_size_t alignment,
1240 	boolean_t might_free,
1241 	unsigned int flags,
1242 	pmap_mapping_type_t mapping_type)
1243 {
1244 	kern_return_t kr;
1245 	vm_offset_t addr;
1246 	vm_offset_t map_addr;
1247 	ppnum_t phys_page;
1248 	unsigned int pmap_flags;
1249 
1250 	/*
1251 	 * Size needs to be aligned to word size.
1252 	 */
1253 	size = (size + sizeof(void *) - 1) & ~(sizeof(void *) - 1);
1254 
1255 	/*
1256 	 * Alignment defaults to word size if not specified.
1257 	 */
1258 	if (alignment == 0) {
1259 		alignment = sizeof(void*);
1260 	}
1261 
1262 	/*
1263 	 * Alignment must be no greater than a page and must be a power of two.
1264 	 */
1265 	assert(alignment <= PAGE_SIZE);
1266 	assert((alignment & (alignment - 1)) == 0);
1267 
1268 	/*
1269 	 * On the first call, get the initial values for virtual address space
1270 	 * and page align them.
1271 	 */
1272 	if (virtual_space_start == virtual_space_end) {
1273 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
1274 		virtual_space_start = round_page(virtual_space_start);
1275 		virtual_space_end = trunc_page(virtual_space_end);
1276 
1277 #if defined(__x86_64__)
1278 		/*
1279 		 * Release remaining unused section of preallocated KVA and the 4K page tables
1280 		 * that map it. This makes the VA available for large page mappings.
1281 		 */
1282 		Idle_PTs_release(virtual_space_start, virtual_space_end);
1283 #endif
1284 	}
1285 
1286 	/*
1287 	 * Allocate the virtual space for this request. On x86, we'll align to a large page
1288 	 * address if the size is big enough to back with at least 1 large page.
1289 	 */
1290 #if defined(__x86_64__)
1291 	if (size >= I386_LPGBYTES) {
1292 		virtual_space_start = ((virtual_space_start + I386_LPGMASK) & ~I386_LPGMASK);
1293 	}
1294 #endif
1295 	virtual_space_start = (virtual_space_start + (alignment - 1)) & ~(alignment - 1);
1296 	addr = virtual_space_start;
1297 	virtual_space_start += size;
1298 
1299 	//kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)size);	/* (TEST/DEBUG) */
1300 
1301 	/*
1302 	 * Allocate and map physical pages to back the new virtual space.
1303 	 */
1304 	map_addr = round_page(addr);
1305 	while (map_addr < addr + size) {
1306 #if defined(__x86_64__)
1307 		/*
1308 		 * Back with a large page if properly aligned on x86
1309 		 */
1310 		if ((map_addr & I386_LPGMASK) == 0 &&
1311 		    map_addr + I386_LPGBYTES <= addr + size &&
1312 		    pmap_pre_expand_large(kernel_pmap, map_addr) == KERN_SUCCESS &&
1313 		    pmap_next_page_large(&phys_page) == KERN_SUCCESS) {
1314 			kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1315 			    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1316 			    VM_WIMG_USE_DEFAULT | VM_MEM_SUPERPAGE, FALSE, mapping_type);
1317 
1318 			if (kr != KERN_SUCCESS) {
1319 				panic("pmap_steal_memory: pmap_enter() large failed, new_addr=%#lx, phys_page=%u",
1320 				    (unsigned long)map_addr, phys_page);
1321 			}
1322 			map_addr += I386_LPGBYTES;
1323 			vm_page_wire_count += I386_LPGBYTES >> PAGE_SHIFT;
1324 			vm_page_stolen_count += I386_LPGBYTES >> PAGE_SHIFT;
1325 			vm_page_kern_lpage_count++;
1326 			continue;
1327 		}
1328 #endif
1329 
1330 		if (!pmap_next_page_hi(&phys_page, might_free)) {
1331 			panic("pmap_steal_memory() size: 0x%llx", (uint64_t)size);
1332 		}
1333 
1334 #if defined(__x86_64__)
1335 		pmap_pre_expand(kernel_pmap, map_addr);
1336 #endif
1337 		pmap_flags = flags ? flags : VM_WIMG_USE_DEFAULT;
1338 
1339 		kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1340 		    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1341 		    pmap_flags, FALSE, mapping_type);
1342 
1343 		if (kr != KERN_SUCCESS) {
1344 			panic("pmap_steal_memory() pmap_enter failed, map_addr=%#lx, phys_page=%u",
1345 			    (unsigned long)map_addr, phys_page);
1346 		}
1347 		map_addr += PAGE_SIZE;
1348 
1349 		/*
1350 		 * Account for newly stolen memory
1351 		 */
1352 		vm_page_wire_count++;
1353 		vm_page_stolen_count++;
1354 	}
1355 
1356 #if defined(__x86_64__)
1357 	/*
1358 	 * The call with might_free is currently the last use of pmap_steal_memory*().
1359 	 * Notify the pmap layer to record which high pages were allocated so far.
1360 	 */
1361 	if (might_free) {
1362 		pmap_hi_pages_done();
1363 	}
1364 #endif
1365 #if KASAN
1366 	kasan_notify_address(round_page(addr), size);
1367 #endif
1368 	return (void *) addr;
1369 }
1370 
1371 void *
pmap_steal_memory(vm_size_t size,vm_size_t alignment)1372 pmap_steal_memory(
1373 	vm_size_t size,
1374 	vm_size_t alignment)
1375 {
1376 	return pmap_steal_memory_internal(size, alignment, FALSE, 0, PMAP_MAPPING_TYPE_RESTRICTED);
1377 }
1378 
1379 void *
pmap_steal_freeable_memory(vm_size_t size)1380 pmap_steal_freeable_memory(
1381 	vm_size_t size)
1382 {
1383 	return pmap_steal_memory_internal(size, 0, TRUE, 0, PMAP_MAPPING_TYPE_RESTRICTED);
1384 }
1385 
1386 
1387 
1388 
1389 #if CONFIG_SECLUDED_MEMORY
1390 /* boot-args to control secluded memory */
1391 TUNABLE_DT(unsigned int, secluded_mem_mb, "/defaults", "kern.secluded_mem_mb", "secluded_mem_mb", 0, TUNABLE_DT_NONE);
1392 /* IOKit can use secluded memory */
1393 TUNABLE(bool, secluded_for_iokit, "secluded_for_iokit", true);
1394 /* apps can use secluded memory */
1395 TUNABLE(bool, secluded_for_apps, "secluded_for_apps", true);
1396 /* filecache can use seclude memory */
1397 TUNABLE(secluded_filecache_mode_t, secluded_for_filecache, "secluded_for_filecache", SECLUDED_FILECACHE_RDONLY);
1398 uint64_t secluded_shutoff_trigger = 0;
1399 uint64_t secluded_shutoff_headroom = 150 * 1024 * 1024; /* original value from N56 */
1400 #endif /* CONFIG_SECLUDED_MEMORY */
1401 
1402 
1403 #if defined(__arm64__)
1404 extern void patch_low_glo_vm_page_info(void *, void *, uint32_t);
1405 unsigned int vm_first_phys_ppnum = 0;
1406 #endif
1407 
1408 void vm_page_release_startup(vm_page_t mem);
1409 void
pmap_startup(vm_offset_t * startp,vm_offset_t * endp)1410 pmap_startup(
1411 	vm_offset_t     *startp,
1412 	vm_offset_t     *endp)
1413 {
1414 	unsigned int    i, npages;
1415 	ppnum_t         phys_page;
1416 	uint64_t        mem_sz;
1417 	uint64_t        start_ns;
1418 	uint64_t        now_ns;
1419 	uint_t          low_page_count = 0;
1420 
1421 #if    defined(__LP64__)
1422 	/*
1423 	 * make sure we are aligned on a 64 byte boundary
1424 	 * for VM_PAGE_PACK_PTR (it clips off the low-order
1425 	 * 6 bits of the pointer)
1426 	 */
1427 	if (virtual_space_start != virtual_space_end) {
1428 		virtual_space_start = round_page(virtual_space_start);
1429 	}
1430 #endif
1431 
1432 	/*
1433 	 * We calculate how many page frames we will have
1434 	 * and then allocate the page structures in one chunk.
1435 	 *
1436 	 * Note that the calculation here doesn't take into account
1437 	 * the memory needed to map what's being allocated, i.e. the page
1438 	 * table entries. So the actual number of pages we get will be
1439 	 * less than this. To do someday: include that in the computation.
1440 	 *
1441 	 * Also for ARM, we don't use the count of free_pages, but rather the
1442 	 * range from last page to first page (ignore holes due to retired pages).
1443 	 */
1444 #if defined(__arm64__)
1445 	mem_sz = pmap_free_pages_span() * (uint64_t)PAGE_SIZE;
1446 #else /* defined(__arm64__) */
1447 	mem_sz = pmap_free_pages() * (uint64_t)PAGE_SIZE;
1448 #endif /* defined(__arm64__) */
1449 	mem_sz += round_page(virtual_space_start) - virtual_space_start;        /* Account for any slop */
1450 	npages = (uint_t)(mem_sz / (PAGE_SIZE + sizeof(*vm_pages)));    /* scaled to include the vm_page_ts */
1451 
1452 
1453 	vm_pages = (vm_page_t) pmap_steal_freeable_memory(npages * sizeof *vm_pages);
1454 
1455 	/*
1456 	 * Check if we want to initialize pages to a known value
1457 	 */
1458 	if (PE_parse_boot_argn("fill", &fillval, sizeof(fillval))) {
1459 		fill = TRUE;
1460 	}
1461 #if     DEBUG
1462 	/* This slows down booting the DEBUG kernel, particularly on
1463 	 * large memory systems, but is worthwhile in deterministically
1464 	 * trapping uninitialized memory usage.
1465 	 */
1466 	if (!fill) {
1467 		fill = TRUE;
1468 		fillval = 0xDEB8F177;
1469 	}
1470 #endif
1471 	if (fill) {
1472 		kprintf("Filling vm_pages with pattern: 0x%x\n", fillval);
1473 	}
1474 
1475 #if CONFIG_SECLUDED_MEMORY
1476 	/*
1477 	 * Figure out how much secluded memory to have before we start
1478 	 * release pages to free lists.
1479 	 * The default, if specified nowhere else, is no secluded mem.
1480 	 */
1481 	vm_page_secluded_target = (unsigned int)atop_64(secluded_mem_mb * 1024ULL * 1024ULL);
1482 
1483 	/*
1484 	 * Allow a really large app to effectively use secluded memory until it exits.
1485 	 */
1486 	if (vm_page_secluded_target != 0) {
1487 		/*
1488 		 * Get an amount from boot-args, else use 1/2 of max_mem.
1489 		 * 1/2 max_mem was chosen from a Peace daemon tentpole test which
1490 		 * used munch to induce jetsam thrashing of false idle daemons on N56.
1491 		 */
1492 		int secluded_shutoff_mb;
1493 		if (PE_parse_boot_argn("secluded_shutoff_mb", &secluded_shutoff_mb,
1494 		    sizeof(secluded_shutoff_mb))) {
1495 			secluded_shutoff_trigger = (uint64_t)secluded_shutoff_mb * 1024 * 1024;
1496 		} else {
1497 			secluded_shutoff_trigger = max_mem / 2;
1498 		}
1499 
1500 		/* ensure the headroom value is sensible and avoid underflows */
1501 		assert(secluded_shutoff_trigger == 0 || secluded_shutoff_trigger > secluded_shutoff_headroom);
1502 	}
1503 
1504 #endif /* CONFIG_SECLUDED_MEMORY */
1505 
1506 #if defined(__x86_64__)
1507 
1508 	/*
1509 	 * Decide how much memory we delay freeing at boot time.
1510 	 */
1511 	uint32_t delay_above_gb;
1512 	if (!PE_parse_boot_argn("delay_above_gb", &delay_above_gb, sizeof(delay_above_gb))) {
1513 		delay_above_gb = DEFAULT_DELAY_ABOVE_PHYS_GB;
1514 	}
1515 
1516 	if (delay_above_gb == 0) {
1517 		delay_above_pnum = PPNUM_MAX;
1518 	} else {
1519 		delay_above_pnum = delay_above_gb * (1024 * 1024 * 1024 / PAGE_SIZE);
1520 	}
1521 
1522 	/* make sure we have sane breathing room: 1G above low memory */
1523 	if (delay_above_pnum <= max_valid_low_ppnum) {
1524 		delay_above_pnum = max_valid_low_ppnum + ((1024 * 1024 * 1024) >> PAGE_SHIFT);
1525 	}
1526 
1527 	if (delay_above_pnum < PPNUM_MAX) {
1528 		printf("pmap_startup() delaying init/free of page nums > 0x%x\n", delay_above_pnum);
1529 	}
1530 
1531 #endif /* defined(__x86_64__) */
1532 
1533 	/*
1534 	 * Initialize and release the page frames.
1535 	 */
1536 	kernel_debug_string_early("page_frame_init");
1537 
1538 	vm_page_array_beginning_addr = &vm_pages[0];
1539 	vm_page_array_ending_addr = &vm_pages[npages];  /* used by ptr packing/unpacking code */
1540 #if VM_PAGE_PACKED_FROM_ARRAY
1541 	if (npages >= VM_PAGE_PACKED_FROM_ARRAY) {
1542 		panic("pmap_startup(): too many pages to support vm_page packing");
1543 	}
1544 #endif
1545 
1546 	vm_delayed_count = 0;
1547 
1548 	absolutetime_to_nanoseconds(mach_absolute_time(), &start_ns);
1549 	vm_pages_count = 0;
1550 	for (i = 0; i < npages; i++) {
1551 		/* Did we run out of pages? */
1552 		if (!pmap_next_page(&phys_page)) {
1553 			break;
1554 		}
1555 
1556 		if (phys_page < max_valid_low_ppnum) {
1557 			++low_page_count;
1558 		}
1559 
1560 		/* Are we at high enough pages to delay the rest? */
1561 		if (low_page_count > vm_lopage_free_limit && phys_page > delay_above_pnum) {
1562 			vm_delayed_count = pmap_free_pages();
1563 			break;
1564 		}
1565 
1566 #if defined(__arm64__)
1567 		if (i == 0) {
1568 			vm_first_phys_ppnum = phys_page;
1569 			patch_low_glo_vm_page_info((void *)vm_page_array_beginning_addr,
1570 			    (void *)vm_page_array_ending_addr, vm_first_phys_ppnum);
1571 		}
1572 #endif /* defined(__arm64__) */
1573 
1574 #if defined(__x86_64__)
1575 		/* The x86 clump freeing code requires increasing ppn's to work correctly */
1576 		if (i > 0) {
1577 			assert(phys_page > vm_pages[i - 1].vmp_phys_page);
1578 		}
1579 #endif
1580 		++vm_pages_count;
1581 		vm_page_init(&vm_pages[i], phys_page, FALSE);
1582 		if (fill) {
1583 			fillPage(phys_page, fillval);
1584 		}
1585 		if (vm_himemory_mode) {
1586 			vm_page_release_startup(&vm_pages[i]);
1587 		}
1588 	}
1589 	vm_page_pages = vm_pages_count; /* used to report to user space */
1590 
1591 	if (!vm_himemory_mode) {
1592 		do {
1593 			if (!VMP_ERROR_GET(&vm_pages[--i])) {               /* skip retired pages */
1594 				vm_page_release_startup(&vm_pages[i]);
1595 			}
1596 		} while (i != 0);
1597 	}
1598 
1599 	absolutetime_to_nanoseconds(mach_absolute_time(), &now_ns);
1600 	printf("pmap_startup() init/release time: %lld microsec\n", (now_ns - start_ns) / NSEC_PER_USEC);
1601 	printf("pmap_startup() delayed init/release of %d pages\n", vm_delayed_count);
1602 
1603 #if defined(__LP64__)
1604 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[0]))) != &vm_pages[0]) {
1605 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[0] - %p", (void *)&vm_pages[0]);
1606 	}
1607 
1608 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[vm_pages_count - 1]))) != &vm_pages[vm_pages_count - 1]) {
1609 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[vm_pages_count-1] - %p", (void *)&vm_pages[vm_pages_count - 1]);
1610 	}
1611 #endif
1612 
1613 	VM_CHECK_MEMORYSTATUS;
1614 
1615 	/*
1616 	 * We have to re-align virtual_space_start,
1617 	 * because pmap_steal_memory has been using it.
1618 	 */
1619 	virtual_space_start = round_page(virtual_space_start);
1620 	*startp = virtual_space_start;
1621 	*endp = virtual_space_end;
1622 }
1623 #endif  /* MACHINE_PAGES */
1624 
1625 /*
1626  * Create the zone that represents the vm_pages[] array. Nothing ever allocates
1627  * or frees to this zone. It's just here for reporting purposes via zprint command.
1628  * This needs to be done after all initially delayed pages are put on the free lists.
1629  */
1630 static void
vm_page_module_init_delayed(void)1631 vm_page_module_init_delayed(void)
1632 {
1633 	(void)zone_create_ext("vm pages array", sizeof(struct vm_page),
1634 	    ZC_KASAN_NOREDZONE | ZC_KASAN_NOQUARANTINE, ZONE_ID_VM_PAGES, ^(zone_t z) {
1635 		uint64_t vm_page_zone_pages, vm_page_array_zone_data_size;
1636 
1637 		zone_set_exhaustible(z, 0, true);
1638 		/*
1639 		 * Reflect size and usage information for vm_pages[].
1640 		 */
1641 
1642 		z->z_elems_avail = (uint32_t)(vm_page_array_ending_addr - vm_pages);
1643 		z->z_elems_free = z->z_elems_avail - vm_pages_count;
1644 		zpercpu_get_cpu(z->z_stats, 0)->zs_mem_allocated =
1645 		vm_pages_count * sizeof(struct vm_page);
1646 		vm_page_array_zone_data_size = (uint64_t)vm_page_array_ending_addr - (uint64_t)vm_pages;
1647 		vm_page_zone_pages = atop(round_page((vm_offset_t)vm_page_array_zone_data_size));
1648 		z->z_wired_cur += vm_page_zone_pages;
1649 		z->z_wired_hwm = z->z_wired_cur;
1650 		z->z_va_cur = z->z_wired_cur;
1651 		/* since zone accounts for these, take them out of stolen */
1652 		VM_PAGE_MOVE_STOLEN(vm_page_zone_pages);
1653 	});
1654 }
1655 
1656 /*
1657  * Create the vm_pages zone. This is used for the vm_page structures for the pages
1658  * that are scavanged from other boot time usages by ml_static_mfree(). As such,
1659  * this needs to happen in early VM bootstrap.
1660  */
1661 
1662 __startup_func
1663 static void
vm_page_module_init(void)1664 vm_page_module_init(void)
1665 {
1666 	vm_size_t vm_page_with_ppnum_size;
1667 
1668 	/*
1669 	 * Since the pointers to elements in this zone will be packed, they
1670 	 * must have appropriate size. Not strictly what sizeof() reports.
1671 	 */
1672 	vm_page_with_ppnum_size =
1673 	    (sizeof(struct vm_page_with_ppnum) + (VM_PAGE_PACKED_PTR_ALIGNMENT - 1)) &
1674 	    ~(VM_PAGE_PACKED_PTR_ALIGNMENT - 1);
1675 
1676 	vm_page_zone = zone_create_ext("vm pages", vm_page_with_ppnum_size,
1677 	    ZC_ALIGNMENT_REQUIRED | ZC_VM | ZC_NO_TBI_TAG,
1678 	    ZONE_ID_ANY, ^(zone_t z) {
1679 		/*
1680 		 * The number "10" is a small number that is larger than the number
1681 		 * of fictitious pages that any single caller will attempt to allocate
1682 		 * without blocking.
1683 		 *
1684 		 * The largest such number at the moment is kmem_alloc()
1685 		 * when 2 guard pages are asked. 10 is simply a somewhat larger number,
1686 		 * taking into account the 50% hysteresis the zone allocator uses.
1687 		 *
1688 		 * Note: this works at all because the zone allocator
1689 		 *       doesn't ever allocate fictitious pages.
1690 		 */
1691 		zone_raise_reserve(z, 10);
1692 	});
1693 }
1694 STARTUP(ZALLOC, STARTUP_RANK_SECOND, vm_page_module_init);
1695 
1696 /*
1697  *	Routine:	vm_page_create
1698  *	Purpose:
1699  *		After the VM system is up, machine-dependent code
1700  *		may stumble across more physical memory.  For example,
1701  *		memory that it was reserving for a frame buffer.
1702  *		vm_page_create turns this memory into available pages.
1703  */
1704 
1705 void
vm_page_create(ppnum_t start,ppnum_t end)1706 vm_page_create(
1707 	ppnum_t start,
1708 	ppnum_t end)
1709 {
1710 	ppnum_t         phys_page;
1711 	vm_page_t       m;
1712 
1713 	for (phys_page = start;
1714 	    phys_page < end;
1715 	    phys_page++) {
1716 		m = vm_page_grab_fictitious_common(phys_page, TRUE);
1717 		m->vmp_fictitious = FALSE;
1718 		pmap_clear_noencrypt(phys_page);
1719 
1720 
1721 		vm_free_page_lock();
1722 		vm_page_pages++;
1723 		vm_free_page_unlock();
1724 		vm_page_release(m, FALSE);
1725 	}
1726 }
1727 
1728 
1729 /*
1730  *	vm_page_hash:
1731  *
1732  *	Distributes the object/offset key pair among hash buckets.
1733  *
1734  *	NOTE:	The bucket count must be a power of 2
1735  */
1736 #define vm_page_hash(object, offset) (\
1737 	( (natural_t)((uintptr_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
1738 	 & vm_page_hash_mask)
1739 
1740 
1741 /*
1742  *	vm_page_insert:		[ internal use only ]
1743  *
1744  *	Inserts the given mem entry into the object/object-page
1745  *	table and object list.
1746  *
1747  *	The object must be locked.
1748  */
1749 void
vm_page_insert(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)1750 vm_page_insert(
1751 	vm_page_t               mem,
1752 	vm_object_t             object,
1753 	vm_object_offset_t      offset)
1754 {
1755 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, FALSE, FALSE, NULL);
1756 }
1757 
1758 void
vm_page_insert_wired(vm_page_t mem,vm_object_t object,vm_object_offset_t offset,vm_tag_t tag)1759 vm_page_insert_wired(
1760 	vm_page_t               mem,
1761 	vm_object_t             object,
1762 	vm_object_offset_t      offset,
1763 	vm_tag_t                tag)
1764 {
1765 	vm_page_insert_internal(mem, object, offset, tag, FALSE, TRUE, FALSE, FALSE, NULL);
1766 }
1767 
1768 void
vm_page_insert_internal(vm_page_t mem,vm_object_t object,vm_object_offset_t offset,vm_tag_t tag,boolean_t queues_lock_held,boolean_t insert_in_hash,boolean_t batch_pmap_op,boolean_t batch_accounting,uint64_t * delayed_ledger_update)1769 vm_page_insert_internal(
1770 	vm_page_t               mem,
1771 	vm_object_t             object,
1772 	vm_object_offset_t      offset,
1773 	vm_tag_t                tag,
1774 	boolean_t               queues_lock_held,
1775 	boolean_t               insert_in_hash,
1776 	boolean_t               batch_pmap_op,
1777 	boolean_t               batch_accounting,
1778 	uint64_t                *delayed_ledger_update)
1779 {
1780 	vm_page_bucket_t        *bucket;
1781 	lck_spin_t              *bucket_lock;
1782 	int                     hash_id;
1783 	task_t                  owner;
1784 	int                     ledger_idx_volatile;
1785 	int                     ledger_idx_nonvolatile;
1786 	int                     ledger_idx_volatile_compressed;
1787 	int                     ledger_idx_nonvolatile_compressed;
1788 	int                     ledger_idx_composite;
1789 	int                     ledger_idx_external_wired;
1790 	boolean_t               do_footprint;
1791 
1792 #if 0
1793 	/*
1794 	 * we may not hold the page queue lock
1795 	 * so this check isn't safe to make
1796 	 */
1797 	VM_PAGE_CHECK(mem);
1798 #endif
1799 
1800 	assertf(page_aligned(offset), "0x%llx\n", offset);
1801 
1802 	assert(!VM_PAGE_WIRED(mem) || mem->vmp_private || mem->vmp_fictitious || (tag != VM_KERN_MEMORY_NONE));
1803 
1804 	vm_object_lock_assert_exclusive(object);
1805 	LCK_MTX_ASSERT(&vm_page_queue_lock,
1806 	    queues_lock_held ? LCK_MTX_ASSERT_OWNED
1807 	    : LCK_MTX_ASSERT_NOTOWNED);
1808 
1809 	if (queues_lock_held == FALSE) {
1810 		assert(!VM_PAGE_PAGEABLE(mem));
1811 	}
1812 
1813 	if (insert_in_hash == TRUE) {
1814 #if DEBUG || VM_PAGE_BUCKETS_CHECK
1815 		if (mem->vmp_tabled || mem->vmp_object) {
1816 			panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
1817 			    "already in (obj=%p,off=0x%llx)",
1818 			    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
1819 		}
1820 #endif
1821 		if (object->internal && (offset >= object->vo_size)) {
1822 			panic("vm_page_insert_internal: (page=%p,obj=%p,off=0x%llx,size=0x%llx) inserted at offset past object bounds",
1823 			    mem, object, offset, object->vo_size);
1824 		}
1825 
1826 		assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
1827 
1828 		/*
1829 		 *	Record the object/offset pair in this page
1830 		 */
1831 
1832 		mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
1833 		mem->vmp_offset = offset;
1834 
1835 #if CONFIG_SECLUDED_MEMORY
1836 		if (object->eligible_for_secluded) {
1837 			vm_page_secluded.eligible_for_secluded++;
1838 		}
1839 #endif /* CONFIG_SECLUDED_MEMORY */
1840 
1841 		/*
1842 		 *	Insert it into the object_object/offset hash table
1843 		 */
1844 		hash_id = vm_page_hash(object, offset);
1845 		bucket = &vm_page_buckets[hash_id];
1846 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
1847 
1848 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
1849 
1850 		mem->vmp_next_m = bucket->page_list;
1851 		bucket->page_list = VM_PAGE_PACK_PTR(mem);
1852 		assert(mem == (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)));
1853 
1854 #if     MACH_PAGE_HASH_STATS
1855 		if (++bucket->cur_count > bucket->hi_count) {
1856 			bucket->hi_count = bucket->cur_count;
1857 		}
1858 #endif /* MACH_PAGE_HASH_STATS */
1859 		mem->vmp_hashed = TRUE;
1860 		lck_spin_unlock(bucket_lock);
1861 	}
1862 
1863 	{
1864 		unsigned int    cache_attr;
1865 
1866 		cache_attr = object->wimg_bits & VM_WIMG_MASK;
1867 
1868 		if (cache_attr != VM_WIMG_USE_DEFAULT) {
1869 			PMAP_SET_CACHE_ATTR(mem, object, cache_attr, batch_pmap_op);
1870 		}
1871 	}
1872 	/*
1873 	 *	Now link into the object's list of backed pages.
1874 	 */
1875 	vm_page_queue_enter(&object->memq, mem, vmp_listq);
1876 	object->memq_hint = mem;
1877 	mem->vmp_tabled = TRUE;
1878 
1879 	/*
1880 	 *	Show that the object has one more resident page.
1881 	 */
1882 
1883 	object->resident_page_count++;
1884 	if (VM_PAGE_WIRED(mem)) {
1885 		assert(mem->vmp_wire_count > 0);
1886 		VM_OBJECT_WIRED_PAGE_UPDATE_START(object);
1887 		VM_OBJECT_WIRED_PAGE_ADD(object, mem);
1888 		VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag);
1889 	}
1890 	assert(object->resident_page_count >= object->wired_page_count);
1891 
1892 #if DEVELOPMENT || DEBUG
1893 	if (object->object_is_shared_cache &&
1894 	    object->pager != NULL &&
1895 	    object->pager->mo_pager_ops == &shared_region_pager_ops) {
1896 		int new, old;
1897 		assert(!object->internal);
1898 		new = OSAddAtomic(+1, &shared_region_pagers_resident_count);
1899 		do {
1900 			old = shared_region_pagers_resident_peak;
1901 		} while (old < new &&
1902 		    !OSCompareAndSwap(old, new, &shared_region_pagers_resident_peak));
1903 	}
1904 #endif /* DEVELOPMENT || DEBUG */
1905 
1906 	if (batch_accounting == FALSE) {
1907 		if (object->internal) {
1908 			OSAddAtomic(1, &vm_page_internal_count);
1909 		} else {
1910 			OSAddAtomic(1, &vm_page_external_count);
1911 		}
1912 	}
1913 
1914 	/*
1915 	 * It wouldn't make sense to insert a "reusable" page in
1916 	 * an object (the page would have been marked "reusable" only
1917 	 * at the time of a madvise(MADV_FREE_REUSABLE) if it was already
1918 	 * in the object at that time).
1919 	 * But a page could be inserted in a "all_reusable" object, if
1920 	 * something faults it in (a vm_read() from another task or a
1921 	 * "use-after-free" issue in user space, for example).  It can
1922 	 * also happen if we're relocating a page from that object to
1923 	 * a different physical page during a physically-contiguous
1924 	 * allocation.
1925 	 */
1926 	assert(!mem->vmp_reusable);
1927 	if (object->all_reusable) {
1928 		OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count);
1929 	}
1930 
1931 	if (object->purgable == VM_PURGABLE_DENY &&
1932 	    !object->vo_ledger_tag) {
1933 		owner = TASK_NULL;
1934 	} else {
1935 		owner = VM_OBJECT_OWNER(object);
1936 		vm_object_ledger_tag_ledgers(object,
1937 		    &ledger_idx_volatile,
1938 		    &ledger_idx_nonvolatile,
1939 		    &ledger_idx_volatile_compressed,
1940 		    &ledger_idx_nonvolatile_compressed,
1941 		    &ledger_idx_composite,
1942 		    &ledger_idx_external_wired,
1943 		    &do_footprint);
1944 	}
1945 	if (owner &&
1946 	    object->internal &&
1947 	    (object->purgable == VM_PURGABLE_NONVOLATILE ||
1948 	    object->purgable == VM_PURGABLE_DENY ||
1949 	    VM_PAGE_WIRED(mem))) {
1950 		if (delayed_ledger_update) {
1951 			*delayed_ledger_update += PAGE_SIZE;
1952 		} else {
1953 			/* more non-volatile bytes */
1954 			ledger_credit(owner->ledger,
1955 			    ledger_idx_nonvolatile,
1956 			    PAGE_SIZE);
1957 			if (do_footprint) {
1958 				/* more footprint */
1959 				ledger_credit(owner->ledger,
1960 				    task_ledgers.phys_footprint,
1961 				    PAGE_SIZE);
1962 			} else if (ledger_idx_composite != -1) {
1963 				ledger_credit(owner->ledger,
1964 				    ledger_idx_composite,
1965 				    PAGE_SIZE);
1966 			}
1967 		}
1968 	} else if (owner &&
1969 	    object->internal &&
1970 	    (object->purgable == VM_PURGABLE_VOLATILE ||
1971 	    object->purgable == VM_PURGABLE_EMPTY)) {
1972 		assert(!VM_PAGE_WIRED(mem));
1973 		/* more volatile bytes */
1974 		ledger_credit(owner->ledger,
1975 		    ledger_idx_volatile,
1976 		    PAGE_SIZE);
1977 	}
1978 
1979 	if (object->purgable == VM_PURGABLE_VOLATILE) {
1980 		if (VM_PAGE_WIRED(mem)) {
1981 			OSAddAtomic(+1, &vm_page_purgeable_wired_count);
1982 		} else {
1983 			OSAddAtomic(+1, &vm_page_purgeable_count);
1984 		}
1985 	} else if (object->purgable == VM_PURGABLE_EMPTY &&
1986 	    mem->vmp_q_state == VM_PAGE_ON_THROTTLED_Q) {
1987 		/*
1988 		 * This page belongs to a purged VM object but hasn't
1989 		 * been purged (because it was "busy").
1990 		 * It's in the "throttled" queue and hence not
1991 		 * visible to vm_pageout_scan().  Move it to a pageable
1992 		 * queue, so that it can eventually be reclaimed, instead
1993 		 * of lingering in the "empty" object.
1994 		 */
1995 		if (queues_lock_held == FALSE) {
1996 			vm_page_lockspin_queues();
1997 		}
1998 		vm_page_deactivate(mem);
1999 		if (queues_lock_held == FALSE) {
2000 			vm_page_unlock_queues();
2001 		}
2002 	}
2003 
2004 #if VM_OBJECT_TRACKING_OP_MODIFIED
2005 	if (vm_object_tracking_btlog &&
2006 	    object->internal &&
2007 	    object->resident_page_count == 0 &&
2008 	    object->pager == NULL &&
2009 	    object->shadow != NULL &&
2010 	    object->shadow->vo_copy == object) {
2011 		btlog_record(vm_object_tracking_btlog, object,
2012 		    VM_OBJECT_TRACKING_OP_MODIFIED,
2013 		    btref_get(__builtin_frame_address(0), 0));
2014 	}
2015 #endif /* VM_OBJECT_TRACKING_OP_MODIFIED */
2016 }
2017 
2018 /*
2019  *	vm_page_replace:
2020  *
2021  *	Exactly like vm_page_insert, except that we first
2022  *	remove any existing page at the given offset in object.
2023  *
2024  *	The object must be locked.
2025  */
2026 void
vm_page_replace(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)2027 vm_page_replace(
2028 	vm_page_t               mem,
2029 	vm_object_t             object,
2030 	vm_object_offset_t      offset)
2031 {
2032 	vm_page_bucket_t *bucket;
2033 	vm_page_t        found_m = VM_PAGE_NULL;
2034 	lck_spin_t      *bucket_lock;
2035 	int             hash_id;
2036 
2037 #if 0
2038 	/*
2039 	 * we don't hold the page queue lock
2040 	 * so this check isn't safe to make
2041 	 */
2042 	VM_PAGE_CHECK(mem);
2043 #endif
2044 	vm_object_lock_assert_exclusive(object);
2045 #if DEBUG || VM_PAGE_BUCKETS_CHECK
2046 	if (mem->vmp_tabled || mem->vmp_object) {
2047 		panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) "
2048 		    "already in (obj=%p,off=0x%llx)",
2049 		    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
2050 	}
2051 #endif
2052 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2053 
2054 	assert(!VM_PAGE_PAGEABLE(mem));
2055 
2056 	/*
2057 	 *	Record the object/offset pair in this page
2058 	 */
2059 	mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
2060 	mem->vmp_offset = offset;
2061 
2062 	/*
2063 	 *	Insert it into the object_object/offset hash table,
2064 	 *	replacing any page that might have been there.
2065 	 */
2066 
2067 	hash_id = vm_page_hash(object, offset);
2068 	bucket = &vm_page_buckets[hash_id];
2069 	bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2070 
2071 	lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2072 
2073 	if (bucket->page_list) {
2074 		vm_page_packed_t *mp = &bucket->page_list;
2075 		vm_page_t m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp));
2076 
2077 		do {
2078 			/*
2079 			 * compare packed object pointers
2080 			 */
2081 			if (m->vmp_object == mem->vmp_object && m->vmp_offset == offset) {
2082 				/*
2083 				 * Remove old page from hash list
2084 				 */
2085 				*mp = m->vmp_next_m;
2086 				m->vmp_hashed = FALSE;
2087 				m->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2088 
2089 				found_m = m;
2090 				break;
2091 			}
2092 			mp = &m->vmp_next_m;
2093 		} while ((m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp))));
2094 
2095 		mem->vmp_next_m = bucket->page_list;
2096 	} else {
2097 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2098 	}
2099 	/*
2100 	 * insert new page at head of hash list
2101 	 */
2102 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
2103 	mem->vmp_hashed = TRUE;
2104 
2105 	lck_spin_unlock(bucket_lock);
2106 
2107 	if (found_m) {
2108 		/*
2109 		 * there was already a page at the specified
2110 		 * offset for this object... remove it from
2111 		 * the object and free it back to the free list
2112 		 */
2113 		vm_page_free_unlocked(found_m, FALSE);
2114 	}
2115 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL);
2116 }
2117 
2118 /*
2119  *	vm_page_remove:		[ internal use only ]
2120  *
2121  *	Removes the given mem entry from the object/offset-page
2122  *	table and the object page list.
2123  *
2124  *	The object must be locked.
2125  */
2126 
2127 void
vm_page_remove(vm_page_t mem,boolean_t remove_from_hash)2128 vm_page_remove(
2129 	vm_page_t       mem,
2130 	boolean_t       remove_from_hash)
2131 {
2132 	vm_page_bucket_t *bucket;
2133 	vm_page_t       this;
2134 	lck_spin_t      *bucket_lock;
2135 	int             hash_id;
2136 	task_t          owner;
2137 	vm_object_t     m_object;
2138 	int             ledger_idx_volatile;
2139 	int             ledger_idx_nonvolatile;
2140 	int             ledger_idx_volatile_compressed;
2141 	int             ledger_idx_nonvolatile_compressed;
2142 	int             ledger_idx_composite;
2143 	int             ledger_idx_external_wired;
2144 	int             do_footprint;
2145 
2146 	m_object = VM_PAGE_OBJECT(mem);
2147 
2148 	vm_object_lock_assert_exclusive(m_object);
2149 	assert(mem->vmp_tabled);
2150 	assert(!mem->vmp_cleaning);
2151 	assert(!mem->vmp_laundry);
2152 
2153 	if (VM_PAGE_PAGEABLE(mem)) {
2154 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2155 	}
2156 #if 0
2157 	/*
2158 	 * we don't hold the page queue lock
2159 	 * so this check isn't safe to make
2160 	 */
2161 	VM_PAGE_CHECK(mem);
2162 #endif
2163 	if (remove_from_hash == TRUE) {
2164 		/*
2165 		 *	Remove from the object_object/offset hash table
2166 		 */
2167 		hash_id = vm_page_hash(m_object, mem->vmp_offset);
2168 		bucket = &vm_page_buckets[hash_id];
2169 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2170 
2171 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2172 
2173 		if ((this = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))) == mem) {
2174 			/* optimize for common case */
2175 
2176 			bucket->page_list = mem->vmp_next_m;
2177 		} else {
2178 			vm_page_packed_t        *prev;
2179 
2180 			for (prev = &this->vmp_next_m;
2181 			    (this = (vm_page_t)(VM_PAGE_UNPACK_PTR(*prev))) != mem;
2182 			    prev = &this->vmp_next_m) {
2183 				continue;
2184 			}
2185 			*prev = this->vmp_next_m;
2186 		}
2187 #if     MACH_PAGE_HASH_STATS
2188 		bucket->cur_count--;
2189 #endif /* MACH_PAGE_HASH_STATS */
2190 		mem->vmp_hashed = FALSE;
2191 		this->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2192 		lck_spin_unlock(bucket_lock);
2193 	}
2194 	/*
2195 	 *	Now remove from the object's list of backed pages.
2196 	 */
2197 
2198 	vm_page_remove_internal(mem);
2199 
2200 	/*
2201 	 *	And show that the object has one fewer resident
2202 	 *	page.
2203 	 */
2204 
2205 	assert(m_object->resident_page_count > 0);
2206 	m_object->resident_page_count--;
2207 
2208 #if DEVELOPMENT || DEBUG
2209 	if (m_object->object_is_shared_cache &&
2210 	    m_object->pager != NULL &&
2211 	    m_object->pager->mo_pager_ops == &shared_region_pager_ops) {
2212 		assert(!m_object->internal);
2213 		OSAddAtomic(-1, &shared_region_pagers_resident_count);
2214 	}
2215 #endif /* DEVELOPMENT || DEBUG */
2216 
2217 	if (m_object->internal) {
2218 #if DEBUG
2219 		assert(vm_page_internal_count);
2220 #endif /* DEBUG */
2221 
2222 		OSAddAtomic(-1, &vm_page_internal_count);
2223 	} else {
2224 		assert(vm_page_external_count);
2225 		OSAddAtomic(-1, &vm_page_external_count);
2226 
2227 		if (mem->vmp_xpmapped) {
2228 			assert(vm_page_xpmapped_external_count);
2229 			OSAddAtomic(-1, &vm_page_xpmapped_external_count);
2230 		}
2231 	}
2232 	if (!m_object->internal &&
2233 	    m_object->cached_list.next &&
2234 	    m_object->cached_list.prev) {
2235 		if (m_object->resident_page_count == 0) {
2236 			vm_object_cache_remove(m_object);
2237 		}
2238 	}
2239 
2240 	if (VM_PAGE_WIRED(mem)) {
2241 		assert(mem->vmp_wire_count > 0);
2242 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
2243 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
2244 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
2245 	}
2246 	assert(m_object->resident_page_count >=
2247 	    m_object->wired_page_count);
2248 	if (mem->vmp_reusable) {
2249 		assert(m_object->reusable_page_count > 0);
2250 		m_object->reusable_page_count--;
2251 		assert(m_object->reusable_page_count <=
2252 		    m_object->resident_page_count);
2253 		mem->vmp_reusable = FALSE;
2254 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2255 		vm_page_stats_reusable.reused_remove++;
2256 	} else if (m_object->all_reusable) {
2257 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2258 		vm_page_stats_reusable.reused_remove++;
2259 	}
2260 
2261 	if (m_object->purgable == VM_PURGABLE_DENY &&
2262 	    !m_object->vo_ledger_tag) {
2263 		owner = TASK_NULL;
2264 	} else {
2265 		owner = VM_OBJECT_OWNER(m_object);
2266 		vm_object_ledger_tag_ledgers(m_object,
2267 		    &ledger_idx_volatile,
2268 		    &ledger_idx_nonvolatile,
2269 		    &ledger_idx_volatile_compressed,
2270 		    &ledger_idx_nonvolatile_compressed,
2271 		    &ledger_idx_composite,
2272 		    &ledger_idx_external_wired,
2273 		    &do_footprint);
2274 	}
2275 	if (owner &&
2276 	    m_object->internal &&
2277 	    (m_object->purgable == VM_PURGABLE_NONVOLATILE ||
2278 	    m_object->purgable == VM_PURGABLE_DENY ||
2279 	    VM_PAGE_WIRED(mem))) {
2280 		/* less non-volatile bytes */
2281 		ledger_debit(owner->ledger,
2282 		    ledger_idx_nonvolatile,
2283 		    PAGE_SIZE);
2284 		if (do_footprint) {
2285 			/* less footprint */
2286 			ledger_debit(owner->ledger,
2287 			    task_ledgers.phys_footprint,
2288 			    PAGE_SIZE);
2289 		} else if (ledger_idx_composite != -1) {
2290 			ledger_debit(owner->ledger,
2291 			    ledger_idx_composite,
2292 			    PAGE_SIZE);
2293 		}
2294 	} else if (owner &&
2295 	    m_object->internal &&
2296 	    (m_object->purgable == VM_PURGABLE_VOLATILE ||
2297 	    m_object->purgable == VM_PURGABLE_EMPTY)) {
2298 		assert(!VM_PAGE_WIRED(mem));
2299 		/* less volatile bytes */
2300 		ledger_debit(owner->ledger,
2301 		    ledger_idx_volatile,
2302 		    PAGE_SIZE);
2303 	}
2304 
2305 	if (m_object->purgable == VM_PURGABLE_VOLATILE) {
2306 		if (VM_PAGE_WIRED(mem)) {
2307 			assert(vm_page_purgeable_wired_count > 0);
2308 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
2309 		} else {
2310 			assert(vm_page_purgeable_count > 0);
2311 			OSAddAtomic(-1, &vm_page_purgeable_count);
2312 		}
2313 	}
2314 
2315 	if (m_object->set_cache_attr == TRUE) {
2316 		pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(mem), 0);
2317 	}
2318 
2319 	mem->vmp_tabled = FALSE;
2320 	mem->vmp_object = 0;
2321 	mem->vmp_offset = (vm_object_offset_t) -1;
2322 }
2323 
2324 
2325 /*
2326  *	vm_page_lookup:
2327  *
2328  *	Returns the page associated with the object/offset
2329  *	pair specified; if none is found, VM_PAGE_NULL is returned.
2330  *
2331  *	The object must be locked.  No side effects.
2332  */
2333 
2334 #define VM_PAGE_HASH_LOOKUP_THRESHOLD   10
2335 
2336 #if DEBUG_VM_PAGE_LOOKUP
2337 
2338 struct {
2339 	uint64_t        vpl_total;
2340 	uint64_t        vpl_empty_obj;
2341 	uint64_t        vpl_bucket_NULL;
2342 	uint64_t        vpl_hit_hint;
2343 	uint64_t        vpl_hit_hint_next;
2344 	uint64_t        vpl_hit_hint_prev;
2345 	uint64_t        vpl_fast;
2346 	uint64_t        vpl_slow;
2347 	uint64_t        vpl_hit;
2348 	uint64_t        vpl_miss;
2349 
2350 	uint64_t        vpl_fast_elapsed;
2351 	uint64_t        vpl_slow_elapsed;
2352 } vm_page_lookup_stats __attribute__((aligned(8)));
2353 
2354 #endif
2355 
2356 #define KDP_VM_PAGE_WALK_MAX    1000
2357 
2358 vm_page_t
kdp_vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2359 kdp_vm_page_lookup(
2360 	vm_object_t             object,
2361 	vm_object_offset_t      offset)
2362 {
2363 	vm_page_t cur_page;
2364 	int num_traversed = 0;
2365 
2366 	if (not_in_kdp) {
2367 		panic("panic: kdp_vm_page_lookup done outside of kernel debugger");
2368 	}
2369 
2370 	vm_page_queue_iterate(&object->memq, cur_page, vmp_listq) {
2371 		if (cur_page->vmp_offset == offset) {
2372 			return cur_page;
2373 		}
2374 		num_traversed++;
2375 
2376 		if (num_traversed >= KDP_VM_PAGE_WALK_MAX) {
2377 			return VM_PAGE_NULL;
2378 		}
2379 	}
2380 
2381 	return VM_PAGE_NULL;
2382 }
2383 
2384 vm_page_t
vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2385 vm_page_lookup(
2386 	vm_object_t             object,
2387 	vm_object_offset_t      offset)
2388 {
2389 	vm_page_t       mem;
2390 	vm_page_bucket_t *bucket;
2391 	vm_page_queue_entry_t   qe;
2392 	lck_spin_t      *bucket_lock = NULL;
2393 	int             hash_id;
2394 #if DEBUG_VM_PAGE_LOOKUP
2395 	uint64_t        start, elapsed;
2396 
2397 	OSAddAtomic64(1, &vm_page_lookup_stats.vpl_total);
2398 #endif
2399 
2400 #if CONFIG_KERNEL_TAGGING
2401 	if (is_kernel_object(object)) {
2402 		offset = vm_memtag_canonicalize_address(offset);
2403 	}
2404 #endif /* CONFIG_KERNEL_TAGGING */
2405 
2406 	vm_object_lock_assert_held(object);
2407 	assertf(page_aligned(offset), "offset 0x%llx\n", offset);
2408 
2409 	if (object->resident_page_count == 0) {
2410 #if DEBUG_VM_PAGE_LOOKUP
2411 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_empty_obj);
2412 #endif
2413 		return VM_PAGE_NULL;
2414 	}
2415 
2416 	mem = object->memq_hint;
2417 
2418 	if (mem != VM_PAGE_NULL) {
2419 		assert(VM_PAGE_OBJECT(mem) == object);
2420 
2421 		if (mem->vmp_offset == offset) {
2422 #if DEBUG_VM_PAGE_LOOKUP
2423 			OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint);
2424 #endif
2425 			return mem;
2426 		}
2427 		qe = (vm_page_queue_entry_t)vm_page_queue_next(&mem->vmp_listq);
2428 
2429 		if (!vm_page_queue_end(&object->memq, qe)) {
2430 			vm_page_t       next_page;
2431 
2432 			next_page = (vm_page_t)((uintptr_t)qe);
2433 			assert(VM_PAGE_OBJECT(next_page) == object);
2434 
2435 			if (next_page->vmp_offset == offset) {
2436 				object->memq_hint = next_page; /* new hint */
2437 #if DEBUG_VM_PAGE_LOOKUP
2438 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next);
2439 #endif
2440 				return next_page;
2441 			}
2442 		}
2443 		qe = (vm_page_queue_entry_t)vm_page_queue_prev(&mem->vmp_listq);
2444 
2445 		if (!vm_page_queue_end(&object->memq, qe)) {
2446 			vm_page_t prev_page;
2447 
2448 			prev_page = (vm_page_t)((uintptr_t)qe);
2449 			assert(VM_PAGE_OBJECT(prev_page) == object);
2450 
2451 			if (prev_page->vmp_offset == offset) {
2452 				object->memq_hint = prev_page; /* new hint */
2453 #if DEBUG_VM_PAGE_LOOKUP
2454 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev);
2455 #endif
2456 				return prev_page;
2457 			}
2458 		}
2459 	}
2460 	/*
2461 	 * Search the hash table for this object/offset pair
2462 	 */
2463 	hash_id = vm_page_hash(object, offset);
2464 	bucket = &vm_page_buckets[hash_id];
2465 
2466 	/*
2467 	 * since we hold the object lock, we are guaranteed that no
2468 	 * new pages can be inserted into this object... this in turn
2469 	 * guarantess that the page we're looking for can't exist
2470 	 * if the bucket it hashes to is currently NULL even when looked
2471 	 * at outside the scope of the hash bucket lock... this is a
2472 	 * really cheap optimiztion to avoid taking the lock
2473 	 */
2474 	if (!bucket->page_list) {
2475 #if DEBUG_VM_PAGE_LOOKUP
2476 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL);
2477 #endif
2478 		return VM_PAGE_NULL;
2479 	}
2480 
2481 #if DEBUG_VM_PAGE_LOOKUP
2482 	start = mach_absolute_time();
2483 #endif
2484 	if (object->resident_page_count <= VM_PAGE_HASH_LOOKUP_THRESHOLD) {
2485 		/*
2486 		 * on average, it's roughly 3 times faster to run a short memq list
2487 		 * than to take the spin lock and go through the hash list
2488 		 */
2489 		mem = (vm_page_t)vm_page_queue_first(&object->memq);
2490 
2491 		while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2492 			if (mem->vmp_offset == offset) {
2493 				break;
2494 			}
2495 
2496 			mem = (vm_page_t)vm_page_queue_next(&mem->vmp_listq);
2497 		}
2498 		if (vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2499 			mem = NULL;
2500 		}
2501 	} else {
2502 		vm_page_object_t        packed_object;
2503 
2504 		packed_object = VM_PAGE_PACK_OBJECT(object);
2505 
2506 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2507 
2508 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2509 
2510 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
2511 		    mem != VM_PAGE_NULL;
2512 		    mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m))) {
2513 #if 0
2514 			/*
2515 			 * we don't hold the page queue lock
2516 			 * so this check isn't safe to make
2517 			 */
2518 			VM_PAGE_CHECK(mem);
2519 #endif
2520 			if ((mem->vmp_object == packed_object) && (mem->vmp_offset == offset)) {
2521 				break;
2522 			}
2523 		}
2524 		lck_spin_unlock(bucket_lock);
2525 	}
2526 
2527 #if DEBUG_VM_PAGE_LOOKUP
2528 	elapsed = mach_absolute_time() - start;
2529 
2530 	if (bucket_lock) {
2531 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_slow);
2532 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_slow_elapsed);
2533 	} else {
2534 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_fast);
2535 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_fast_elapsed);
2536 	}
2537 	if (mem != VM_PAGE_NULL) {
2538 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit);
2539 	} else {
2540 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss);
2541 	}
2542 #endif
2543 	if (mem != VM_PAGE_NULL) {
2544 		assert(VM_PAGE_OBJECT(mem) == object);
2545 
2546 		object->memq_hint = mem;
2547 	}
2548 	return mem;
2549 }
2550 
2551 
2552 /*
2553  *	vm_page_rename:
2554  *
2555  *	Move the given memory entry from its
2556  *	current object to the specified target object/offset.
2557  *
2558  *	The object must be locked.
2559  */
2560 void
vm_page_rename(vm_page_t mem,vm_object_t new_object,vm_object_offset_t new_offset)2561 vm_page_rename(
2562 	vm_page_t               mem,
2563 	vm_object_t             new_object,
2564 	vm_object_offset_t      new_offset)
2565 {
2566 	boolean_t       internal_to_external, external_to_internal;
2567 	vm_tag_t        tag;
2568 	vm_object_t     m_object;
2569 
2570 	m_object = VM_PAGE_OBJECT(mem);
2571 
2572 	assert(m_object != new_object);
2573 	assert(m_object);
2574 
2575 	/*
2576 	 *	Changes to mem->vmp_object require the page lock because
2577 	 *	the pageout daemon uses that lock to get the object.
2578 	 */
2579 	vm_page_lockspin_queues();
2580 
2581 	internal_to_external = FALSE;
2582 	external_to_internal = FALSE;
2583 
2584 	if (mem->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q) {
2585 		/*
2586 		 * it's much easier to get the vm_page_pageable_xxx accounting correct
2587 		 * if we first move the page to the active queue... it's going to end
2588 		 * up there anyway, and we don't do vm_page_rename's frequently enough
2589 		 * for this to matter.
2590 		 */
2591 		vm_page_queues_remove(mem, FALSE);
2592 		vm_page_activate(mem);
2593 	}
2594 	if (VM_PAGE_PAGEABLE(mem)) {
2595 		if (m_object->internal && !new_object->internal) {
2596 			internal_to_external = TRUE;
2597 		}
2598 		if (!m_object->internal && new_object->internal) {
2599 			external_to_internal = TRUE;
2600 		}
2601 	}
2602 
2603 	tag = m_object->wire_tag;
2604 	vm_page_remove(mem, TRUE);
2605 	vm_page_insert_internal(mem, new_object, new_offset, tag, TRUE, TRUE, FALSE, FALSE, NULL);
2606 
2607 	if (internal_to_external) {
2608 		vm_page_pageable_internal_count--;
2609 		vm_page_pageable_external_count++;
2610 	} else if (external_to_internal) {
2611 		vm_page_pageable_external_count--;
2612 		vm_page_pageable_internal_count++;
2613 	}
2614 
2615 	vm_page_unlock_queues();
2616 }
2617 
2618 /*
2619  *	vm_page_init:
2620  *
2621  *	Initialize the fields in a new page.
2622  *	This takes a structure with random values and initializes it
2623  *	so that it can be given to vm_page_release or vm_page_insert.
2624  */
2625 void
vm_page_init(vm_page_t mem,ppnum_t phys_page,boolean_t lopage)2626 vm_page_init(
2627 	vm_page_t mem,
2628 	ppnum_t   phys_page,
2629 	boolean_t lopage)
2630 {
2631 	uint_t    i;
2632 	uintptr_t *p;
2633 
2634 	assert(phys_page);
2635 
2636 #if DEBUG
2637 	if ((phys_page != vm_page_fictitious_addr) && (phys_page != vm_page_guard_addr)) {
2638 		if (!(pmap_valid_page(phys_page))) {
2639 			panic("vm_page_init: non-DRAM phys_page 0x%x", phys_page);
2640 		}
2641 	}
2642 #endif /* DEBUG */
2643 
2644 	/*
2645 	 * Initialize the fields of the vm_page. If adding any new fields to vm_page,
2646 	 * try to use initial values which match 0. This minimizes the number of writes
2647 	 * needed for boot-time initialization.
2648 	 *
2649 	 * Kernel bzero() isn't an inline yet, so do it by hand for performance.
2650 	 */
2651 	assert(VM_PAGE_NOT_ON_Q == 0);
2652 	assert(sizeof(*mem) % sizeof(uintptr_t) == 0);
2653 	for (p = (uintptr_t *)(void *)mem, i = sizeof(*mem) / sizeof(uintptr_t); i != 0; --i) {
2654 		*p++ = 0;
2655 	}
2656 	mem->vmp_offset = (vm_object_offset_t)-1;
2657 	mem->vmp_busy = TRUE;
2658 	mem->vmp_lopage = lopage;
2659 
2660 	VM_PAGE_SET_PHYS_PAGE(mem, phys_page);
2661 #if 0
2662 	/*
2663 	 * we're leaving this turned off for now... currently pages
2664 	 * come off the free list and are either immediately dirtied/referenced
2665 	 * due to zero-fill or COW faults, or are used to read or write files...
2666 	 * in the file I/O case, the UPL mechanism takes care of clearing
2667 	 * the state of the HW ref/mod bits in a somewhat fragile way.
2668 	 * Since we may change the way this works in the future (to toughen it up),
2669 	 * I'm leaving this as a reminder of where these bits could get cleared
2670 	 */
2671 
2672 	/*
2673 	 * make sure both the h/w referenced and modified bits are
2674 	 * clear at this point... we are especially dependent on
2675 	 * not finding a 'stale' h/w modified in a number of spots
2676 	 * once this page goes back into use
2677 	 */
2678 	pmap_clear_refmod(phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
2679 #endif
2680 }
2681 
2682 /*
2683  *	vm_page_grab_fictitious:
2684  *
2685  *	Remove a fictitious page from the free list.
2686  *	Returns VM_PAGE_NULL if there are no free pages.
2687  */
2688 
2689 static vm_page_t
vm_page_grab_fictitious_common(ppnum_t phys_addr,boolean_t canwait)2690 vm_page_grab_fictitious_common(ppnum_t phys_addr, boolean_t canwait)
2691 {
2692 	vm_page_t m;
2693 
2694 	m = zalloc_flags(vm_page_zone, canwait ? Z_WAITOK : Z_NOWAIT);
2695 	if (m) {
2696 		vm_page_init(m, phys_addr, FALSE);
2697 		m->vmp_fictitious = TRUE;
2698 	}
2699 	return m;
2700 }
2701 
2702 vm_page_t
vm_page_grab_fictitious(boolean_t canwait)2703 vm_page_grab_fictitious(boolean_t canwait)
2704 {
2705 	return vm_page_grab_fictitious_common(vm_page_fictitious_addr, canwait);
2706 }
2707 
2708 int vm_guard_count;
2709 
2710 
2711 vm_page_t
vm_page_grab_guard(boolean_t canwait)2712 vm_page_grab_guard(boolean_t canwait)
2713 {
2714 	vm_page_t page;
2715 	page = vm_page_grab_fictitious_common(vm_page_guard_addr, canwait);
2716 	if (page) {
2717 		OSAddAtomic(1, &vm_guard_count);
2718 	}
2719 	return page;
2720 }
2721 
2722 
2723 /*
2724  *	vm_page_release_fictitious:
2725  *
2726  *	Release a fictitious page to the zone pool
2727  */
2728 void
vm_page_release_fictitious(vm_page_t m)2729 vm_page_release_fictitious(
2730 	vm_page_t m)
2731 {
2732 	assert((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || (m->vmp_q_state == VM_PAGE_IS_WIRED));
2733 	assert(m->vmp_fictitious);
2734 	assert(VM_PAGE_GET_PHYS_PAGE(m) == vm_page_fictitious_addr ||
2735 	    VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr);
2736 	assert(!m->vmp_realtime);
2737 
2738 	if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) {
2739 		OSAddAtomic(-1, &vm_guard_count);
2740 	}
2741 
2742 	zfree(vm_page_zone, m);
2743 }
2744 
2745 /*
2746  *	vm_pool_low():
2747  *
2748  *	Return true if it is not likely that a non-vm_privileged thread
2749  *	can get memory without blocking.  Advisory only, since the
2750  *	situation may change under us.
2751  */
2752 bool
vm_pool_low(void)2753 vm_pool_low(void)
2754 {
2755 	/* No locking, at worst we will fib. */
2756 	return vm_page_free_count <= vm_page_free_reserved;
2757 }
2758 
2759 boolean_t vm_darkwake_mode = FALSE;
2760 
2761 /*
2762  * vm_update_darkwake_mode():
2763  *
2764  * Tells the VM that the system is in / out of darkwake.
2765  *
2766  * Today, the VM only lowers/raises the background queue target
2767  * so as to favor consuming more/less background pages when
2768  * darwake is ON/OFF.
2769  *
2770  * We might need to do more things in the future.
2771  */
2772 
2773 void
vm_update_darkwake_mode(boolean_t darkwake_mode)2774 vm_update_darkwake_mode(boolean_t darkwake_mode)
2775 {
2776 #if XNU_TARGET_OS_OSX && defined(__arm64__)
2777 #pragma unused(darkwake_mode)
2778 	assert(vm_darkwake_mode == FALSE);
2779 	/*
2780 	 * Darkwake mode isn't supported for AS macOS.
2781 	 */
2782 	return;
2783 #else /* XNU_TARGET_OS_OSX && __arm64__ */
2784 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2785 
2786 	vm_page_lockspin_queues();
2787 
2788 	if (vm_darkwake_mode == darkwake_mode) {
2789 		/*
2790 		 * No change.
2791 		 */
2792 		vm_page_unlock_queues();
2793 		return;
2794 	}
2795 
2796 	vm_darkwake_mode = darkwake_mode;
2797 
2798 	if (vm_darkwake_mode == TRUE) {
2799 		/* save background target to restore later */
2800 		vm_page_background_target_snapshot = vm_page_background_target;
2801 
2802 		/* target is set to 0...no protection for background pages */
2803 		vm_page_background_target = 0;
2804 	} else if (vm_darkwake_mode == FALSE) {
2805 		if (vm_page_background_target_snapshot) {
2806 			vm_page_background_target = vm_page_background_target_snapshot;
2807 		}
2808 	}
2809 	vm_page_unlock_queues();
2810 #endif
2811 }
2812 
2813 void
vm_page_update_special_state(vm_page_t mem)2814 vm_page_update_special_state(vm_page_t mem)
2815 {
2816 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR || mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
2817 		return;
2818 	}
2819 
2820 	int mode = mem->vmp_on_specialq;
2821 
2822 	switch (mode) {
2823 	case VM_PAGE_SPECIAL_Q_BG:
2824 	{
2825 		task_t  my_task = current_task_early();
2826 
2827 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2828 			return;
2829 		}
2830 
2831 		if (my_task) {
2832 			if (task_get_darkwake_mode(my_task)) {
2833 				return;
2834 			}
2835 		}
2836 
2837 		if (my_task) {
2838 			if (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG)) {
2839 				return;
2840 			}
2841 		}
2842 		vm_page_lockspin_queues();
2843 
2844 		vm_page_background_promoted_count++;
2845 
2846 		vm_page_remove_from_specialq(mem);
2847 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2848 
2849 		vm_page_unlock_queues();
2850 		break;
2851 	}
2852 
2853 	case VM_PAGE_SPECIAL_Q_DONATE:
2854 	{
2855 		task_t  my_task = current_task_early();
2856 
2857 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2858 			return;
2859 		}
2860 
2861 		if (my_task->donates_own_pages == false) {
2862 			vm_page_lockspin_queues();
2863 
2864 			vm_page_remove_from_specialq(mem);
2865 			mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2866 
2867 			vm_page_unlock_queues();
2868 		}
2869 		break;
2870 	}
2871 
2872 	default:
2873 	{
2874 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2875 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2876 		break;
2877 	}
2878 	}
2879 }
2880 
2881 
2882 void
vm_page_assign_special_state(vm_page_t mem,int mode)2883 vm_page_assign_special_state(vm_page_t mem, int mode)
2884 {
2885 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
2886 		return;
2887 	}
2888 
2889 	switch (mode) {
2890 	case VM_PAGE_SPECIAL_Q_BG:
2891 	{
2892 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2893 			return;
2894 		}
2895 
2896 		task_t  my_task = current_task_early();
2897 
2898 		if (my_task) {
2899 			if (task_get_darkwake_mode(my_task)) {
2900 				mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
2901 				return;
2902 			}
2903 		}
2904 
2905 		if (my_task) {
2906 			mem->vmp_on_specialq = (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG) ? VM_PAGE_SPECIAL_Q_BG : VM_PAGE_SPECIAL_Q_EMPTY);
2907 		}
2908 		break;
2909 	}
2910 
2911 	case VM_PAGE_SPECIAL_Q_DONATE:
2912 	{
2913 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2914 			return;
2915 		}
2916 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
2917 		break;
2918 	}
2919 
2920 	default:
2921 		break;
2922 	}
2923 }
2924 
2925 
2926 void
vm_page_remove_from_specialq(vm_page_t mem)2927 vm_page_remove_from_specialq(
2928 	vm_page_t       mem)
2929 {
2930 	vm_object_t     m_object;
2931 	unsigned short  mode;
2932 
2933 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2934 
2935 	mode = mem->vmp_on_specialq;
2936 
2937 	switch (mode) {
2938 	case VM_PAGE_SPECIAL_Q_BG:
2939 	{
2940 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2941 			vm_page_queue_remove(&vm_page_queue_background, mem, vmp_specialq);
2942 
2943 			mem->vmp_specialq.next = 0;
2944 			mem->vmp_specialq.prev = 0;
2945 
2946 			vm_page_background_count--;
2947 
2948 			m_object = VM_PAGE_OBJECT(mem);
2949 
2950 			if (m_object->internal) {
2951 				vm_page_background_internal_count--;
2952 			} else {
2953 				vm_page_background_external_count--;
2954 			}
2955 		}
2956 		break;
2957 	}
2958 
2959 	case VM_PAGE_SPECIAL_Q_DONATE:
2960 	{
2961 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2962 			vm_page_queue_remove((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
2963 			mem->vmp_specialq.next = 0;
2964 			mem->vmp_specialq.prev = 0;
2965 			vm_page_donate_count--;
2966 			if (vm_page_donate_queue_ripe && (vm_page_donate_count < vm_page_donate_target)) {
2967 				assert(vm_page_donate_target == vm_page_donate_target_low);
2968 				vm_page_donate_target = vm_page_donate_target_high;
2969 				vm_page_donate_queue_ripe = false;
2970 			}
2971 		}
2972 
2973 		break;
2974 	}
2975 
2976 	default:
2977 	{
2978 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2979 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2980 		break;
2981 	}
2982 	}
2983 }
2984 
2985 
2986 void
vm_page_add_to_specialq(vm_page_t mem,boolean_t first)2987 vm_page_add_to_specialq(
2988 	vm_page_t       mem,
2989 	boolean_t       first)
2990 {
2991 	vm_object_t     m_object;
2992 
2993 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2994 
2995 	if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2996 		return;
2997 	}
2998 
2999 	int mode = mem->vmp_on_specialq;
3000 
3001 	switch (mode) {
3002 	case VM_PAGE_SPECIAL_Q_BG:
3003 	{
3004 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
3005 			return;
3006 		}
3007 
3008 		m_object = VM_PAGE_OBJECT(mem);
3009 
3010 		if (vm_page_background_exclude_external && !m_object->internal) {
3011 			return;
3012 		}
3013 
3014 		if (first == TRUE) {
3015 			vm_page_queue_enter_first(&vm_page_queue_background, mem, vmp_specialq);
3016 		} else {
3017 			vm_page_queue_enter(&vm_page_queue_background, mem, vmp_specialq);
3018 		}
3019 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
3020 
3021 		vm_page_background_count++;
3022 
3023 		if (m_object->internal) {
3024 			vm_page_background_internal_count++;
3025 		} else {
3026 			vm_page_background_external_count++;
3027 		}
3028 		break;
3029 	}
3030 
3031 	case VM_PAGE_SPECIAL_Q_DONATE:
3032 	{
3033 		if (first == TRUE) {
3034 			vm_page_queue_enter_first((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3035 		} else {
3036 			vm_page_queue_enter((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3037 		}
3038 		vm_page_donate_count++;
3039 		if (!vm_page_donate_queue_ripe && (vm_page_donate_count > vm_page_donate_target)) {
3040 			assert(vm_page_donate_target == vm_page_donate_target_high);
3041 			vm_page_donate_target = vm_page_donate_target_low;
3042 			vm_page_donate_queue_ripe = true;
3043 		}
3044 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
3045 		break;
3046 	}
3047 
3048 	default:
3049 		break;
3050 	}
3051 }
3052 
3053 /*
3054  * This can be switched to FALSE to help debug drivers
3055  * that are having problems with memory > 4G.
3056  */
3057 boolean_t       vm_himemory_mode = TRUE;
3058 
3059 /*
3060  * this interface exists to support hardware controllers
3061  * incapable of generating DMAs with more than 32 bits
3062  * of address on platforms with physical memory > 4G...
3063  */
3064 unsigned int    vm_lopages_allocated_q = 0;
3065 unsigned int    vm_lopages_allocated_cpm_success = 0;
3066 unsigned int    vm_lopages_allocated_cpm_failed = 0;
3067 vm_page_queue_head_t    vm_lopage_queue_free VM_PAGE_PACKED_ALIGNED;
3068 
3069 vm_page_t
vm_page_grablo(void)3070 vm_page_grablo(void)
3071 {
3072 	vm_page_t       mem;
3073 
3074 	if (vm_lopage_needed == FALSE) {
3075 		return vm_page_grab();
3076 	}
3077 
3078 	vm_free_page_lock_spin();
3079 
3080 	if (!vm_page_queue_empty(&vm_lopage_queue_free)) {
3081 		vm_page_queue_remove_first(&vm_lopage_queue_free, mem, vmp_pageq);
3082 		assert(vm_lopage_free_count);
3083 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
3084 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3085 
3086 		vm_lopage_free_count--;
3087 		vm_lopages_allocated_q++;
3088 
3089 		if (vm_lopage_free_count < vm_lopage_lowater) {
3090 			vm_lopage_refill = TRUE;
3091 		}
3092 
3093 		vm_free_page_unlock();
3094 
3095 		if (current_task()->donates_own_pages) {
3096 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3097 		} else {
3098 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3099 		}
3100 	} else {
3101 		vm_free_page_unlock();
3102 
3103 		if (cpm_allocate(PAGE_SIZE, &mem, atop(PPNUM_MAX), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) {
3104 			vm_free_page_lock_spin();
3105 			vm_lopages_allocated_cpm_failed++;
3106 			vm_free_page_unlock();
3107 
3108 			return VM_PAGE_NULL;
3109 		}
3110 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3111 
3112 		mem->vmp_busy = TRUE;
3113 
3114 		vm_page_lockspin_queues();
3115 
3116 		mem->vmp_gobbled = FALSE;
3117 		vm_page_gobble_count--;
3118 		vm_page_wire_count--;
3119 
3120 		vm_lopages_allocated_cpm_success++;
3121 		vm_page_unlock_queues();
3122 	}
3123 	assert(mem->vmp_busy);
3124 	assert(!mem->vmp_pmapped);
3125 	assert(!mem->vmp_wpmapped);
3126 	assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3127 
3128 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3129 
3130 	counter_inc(&vm_page_grab_count);
3131 	VM_DEBUG_EVENT(vm_page_grab, DBG_VM_PAGE_GRAB, DBG_FUNC_NONE, 0, 1, 0, 0);
3132 
3133 	return mem;
3134 }
3135 
3136 /*
3137  *	vm_page_grab:
3138  *
3139  *	first try to grab a page from the per-cpu free list...
3140  *	this must be done while pre-emption is disabled... if
3141  *      a page is available, we're done...
3142  *	if no page is available, grab the vm_page_queue_free_lock
3143  *	and see if current number of free pages would allow us
3144  *      to grab at least 1... if not, return VM_PAGE_NULL as before...
3145  *	if there are pages available, disable preemption and
3146  *      recheck the state of the per-cpu free list... we could
3147  *	have been preempted and moved to a different cpu, or
3148  *      some other thread could have re-filled it... if still
3149  *	empty, figure out how many pages we can steal from the
3150  *	global free queue and move to the per-cpu queue...
3151  *	return 1 of these pages when done... only wakeup the
3152  *      pageout_scan thread if we moved pages from the global
3153  *	list... no need for the wakeup if we've satisfied the
3154  *	request from the per-cpu queue.
3155  */
3156 
3157 #if CONFIG_SECLUDED_MEMORY
3158 vm_page_t vm_page_grab_secluded(void);
3159 #endif /* CONFIG_SECLUDED_MEMORY */
3160 
3161 static inline void
3162 vm_page_grab_diags(void);
3163 
3164 /*
3165  *	vm_page_validate_no_references:
3166  *
3167  *	Make sure the physical page has no refcounts.
3168  *
3169  */
3170 static inline void
vm_page_validate_no_references(vm_page_t mem)3171 vm_page_validate_no_references(
3172 	vm_page_t       mem)
3173 {
3174 	bool is_freed;
3175 
3176 	if (mem->vmp_fictitious) {
3177 		return;
3178 	}
3179 
3180 	pmap_paddr_t paddr = ptoa(VM_PAGE_GET_PHYS_PAGE(mem));
3181 
3182 #if CONFIG_SPTM
3183 	is_freed = pmap_is_page_free(paddr);
3184 #else
3185 	is_freed = pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem));
3186 #endif /* CONFIG_SPTM */
3187 
3188 	if (!is_freed) {
3189 		/*
3190 		 * There is a redundancy here, but we are going to panic anyways,
3191 		 * and ASSERT_PMAP_FREE traces useful information. So, we keep this
3192 		 * behavior.
3193 		 */
3194 		ASSERT_PMAP_FREE(mem);
3195 		panic("%s: page 0x%llx is referenced", __func__, paddr);
3196 	}
3197 }
3198 
3199 vm_page_t
vm_page_grab(void)3200 vm_page_grab(void)
3201 {
3202 	return vm_page_grab_options(VM_PAGE_GRAB_OPTIONS_NONE);
3203 }
3204 
3205 #if HIBERNATION
3206 boolean_t       hibernate_rebuild_needed = FALSE;
3207 #endif /* HIBERNATION */
3208 
3209 static void
vm_page_finalize_grabed_page(vm_page_t mem)3210 vm_page_finalize_grabed_page(vm_page_t mem)
3211 {
3212 	task_t cur_task = current_task_early();
3213 	if (cur_task && cur_task != kernel_task) {
3214 		/* tag:DONATE this is where the donate state of the page is decided according to what task grabs it */
3215 		if (cur_task->donates_own_pages) {
3216 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3217 		} else {
3218 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3219 		}
3220 	}
3221 }
3222 
3223 vm_page_t
vm_page_grab_options(int grab_options)3224 vm_page_grab_options(
3225 	int grab_options)
3226 {
3227 	vm_page_t       mem;
3228 
3229 restart:
3230 	disable_preemption();
3231 
3232 	if ((mem = *PERCPU_GET(free_pages))) {
3233 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
3234 
3235 #if HIBERNATION
3236 		if (hibernate_rebuild_needed) {
3237 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3238 		}
3239 #endif /* HIBERNATION */
3240 
3241 		vm_page_grab_diags();
3242 
3243 		vm_offset_t pcpu_base = current_percpu_base();
3244 		counter_inc_preemption_disabled(&vm_page_grab_count);
3245 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = mem->vmp_snext;
3246 		VM_DEBUG_EVENT(vm_page_grab, DBG_VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3247 
3248 		VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3249 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3250 		enable_preemption();
3251 
3252 		assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3253 		assert(mem->vmp_tabled == FALSE);
3254 		assert(mem->vmp_object == 0);
3255 		assert(!mem->vmp_laundry);
3256 		assert(mem->vmp_busy);
3257 		assert(!mem->vmp_pmapped);
3258 		assert(!mem->vmp_wpmapped);
3259 		assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3260 		assert(!mem->vmp_realtime);
3261 
3262 #if MACH_ASSERT
3263 		if (vm_check_refs_on_alloc) {
3264 			vm_page_validate_no_references(mem);
3265 		}
3266 #endif /* MACH_ASSERT */
3267 		vm_page_finalize_grabed_page(mem);
3268 		return mem;
3269 	}
3270 	enable_preemption();
3271 
3272 
3273 	/*
3274 	 *	Optionally produce warnings if the wire or gobble
3275 	 *	counts exceed some threshold.
3276 	 */
3277 #if VM_PAGE_WIRE_COUNT_WARNING
3278 	if (vm_page_wire_count >= VM_PAGE_WIRE_COUNT_WARNING) {
3279 		printf("mk: vm_page_grab(): high wired page count of %d\n",
3280 		    vm_page_wire_count);
3281 	}
3282 #endif
3283 #if VM_PAGE_GOBBLE_COUNT_WARNING
3284 	if (vm_page_gobble_count >= VM_PAGE_GOBBLE_COUNT_WARNING) {
3285 		printf("mk: vm_page_grab(): high gobbled page count of %d\n",
3286 		    vm_page_gobble_count);
3287 	}
3288 #endif
3289 
3290 	/*
3291 	 * If free count is low and we have delayed pages from early boot,
3292 	 * get one of those instead.
3293 	 */
3294 	if (__improbable(vm_delayed_count > 0 &&
3295 	    vm_page_free_count <= vm_page_free_target &&
3296 	    (mem = vm_get_delayed_page(grab_options)) != NULL)) {
3297 		assert(!mem->vmp_realtime);
3298 		// TODO: missing vm_page_finalize_grabed_page()?
3299 		return mem;
3300 	}
3301 
3302 	vm_free_page_lock_spin();
3303 
3304 	/*
3305 	 *	Only let privileged threads (involved in pageout)
3306 	 *	dip into the reserved pool.
3307 	 */
3308 	if ((vm_page_free_count < vm_page_free_reserved) &&
3309 	    !(current_thread()->options & TH_OPT_VMPRIV)) {
3310 		/* no page for us in the free queue... */
3311 		vm_free_page_unlock();
3312 		mem = VM_PAGE_NULL;
3313 
3314 #if CONFIG_SECLUDED_MEMORY
3315 		/* ... but can we try and grab from the secluded queue? */
3316 		if (vm_page_secluded_count > 0 &&
3317 		    ((grab_options & VM_PAGE_GRAB_SECLUDED) ||
3318 		    task_can_use_secluded_mem(current_task(), TRUE))) {
3319 			mem = vm_page_grab_secluded();
3320 			if (grab_options & VM_PAGE_GRAB_SECLUDED) {
3321 				vm_page_secluded.grab_for_iokit++;
3322 				if (mem) {
3323 					vm_page_secluded.grab_for_iokit_success++;
3324 				}
3325 			}
3326 			if (mem) {
3327 				VM_CHECK_MEMORYSTATUS;
3328 
3329 				vm_page_grab_diags();
3330 				counter_inc(&vm_page_grab_count);
3331 				VM_DEBUG_EVENT(vm_page_grab, DBG_VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3332 
3333 				assert(!mem->vmp_realtime);
3334 				// TODO: missing vm_page_finalize_grabed_page()?
3335 				return mem;
3336 			}
3337 		}
3338 #else /* CONFIG_SECLUDED_MEMORY */
3339 		(void) grab_options;
3340 #endif /* CONFIG_SECLUDED_MEMORY */
3341 	} else {
3342 		vm_page_t        head;
3343 		vm_page_t        tail;
3344 		unsigned int     pages_to_steal;
3345 		unsigned int     color;
3346 		unsigned int clump_end, sub_count;
3347 
3348 		/*
3349 		 * Replenishing our per-CPU cache of free pages might take
3350 		 * too long to keep holding the "free_page" lock as a spinlock,
3351 		 * so convert to the full mutex to prevent other threads trying
3352 		 * to acquire the "free_page" lock from timing out spinning on
3353 		 * the mutex interlock.
3354 		 */
3355 		vm_free_page_lock_convert();
3356 
3357 		while (vm_page_free_count == 0) {
3358 			vm_free_page_unlock();
3359 			/*
3360 			 * must be a privileged thread to be
3361 			 * in this state since a non-privileged
3362 			 * thread would have bailed if we were
3363 			 * under the vm_page_free_reserved mark
3364 			 */
3365 			VM_PAGE_WAIT();
3366 			vm_free_page_lock();
3367 		}
3368 
3369 		/*
3370 		 * Need to repopulate the per-CPU free list from the global free list.
3371 		 * Note we don't do any processing of pending retirement pages here.
3372 		 * That'll happen in the code above when the page comes off the per-CPU list.
3373 		 */
3374 		disable_preemption();
3375 
3376 		/*
3377 		 * If we got preempted the cache might now have pages.
3378 		 */
3379 		if ((mem = *PERCPU_GET(free_pages))) {
3380 			vm_free_page_unlock();
3381 			enable_preemption();
3382 			goto restart;
3383 		}
3384 
3385 		if (vm_page_free_count <= vm_page_free_reserved) {
3386 			pages_to_steal = 1;
3387 		} else {
3388 			if (vm_free_magazine_refill_limit <= (vm_page_free_count - vm_page_free_reserved)) {
3389 				pages_to_steal = vm_free_magazine_refill_limit;
3390 			} else {
3391 				pages_to_steal = (vm_page_free_count - vm_page_free_reserved);
3392 			}
3393 		}
3394 		color = *PERCPU_GET(start_color);
3395 		head = tail = NULL;
3396 
3397 		vm_page_free_count -= pages_to_steal;
3398 		clump_end = sub_count = 0;
3399 
3400 		while (pages_to_steal--) {
3401 			while (vm_page_queue_empty(&vm_page_queue_free[color].qhead)) {
3402 				color = (color + 1) & vm_color_mask;
3403 			}
3404 #if defined(__x86_64__)
3405 			vm_page_queue_remove_first_with_clump(&vm_page_queue_free[color].qhead,
3406 			    mem, clump_end);
3407 #else
3408 			vm_page_queue_remove_first(&vm_page_queue_free[color].qhead,
3409 			    mem, vmp_pageq);
3410 #endif
3411 
3412 			assert(mem->vmp_q_state == VM_PAGE_ON_FREE_Q);
3413 
3414 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3415 
3416 #if defined(__arm64__)
3417 			color = (color + 1) & vm_color_mask;
3418 #else
3419 
3420 #if DEVELOPMENT || DEBUG
3421 
3422 			sub_count++;
3423 			if (clump_end) {
3424 				vm_clump_update_stats(sub_count);
3425 				sub_count = 0;
3426 				color = (color + 1) & vm_color_mask;
3427 			}
3428 #else
3429 			if (clump_end) {
3430 				color = (color + 1) & vm_color_mask;
3431 			}
3432 
3433 #endif /* if DEVELOPMENT || DEBUG */
3434 
3435 #endif  /* if defined(__arm64__) */
3436 
3437 			if (head == NULL) {
3438 				head = mem;
3439 			} else {
3440 				tail->vmp_snext = mem;
3441 			}
3442 			tail = mem;
3443 
3444 			assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3445 			assert(mem->vmp_tabled == FALSE);
3446 			assert(mem->vmp_object == 0);
3447 			assert(!mem->vmp_laundry);
3448 
3449 			mem->vmp_q_state = VM_PAGE_ON_FREE_LOCAL_Q;
3450 
3451 			assert(mem->vmp_busy);
3452 			assert(!mem->vmp_pmapped);
3453 			assert(!mem->vmp_wpmapped);
3454 			assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3455 			assert(!mem->vmp_realtime);
3456 #if MACH_ASSERT
3457 			if (vm_check_refs_on_alloc) {
3458 				vm_page_validate_no_references(mem);
3459 			}
3460 #endif /* MACH_ASSERT */
3461 		}
3462 #if defined (__x86_64__) && (DEVELOPMENT || DEBUG)
3463 		vm_clump_update_stats(sub_count);
3464 #endif
3465 
3466 #if HIBERNATION
3467 		if (hibernate_rebuild_needed) {
3468 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3469 		}
3470 #endif /* HIBERNATION */
3471 		vm_offset_t pcpu_base = current_percpu_base();
3472 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = head;
3473 		*PERCPU_GET_WITH_BASE(pcpu_base, start_color) = color;
3474 
3475 		/*
3476 		 * We decremented vm_page_free_count above
3477 		 * so we must wake up vm_pageout_scan() if
3478 		 * we brought it down below vm_page_free_min.
3479 		 */
3480 		bool wakeup_pageout_scan = false;
3481 		if (vm_page_free_count < vm_page_free_min &&
3482 		    !vm_pageout_running) {
3483 			wakeup_pageout_scan = true;
3484 		}
3485 		vm_free_page_unlock();
3486 
3487 		enable_preemption();
3488 
3489 		if (wakeup_pageout_scan) {
3490 			thread_wakeup((event_t) &vm_page_free_wanted);
3491 		}
3492 		VM_CHECK_MEMORYSTATUS;
3493 
3494 		goto restart;
3495 	}
3496 
3497 	/*
3498 	 *	Decide if we should poke the pageout daemon.
3499 	 *	We do this if the free count is less than the low
3500 	 *	water mark. VM Pageout Scan will keep running till
3501 	 *	the free_count > free_target (& hence above free_min).
3502 	 *	This wakeup is to catch the possibility of the counts
3503 	 *	dropping between VM Pageout Scan parking and this check.
3504 	 *
3505 	 *	We don't have the counts locked ... if they change a little,
3506 	 *	it doesn't really matter.
3507 	 */
3508 	if (vm_page_free_count < vm_page_free_min) {
3509 		vm_free_page_lock();
3510 		if (vm_pageout_running == FALSE) {
3511 			vm_free_page_unlock();
3512 			thread_wakeup((event_t) &vm_page_free_wanted);
3513 		} else {
3514 			vm_free_page_unlock();
3515 		}
3516 	}
3517 
3518 	VM_CHECK_MEMORYSTATUS;
3519 
3520 	if (mem) {
3521 		assert(!mem->vmp_realtime);
3522 //		dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 4);	/* (TEST/DEBUG) */
3523 
3524 		vm_page_finalize_grabed_page(mem);
3525 	}
3526 	return mem;
3527 }
3528 
3529 #if CONFIG_SECLUDED_MEMORY
3530 vm_page_t
vm_page_grab_secluded(void)3531 vm_page_grab_secluded(void)
3532 {
3533 	vm_page_t       mem;
3534 	vm_object_t     object;
3535 	int             refmod_state;
3536 
3537 	if (vm_page_secluded_count == 0) {
3538 		/* no secluded pages to grab... */
3539 		return VM_PAGE_NULL;
3540 	}
3541 
3542 	/* secluded queue is protected by the VM page queue lock */
3543 	vm_page_lock_queues();
3544 
3545 	if (vm_page_secluded_count == 0) {
3546 		/* no secluded pages to grab... */
3547 		vm_page_unlock_queues();
3548 		return VM_PAGE_NULL;
3549 	}
3550 
3551 #if 00
3552 	/* can we grab from the secluded queue? */
3553 	if (vm_page_secluded_count > vm_page_secluded_target ||
3554 	    (vm_page_secluded_count > 0 &&
3555 	    task_can_use_secluded_mem(current_task(), TRUE))) {
3556 		/* OK */
3557 	} else {
3558 		/* can't grab from secluded queue... */
3559 		vm_page_unlock_queues();
3560 		return VM_PAGE_NULL;
3561 	}
3562 #endif
3563 
3564 	/* we can grab a page from secluded queue! */
3565 	assert((vm_page_secluded_count_free +
3566 	    vm_page_secluded_count_inuse) ==
3567 	    vm_page_secluded_count);
3568 	if (current_task()->task_can_use_secluded_mem) {
3569 		assert(num_tasks_can_use_secluded_mem > 0);
3570 	}
3571 	assert(!vm_page_queue_empty(&vm_page_queue_secluded));
3572 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3573 	mem = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3574 	assert(mem->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3575 	vm_page_queues_remove(mem, TRUE);
3576 
3577 	object = VM_PAGE_OBJECT(mem);
3578 
3579 	assert(!mem->vmp_fictitious);
3580 	assert(!VM_PAGE_WIRED(mem));
3581 	if (object == VM_OBJECT_NULL) {
3582 		/* free for grab! */
3583 		vm_page_unlock_queues();
3584 		vm_page_secluded.grab_success_free++;
3585 
3586 		assert(mem->vmp_busy);
3587 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3588 		assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3589 		assert(mem->vmp_pageq.next == 0);
3590 		assert(mem->vmp_pageq.prev == 0);
3591 		assert(mem->vmp_listq.next == 0);
3592 		assert(mem->vmp_listq.prev == 0);
3593 		assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3594 		assert(mem->vmp_specialq.next == 0);
3595 		assert(mem->vmp_specialq.prev == 0);
3596 		return mem;
3597 	}
3598 
3599 	assert(!object->internal);
3600 //	vm_page_pageable_external_count--;
3601 
3602 	if (!vm_object_lock_try(object)) {
3603 //		printf("SECLUDED: page %p: object %p locked\n", mem, object);
3604 		vm_page_secluded.grab_failure_locked++;
3605 reactivate_secluded_page:
3606 		vm_page_activate(mem);
3607 		vm_page_unlock_queues();
3608 		return VM_PAGE_NULL;
3609 	}
3610 	if (mem->vmp_busy ||
3611 	    mem->vmp_cleaning ||
3612 	    mem->vmp_laundry) {
3613 		/* can't steal page in this state... */
3614 		vm_object_unlock(object);
3615 		vm_page_secluded.grab_failure_state++;
3616 		goto reactivate_secluded_page;
3617 	}
3618 	if (mem->vmp_realtime) {
3619 		/* don't steal pages used by realtime threads... */
3620 		vm_object_unlock(object);
3621 		vm_page_secluded.grab_failure_realtime++;
3622 		goto reactivate_secluded_page;
3623 	}
3624 
3625 	mem->vmp_busy = TRUE;
3626 	refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem));
3627 	if (refmod_state & VM_MEM_REFERENCED) {
3628 		mem->vmp_reference = TRUE;
3629 	}
3630 	if (refmod_state & VM_MEM_MODIFIED) {
3631 		SET_PAGE_DIRTY(mem, FALSE);
3632 	}
3633 	if (mem->vmp_dirty || mem->vmp_precious) {
3634 		/* can't grab a dirty page; re-activate */
3635 //		printf("SECLUDED: dirty page %p\n", mem);
3636 		vm_page_wakeup_done(object, mem);
3637 		vm_page_secluded.grab_failure_dirty++;
3638 		vm_object_unlock(object);
3639 		goto reactivate_secluded_page;
3640 	}
3641 	if (mem->vmp_reference) {
3642 		/* it's been used but we do need to grab a page... */
3643 	}
3644 
3645 	vm_page_unlock_queues();
3646 
3647 
3648 	/* finish what vm_page_free() would have done... */
3649 	vm_page_free_prepare_object(mem, TRUE);
3650 	vm_object_unlock(object);
3651 	object = VM_OBJECT_NULL;
3652 
3653 #if MACH_ASSERT
3654 	if (vm_check_refs_on_alloc) {
3655 		vm_page_validate_no_references(mem);
3656 	}
3657 #endif /* MACH_ASSERT */
3658 
3659 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3660 	vm_page_secluded.grab_success_other++;
3661 
3662 	assert(mem->vmp_busy);
3663 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3664 	assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3665 	assert(mem->vmp_pageq.next == 0);
3666 	assert(mem->vmp_pageq.prev == 0);
3667 	assert(mem->vmp_listq.next == 0);
3668 	assert(mem->vmp_listq.prev == 0);
3669 	assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3670 	assert(mem->vmp_specialq.next == 0);
3671 	assert(mem->vmp_specialq.prev == 0);
3672 
3673 	return mem;
3674 }
3675 
3676 uint64_t
vm_page_secluded_drain(void)3677 vm_page_secluded_drain(void)
3678 {
3679 	vm_page_t local_freeq;
3680 	int local_freed;
3681 	uint64_t num_reclaimed;
3682 	unsigned int saved_secluded_count, saved_secluded_target;
3683 
3684 	num_reclaimed = 0;
3685 	local_freeq = NULL;
3686 	local_freed = 0;
3687 
3688 	vm_page_lock_queues();
3689 
3690 	saved_secluded_count = vm_page_secluded_count;
3691 	saved_secluded_target = vm_page_secluded_target;
3692 	vm_page_secluded_target = 0;
3693 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3694 	while (vm_page_secluded_count) {
3695 		vm_page_t secluded_page;
3696 
3697 		assert((vm_page_secluded_count_free +
3698 		    vm_page_secluded_count_inuse) ==
3699 		    vm_page_secluded_count);
3700 		secluded_page = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3701 		assert(secluded_page->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3702 
3703 		vm_page_queues_remove(secluded_page, FALSE);
3704 		assert(!secluded_page->vmp_fictitious);
3705 		assert(!VM_PAGE_WIRED(secluded_page));
3706 
3707 		if (secluded_page->vmp_object == 0) {
3708 			/* transfer to free queue */
3709 			assert(secluded_page->vmp_busy);
3710 			secluded_page->vmp_snext = local_freeq;
3711 			local_freeq = secluded_page;
3712 			local_freed += 1;
3713 		} else {
3714 			/* transfer to head of active queue */
3715 			vm_page_enqueue_active(secluded_page, FALSE);
3716 			secluded_page = VM_PAGE_NULL;
3717 		}
3718 		num_reclaimed++;
3719 	}
3720 	vm_page_secluded_target = saved_secluded_target;
3721 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3722 
3723 //	printf("FBDP %s:%d secluded_count %d->%d, target %d, reclaimed %lld\n", __FUNCTION__, __LINE__, saved_secluded_count, vm_page_secluded_count, vm_page_secluded_target, num_reclaimed);
3724 
3725 	vm_page_unlock_queues();
3726 
3727 	if (local_freed) {
3728 		vm_page_free_list(local_freeq, TRUE);
3729 		local_freeq = NULL;
3730 		local_freed = 0;
3731 	}
3732 
3733 	return num_reclaimed;
3734 }
3735 #endif /* CONFIG_SECLUDED_MEMORY */
3736 
3737 static inline void
vm_page_grab_diags()3738 vm_page_grab_diags()
3739 {
3740 #if DEVELOPMENT || DEBUG
3741 	task_t task = current_task_early();
3742 	if (task == NULL) {
3743 		return;
3744 	}
3745 
3746 	ledger_credit(task->ledger, task_ledgers.pages_grabbed, 1);
3747 #endif /* DEVELOPMENT || DEBUG */
3748 }
3749 
3750 /*
3751  *	vm_page_release:
3752  *
3753  *	Return a page to the free list.
3754  */
3755 
3756 void
vm_page_release(vm_page_t mem,boolean_t page_queues_locked)3757 vm_page_release(
3758 	vm_page_t       mem,
3759 	boolean_t       page_queues_locked)
3760 {
3761 	unsigned int    color;
3762 	int     need_wakeup = 0;
3763 	int     need_priv_wakeup = 0;
3764 #if CONFIG_SECLUDED_MEMORY
3765 	int     need_secluded_wakeup = 0;
3766 #endif /* CONFIG_SECLUDED_MEMORY */
3767 	event_t wakeup_event = NULL;
3768 
3769 	if (page_queues_locked) {
3770 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3771 	} else {
3772 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
3773 	}
3774 
3775 	assert(!mem->vmp_private && !mem->vmp_fictitious);
3776 
3777 	vm_page_validate_no_references(mem);
3778 
3779 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 5);	/* (TEST/DEBUG) */
3780 
3781 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3782 
3783 	if (__improbable(mem->vmp_realtime)) {
3784 		if (!page_queues_locked) {
3785 			vm_page_lock_queues();
3786 		}
3787 		if (mem->vmp_realtime) {
3788 			mem->vmp_realtime = false;
3789 			vm_page_realtime_count--;
3790 		}
3791 		if (!page_queues_locked) {
3792 			vm_page_unlock_queues();
3793 		}
3794 	}
3795 
3796 	vm_free_page_lock_spin();
3797 
3798 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3799 	assert(mem->vmp_busy);
3800 	assert(!mem->vmp_laundry);
3801 	assert(mem->vmp_object == 0);
3802 	assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
3803 	assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3804 	assert(mem->vmp_specialq.next == 0 && mem->vmp_specialq.prev == 0);
3805 
3806 	/* Clear any specialQ hints before releasing page to the free pool*/
3807 	mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
3808 
3809 	if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
3810 	    vm_lopage_free_count < vm_lopage_free_limit &&
3811 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3812 		/*
3813 		 * this exists to support hardware controllers
3814 		 * incapable of generating DMAs with more than 32 bits
3815 		 * of address on platforms with physical memory > 4G...
3816 		 */
3817 		vm_page_queue_enter_first(&vm_lopage_queue_free, mem, vmp_pageq);
3818 		vm_lopage_free_count++;
3819 
3820 		if (vm_lopage_free_count >= vm_lopage_free_limit) {
3821 			vm_lopage_refill = FALSE;
3822 		}
3823 
3824 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3825 		mem->vmp_lopage = TRUE;
3826 #if CONFIG_SECLUDED_MEMORY
3827 	} else if (vm_page_free_count > vm_page_free_reserved &&
3828 	    vm_page_secluded_count < vm_page_secluded_target &&
3829 	    num_tasks_can_use_secluded_mem == 0) {
3830 		/*
3831 		 * XXX FBDP TODO: also avoid refilling secluded queue
3832 		 * when some IOKit objects are already grabbing from it...
3833 		 */
3834 		if (!page_queues_locked) {
3835 			if (!vm_page_trylock_queues()) {
3836 				/* take locks in right order */
3837 				vm_free_page_unlock();
3838 				vm_page_lock_queues();
3839 				vm_free_page_lock_spin();
3840 			}
3841 		}
3842 		mem->vmp_lopage = FALSE;
3843 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3844 		vm_page_queue_enter_first(&vm_page_queue_secluded, mem, vmp_pageq);
3845 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3846 		vm_page_secluded_count++;
3847 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3848 		vm_page_secluded_count_free++;
3849 		if (!page_queues_locked) {
3850 			vm_page_unlock_queues();
3851 		}
3852 		LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_OWNED);
3853 		if (vm_page_free_wanted_secluded > 0) {
3854 			vm_page_free_wanted_secluded--;
3855 			need_secluded_wakeup = 1;
3856 		}
3857 #endif /* CONFIG_SECLUDED_MEMORY */
3858 	} else {
3859 		mem->vmp_lopage = FALSE;
3860 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3861 
3862 		color = VM_PAGE_GET_COLOR(mem);
3863 #if defined(__x86_64__)
3864 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
3865 #else
3866 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
3867 #endif
3868 		vm_page_free_count++;
3869 		/*
3870 		 *	Check if we should wake up someone waiting for page.
3871 		 *	But don't bother waking them unless they can allocate.
3872 		 *
3873 		 *	We wakeup only one thread, to prevent starvation.
3874 		 *	Because the scheduling system handles wait queues FIFO,
3875 		 *	if we wakeup all waiting threads, one greedy thread
3876 		 *	can starve multiple niceguy threads.  When the threads
3877 		 *	all wakeup, the greedy threads runs first, grabs the page,
3878 		 *	and waits for another page.  It will be the first to run
3879 		 *	when the next page is freed.
3880 		 *
3881 		 *	However, there is a slight danger here.
3882 		 *	The thread we wake might not use the free page.
3883 		 *	Then the other threads could wait indefinitely
3884 		 *	while the page goes unused.  To forestall this,
3885 		 *	the pageout daemon will keep making free pages
3886 		 *	as long as vm_page_free_wanted is non-zero.
3887 		 */
3888 
3889 		assert(vm_page_free_count > 0);
3890 		if (vm_page_free_wanted_privileged > 0) {
3891 			vm_page_free_wanted_privileged--;
3892 			need_priv_wakeup = 1;
3893 #if CONFIG_SECLUDED_MEMORY
3894 		} else if (vm_page_free_wanted_secluded > 0 &&
3895 		    vm_page_free_count > vm_page_free_reserved) {
3896 			vm_page_free_wanted_secluded--;
3897 			need_secluded_wakeup = 1;
3898 #endif /* CONFIG_SECLUDED_MEMORY */
3899 		} else if (vm_page_free_wanted > 0 &&
3900 		    vm_page_free_count > vm_page_free_reserved) {
3901 			vm_page_free_wanted--;
3902 			need_wakeup = 1;
3903 		}
3904 	}
3905 	vm_pageout_vminfo.vm_page_pages_freed++;
3906 
3907 	vm_free_page_unlock();
3908 
3909 	VM_DEBUG_CONSTANT_EVENT(vm_page_release, DBG_VM_PAGE_RELEASE, DBG_FUNC_NONE, 1, 0, 0, 0);
3910 
3911 	if (need_priv_wakeup) {
3912 		wakeup_event = &vm_page_free_wanted_privileged;
3913 	}
3914 #if CONFIG_SECLUDED_MEMORY
3915 	else if (need_secluded_wakeup) {
3916 		wakeup_event = &vm_page_free_wanted_secluded;
3917 	}
3918 #endif /* CONFIG_SECLUDED_MEMORY */
3919 	else if (need_wakeup) {
3920 		wakeup_event = &vm_page_free_count;
3921 	}
3922 
3923 	if (wakeup_event) {
3924 		if (vps_dynamic_priority_enabled) {
3925 			wakeup_one_with_inheritor((event_t) wakeup_event,
3926 			    THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH,
3927 			    NULL);
3928 		} else {
3929 			thread_wakeup_one((event_t) wakeup_event);
3930 		}
3931 	}
3932 
3933 	VM_CHECK_MEMORYSTATUS;
3934 }
3935 
3936 /*
3937  * This version of vm_page_release() is used only at startup
3938  * when we are single-threaded and pages are being released
3939  * for the first time. Hence, no locking or unnecessary checks are made.
3940  * Note: VM_CHECK_MEMORYSTATUS invoked by the caller.
3941  */
3942 void
vm_page_release_startup(vm_page_t mem)3943 vm_page_release_startup(
3944 	vm_page_t       mem)
3945 {
3946 	vm_page_queue_t queue_free;
3947 
3948 	if (vm_lopage_free_count < vm_lopage_free_limit &&
3949 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3950 		mem->vmp_lopage = TRUE;
3951 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3952 		vm_lopage_free_count++;
3953 		queue_free = &vm_lopage_queue_free;
3954 #if CONFIG_SECLUDED_MEMORY
3955 	} else if (vm_page_secluded_count < vm_page_secluded_target) {
3956 		mem->vmp_lopage = FALSE;
3957 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3958 		vm_page_secluded_count++;
3959 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3960 		vm_page_secluded_count_free++;
3961 		queue_free = &vm_page_queue_secluded;
3962 #endif /* CONFIG_SECLUDED_MEMORY */
3963 	} else {
3964 		mem->vmp_lopage = FALSE;
3965 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3966 		vm_page_free_count++;
3967 		queue_free = &vm_page_queue_free[VM_PAGE_GET_COLOR(mem)].qhead;
3968 	}
3969 	if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
3970 #if defined(__x86_64__)
3971 		vm_page_queue_enter_clump(queue_free, mem);
3972 #else
3973 		vm_page_queue_enter(queue_free, mem, vmp_pageq);
3974 #endif
3975 	} else {
3976 		vm_page_queue_enter_first(queue_free, mem, vmp_pageq);
3977 	}
3978 }
3979 
3980 /*
3981  *	vm_page_wait:
3982  *
3983  *	Wait for a page to become available.
3984  *	If there are plenty of free pages, then we don't sleep.
3985  *
3986  *	Returns:
3987  *		TRUE:  There may be another page, try again
3988  *		FALSE: We were interrupted out of our wait, don't try again
3989  */
3990 
3991 boolean_t
vm_page_wait(int interruptible)3992 vm_page_wait(
3993 	int     interruptible )
3994 {
3995 	/*
3996 	 *	We can't use vm_page_free_reserved to make this
3997 	 *	determination.  Consider: some thread might
3998 	 *	need to allocate two pages.  The first allocation
3999 	 *	succeeds, the second fails.  After the first page is freed,
4000 	 *	a call to vm_page_wait must really block.
4001 	 */
4002 	kern_return_t   wait_result;
4003 	int             need_wakeup = 0;
4004 	int             is_privileged = current_thread()->options & TH_OPT_VMPRIV;
4005 	event_t         wait_event = NULL;
4006 
4007 	vm_free_page_lock_spin();
4008 
4009 	if (is_privileged && vm_page_free_count) {
4010 		vm_free_page_unlock();
4011 		return TRUE;
4012 	}
4013 
4014 	if (vm_page_free_count >= vm_page_free_target) {
4015 		vm_free_page_unlock();
4016 		return TRUE;
4017 	}
4018 
4019 	if (is_privileged) {
4020 		if (vm_page_free_wanted_privileged++ == 0) {
4021 			need_wakeup = 1;
4022 		}
4023 		wait_event = (event_t)&vm_page_free_wanted_privileged;
4024 #if CONFIG_SECLUDED_MEMORY
4025 	} else if (secluded_for_apps &&
4026 	    task_can_use_secluded_mem(current_task(), FALSE)) {
4027 #if 00
4028 		/* XXX FBDP: need pageq lock for this... */
4029 		/* XXX FBDP: might wait even if pages available, */
4030 		/* XXX FBDP: hopefully not for too long... */
4031 		if (vm_page_secluded_count > 0) {
4032 			vm_free_page_unlock();
4033 			return TRUE;
4034 		}
4035 #endif
4036 		if (vm_page_free_wanted_secluded++ == 0) {
4037 			need_wakeup = 1;
4038 		}
4039 		wait_event = (event_t)&vm_page_free_wanted_secluded;
4040 #endif /* CONFIG_SECLUDED_MEMORY */
4041 	} else {
4042 		if (vm_page_free_wanted++ == 0) {
4043 			need_wakeup = 1;
4044 		}
4045 		wait_event = (event_t)&vm_page_free_count;
4046 	}
4047 
4048 	/*
4049 	 * We don't do a vm_pageout_scan wakeup if we already have
4050 	 * some waiters because vm_pageout_scan checks for waiters
4051 	 * before it returns and does so behind the vm_page_queue_free_lock,
4052 	 * which we own when we bump the waiter counts.
4053 	 */
4054 
4055 	if (vps_dynamic_priority_enabled) {
4056 		/*
4057 		 * We are waking up vm_pageout_scan here. If it needs
4058 		 * the vm_page_queue_free_lock before we unlock it
4059 		 * we'll end up just blocking and incur an extra
4060 		 * context switch. Could be a perf. issue.
4061 		 */
4062 
4063 		if (need_wakeup) {
4064 			thread_wakeup((event_t)&vm_page_free_wanted);
4065 		}
4066 
4067 		/*
4068 		 * LD: This event is going to get recorded every time because
4069 		 * we don't get back THREAD_WAITING from lck_mtx_sleep_with_inheritor.
4070 		 * We just block in that routine.
4071 		 */
4072 		VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, DBG_VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
4073 		    vm_page_free_wanted_privileged,
4074 		    vm_page_free_wanted,
4075 #if CONFIG_SECLUDED_MEMORY
4076 		    vm_page_free_wanted_secluded,
4077 #else /* CONFIG_SECLUDED_MEMORY */
4078 		    0,
4079 #endif /* CONFIG_SECLUDED_MEMORY */
4080 		    0);
4081 		wait_result =  lck_mtx_sleep_with_inheritor(&vm_page_queue_free_lock,
4082 		    LCK_SLEEP_UNLOCK,
4083 		    wait_event,
4084 		    vm_pageout_scan_thread,
4085 		    interruptible,
4086 		    0);
4087 	} else {
4088 		wait_result = assert_wait(wait_event, interruptible);
4089 
4090 		vm_free_page_unlock();
4091 
4092 		if (need_wakeup) {
4093 			thread_wakeup((event_t)&vm_page_free_wanted);
4094 		}
4095 
4096 		if (wait_result == THREAD_WAITING) {
4097 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, DBG_VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
4098 			    vm_page_free_wanted_privileged,
4099 			    vm_page_free_wanted,
4100 #if CONFIG_SECLUDED_MEMORY
4101 			    vm_page_free_wanted_secluded,
4102 #else /* CONFIG_SECLUDED_MEMORY */
4103 			    0,
4104 #endif /* CONFIG_SECLUDED_MEMORY */
4105 			    0);
4106 			wait_result = thread_block(THREAD_CONTINUE_NULL);
4107 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block,
4108 			    DBG_VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0);
4109 		}
4110 	}
4111 
4112 	return (wait_result == THREAD_AWAKENED) || (wait_result == THREAD_NOT_WAITING);
4113 }
4114 
4115 /*
4116  *	vm_page_alloc:
4117  *
4118  *	Allocate and return a memory cell associated
4119  *	with this VM object/offset pair.
4120  *
4121  *	Object must be locked.
4122  */
4123 
4124 vm_page_t
vm_page_alloc(vm_object_t object,vm_object_offset_t offset)4125 vm_page_alloc(
4126 	vm_object_t             object,
4127 	vm_object_offset_t      offset)
4128 {
4129 	vm_page_t       mem;
4130 	int             grab_options;
4131 
4132 	vm_object_lock_assert_exclusive(object);
4133 	grab_options = 0;
4134 #if CONFIG_SECLUDED_MEMORY
4135 	if (object->can_grab_secluded) {
4136 		grab_options |= VM_PAGE_GRAB_SECLUDED;
4137 	}
4138 #endif /* CONFIG_SECLUDED_MEMORY */
4139 	mem = vm_page_grab_options(grab_options);
4140 	if (mem == VM_PAGE_NULL) {
4141 		return VM_PAGE_NULL;
4142 	}
4143 
4144 	vm_page_insert(mem, object, offset);
4145 
4146 	return mem;
4147 }
4148 
4149 /*
4150  *	vm_page_free_prepare:
4151  *
4152  *	Removes page from any queue it may be on
4153  *	and disassociates it from its VM object.
4154  *
4155  *	Object and page queues must be locked prior to entry.
4156  */
4157 static void
vm_page_free_prepare(vm_page_t mem)4158 vm_page_free_prepare(
4159 	vm_page_t       mem)
4160 {
4161 #if CONFIG_SPTM
4162 	/**
4163 	 * SPTM TODO: The pmap should retype frames automatically as mappings to them are
4164 	 *            created and destroyed. In order to catch potential cases where this
4165 	 *            does not happen, add an appropriate assert here. This code should be
4166 	 *            executed on every frame that is about to be released to the VM.
4167 	 */
4168 	const sptm_paddr_t paddr = ((uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)) << PAGE_SHIFT;
4169 	__unused const sptm_frame_type_t frame_type = sptm_get_frame_type(paddr);
4170 
4171 	assert(frame_type == XNU_DEFAULT);
4172 #endif /* CONFIG_SPTM */
4173 
4174 	vm_page_free_prepare_queues(mem);
4175 	vm_page_free_prepare_object(mem, TRUE);
4176 }
4177 
4178 
4179 void
vm_page_free_prepare_queues(vm_page_t mem)4180 vm_page_free_prepare_queues(
4181 	vm_page_t       mem)
4182 {
4183 	vm_object_t     m_object;
4184 
4185 	VM_PAGE_CHECK(mem);
4186 
4187 	assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
4188 	assert(!mem->vmp_cleaning);
4189 	m_object = VM_PAGE_OBJECT(mem);
4190 
4191 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4192 	if (m_object) {
4193 		vm_object_lock_assert_exclusive(m_object);
4194 	}
4195 	if (mem->vmp_laundry) {
4196 		/*
4197 		 * We may have to free a page while it's being laundered
4198 		 * if we lost its pager (due to a forced unmount, for example).
4199 		 * We need to call vm_pageout_steal_laundry() before removing
4200 		 * the page from its VM object, so that we can remove it
4201 		 * from its pageout queue and adjust the laundry accounting
4202 		 */
4203 		vm_pageout_steal_laundry(mem, TRUE);
4204 	}
4205 
4206 	vm_page_queues_remove(mem, TRUE);
4207 
4208 	if (__improbable(mem->vmp_realtime)) {
4209 		mem->vmp_realtime = false;
4210 		vm_page_realtime_count--;
4211 	}
4212 
4213 	if (VM_PAGE_WIRED(mem)) {
4214 		assert(mem->vmp_wire_count > 0);
4215 
4216 		if (m_object) {
4217 			task_t          owner;
4218 			int             ledger_idx_volatile;
4219 			int             ledger_idx_nonvolatile;
4220 			int             ledger_idx_volatile_compressed;
4221 			int             ledger_idx_nonvolatile_compressed;
4222 			int             ledger_idx_composite;
4223 			int             ledger_idx_external_wired;
4224 			boolean_t       do_footprint;
4225 
4226 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4227 			VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4228 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4229 
4230 			assert(m_object->resident_page_count >=
4231 			    m_object->wired_page_count);
4232 
4233 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4234 				OSAddAtomic(+1, &vm_page_purgeable_count);
4235 				assert(vm_page_purgeable_wired_count > 0);
4236 				OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4237 			}
4238 			if (m_object->internal &&
4239 			    m_object->vo_owner != TASK_NULL &&
4240 			    (m_object->purgable == VM_PURGABLE_VOLATILE ||
4241 			    m_object->purgable == VM_PURGABLE_EMPTY)) {
4242 				owner = VM_OBJECT_OWNER(m_object);
4243 				vm_object_ledger_tag_ledgers(
4244 					m_object,
4245 					&ledger_idx_volatile,
4246 					&ledger_idx_nonvolatile,
4247 					&ledger_idx_volatile_compressed,
4248 					&ledger_idx_nonvolatile_compressed,
4249 					&ledger_idx_composite,
4250 					&ledger_idx_external_wired,
4251 					&do_footprint);
4252 				/*
4253 				 * While wired, this page was accounted
4254 				 * as "non-volatile" but it should now
4255 				 * be accounted as "volatile".
4256 				 */
4257 				/* one less "non-volatile"... */
4258 				ledger_debit(owner->ledger,
4259 				    ledger_idx_nonvolatile,
4260 				    PAGE_SIZE);
4261 				if (do_footprint) {
4262 					/* ... and "phys_footprint" */
4263 					ledger_debit(owner->ledger,
4264 					    task_ledgers.phys_footprint,
4265 					    PAGE_SIZE);
4266 				} else if (ledger_idx_composite != -1) {
4267 					ledger_debit(owner->ledger,
4268 					    ledger_idx_composite,
4269 					    PAGE_SIZE);
4270 				}
4271 				/* one more "volatile" */
4272 				ledger_credit(owner->ledger,
4273 				    ledger_idx_volatile,
4274 				    PAGE_SIZE);
4275 			}
4276 		}
4277 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4278 			vm_page_wire_count--;
4279 		}
4280 
4281 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4282 		mem->vmp_wire_count = 0;
4283 		assert(!mem->vmp_gobbled);
4284 	} else if (mem->vmp_gobbled) {
4285 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4286 			vm_page_wire_count--;
4287 		}
4288 		vm_page_gobble_count--;
4289 	}
4290 }
4291 
4292 
4293 void
vm_page_free_prepare_object(vm_page_t mem,boolean_t remove_from_hash)4294 vm_page_free_prepare_object(
4295 	vm_page_t       mem,
4296 	boolean_t       remove_from_hash)
4297 {
4298 	assert(!mem->vmp_realtime);
4299 	if (mem->vmp_tabled) {
4300 		vm_page_remove(mem, remove_from_hash);  /* clears tabled, object, offset */
4301 	}
4302 	vm_page_wakeup(VM_OBJECT_NULL, mem);               /* clears wanted */
4303 
4304 	if (mem->vmp_private) {
4305 		mem->vmp_private = FALSE;
4306 		mem->vmp_fictitious = TRUE;
4307 		VM_PAGE_SET_PHYS_PAGE(mem, vm_page_fictitious_addr);
4308 	}
4309 	if (!mem->vmp_fictitious) {
4310 		assert(mem->vmp_pageq.next == 0);
4311 		assert(mem->vmp_pageq.prev == 0);
4312 		assert(mem->vmp_listq.next == 0);
4313 		assert(mem->vmp_listq.prev == 0);
4314 		assert(mem->vmp_specialq.next == 0);
4315 		assert(mem->vmp_specialq.prev == 0);
4316 		assert(mem->vmp_next_m == 0);
4317 
4318 		vm_page_validate_no_references(mem);
4319 
4320 		{
4321 			vm_page_init(mem, VM_PAGE_GET_PHYS_PAGE(mem), mem->vmp_lopage);
4322 		}
4323 	}
4324 }
4325 
4326 /*
4327  *	vm_page_free:
4328  *
4329  *	Returns the given page to the free list,
4330  *	disassociating it with any VM object.
4331  *
4332  *	Object and page queues must be locked prior to entry.
4333  */
4334 void
vm_page_free(vm_page_t mem)4335 vm_page_free(
4336 	vm_page_t       mem)
4337 {
4338 	vm_page_free_prepare(mem);
4339 
4340 	if (mem->vmp_fictitious) {
4341 		vm_page_release_fictitious(mem);
4342 	} else {
4343 		vm_page_release(mem, TRUE);  /* page queues are locked */
4344 	}
4345 }
4346 
4347 
4348 void
vm_page_free_unlocked(vm_page_t mem,boolean_t remove_from_hash)4349 vm_page_free_unlocked(
4350 	vm_page_t       mem,
4351 	boolean_t       remove_from_hash)
4352 {
4353 	vm_page_lockspin_queues();
4354 	vm_page_free_prepare_queues(mem);
4355 	vm_page_unlock_queues();
4356 
4357 	vm_page_free_prepare_object(mem, remove_from_hash);
4358 
4359 	if (mem->vmp_fictitious) {
4360 		vm_page_release_fictitious(mem);
4361 	} else {
4362 		vm_page_release(mem, FALSE); /* page queues are not locked */
4363 	}
4364 }
4365 
4366 
4367 /*
4368  * Free a list of pages.  The list can be up to several hundred pages,
4369  * as blocked up by vm_pageout_scan().
4370  * The big win is not having to take the free list lock once
4371  * per page.
4372  *
4373  * The VM page queues lock (vm_page_queue_lock) should NOT be held.
4374  * The VM page free queues lock (vm_page_queue_free_lock) should NOT be held.
4375  */
4376 void
vm_page_free_list(vm_page_t freeq,boolean_t prepare_object)4377 vm_page_free_list(
4378 	vm_page_t       freeq,
4379 	boolean_t       prepare_object)
4380 {
4381 	vm_page_t       mem;
4382 	vm_page_t       nxt;
4383 	vm_page_t       local_freeq;
4384 	int             pg_count;
4385 
4386 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
4387 	LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_NOTOWNED);
4388 
4389 	while (freeq) {
4390 		pg_count = 0;
4391 		local_freeq = VM_PAGE_NULL;
4392 		mem = freeq;
4393 
4394 		/*
4395 		 * break up the processing into smaller chunks so
4396 		 * that we can 'pipeline' the pages onto the
4397 		 * free list w/o introducing too much
4398 		 * contention on the global free queue lock
4399 		 */
4400 		while (mem && pg_count < 64) {
4401 			assert((mem->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
4402 			    (mem->vmp_q_state == VM_PAGE_IS_WIRED));
4403 			assert(mem->vmp_specialq.next == 0 &&
4404 			    mem->vmp_specialq.prev == 0);
4405 			/*
4406 			 * &&
4407 			 *   mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
4408 			 */
4409 			nxt = mem->vmp_snext;
4410 			mem->vmp_snext = NULL;
4411 			assert(mem->vmp_pageq.prev == 0);
4412 
4413 			if (!mem->vmp_fictitious && !mem->vmp_private) {
4414 				vm_page_validate_no_references(mem);
4415 			}
4416 
4417 			if (__improbable(mem->vmp_realtime)) {
4418 				vm_page_lock_queues();
4419 				if (mem->vmp_realtime) {
4420 					mem->vmp_realtime = false;
4421 					vm_page_realtime_count--;
4422 				}
4423 				vm_page_unlock_queues();
4424 			}
4425 
4426 			if (prepare_object == TRUE) {
4427 				vm_page_free_prepare_object(mem, TRUE);
4428 			}
4429 
4430 			if (!mem->vmp_fictitious) {
4431 				assert(mem->vmp_busy);
4432 
4433 				if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
4434 				    vm_lopage_free_count < vm_lopage_free_limit &&
4435 				    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
4436 					vm_page_release(mem, FALSE); /* page queues are not locked */
4437 #if CONFIG_SECLUDED_MEMORY
4438 				} else if (vm_page_secluded_count < vm_page_secluded_target &&
4439 				    num_tasks_can_use_secluded_mem == 0) {
4440 					vm_page_release(mem,
4441 					    FALSE);             /* page queues are not locked */
4442 #endif /* CONFIG_SECLUDED_MEMORY */
4443 				} else {
4444 					/*
4445 					 * IMPORTANT: we can't set the page "free" here
4446 					 * because that would make the page eligible for
4447 					 * a physically-contiguous allocation (see
4448 					 * vm_page_find_contiguous()) right away (we don't
4449 					 * hold the vm_page_queue_free lock).  That would
4450 					 * cause trouble because the page is not actually
4451 					 * in the free queue yet...
4452 					 */
4453 					mem->vmp_snext = local_freeq;
4454 					local_freeq = mem;
4455 					pg_count++;
4456 
4457 					pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
4458 				}
4459 			} else {
4460 				assert(VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_fictitious_addr ||
4461 				    VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_guard_addr);
4462 				vm_page_release_fictitious(mem);
4463 			}
4464 			mem = nxt;
4465 		}
4466 		freeq = mem;
4467 
4468 		if ((mem = local_freeq)) {
4469 			unsigned int    avail_free_count;
4470 			unsigned int    need_wakeup = 0;
4471 			unsigned int    need_priv_wakeup = 0;
4472 #if CONFIG_SECLUDED_MEMORY
4473 			unsigned int    need_wakeup_secluded = 0;
4474 #endif /* CONFIG_SECLUDED_MEMORY */
4475 			event_t         priv_wakeup_event, secluded_wakeup_event, normal_wakeup_event;
4476 			boolean_t       priv_wakeup_all, secluded_wakeup_all, normal_wakeup_all;
4477 
4478 			vm_free_page_lock_spin();
4479 
4480 			while (mem) {
4481 				int     color;
4482 
4483 				nxt = mem->vmp_snext;
4484 
4485 				assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
4486 				assert(mem->vmp_busy);
4487 				assert(!mem->vmp_realtime);
4488 				mem->vmp_lopage = FALSE;
4489 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
4490 
4491 				color = VM_PAGE_GET_COLOR(mem);
4492 #if defined(__x86_64__)
4493 				vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
4494 #else
4495 				vm_page_queue_enter(&vm_page_queue_free[color].qhead,
4496 				    mem, vmp_pageq);
4497 #endif
4498 				mem = nxt;
4499 			}
4500 			vm_pageout_vminfo.vm_page_pages_freed += pg_count;
4501 			vm_page_free_count += pg_count;
4502 			avail_free_count = vm_page_free_count;
4503 
4504 			VM_DEBUG_CONSTANT_EVENT(vm_page_release, DBG_VM_PAGE_RELEASE, DBG_FUNC_NONE, pg_count, 0, 0, 0);
4505 
4506 			if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) {
4507 				if (avail_free_count < vm_page_free_wanted_privileged) {
4508 					need_priv_wakeup = avail_free_count;
4509 					vm_page_free_wanted_privileged -= avail_free_count;
4510 					avail_free_count = 0;
4511 				} else {
4512 					need_priv_wakeup = vm_page_free_wanted_privileged;
4513 					avail_free_count -= vm_page_free_wanted_privileged;
4514 					vm_page_free_wanted_privileged = 0;
4515 				}
4516 			}
4517 #if CONFIG_SECLUDED_MEMORY
4518 			if (vm_page_free_wanted_secluded > 0 &&
4519 			    avail_free_count > vm_page_free_reserved) {
4520 				unsigned int available_pages;
4521 				available_pages = (avail_free_count -
4522 				    vm_page_free_reserved);
4523 				if (available_pages <
4524 				    vm_page_free_wanted_secluded) {
4525 					need_wakeup_secluded = available_pages;
4526 					vm_page_free_wanted_secluded -=
4527 					    available_pages;
4528 					avail_free_count -= available_pages;
4529 				} else {
4530 					need_wakeup_secluded =
4531 					    vm_page_free_wanted_secluded;
4532 					avail_free_count -=
4533 					    vm_page_free_wanted_secluded;
4534 					vm_page_free_wanted_secluded = 0;
4535 				}
4536 			}
4537 #endif /* CONFIG_SECLUDED_MEMORY */
4538 			if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) {
4539 				unsigned int  available_pages;
4540 
4541 				available_pages = avail_free_count - vm_page_free_reserved;
4542 
4543 				if (available_pages >= vm_page_free_wanted) {
4544 					need_wakeup = vm_page_free_wanted;
4545 					vm_page_free_wanted = 0;
4546 				} else {
4547 					need_wakeup = available_pages;
4548 					vm_page_free_wanted -= available_pages;
4549 				}
4550 			}
4551 			vm_free_page_unlock();
4552 
4553 			priv_wakeup_event = NULL;
4554 			secluded_wakeup_event = NULL;
4555 			normal_wakeup_event = NULL;
4556 
4557 			priv_wakeup_all = FALSE;
4558 			secluded_wakeup_all = FALSE;
4559 			normal_wakeup_all = FALSE;
4560 
4561 
4562 			if (need_priv_wakeup != 0) {
4563 				/*
4564 				 * There shouldn't be that many VM-privileged threads,
4565 				 * so let's wake them all up, even if we don't quite
4566 				 * have enough pages to satisfy them all.
4567 				 */
4568 				priv_wakeup_event = (event_t)&vm_page_free_wanted_privileged;
4569 				priv_wakeup_all = TRUE;
4570 			}
4571 #if CONFIG_SECLUDED_MEMORY
4572 			if (need_wakeup_secluded != 0 &&
4573 			    vm_page_free_wanted_secluded == 0) {
4574 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4575 				secluded_wakeup_all = TRUE;
4576 				need_wakeup_secluded = 0;
4577 			} else {
4578 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4579 			}
4580 #endif /* CONFIG_SECLUDED_MEMORY */
4581 			if (need_wakeup != 0 && vm_page_free_wanted == 0) {
4582 				/*
4583 				 * We don't expect to have any more waiters
4584 				 * after this, so let's wake them all up at
4585 				 * once.
4586 				 */
4587 				normal_wakeup_event = (event_t) &vm_page_free_count;
4588 				normal_wakeup_all = TRUE;
4589 				need_wakeup = 0;
4590 			} else {
4591 				normal_wakeup_event = (event_t) &vm_page_free_count;
4592 			}
4593 
4594 			if (priv_wakeup_event ||
4595 #if CONFIG_SECLUDED_MEMORY
4596 			    secluded_wakeup_event ||
4597 #endif /* CONFIG_SECLUDED_MEMORY */
4598 			    normal_wakeup_event) {
4599 				if (vps_dynamic_priority_enabled) {
4600 					if (priv_wakeup_all == TRUE) {
4601 						wakeup_all_with_inheritor(priv_wakeup_event, THREAD_AWAKENED);
4602 					}
4603 
4604 #if CONFIG_SECLUDED_MEMORY
4605 					if (secluded_wakeup_all == TRUE) {
4606 						wakeup_all_with_inheritor(secluded_wakeup_event, THREAD_AWAKENED);
4607 					}
4608 
4609 					while (need_wakeup_secluded-- != 0) {
4610 						/*
4611 						 * Wake up one waiter per page we just released.
4612 						 */
4613 						wakeup_one_with_inheritor(secluded_wakeup_event,
4614 						    THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, NULL);
4615 					}
4616 #endif /* CONFIG_SECLUDED_MEMORY */
4617 
4618 					if (normal_wakeup_all == TRUE) {
4619 						wakeup_all_with_inheritor(normal_wakeup_event, THREAD_AWAKENED);
4620 					}
4621 
4622 					while (need_wakeup-- != 0) {
4623 						/*
4624 						 * Wake up one waiter per page we just released.
4625 						 */
4626 						wakeup_one_with_inheritor(normal_wakeup_event,
4627 						    THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH,
4628 						    NULL);
4629 					}
4630 				} else {
4631 					/*
4632 					 * Non-priority-aware wakeups.
4633 					 */
4634 
4635 					if (priv_wakeup_all == TRUE) {
4636 						thread_wakeup(priv_wakeup_event);
4637 					}
4638 
4639 #if CONFIG_SECLUDED_MEMORY
4640 					if (secluded_wakeup_all == TRUE) {
4641 						thread_wakeup(secluded_wakeup_event);
4642 					}
4643 
4644 					while (need_wakeup_secluded-- != 0) {
4645 						/*
4646 						 * Wake up one waiter per page we just released.
4647 						 */
4648 						thread_wakeup_one(secluded_wakeup_event);
4649 					}
4650 
4651 #endif /* CONFIG_SECLUDED_MEMORY */
4652 					if (normal_wakeup_all == TRUE) {
4653 						thread_wakeup(normal_wakeup_event);
4654 					}
4655 
4656 					while (need_wakeup-- != 0) {
4657 						/*
4658 						 * Wake up one waiter per page we just released.
4659 						 */
4660 						thread_wakeup_one(normal_wakeup_event);
4661 					}
4662 				}
4663 			}
4664 
4665 			VM_CHECK_MEMORYSTATUS;
4666 		}
4667 	}
4668 }
4669 
4670 
4671 /*
4672  *	vm_page_wire:
4673  *
4674  *	Mark this page as wired down by yet
4675  *	another map, removing it from paging queues
4676  *	as necessary.
4677  *
4678  *	The page's object and the page queues must be locked.
4679  */
4680 
4681 
4682 void
vm_page_wire(vm_page_t mem,vm_tag_t tag,boolean_t check_memorystatus)4683 vm_page_wire(
4684 	vm_page_t mem,
4685 	vm_tag_t           tag,
4686 	boolean_t          check_memorystatus)
4687 {
4688 	vm_object_t     m_object;
4689 
4690 	m_object = VM_PAGE_OBJECT(mem);
4691 
4692 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 1);	/* (TEST/DEBUG) */
4693 
4694 	VM_PAGE_CHECK(mem);
4695 	if (m_object) {
4696 		vm_object_lock_assert_exclusive(m_object);
4697 	} else {
4698 		/*
4699 		 * In theory, the page should be in an object before it
4700 		 * gets wired, since we need to hold the object lock
4701 		 * to update some fields in the page structure.
4702 		 * However, some code (i386 pmap, for example) might want
4703 		 * to wire a page before it gets inserted into an object.
4704 		 * That's somewhat OK, as long as nobody else can get to
4705 		 * that page and update it at the same time.
4706 		 */
4707 	}
4708 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4709 	if (!VM_PAGE_WIRED(mem)) {
4710 		if (mem->vmp_laundry) {
4711 			vm_pageout_steal_laundry(mem, TRUE);
4712 		}
4713 
4714 		vm_page_queues_remove(mem, TRUE);
4715 
4716 		assert(mem->vmp_wire_count == 0);
4717 		mem->vmp_q_state = VM_PAGE_IS_WIRED;
4718 
4719 #if CONFIG_TRACK_UNMODIFIED_ANON_PAGES
4720 		if (mem->vmp_unmodified_ro == true) {
4721 			/* Object and PageQ locks are held*/
4722 			mem->vmp_unmodified_ro = false;
4723 			os_atomic_dec(&compressor_ro_uncompressed, relaxed);
4724 			vm_object_compressor_pager_state_clr(VM_PAGE_OBJECT(mem), mem->vmp_offset);
4725 		}
4726 #endif /* CONFIG_TRACK_UNMODIFIED_ANON_PAGES */
4727 
4728 		if (m_object) {
4729 			task_t          owner;
4730 			int             ledger_idx_volatile;
4731 			int             ledger_idx_nonvolatile;
4732 			int             ledger_idx_volatile_compressed;
4733 			int             ledger_idx_nonvolatile_compressed;
4734 			int             ledger_idx_composite;
4735 			int             ledger_idx_external_wired;
4736 			boolean_t       do_footprint;
4737 
4738 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4739 			VM_OBJECT_WIRED_PAGE_ADD(m_object, mem);
4740 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, tag);
4741 
4742 			assert(m_object->resident_page_count >=
4743 			    m_object->wired_page_count);
4744 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4745 				assert(vm_page_purgeable_count > 0);
4746 				OSAddAtomic(-1, &vm_page_purgeable_count);
4747 				OSAddAtomic(1, &vm_page_purgeable_wired_count);
4748 			}
4749 			if (m_object->internal &&
4750 			    m_object->vo_owner != TASK_NULL &&
4751 			    (m_object->purgable == VM_PURGABLE_VOLATILE ||
4752 			    m_object->purgable == VM_PURGABLE_EMPTY)) {
4753 				owner = VM_OBJECT_OWNER(m_object);
4754 				vm_object_ledger_tag_ledgers(
4755 					m_object,
4756 					&ledger_idx_volatile,
4757 					&ledger_idx_nonvolatile,
4758 					&ledger_idx_volatile_compressed,
4759 					&ledger_idx_nonvolatile_compressed,
4760 					&ledger_idx_composite,
4761 					&ledger_idx_external_wired,
4762 					&do_footprint);
4763 				/* less volatile bytes */
4764 				ledger_debit(owner->ledger,
4765 				    ledger_idx_volatile,
4766 				    PAGE_SIZE);
4767 				/* more not-quite-volatile bytes */
4768 				ledger_credit(owner->ledger,
4769 				    ledger_idx_nonvolatile,
4770 				    PAGE_SIZE);
4771 				if (do_footprint) {
4772 					/* more footprint */
4773 					ledger_credit(owner->ledger,
4774 					    task_ledgers.phys_footprint,
4775 					    PAGE_SIZE);
4776 				} else if (ledger_idx_composite != -1) {
4777 					ledger_credit(owner->ledger,
4778 					    ledger_idx_composite,
4779 					    PAGE_SIZE);
4780 				}
4781 			}
4782 
4783 			if (m_object->all_reusable) {
4784 				/*
4785 				 * Wired pages are not counted as "re-usable"
4786 				 * in "all_reusable" VM objects, so nothing
4787 				 * to do here.
4788 				 */
4789 			} else if (mem->vmp_reusable) {
4790 				/*
4791 				 * This page is not "re-usable" when it's
4792 				 * wired, so adjust its state and the
4793 				 * accounting.
4794 				 */
4795 				vm_page_lockconvert_queues();
4796 				vm_object_reuse_pages(m_object,
4797 				    mem->vmp_offset,
4798 				    mem->vmp_offset + PAGE_SIZE_64,
4799 				    FALSE);
4800 			}
4801 		}
4802 		assert(!mem->vmp_reusable);
4803 
4804 		if (!mem->vmp_private && !mem->vmp_fictitious && !mem->vmp_gobbled) {
4805 			vm_page_wire_count++;
4806 		}
4807 		if (mem->vmp_gobbled) {
4808 			vm_page_gobble_count--;
4809 		}
4810 		mem->vmp_gobbled = FALSE;
4811 
4812 		if (check_memorystatus == TRUE) {
4813 			VM_CHECK_MEMORYSTATUS;
4814 		}
4815 	}
4816 	assert(!mem->vmp_gobbled);
4817 	assert(mem->vmp_q_state == VM_PAGE_IS_WIRED);
4818 	mem->vmp_wire_count++;
4819 	if (__improbable(mem->vmp_wire_count == 0)) {
4820 		panic("vm_page_wire(%p): wire_count overflow", mem);
4821 	}
4822 	VM_PAGE_CHECK(mem);
4823 }
4824 
4825 /*
4826  *	vm_page_unwire:
4827  *
4828  *	Release one wiring of this page, potentially
4829  *	enabling it to be paged again.
4830  *
4831  *	The page's object and the page queues must be locked.
4832  */
4833 void
vm_page_unwire(vm_page_t mem,boolean_t queueit)4834 vm_page_unwire(
4835 	vm_page_t       mem,
4836 	boolean_t       queueit)
4837 {
4838 	vm_object_t     m_object;
4839 
4840 	m_object = VM_PAGE_OBJECT(mem);
4841 
4842 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 0);	/* (TEST/DEBUG) */
4843 
4844 	VM_PAGE_CHECK(mem);
4845 	assert(VM_PAGE_WIRED(mem));
4846 	assert(mem->vmp_wire_count > 0);
4847 	assert(!mem->vmp_gobbled);
4848 	assert(m_object != VM_OBJECT_NULL);
4849 	vm_object_lock_assert_exclusive(m_object);
4850 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4851 	if (--mem->vmp_wire_count == 0) {
4852 		task_t          owner;
4853 		int             ledger_idx_volatile;
4854 		int             ledger_idx_nonvolatile;
4855 		int             ledger_idx_volatile_compressed;
4856 		int             ledger_idx_nonvolatile_compressed;
4857 		int             ledger_idx_composite;
4858 		int             ledger_idx_external_wired;
4859 		boolean_t       do_footprint;
4860 
4861 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4862 
4863 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4864 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4865 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4866 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4867 			vm_page_wire_count--;
4868 		}
4869 
4870 		assert(m_object->resident_page_count >=
4871 		    m_object->wired_page_count);
4872 		if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4873 			OSAddAtomic(+1, &vm_page_purgeable_count);
4874 			assert(vm_page_purgeable_wired_count > 0);
4875 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4876 		}
4877 		if (m_object->internal &&
4878 		    m_object->vo_owner != TASK_NULL &&
4879 		    (m_object->purgable == VM_PURGABLE_VOLATILE ||
4880 		    m_object->purgable == VM_PURGABLE_EMPTY)) {
4881 			owner = VM_OBJECT_OWNER(m_object);
4882 			vm_object_ledger_tag_ledgers(
4883 				m_object,
4884 				&ledger_idx_volatile,
4885 				&ledger_idx_nonvolatile,
4886 				&ledger_idx_volatile_compressed,
4887 				&ledger_idx_nonvolatile_compressed,
4888 				&ledger_idx_composite,
4889 				&ledger_idx_external_wired,
4890 				&do_footprint);
4891 			/* more volatile bytes */
4892 			ledger_credit(owner->ledger,
4893 			    ledger_idx_volatile,
4894 			    PAGE_SIZE);
4895 			/* less not-quite-volatile bytes */
4896 			ledger_debit(owner->ledger,
4897 			    ledger_idx_nonvolatile,
4898 			    PAGE_SIZE);
4899 			if (do_footprint) {
4900 				/* less footprint */
4901 				ledger_debit(owner->ledger,
4902 				    task_ledgers.phys_footprint,
4903 				    PAGE_SIZE);
4904 			} else if (ledger_idx_composite != -1) {
4905 				ledger_debit(owner->ledger,
4906 				    ledger_idx_composite,
4907 				    PAGE_SIZE);
4908 			}
4909 		}
4910 		assert(!is_kernel_object(m_object));
4911 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
4912 
4913 		if (queueit == TRUE) {
4914 			if (m_object->purgable == VM_PURGABLE_EMPTY) {
4915 				vm_page_deactivate(mem);
4916 			} else {
4917 				vm_page_activate(mem);
4918 			}
4919 		}
4920 
4921 		VM_CHECK_MEMORYSTATUS;
4922 	}
4923 	VM_PAGE_CHECK(mem);
4924 }
4925 
4926 /*
4927  *	vm_page_deactivate:
4928  *
4929  *	Returns the given page to the inactive list,
4930  *	indicating that no physical maps have access
4931  *	to this page.  [Used by the physical mapping system.]
4932  *
4933  *	The page queues must be locked.
4934  */
4935 void
vm_page_deactivate(vm_page_t m)4936 vm_page_deactivate(
4937 	vm_page_t       m)
4938 {
4939 	vm_page_deactivate_internal(m, TRUE);
4940 }
4941 
4942 
4943 void
vm_page_deactivate_internal(vm_page_t m,boolean_t clear_hw_reference)4944 vm_page_deactivate_internal(
4945 	vm_page_t       m,
4946 	boolean_t       clear_hw_reference)
4947 {
4948 	vm_object_t     m_object;
4949 
4950 	m_object = VM_PAGE_OBJECT(m);
4951 
4952 	VM_PAGE_CHECK(m);
4953 	assert(!is_kernel_object(m_object));
4954 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4955 
4956 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(m), vm_page_free_count, vm_page_wire_count, 6);	/* (TEST/DEBUG) */
4957 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4958 	/*
4959 	 *	This page is no longer very interesting.  If it was
4960 	 *	interesting (active or inactive/referenced), then we
4961 	 *	clear the reference bit and (re)enter it in the
4962 	 *	inactive queue.  Note wired pages should not have
4963 	 *	their reference bit cleared.
4964 	 */
4965 	assert( !(m->vmp_absent && !m->vmp_unusual));
4966 
4967 	if (m->vmp_gobbled) {           /* can this happen? */
4968 		assert( !VM_PAGE_WIRED(m));
4969 
4970 		if (!m->vmp_private && !m->vmp_fictitious) {
4971 			vm_page_wire_count--;
4972 		}
4973 		vm_page_gobble_count--;
4974 		m->vmp_gobbled = FALSE;
4975 	}
4976 	/*
4977 	 * if this page is currently on the pageout queue, we can't do the
4978 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4979 	 * and we can't remove it manually since we would need the object lock
4980 	 * (which is not required here) to decrement the activity_in_progress
4981 	 * reference which is held on the object while the page is in the pageout queue...
4982 	 * just let the normal laundry processing proceed
4983 	 */
4984 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4985 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
4986 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
4987 	    VM_PAGE_WIRED(m)) {
4988 		return;
4989 	}
4990 	if (!m->vmp_absent && clear_hw_reference == TRUE) {
4991 		vm_page_lockconvert_queues();
4992 		pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(m));
4993 	}
4994 
4995 	m->vmp_reference = FALSE;
4996 	m->vmp_no_cache = FALSE;
4997 
4998 	if (!VM_PAGE_INACTIVE(m)) {
4999 		vm_page_queues_remove(m, FALSE);
5000 
5001 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
5002 		    m->vmp_dirty && m_object->internal &&
5003 		    (m_object->purgable == VM_PURGABLE_DENY ||
5004 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
5005 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
5006 			vm_page_check_pageable_safe(m);
5007 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
5008 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
5009 			vm_page_throttled_count++;
5010 		} else {
5011 			if (m_object->named &&
5012 			    os_ref_get_count_raw(&m_object->ref_count) == 1) {
5013 				vm_page_speculate(m, FALSE);
5014 #if DEVELOPMENT || DEBUG
5015 				vm_page_speculative_recreated++;
5016 #endif
5017 			} else {
5018 				vm_page_enqueue_inactive(m, FALSE);
5019 			}
5020 		}
5021 	}
5022 }
5023 
5024 /*
5025  * vm_page_enqueue_cleaned
5026  *
5027  * Put the page on the cleaned queue, mark it cleaned, etc.
5028  * Being on the cleaned queue (and having m->clean_queue set)
5029  * does ** NOT ** guarantee that the page is clean!
5030  *
5031  * Call with the queues lock held.
5032  */
5033 
5034 void
vm_page_enqueue_cleaned(vm_page_t m)5035 vm_page_enqueue_cleaned(vm_page_t m)
5036 {
5037 	vm_object_t     m_object;
5038 
5039 	m_object = VM_PAGE_OBJECT(m);
5040 
5041 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5042 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5043 	assert( !(m->vmp_absent && !m->vmp_unusual));
5044 
5045 	if (VM_PAGE_WIRED(m)) {
5046 		return;
5047 	}
5048 
5049 	if (m->vmp_gobbled) {
5050 		if (!m->vmp_private && !m->vmp_fictitious) {
5051 			vm_page_wire_count--;
5052 		}
5053 		vm_page_gobble_count--;
5054 		m->vmp_gobbled = FALSE;
5055 	}
5056 	/*
5057 	 * if this page is currently on the pageout queue, we can't do the
5058 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5059 	 * and we can't remove it manually since we would need the object lock
5060 	 * (which is not required here) to decrement the activity_in_progress
5061 	 * reference which is held on the object while the page is in the pageout queue...
5062 	 * just let the normal laundry processing proceed
5063 	 */
5064 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
5065 	    (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) ||
5066 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5067 		return;
5068 	}
5069 	vm_page_queues_remove(m, FALSE);
5070 
5071 	vm_page_check_pageable_safe(m);
5072 	vm_page_queue_enter(&vm_page_queue_cleaned, m, vmp_pageq);
5073 	m->vmp_q_state = VM_PAGE_ON_INACTIVE_CLEANED_Q;
5074 	vm_page_cleaned_count++;
5075 
5076 	vm_page_inactive_count++;
5077 	if (m_object->internal) {
5078 		vm_page_pageable_internal_count++;
5079 	} else {
5080 		vm_page_pageable_external_count++;
5081 	}
5082 	vm_page_add_to_specialq(m, TRUE);
5083 	VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
5084 }
5085 
5086 /*
5087  *	vm_page_activate:
5088  *
5089  *	Put the specified page on the active list (if appropriate).
5090  *
5091  *	The page queues must be locked.
5092  */
5093 
5094 void
vm_page_activate(vm_page_t m)5095 vm_page_activate(
5096 	vm_page_t       m)
5097 {
5098 	vm_object_t     m_object;
5099 
5100 	m_object = VM_PAGE_OBJECT(m);
5101 
5102 	VM_PAGE_CHECK(m);
5103 #ifdef  FIXME_4778297
5104 	assert(!is_kernel_object(m_object));
5105 #endif
5106 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5107 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5108 	assert( !(m->vmp_absent && !m->vmp_unusual));
5109 
5110 	if (m->vmp_gobbled) {
5111 		assert( !VM_PAGE_WIRED(m));
5112 		if (!m->vmp_private && !m->vmp_fictitious) {
5113 			vm_page_wire_count--;
5114 		}
5115 		vm_page_gobble_count--;
5116 		m->vmp_gobbled = FALSE;
5117 	}
5118 	/*
5119 	 * if this page is currently on the pageout queue, we can't do the
5120 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5121 	 * and we can't remove it manually since we would need the object lock
5122 	 * (which is not required here) to decrement the activity_in_progress
5123 	 * reference which is held on the object while the page is in the pageout queue...
5124 	 * just let the normal laundry processing proceed
5125 	 */
5126 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
5127 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5128 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5129 		return;
5130 	}
5131 
5132 #if DEBUG
5133 	if (m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q) {
5134 		panic("vm_page_activate: already active");
5135 	}
5136 #endif
5137 
5138 	if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) {
5139 		DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL);
5140 		DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL);
5141 	}
5142 
5143 	/*
5144 	 * A freshly activated page should be promoted in the donation queue.
5145 	 * So we remove it here while preserving its hint and we will enqueue
5146 	 * it again in vm_page_enqueue_active.
5147 	 */
5148 	vm_page_queues_remove(m, ((m->vmp_on_specialq == VM_PAGE_SPECIAL_Q_DONATE) ? TRUE : FALSE));
5149 
5150 	if (!VM_PAGE_WIRED(m)) {
5151 		vm_page_check_pageable_safe(m);
5152 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
5153 		    m->vmp_dirty && m_object->internal &&
5154 		    (m_object->purgable == VM_PURGABLE_DENY ||
5155 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
5156 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
5157 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
5158 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
5159 			vm_page_throttled_count++;
5160 		} else {
5161 #if CONFIG_SECLUDED_MEMORY
5162 			if (secluded_for_filecache &&
5163 			    vm_page_secluded_target != 0 &&
5164 			    num_tasks_can_use_secluded_mem == 0 &&
5165 			    m_object->eligible_for_secluded &&
5166 			    !m->vmp_realtime) {
5167 				vm_page_queue_enter(&vm_page_queue_secluded, m, vmp_pageq);
5168 				m->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
5169 				vm_page_secluded_count++;
5170 				VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
5171 				vm_page_secluded_count_inuse++;
5172 				assert(!m_object->internal);
5173 //				vm_page_pageable_external_count++;
5174 			} else
5175 #endif /* CONFIG_SECLUDED_MEMORY */
5176 			vm_page_enqueue_active(m, FALSE);
5177 		}
5178 		m->vmp_reference = TRUE;
5179 		m->vmp_no_cache = FALSE;
5180 	}
5181 	VM_PAGE_CHECK(m);
5182 }
5183 
5184 
5185 /*
5186  *      vm_page_speculate:
5187  *
5188  *      Put the specified page on the speculative list (if appropriate).
5189  *
5190  *      The page queues must be locked.
5191  */
5192 void
vm_page_speculate(vm_page_t m,boolean_t new)5193 vm_page_speculate(
5194 	vm_page_t       m,
5195 	boolean_t       new)
5196 {
5197 	struct vm_speculative_age_q     *aq;
5198 	vm_object_t     m_object;
5199 
5200 	m_object = VM_PAGE_OBJECT(m);
5201 
5202 	VM_PAGE_CHECK(m);
5203 	vm_page_check_pageable_safe(m);
5204 
5205 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5206 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5207 	assert( !(m->vmp_absent && !m->vmp_unusual));
5208 	assert(m_object->internal == FALSE);
5209 
5210 	/*
5211 	 * if this page is currently on the pageout queue, we can't do the
5212 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5213 	 * and we can't remove it manually since we would need the object lock
5214 	 * (which is not required here) to decrement the activity_in_progress
5215 	 * reference which is held on the object while the page is in the pageout queue...
5216 	 * just let the normal laundry processing proceed
5217 	 */
5218 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
5219 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5220 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5221 		return;
5222 	}
5223 
5224 	vm_page_queues_remove(m, FALSE);
5225 
5226 	if (!VM_PAGE_WIRED(m)) {
5227 		mach_timespec_t         ts;
5228 		clock_sec_t sec;
5229 		clock_nsec_t nsec;
5230 
5231 		clock_get_system_nanotime(&sec, &nsec);
5232 		ts.tv_sec = (unsigned int) sec;
5233 		ts.tv_nsec = nsec;
5234 
5235 		if (vm_page_speculative_count == 0) {
5236 			speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5237 			speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5238 
5239 			aq = &vm_page_queue_speculative[speculative_age_index];
5240 
5241 			/*
5242 			 * set the timer to begin a new group
5243 			 */
5244 			aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5245 			aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5246 
5247 			ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5248 		} else {
5249 			aq = &vm_page_queue_speculative[speculative_age_index];
5250 
5251 			if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) {
5252 				speculative_age_index++;
5253 
5254 				if (speculative_age_index > vm_page_max_speculative_age_q) {
5255 					speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5256 				}
5257 				if (speculative_age_index == speculative_steal_index) {
5258 					speculative_steal_index = speculative_age_index + 1;
5259 
5260 					if (speculative_steal_index > vm_page_max_speculative_age_q) {
5261 						speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5262 					}
5263 				}
5264 				aq = &vm_page_queue_speculative[speculative_age_index];
5265 
5266 				if (!vm_page_queue_empty(&aq->age_q)) {
5267 					vm_page_speculate_ageit(aq);
5268 				}
5269 
5270 				aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5271 				aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5272 
5273 				ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5274 			}
5275 		}
5276 		vm_page_enqueue_tail(&aq->age_q, &m->vmp_pageq);
5277 		m->vmp_q_state = VM_PAGE_ON_SPECULATIVE_Q;
5278 		vm_page_speculative_count++;
5279 		vm_page_pageable_external_count++;
5280 
5281 		if (new == TRUE) {
5282 			vm_object_lock_assert_exclusive(m_object);
5283 
5284 			m_object->pages_created++;
5285 #if DEVELOPMENT || DEBUG
5286 			vm_page_speculative_created++;
5287 #endif
5288 		}
5289 	}
5290 	VM_PAGE_CHECK(m);
5291 }
5292 
5293 
5294 /*
5295  * move pages from the specified aging bin to
5296  * the speculative bin that pageout_scan claims from
5297  *
5298  *      The page queues must be locked.
5299  */
5300 void
vm_page_speculate_ageit(struct vm_speculative_age_q * aq)5301 vm_page_speculate_ageit(struct vm_speculative_age_q *aq)
5302 {
5303 	struct vm_speculative_age_q     *sq;
5304 	vm_page_t       t;
5305 
5306 	sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q];
5307 
5308 	if (vm_page_queue_empty(&sq->age_q)) {
5309 		sq->age_q.next = aq->age_q.next;
5310 		sq->age_q.prev = aq->age_q.prev;
5311 
5312 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.next);
5313 		t->vmp_pageq.prev = VM_PAGE_PACK_PTR(&sq->age_q);
5314 
5315 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5316 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5317 	} else {
5318 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5319 		t->vmp_pageq.next = aq->age_q.next;
5320 
5321 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.next);
5322 		t->vmp_pageq.prev = sq->age_q.prev;
5323 
5324 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.prev);
5325 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5326 
5327 		sq->age_q.prev = aq->age_q.prev;
5328 	}
5329 	vm_page_queue_init(&aq->age_q);
5330 }
5331 
5332 
5333 void
vm_page_lru(vm_page_t m)5334 vm_page_lru(
5335 	vm_page_t       m)
5336 {
5337 	VM_PAGE_CHECK(m);
5338 	assert(!is_kernel_object(VM_PAGE_OBJECT(m)));
5339 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5340 
5341 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5342 
5343 	if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q) {
5344 		/*
5345 		 * we don't need to do all the other work that
5346 		 * vm_page_queues_remove and vm_page_enqueue_inactive
5347 		 * bring along for the ride
5348 		 */
5349 		assert(!m->vmp_laundry);
5350 		assert(!m->vmp_private);
5351 
5352 		m->vmp_no_cache = FALSE;
5353 
5354 		vm_page_queue_remove(&vm_page_queue_inactive, m, vmp_pageq);
5355 		vm_page_queue_enter(&vm_page_queue_inactive, m, vmp_pageq);
5356 
5357 		return;
5358 	}
5359 	/*
5360 	 * if this page is currently on the pageout queue, we can't do the
5361 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5362 	 * and we can't remove it manually since we would need the object lock
5363 	 * (which is not required here) to decrement the activity_in_progress
5364 	 * reference which is held on the object while the page is in the pageout queue...
5365 	 * just let the normal laundry processing proceed
5366 	 */
5367 	if (m->vmp_laundry || m->vmp_private ||
5368 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5369 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
5370 	    VM_PAGE_WIRED(m)) {
5371 		return;
5372 	}
5373 
5374 	m->vmp_no_cache = FALSE;
5375 
5376 	vm_page_queues_remove(m, FALSE);
5377 
5378 	vm_page_enqueue_inactive(m, FALSE);
5379 }
5380 
5381 
5382 void
vm_page_reactivate_all_throttled(void)5383 vm_page_reactivate_all_throttled(void)
5384 {
5385 	vm_page_t       first_throttled, last_throttled;
5386 	vm_page_t       first_active;
5387 	vm_page_t       m;
5388 	int             extra_active_count;
5389 	int             extra_internal_count, extra_external_count;
5390 	vm_object_t     m_object;
5391 
5392 	if (!VM_DYNAMIC_PAGING_ENABLED()) {
5393 		return;
5394 	}
5395 
5396 	extra_active_count = 0;
5397 	extra_internal_count = 0;
5398 	extra_external_count = 0;
5399 	vm_page_lock_queues();
5400 	if (!vm_page_queue_empty(&vm_page_queue_throttled)) {
5401 		/*
5402 		 * Switch "throttled" pages to "active".
5403 		 */
5404 		vm_page_queue_iterate(&vm_page_queue_throttled, m, vmp_pageq) {
5405 			VM_PAGE_CHECK(m);
5406 			assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
5407 
5408 			m_object = VM_PAGE_OBJECT(m);
5409 
5410 			extra_active_count++;
5411 			if (m_object->internal) {
5412 				extra_internal_count++;
5413 			} else {
5414 				extra_external_count++;
5415 			}
5416 
5417 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5418 			VM_PAGE_CHECK(m);
5419 			vm_page_add_to_specialq(m, FALSE);
5420 		}
5421 
5422 		/*
5423 		 * Transfer the entire throttled queue to a regular LRU page queues.
5424 		 * We insert it at the head of the active queue, so that these pages
5425 		 * get re-evaluated by the LRU algorithm first, since they've been
5426 		 * completely out of it until now.
5427 		 */
5428 		first_throttled = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
5429 		last_throttled = (vm_page_t) vm_page_queue_last(&vm_page_queue_throttled);
5430 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5431 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5432 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5433 		} else {
5434 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5435 		}
5436 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_throttled);
5437 		first_throttled->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5438 		last_throttled->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5439 
5440 #if DEBUG
5441 		printf("reactivated %d throttled pages\n", vm_page_throttled_count);
5442 #endif
5443 		vm_page_queue_init(&vm_page_queue_throttled);
5444 		/*
5445 		 * Adjust the global page counts.
5446 		 */
5447 		vm_page_active_count += extra_active_count;
5448 		vm_page_pageable_internal_count += extra_internal_count;
5449 		vm_page_pageable_external_count += extra_external_count;
5450 		vm_page_throttled_count = 0;
5451 	}
5452 	assert(vm_page_throttled_count == 0);
5453 	assert(vm_page_queue_empty(&vm_page_queue_throttled));
5454 	vm_page_unlock_queues();
5455 }
5456 
5457 
5458 /*
5459  * move pages from the indicated local queue to the global active queue
5460  * its ok to fail if we're below the hard limit and force == FALSE
5461  * the nolocks == TRUE case is to allow this function to be run on
5462  * the hibernate path
5463  */
5464 
5465 void
vm_page_reactivate_local(uint32_t lid,boolean_t force,boolean_t nolocks)5466 vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks)
5467 {
5468 	struct vpl      *lq;
5469 	vm_page_t       first_local, last_local;
5470 	vm_page_t       first_active;
5471 	vm_page_t       m;
5472 	uint32_t        count = 0;
5473 
5474 	if (vm_page_local_q == NULL) {
5475 		return;
5476 	}
5477 
5478 	lq = zpercpu_get_cpu(vm_page_local_q, lid);
5479 
5480 	if (nolocks == FALSE) {
5481 		if (lq->vpl_count < vm_page_local_q_hard_limit && force == FALSE) {
5482 			if (!vm_page_trylockspin_queues()) {
5483 				return;
5484 			}
5485 		} else {
5486 			vm_page_lockspin_queues();
5487 		}
5488 
5489 		VPL_LOCK(&lq->vpl_lock);
5490 	}
5491 	if (lq->vpl_count) {
5492 		/*
5493 		 * Switch "local" pages to "active".
5494 		 */
5495 		assert(!vm_page_queue_empty(&lq->vpl_queue));
5496 
5497 		vm_page_queue_iterate(&lq->vpl_queue, m, vmp_pageq) {
5498 			VM_PAGE_CHECK(m);
5499 			vm_page_check_pageable_safe(m);
5500 			assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q);
5501 			assert(!m->vmp_fictitious);
5502 
5503 			if (m->vmp_local_id != lid) {
5504 				panic("vm_page_reactivate_local: found vm_page_t(%p) with wrong cpuid", m);
5505 			}
5506 
5507 			m->vmp_local_id = 0;
5508 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5509 			VM_PAGE_CHECK(m);
5510 			vm_page_add_to_specialq(m, FALSE);
5511 			count++;
5512 		}
5513 		if (count != lq->vpl_count) {
5514 			panic("vm_page_reactivate_local: count = %d, vm_page_local_count = %d", count, lq->vpl_count);
5515 		}
5516 
5517 		/*
5518 		 * Transfer the entire local queue to a regular LRU page queues.
5519 		 */
5520 		first_local = (vm_page_t) vm_page_queue_first(&lq->vpl_queue);
5521 		last_local = (vm_page_t) vm_page_queue_last(&lq->vpl_queue);
5522 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5523 
5524 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5525 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5526 		} else {
5527 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5528 		}
5529 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_local);
5530 		first_local->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5531 		last_local->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5532 
5533 		vm_page_queue_init(&lq->vpl_queue);
5534 		/*
5535 		 * Adjust the global page counts.
5536 		 */
5537 		vm_page_active_count += lq->vpl_count;
5538 		vm_page_pageable_internal_count += lq->vpl_internal_count;
5539 		vm_page_pageable_external_count += lq->vpl_external_count;
5540 		lq->vpl_count = 0;
5541 		lq->vpl_internal_count = 0;
5542 		lq->vpl_external_count = 0;
5543 	}
5544 	assert(vm_page_queue_empty(&lq->vpl_queue));
5545 
5546 	if (nolocks == FALSE) {
5547 		VPL_UNLOCK(&lq->vpl_lock);
5548 
5549 		vm_page_balance_inactive(count / 4);
5550 		vm_page_unlock_queues();
5551 	}
5552 }
5553 
5554 /*
5555  *	vm_page_part_zero_fill:
5556  *
5557  *	Zero-fill a part of the page.
5558  */
5559 #define PMAP_ZERO_PART_PAGE_IMPLEMENTED
5560 void
vm_page_part_zero_fill(vm_page_t m,vm_offset_t m_pa,vm_size_t len)5561 vm_page_part_zero_fill(
5562 	vm_page_t       m,
5563 	vm_offset_t     m_pa,
5564 	vm_size_t       len)
5565 {
5566 #if 0
5567 	/*
5568 	 * we don't hold the page queue lock
5569 	 * so this check isn't safe to make
5570 	 */
5571 	VM_PAGE_CHECK(m);
5572 #endif
5573 
5574 #ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED
5575 	pmap_zero_part_page(VM_PAGE_GET_PHYS_PAGE(m), m_pa, len);
5576 #else
5577 	vm_page_t       tmp;
5578 	while (1) {
5579 		tmp = vm_page_grab();
5580 		if (tmp == VM_PAGE_NULL) {
5581 			vm_page_wait(THREAD_UNINT);
5582 			continue;
5583 		}
5584 		break;
5585 	}
5586 	vm_page_zero_fill(tmp);
5587 	if (m_pa != 0) {
5588 		vm_page_part_copy(m, 0, tmp, 0, m_pa);
5589 	}
5590 	if ((m_pa + len) < PAGE_SIZE) {
5591 		vm_page_part_copy(m, m_pa + len, tmp,
5592 		    m_pa + len, PAGE_SIZE - (m_pa + len));
5593 	}
5594 	vm_page_copy(tmp, m);
5595 	VM_PAGE_FREE(tmp);
5596 #endif
5597 }
5598 
5599 /*
5600  *	vm_page_zero_fill:
5601  *
5602  *	Zero-fill the specified page.
5603  */
5604 void
vm_page_zero_fill(vm_page_t m)5605 vm_page_zero_fill(
5606 	vm_page_t       m)
5607 {
5608 #if 0
5609 	/*
5610 	 * we don't hold the page queue lock
5611 	 * so this check isn't safe to make
5612 	 */
5613 	VM_PAGE_CHECK(m);
5614 #endif
5615 
5616 //	dbgTrace(0xAEAEAEAE, VM_PAGE_GET_PHYS_PAGE(m), 0);		/* (BRINGUP) */
5617 	pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m));
5618 }
5619 
5620 /*
5621  *	vm_page_part_copy:
5622  *
5623  *	copy part of one page to another
5624  */
5625 
5626 void
vm_page_part_copy(vm_page_t src_m,vm_offset_t src_pa,vm_page_t dst_m,vm_offset_t dst_pa,vm_size_t len)5627 vm_page_part_copy(
5628 	vm_page_t       src_m,
5629 	vm_offset_t     src_pa,
5630 	vm_page_t       dst_m,
5631 	vm_offset_t     dst_pa,
5632 	vm_size_t       len)
5633 {
5634 #if 0
5635 	/*
5636 	 * we don't hold the page queue lock
5637 	 * so this check isn't safe to make
5638 	 */
5639 	VM_PAGE_CHECK(src_m);
5640 	VM_PAGE_CHECK(dst_m);
5641 #endif
5642 	pmap_copy_part_page(VM_PAGE_GET_PHYS_PAGE(src_m), src_pa,
5643 	    VM_PAGE_GET_PHYS_PAGE(dst_m), dst_pa, len);
5644 }
5645 
5646 /*
5647  *	vm_page_copy:
5648  *
5649  *	Copy one page to another
5650  */
5651 
5652 int vm_page_copy_cs_validations = 0;
5653 int vm_page_copy_cs_tainted = 0;
5654 
5655 void
vm_page_copy(vm_page_t src_m,vm_page_t dest_m)5656 vm_page_copy(
5657 	vm_page_t       src_m,
5658 	vm_page_t       dest_m)
5659 {
5660 	vm_object_t     src_m_object;
5661 
5662 	src_m_object = VM_PAGE_OBJECT(src_m);
5663 
5664 #if 0
5665 	/*
5666 	 * we don't hold the page queue lock
5667 	 * so this check isn't safe to make
5668 	 */
5669 	VM_PAGE_CHECK(src_m);
5670 	VM_PAGE_CHECK(dest_m);
5671 #endif
5672 	vm_object_lock_assert_held(src_m_object);
5673 
5674 #if CONFIG_SPTM
5675 	sptm_paddr_t src_paddr = ptoa(VM_PAGE_GET_PHYS_PAGE(src_m));
5676 	sptm_frame_type_t src_frame_type = sptm_get_frame_type(src_paddr);
5677 	if (src_frame_type == XNU_KERNEL_RESTRICTED) {
5678 		panic("%s: cannot copy from a restricted page", __func__);
5679 	}
5680 #endif /* CONFIG_SPTM */
5681 
5682 	if (src_m_object != VM_OBJECT_NULL &&
5683 	    src_m_object->code_signed) {
5684 		/*
5685 		 * We're copying a page from a code-signed object.
5686 		 * Whoever ends up mapping the copy page might care about
5687 		 * the original page's integrity, so let's validate the
5688 		 * source page now.
5689 		 */
5690 		vm_page_copy_cs_validations++;
5691 		vm_page_validate_cs(src_m, PAGE_SIZE, 0);
5692 #if DEVELOPMENT || DEBUG
5693 		DTRACE_VM4(codesigned_copy,
5694 		    vm_object_t, src_m_object,
5695 		    vm_object_offset_t, src_m->vmp_offset,
5696 		    int, src_m->vmp_cs_validated,
5697 		    int, src_m->vmp_cs_tainted);
5698 #endif /* DEVELOPMENT || DEBUG */
5699 	}
5700 
5701 	/*
5702 	 * Propagate the cs_tainted bit to the copy page. Do not propagate
5703 	 * the cs_validated bit.
5704 	 */
5705 	dest_m->vmp_cs_tainted = src_m->vmp_cs_tainted;
5706 	dest_m->vmp_cs_nx = src_m->vmp_cs_nx;
5707 	if (dest_m->vmp_cs_tainted) {
5708 		vm_page_copy_cs_tainted++;
5709 	}
5710 	dest_m->vmp_error = VMP_ERROR_GET(src_m); /* sliding src_m might have failed... */
5711 	pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(src_m), VM_PAGE_GET_PHYS_PAGE(dest_m));
5712 }
5713 
5714 #if MACH_ASSERT
5715 static void
_vm_page_print(vm_page_t p)5716 _vm_page_print(
5717 	vm_page_t       p)
5718 {
5719 	printf("vm_page %p: \n", p);
5720 	printf("  pageq: next=%p prev=%p\n",
5721 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.next),
5722 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev));
5723 	printf("  listq: next=%p prev=%p\n",
5724 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.next)),
5725 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.prev)));
5726 	printf("  next=%p\n", (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m)));
5727 	printf("  object=%p offset=0x%llx\n", VM_PAGE_OBJECT(p), p->vmp_offset);
5728 	printf("  wire_count=%u\n", p->vmp_wire_count);
5729 	printf("  q_state=%u\n", p->vmp_q_state);
5730 
5731 	printf("  %slaundry, %sref, %sgobbled, %sprivate\n",
5732 	    (p->vmp_laundry ? "" : "!"),
5733 	    (p->vmp_reference ? "" : "!"),
5734 	    (p->vmp_gobbled ? "" : "!"),
5735 	    (p->vmp_private ? "" : "!"));
5736 	printf("  %sbusy, %swanted, %stabled, %sfictitious, %spmapped, %swpmapped\n",
5737 	    (p->vmp_busy ? "" : "!"),
5738 	    (p->vmp_wanted ? "" : "!"),
5739 	    (p->vmp_tabled ? "" : "!"),
5740 	    (p->vmp_fictitious ? "" : "!"),
5741 	    (p->vmp_pmapped ? "" : "!"),
5742 	    (p->vmp_wpmapped ? "" : "!"));
5743 	printf("  %sfree_when_done, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n",
5744 	    (p->vmp_free_when_done ? "" : "!"),
5745 	    (p->vmp_absent ? "" : "!"),
5746 	    (VMP_ERROR_GET(p) ? "" : "!"),
5747 	    (p->vmp_dirty ? "" : "!"),
5748 	    (p->vmp_cleaning ? "" : "!"),
5749 	    (p->vmp_precious ? "" : "!"),
5750 	    (p->vmp_clustered ? "" : "!"));
5751 	printf("  %soverwriting, %srestart, %sunusual\n",
5752 	    (p->vmp_overwriting ? "" : "!"),
5753 	    (p->vmp_restart ? "" : "!"),
5754 	    (p->vmp_unusual ? "" : "!"));
5755 	printf("  cs_validated=%d, cs_tainted=%d, cs_nx=%d, %sno_cache\n",
5756 	    p->vmp_cs_validated,
5757 	    p->vmp_cs_tainted,
5758 	    p->vmp_cs_nx,
5759 	    (p->vmp_no_cache ? "" : "!"));
5760 
5761 	printf("phys_page=0x%x\n", VM_PAGE_GET_PHYS_PAGE(p));
5762 }
5763 
5764 /*
5765  *	Check that the list of pages is ordered by
5766  *	ascending physical address and has no holes.
5767  */
5768 static int
vm_page_verify_contiguous(vm_page_t pages,unsigned int npages)5769 vm_page_verify_contiguous(
5770 	vm_page_t       pages,
5771 	unsigned int    npages)
5772 {
5773 	vm_page_t               m;
5774 	unsigned int            page_count;
5775 	vm_offset_t             prev_addr;
5776 
5777 	prev_addr = VM_PAGE_GET_PHYS_PAGE(pages);
5778 	page_count = 1;
5779 	for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
5780 		if (VM_PAGE_GET_PHYS_PAGE(m) != prev_addr + 1) {
5781 			printf("m %p prev_addr 0x%lx, current addr 0x%x\n",
5782 			    m, (long)prev_addr, VM_PAGE_GET_PHYS_PAGE(m));
5783 			printf("pages %p page_count %d npages %d\n", pages, page_count, npages);
5784 			panic("vm_page_verify_contiguous:  not contiguous!");
5785 		}
5786 		prev_addr = VM_PAGE_GET_PHYS_PAGE(m);
5787 		++page_count;
5788 	}
5789 	if (page_count != npages) {
5790 		printf("pages %p actual count 0x%x but requested 0x%x\n",
5791 		    pages, page_count, npages);
5792 		panic("vm_page_verify_contiguous:  count error");
5793 	}
5794 	return 1;
5795 }
5796 
5797 
5798 /*
5799  *	Check the free lists for proper length etc.
5800  */
5801 static boolean_t vm_page_verify_this_free_list_enabled = FALSE;
5802 static unsigned int
vm_page_verify_free_list(vm_page_queue_head_t * vm_page_queue,unsigned int color,vm_page_t look_for_page,boolean_t expect_page)5803 vm_page_verify_free_list(
5804 	vm_page_queue_head_t    *vm_page_queue,
5805 	unsigned int    color,
5806 	vm_page_t       look_for_page,
5807 	boolean_t       expect_page)
5808 {
5809 	unsigned int    npages;
5810 	vm_page_t       m;
5811 	vm_page_t       prev_m;
5812 	boolean_t       found_page;
5813 
5814 	if (!vm_page_verify_this_free_list_enabled) {
5815 		return 0;
5816 	}
5817 
5818 	found_page = FALSE;
5819 	npages = 0;
5820 	prev_m = (vm_page_t)((uintptr_t)vm_page_queue);
5821 
5822 	vm_page_queue_iterate(vm_page_queue, m, vmp_pageq) {
5823 		if (m == look_for_page) {
5824 			found_page = TRUE;
5825 		}
5826 		if ((vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev) != prev_m) {
5827 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p corrupted prev ptr %p instead of %p",
5828 			    color, npages, m, (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev), prev_m);
5829 		}
5830 		if (!m->vmp_busy) {
5831 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy",
5832 			    color, npages, m);
5833 		}
5834 		if (color != (unsigned int) -1) {
5835 			if (VM_PAGE_GET_COLOR(m) != color) {
5836 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u",
5837 				    color, npages, m, VM_PAGE_GET_COLOR(m), color);
5838 			}
5839 			if (m->vmp_q_state != VM_PAGE_ON_FREE_Q) {
5840 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p - expecting q_state == VM_PAGE_ON_FREE_Q, found %d",
5841 				    color, npages, m, m->vmp_q_state);
5842 			}
5843 		} else {
5844 			if (m->vmp_q_state != VM_PAGE_ON_FREE_LOCAL_Q) {
5845 				panic("vm_page_verify_free_list(npages=%u): local page %p - expecting q_state == VM_PAGE_ON_FREE_LOCAL_Q, found %d",
5846 				    npages, m, m->vmp_q_state);
5847 			}
5848 		}
5849 		++npages;
5850 		prev_m = m;
5851 	}
5852 	if (look_for_page != VM_PAGE_NULL) {
5853 		unsigned int other_color;
5854 
5855 		if (expect_page && !found_page) {
5856 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p not found phys=%u\n",
5857 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5858 			_vm_page_print(look_for_page);
5859 			for (other_color = 0;
5860 			    other_color < vm_colors;
5861 			    other_color++) {
5862 				if (other_color == color) {
5863 					continue;
5864 				}
5865 				vm_page_verify_free_list(&vm_page_queue_free[other_color].qhead,
5866 				    other_color, look_for_page, FALSE);
5867 			}
5868 			if (color == (unsigned int) -1) {
5869 				vm_page_verify_free_list(&vm_lopage_queue_free,
5870 				    (unsigned int) -1, look_for_page, FALSE);
5871 			}
5872 			panic("vm_page_verify_free_list(color=%u)", color);
5873 		}
5874 		if (!expect_page && found_page) {
5875 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p found phys=%u\n",
5876 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5877 		}
5878 	}
5879 	return npages;
5880 }
5881 
5882 static boolean_t vm_page_verify_all_free_lists_enabled = FALSE;
5883 static void
vm_page_verify_free_lists(void)5884 vm_page_verify_free_lists( void )
5885 {
5886 	unsigned int    color, npages, nlopages;
5887 	boolean_t       toggle = TRUE;
5888 
5889 	if (!vm_page_verify_all_free_lists_enabled) {
5890 		return;
5891 	}
5892 
5893 	npages = 0;
5894 
5895 	vm_free_page_lock();
5896 
5897 	if (vm_page_verify_this_free_list_enabled == TRUE) {
5898 		/*
5899 		 * This variable has been set globally for extra checking of
5900 		 * each free list Q. Since we didn't set it, we don't own it
5901 		 * and we shouldn't toggle it.
5902 		 */
5903 		toggle = FALSE;
5904 	}
5905 
5906 	if (toggle == TRUE) {
5907 		vm_page_verify_this_free_list_enabled = TRUE;
5908 	}
5909 
5910 	for (color = 0; color < vm_colors; color++) {
5911 		npages += vm_page_verify_free_list(&vm_page_queue_free[color].qhead,
5912 		    color, VM_PAGE_NULL, FALSE);
5913 	}
5914 	nlopages = vm_page_verify_free_list(&vm_lopage_queue_free,
5915 	    (unsigned int) -1,
5916 	    VM_PAGE_NULL, FALSE);
5917 	if (npages != vm_page_free_count || nlopages != vm_lopage_free_count) {
5918 		panic("vm_page_verify_free_lists:  "
5919 		    "npages %u free_count %d nlopages %u lo_free_count %u",
5920 		    npages, vm_page_free_count, nlopages, vm_lopage_free_count);
5921 	}
5922 
5923 	if (toggle == TRUE) {
5924 		vm_page_verify_this_free_list_enabled = FALSE;
5925 	}
5926 
5927 	vm_free_page_unlock();
5928 }
5929 
5930 #endif  /* MACH_ASSERT */
5931 
5932 /*
5933  * wrapper for pmap_enter()
5934  */
5935 kern_return_t
pmap_enter_check(pmap_t pmap,vm_map_address_t virtual_address,vm_page_t page,vm_prot_t protection,vm_prot_t fault_type,unsigned int flags,boolean_t wired)5936 pmap_enter_check(
5937 	pmap_t           pmap,
5938 	vm_map_address_t virtual_address,
5939 	vm_page_t        page,
5940 	vm_prot_t        protection,
5941 	vm_prot_t        fault_type,
5942 	unsigned int     flags,
5943 	boolean_t        wired)
5944 {
5945 	int             options = 0;
5946 	vm_object_t     obj;
5947 
5948 	if (VMP_ERROR_GET(page)) {
5949 		return KERN_MEMORY_FAILURE;
5950 	}
5951 	obj = VM_PAGE_OBJECT(page);
5952 	if (obj->internal) {
5953 		options |= PMAP_OPTIONS_INTERNAL;
5954 	}
5955 	if (page->vmp_reusable || obj->all_reusable) {
5956 		options |= PMAP_OPTIONS_REUSABLE;
5957 	}
5958 	return pmap_enter_options(pmap,
5959 	           virtual_address,
5960 	           VM_PAGE_GET_PHYS_PAGE(page),
5961 	           protection,
5962 	           fault_type,
5963 	           flags,
5964 	           wired,
5965 	           options,
5966 	           NULL,
5967 	           PMAP_MAPPING_TYPE_INFER);
5968 }
5969 
5970 
5971 extern boolean_t(*volatile consider_buffer_cache_collect)(int);
5972 
5973 /*
5974  *	CONTIGUOUS PAGE ALLOCATION AND HELPER FUNCTIONS
5975  */
5976 
5977 /*
5978  * Helper function used to determine if a page can be relocated
5979  * A page is relocatable if it is in a stable non-transient state
5980  */
5981 static inline boolean_t
vm_page_is_relocatable(vm_page_t m)5982 vm_page_is_relocatable(vm_page_t m)
5983 {
5984 
5985 	if (VM_PAGE_WIRED(m) || m->vmp_gobbled || m->vmp_laundry || m->vmp_wanted ||
5986 	    m->vmp_cleaning || m->vmp_overwriting || m->vmp_free_when_done) {
5987 		/*
5988 		 * Page is in a transient state
5989 		 * or a state we don't want to deal with.
5990 		 */
5991 		return FALSE;
5992 	} else if ((m->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
5993 	    (m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q) ||
5994 	    (m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q) ||
5995 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5996 		/*
5997 		 * Page needs to be on one of our queues (other then the pageout or special
5998 		 * free queues) or it needs to belong to the compressor pool (which is now
5999 		 * indicated by vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR and falls out from
6000 		 * the check for VM_PAGE_NOT_ON_Q) in order for it to be stable behind the
6001 		 * locks we hold at this point...
6002 		 */
6003 		return FALSE;
6004 	} else if ((m->vmp_q_state != VM_PAGE_ON_FREE_Q) &&
6005 	    (!m->vmp_tabled || m->vmp_busy)) {
6006 		/*
6007 		 * pages on the free list are always 'busy'
6008 		 * so we couldn't test for 'busy' in the check
6009 		 * for the transient states... pages that are
6010 		 * 'free' are never 'tabled', so we also couldn't
6011 		 * test for 'tabled'.  So we check here to make
6012 		 * sure that a non-free page is not busy and is
6013 		 * tabled on an object...
6014 		 */
6015 		return FALSE;
6016 	}
6017 	return TRUE;
6018 }
6019 
6020 /*
6021  * Free up the given page by possibily relocating its contents to a new page
6022  * If the page is on an object the object lock must be held.
6023  */
6024 static kern_return_t
vm_page_relocate(vm_page_t m1,int * compressed_pages)6025 vm_page_relocate(vm_page_t m1, int *compressed_pages)
6026 {
6027 	int refmod = 0;
6028 	vm_object_t object = VM_PAGE_OBJECT(m1);
6029 	kern_return_t kr;
6030 
6031 	if (object == VM_OBJECT_NULL) {
6032 		return KERN_FAILURE;
6033 	}
6034 
6035 	vm_object_lock_assert_held(object);
6036 
6037 	if (VM_PAGE_WIRED(m1) ||
6038 	    m1->vmp_gobbled ||
6039 	    m1->vmp_laundry ||
6040 	    m1->vmp_wanted ||
6041 	    m1->vmp_cleaning ||
6042 	    m1->vmp_overwriting ||
6043 	    m1->vmp_free_when_done ||
6044 	    m1->vmp_busy ||
6045 	    m1->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) {
6046 		return KERN_FAILURE;
6047 	}
6048 
6049 	boolean_t disconnected = FALSE;
6050 	boolean_t reusable = FALSE;
6051 
6052 	/*
6053 	 * Pages from reusable objects can be reclaimed directly.
6054 	 */
6055 	if ((m1->vmp_reusable || object->all_reusable) &&
6056 	    m1->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q && !m1->vmp_dirty &&
6057 	    !m1->vmp_reference) {
6058 		/*
6059 		 * reusable page...
6060 		 */
6061 
6062 		refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6063 		disconnected = TRUE;
6064 		if (refmod == 0) {
6065 			/*
6066 			 * ... not reused: can steal without relocating contents.
6067 			 */
6068 			reusable = TRUE;
6069 		}
6070 	}
6071 
6072 	if ((m1->vmp_pmapped && !reusable) || m1->vmp_dirty || m1->vmp_precious) {
6073 		vm_object_offset_t offset;
6074 
6075 		/* page is not reusable, we need to allocate a new page
6076 		 * and move its contents there.
6077 		 */
6078 		vm_page_t m2 = vm_page_grab_options(VM_PAGE_GRAB_Q_LOCK_HELD);
6079 
6080 		if (m2 == VM_PAGE_NULL) {
6081 			return KERN_RESOURCE_SHORTAGE;
6082 		}
6083 
6084 		if (!disconnected) {
6085 			if (m1->vmp_pmapped) {
6086 				refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6087 			} else {
6088 				refmod = 0;
6089 			}
6090 		}
6091 
6092 		/* copy the page's contents */
6093 		pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(m1), VM_PAGE_GET_PHYS_PAGE(m2));
6094 
6095 		/* copy the page's state */
6096 		assert(!VM_PAGE_WIRED(m1));
6097 		assert(m1->vmp_q_state != VM_PAGE_ON_FREE_Q);
6098 		assert(m1->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q);
6099 		assert(!m1->vmp_laundry);
6100 		m2->vmp_reference = m1->vmp_reference;
6101 		assert(!m1->vmp_gobbled);
6102 		assert(!m1->vmp_private);
6103 		m2->vmp_no_cache = m1->vmp_no_cache;
6104 		m2->vmp_xpmapped = 0;
6105 		assert(!m1->vmp_busy);
6106 		assert(!m1->vmp_wanted);
6107 		assert(!m1->vmp_fictitious);
6108 		m2->vmp_pmapped = m1->vmp_pmapped; /* should flush cache ? */
6109 		m2->vmp_wpmapped = m1->vmp_wpmapped;
6110 		assert(!m1->vmp_free_when_done);
6111 		m2->vmp_absent = m1->vmp_absent;
6112 		m2->vmp_error = VMP_ERROR_GET(m1);
6113 		m2->vmp_dirty = m1->vmp_dirty;
6114 		assert(!m1->vmp_cleaning);
6115 		m2->vmp_precious = m1->vmp_precious;
6116 		m2->vmp_clustered = m1->vmp_clustered;
6117 		assert(!m1->vmp_overwriting);
6118 		m2->vmp_restart = m1->vmp_restart;
6119 		m2->vmp_unusual = m1->vmp_unusual;
6120 		m2->vmp_cs_validated = m1->vmp_cs_validated;
6121 		m2->vmp_cs_tainted = m1->vmp_cs_tainted;
6122 		m2->vmp_cs_nx = m1->vmp_cs_nx;
6123 
6124 		m2->vmp_realtime = m1->vmp_realtime;
6125 		m1->vmp_realtime = false;
6126 
6127 		/*
6128 		 * If m1 had really been reusable,
6129 		 * we would have just stolen it, so
6130 		 * let's not propagate its "reusable"
6131 		 * bit and assert that m2 is not
6132 		 * marked as "reusable".
6133 		 */
6134 		// m2->vmp_reusable	= m1->vmp_reusable;
6135 		assert(!m2->vmp_reusable);
6136 
6137 		// assert(!m1->vmp_lopage);
6138 
6139 		if (m1->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6140 			m2->vmp_q_state = VM_PAGE_USED_BY_COMPRESSOR;
6141 			/*
6142 			 * We just grabbed m2 up above and so it isn't
6143 			 * going to be on any special Q as yet and so
6144 			 * we don't need to 'remove' it from the special
6145 			 * queues. Just resetting the state should be enough.
6146 			 */
6147 			m2->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
6148 		}
6149 
6150 		/*
6151 		 * page may need to be flushed if
6152 		 * it is marshalled into a UPL
6153 		 * that is going to be used by a device
6154 		 * that doesn't support coherency
6155 		 */
6156 		m2->vmp_written_by_kernel = TRUE;
6157 
6158 		/*
6159 		 * make sure we clear the ref/mod state
6160 		 * from the pmap layer... else we risk
6161 		 * inheriting state from the last time
6162 		 * this page was used...
6163 		 */
6164 		pmap_clear_refmod(VM_PAGE_GET_PHYS_PAGE(m2),
6165 		    VM_MEM_MODIFIED | VM_MEM_REFERENCED);
6166 
6167 		if (refmod & VM_MEM_REFERENCED) {
6168 			m2->vmp_reference = TRUE;
6169 		}
6170 		if (refmod & VM_MEM_MODIFIED) {
6171 			SET_PAGE_DIRTY(m2, TRUE);
6172 		}
6173 		offset = m1->vmp_offset;
6174 
6175 		/*
6176 		 * completely cleans up the state
6177 		 * of the page so that it is ready
6178 		 * to be put onto the free list, or
6179 		 * for this purpose it looks like it
6180 		 * just came off of the free list
6181 		 */
6182 		vm_page_free_prepare(m1);
6183 
6184 		/*
6185 		 * now put the substitute page on the object
6186 		 */
6187 		vm_page_insert_internal(m2, object, offset, VM_KERN_MEMORY_NONE, TRUE,
6188 		    TRUE, FALSE, FALSE, NULL);
6189 
6190 		if (m2->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6191 			m2->vmp_pmapped = TRUE;
6192 			m2->vmp_wpmapped = TRUE;
6193 
6194 			kr = pmap_enter_check(kernel_pmap, (vm_map_offset_t)m2->vmp_offset, m2,
6195 			    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE);
6196 
6197 			assert(kr == KERN_SUCCESS);
6198 
6199 			if (compressed_pages) {
6200 				++*compressed_pages;
6201 			}
6202 		} else {
6203 			/* relocated page was not used by the compressor
6204 			 * put it on either the active or inactive lists */
6205 			if (m2->vmp_reference) {
6206 				vm_page_activate(m2);
6207 			} else {
6208 				vm_page_deactivate(m2);
6209 			}
6210 		}
6211 
6212 		/* unset the busy flag (pages on the free queue are busy) and notify if wanted */
6213 		vm_page_wakeup_done(object, m2);
6214 
6215 		return KERN_SUCCESS;
6216 	} else {
6217 		assert(m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR);
6218 
6219 		/*
6220 		 * completely cleans up the state
6221 		 * of the page so that it is ready
6222 		 * to be put onto the free list, or
6223 		 * for this purpose it looks like it
6224 		 * just came off of the free list
6225 		 */
6226 		vm_page_free_prepare(m1);
6227 
6228 		/* we're done here */
6229 		return KERN_SUCCESS;
6230 	}
6231 
6232 	return KERN_FAILURE;
6233 }
6234 
6235 /*
6236  *	CONTIGUOUS PAGE ALLOCATION
6237  *
6238  *	Find a region large enough to contain at least n pages
6239  *	of contiguous physical memory.
6240  *
6241  *	This is done by traversing the vm_page_t array in a linear fashion
6242  *	we assume that the vm_page_t array has the avaiable physical pages in an
6243  *	ordered, ascending list... this is currently true of all our implementations
6244  *      and must remain so... there can be 'holes' in the array...  we also can
6245  *	no longer tolerate the vm_page_t's in the list being 'freed' and reclaimed
6246  *      which use to happen via 'vm_page_convert'... that function was no longer
6247  *      being called and was removed...
6248  *
6249  *	The basic flow consists of stabilizing some of the interesting state of
6250  *	a vm_page_t behind the vm_page_queue and vm_page_free locks... we start our
6251  *	sweep at the beginning of the array looking for pages that meet our criterea
6252  *	for a 'stealable' page... currently we are pretty conservative... if the page
6253  *	meets this criterea and is physically contiguous to the previous page in the 'run'
6254  *      we keep developing it.  If we hit a page that doesn't fit, we reset our state
6255  *	and start to develop a new run... if at this point we've already considered
6256  *      at least MAX_CONSIDERED_BEFORE_YIELD pages, we'll drop the 2 locks we hold,
6257  *	and mutex_pause (which will yield the processor), to keep the latency low w/r
6258  *	to other threads trying to acquire free pages (or move pages from q to q),
6259  *	and then continue from the spot we left off... we only make 1 pass through the
6260  *	array.  Once we have a 'run' that is long enough, we'll go into the loop which
6261  *      which steals the pages from the queues they're currently on... pages on the free
6262  *	queue can be stolen directly... pages that are on any of the other queues
6263  *	must be removed from the object they are tabled on... this requires taking the
6264  *      object lock... we do this as a 'try' to prevent deadlocks... if the 'try' fails
6265  *	or if the state of the page behind the vm_object lock is no longer viable, we'll
6266  *	dump the pages we've currently stolen back to the free list, and pick up our
6267  *	scan from the point where we aborted the 'current' run.
6268  *
6269  *
6270  *	Requirements:
6271  *		- neither vm_page_queue nor vm_free_list lock can be held on entry
6272  *
6273  *	Returns a pointer to a list of gobbled/wired pages or VM_PAGE_NULL.
6274  *
6275  * Algorithm:
6276  */
6277 
6278 #define MAX_CONSIDERED_BEFORE_YIELD     1000
6279 
6280 
6281 #define RESET_STATE_OF_RUN()    \
6282 	MACRO_BEGIN             \
6283 	prevcontaddr = -2;      \
6284 	start_pnum = -1;        \
6285 	free_considered = 0;    \
6286 	substitute_needed = 0;  \
6287 	npages = 0;             \
6288 	MACRO_END
6289 
6290 /*
6291  * Can we steal in-use (i.e. not free) pages when searching for
6292  * physically-contiguous pages ?
6293  */
6294 #define VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL 1
6295 
6296 static unsigned int vm_page_find_contiguous_last_idx = 0, vm_page_lomem_find_contiguous_last_idx = 0;
6297 #if DEBUG
6298 int vm_page_find_contig_debug = 0;
6299 #endif
6300 
6301 static vm_page_t
vm_page_find_contiguous(unsigned int contig_pages,ppnum_t max_pnum,ppnum_t pnum_mask,boolean_t wire,int flags)6302 vm_page_find_contiguous(
6303 	unsigned int    contig_pages,
6304 	ppnum_t         max_pnum,
6305 	ppnum_t         pnum_mask,
6306 	boolean_t       wire,
6307 	int             flags)
6308 {
6309 	vm_page_t       m = NULL;
6310 	ppnum_t         prevcontaddr = 0;
6311 	ppnum_t         start_pnum = 0;
6312 	unsigned int    npages = 0, considered = 0, scanned = 0;
6313 	unsigned int    page_idx = 0, start_idx = 0, last_idx = 0, orig_last_idx = 0;
6314 	unsigned int    idx_last_contig_page_found = 0;
6315 	int             free_considered = 0, free_available = 0;
6316 	int             substitute_needed = 0;
6317 	int             zone_gc_called = 0;
6318 	boolean_t       wrapped;
6319 	kern_return_t   kr;
6320 #if DEBUG
6321 	clock_sec_t     tv_start_sec = 0, tv_end_sec = 0;
6322 	clock_usec_t    tv_start_usec = 0, tv_end_usec = 0;
6323 #endif
6324 
6325 	int             yielded = 0;
6326 	int             dumped_run = 0;
6327 	int             stolen_pages = 0;
6328 	int             compressed_pages = 0;
6329 
6330 
6331 	if (contig_pages == 0) {
6332 		return VM_PAGE_NULL;
6333 	}
6334 
6335 full_scan_again:
6336 
6337 #if MACH_ASSERT
6338 	vm_page_verify_free_lists();
6339 #endif
6340 #if DEBUG
6341 	clock_get_system_microtime(&tv_start_sec, &tv_start_usec);
6342 #endif
6343 	PAGE_REPLACEMENT_ALLOWED(TRUE);
6344 
6345 	/*
6346 	 * If there are still delayed pages, try to free up some that match.
6347 	 */
6348 	if (__improbable(vm_delayed_count != 0 && contig_pages != 0)) {
6349 		vm_free_delayed_pages_contig(contig_pages, max_pnum, pnum_mask);
6350 	}
6351 
6352 	vm_page_lock_queues();
6353 	vm_free_page_lock();
6354 
6355 	RESET_STATE_OF_RUN();
6356 
6357 	scanned = 0;
6358 	considered = 0;
6359 	free_available = vm_page_free_count - vm_page_free_reserved;
6360 
6361 	wrapped = FALSE;
6362 
6363 	if (flags & KMA_LOMEM) {
6364 		idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx;
6365 	} else {
6366 		idx_last_contig_page_found =  vm_page_find_contiguous_last_idx;
6367 	}
6368 
6369 	orig_last_idx = idx_last_contig_page_found;
6370 	last_idx = orig_last_idx;
6371 
6372 	for (page_idx = last_idx, start_idx = last_idx;
6373 	    npages < contig_pages && page_idx < vm_pages_count;
6374 	    page_idx++) {
6375 retry:
6376 		if (wrapped &&
6377 		    npages == 0 &&
6378 		    page_idx >= orig_last_idx) {
6379 			/*
6380 			 * We're back where we started and we haven't
6381 			 * found any suitable contiguous range.  Let's
6382 			 * give up.
6383 			 */
6384 			break;
6385 		}
6386 		scanned++;
6387 		m = &vm_pages[page_idx];
6388 
6389 		assert(!m->vmp_fictitious);
6390 		assert(!m->vmp_private);
6391 
6392 		if (max_pnum && VM_PAGE_GET_PHYS_PAGE(m) > max_pnum) {
6393 			/* no more low pages... */
6394 			break;
6395 		}
6396 		if (!npages & ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0)) {
6397 			/*
6398 			 * not aligned
6399 			 */
6400 			RESET_STATE_OF_RUN();
6401 		} else if (!vm_page_is_relocatable(m)) {
6402 			/*
6403 			 * page is not relocatable */
6404 			RESET_STATE_OF_RUN();
6405 		} else {
6406 			if (VM_PAGE_GET_PHYS_PAGE(m) != prevcontaddr + 1) {
6407 				if ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0) {
6408 					RESET_STATE_OF_RUN();
6409 					goto did_consider;
6410 				} else {
6411 					npages = 1;
6412 					start_idx = page_idx;
6413 					start_pnum = VM_PAGE_GET_PHYS_PAGE(m);
6414 				}
6415 			} else {
6416 				npages++;
6417 			}
6418 			prevcontaddr = VM_PAGE_GET_PHYS_PAGE(m);
6419 
6420 			VM_PAGE_CHECK(m);
6421 			if (m->vmp_q_state == VM_PAGE_ON_FREE_Q) {
6422 				free_considered++;
6423 			} else {
6424 				/*
6425 				 * This page is not free.
6426 				 * If we can't steal used pages,
6427 				 * we have to give up this run
6428 				 * and keep looking.
6429 				 * Otherwise, we might need to
6430 				 * move the contents of this page
6431 				 * into a substitute page.
6432 				 */
6433 #if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
6434 				if (m->vmp_pmapped || m->vmp_dirty || m->vmp_precious) {
6435 					substitute_needed++;
6436 				}
6437 #else
6438 				RESET_STATE_OF_RUN();
6439 #endif
6440 			}
6441 
6442 			if ((free_considered + substitute_needed) > free_available) {
6443 				/*
6444 				 * if we let this run continue
6445 				 * we will end up dropping the vm_page_free_count
6446 				 * below the reserve limit... we need to abort
6447 				 * this run, but we can at least re-consider this
6448 				 * page... thus the jump back to 'retry'
6449 				 */
6450 				RESET_STATE_OF_RUN();
6451 
6452 				if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) {
6453 					considered++;
6454 					goto retry;
6455 				}
6456 				/*
6457 				 * free_available == 0
6458 				 * so can't consider any free pages... if
6459 				 * we went to retry in this case, we'd
6460 				 * get stuck looking at the same page
6461 				 * w/o making any forward progress
6462 				 * we also want to take this path if we've already
6463 				 * reached our limit that controls the lock latency
6464 				 */
6465 			}
6466 		}
6467 did_consider:
6468 		if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) {
6469 			PAGE_REPLACEMENT_ALLOWED(FALSE);
6470 
6471 			vm_free_page_unlock();
6472 			vm_page_unlock_queues();
6473 
6474 			mutex_pause(0);
6475 
6476 			PAGE_REPLACEMENT_ALLOWED(TRUE);
6477 
6478 			vm_page_lock_queues();
6479 			vm_free_page_lock();
6480 
6481 			RESET_STATE_OF_RUN();
6482 			/*
6483 			 * reset our free page limit since we
6484 			 * dropped the lock protecting the vm_page_free_queue
6485 			 */
6486 			free_available = vm_page_free_count - vm_page_free_reserved;
6487 			considered = 0;
6488 
6489 			yielded++;
6490 
6491 			goto retry;
6492 		}
6493 		considered++;
6494 	} /* main for-loop end */
6495 
6496 	m = VM_PAGE_NULL;
6497 
6498 	if (npages != contig_pages) {
6499 		if (!wrapped) {
6500 			/*
6501 			 * We didn't find a contiguous range but we didn't
6502 			 * start from the very first page.
6503 			 * Start again from the very first page.
6504 			 */
6505 			RESET_STATE_OF_RUN();
6506 			if (flags & KMA_LOMEM) {
6507 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = 0;
6508 			} else {
6509 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0;
6510 			}
6511 			last_idx = 0;
6512 			page_idx = last_idx;
6513 			wrapped = TRUE;
6514 			goto retry;
6515 		}
6516 		vm_free_page_unlock();
6517 	} else {
6518 		vm_page_t m1;
6519 		unsigned int cur_idx;
6520 		unsigned int tmp_start_idx;
6521 		vm_object_t locked_object = VM_OBJECT_NULL;
6522 		boolean_t abort_run = FALSE;
6523 
6524 		assert(page_idx - start_idx == contig_pages);
6525 
6526 		tmp_start_idx = start_idx;
6527 
6528 		/*
6529 		 * first pass through to pull the free pages
6530 		 * off of the free queue so that in case we
6531 		 * need substitute pages, we won't grab any
6532 		 * of the free pages in the run... we'll clear
6533 		 * the 'free' bit in the 2nd pass, and even in
6534 		 * an abort_run case, we'll collect all of the
6535 		 * free pages in this run and return them to the free list
6536 		 */
6537 		while (start_idx < page_idx) {
6538 			m1 = &vm_pages[start_idx++];
6539 
6540 #if !VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
6541 			assert(m1->vmp_q_state == VM_PAGE_ON_FREE_Q);
6542 #endif
6543 
6544 			if (m1->vmp_q_state == VM_PAGE_ON_FREE_Q) {
6545 				unsigned int color;
6546 
6547 				color = VM_PAGE_GET_COLOR(m1);
6548 #if MACH_ASSERT
6549 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, m1, TRUE);
6550 #endif
6551 				vm_page_queue_remove(&vm_page_queue_free[color].qhead, m1, vmp_pageq);
6552 
6553 				VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6554 #if MACH_ASSERT
6555 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, VM_PAGE_NULL, FALSE);
6556 #endif
6557 				/*
6558 				 * Clear the "free" bit so that this page
6559 				 * does not get considered for another
6560 				 * concurrent physically-contiguous allocation.
6561 				 */
6562 				m1->vmp_q_state = VM_PAGE_NOT_ON_Q;
6563 				assert(m1->vmp_busy);
6564 
6565 				vm_page_free_count--;
6566 			}
6567 		}
6568 		if (flags & KMA_LOMEM) {
6569 			vm_page_lomem_find_contiguous_last_idx = page_idx;
6570 		} else {
6571 			vm_page_find_contiguous_last_idx = page_idx;
6572 		}
6573 
6574 		/*
6575 		 * we can drop the free queue lock at this point since
6576 		 * we've pulled any 'free' candidates off of the list
6577 		 * we need it dropped so that we can do a vm_page_grab
6578 		 * when substituing for pmapped/dirty pages
6579 		 */
6580 		vm_free_page_unlock();
6581 
6582 		start_idx = tmp_start_idx;
6583 		cur_idx = page_idx - 1;
6584 
6585 		while (start_idx++ < page_idx) {
6586 			/*
6587 			 * must go through the list from back to front
6588 			 * so that the page list is created in the
6589 			 * correct order - low -> high phys addresses
6590 			 */
6591 			m1 = &vm_pages[cur_idx--];
6592 
6593 			if (m1->vmp_object == 0) {
6594 				/*
6595 				 * page has already been removed from
6596 				 * the free list in the 1st pass
6597 				 */
6598 				assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6599 				assert(m1->vmp_offset == (vm_object_offset_t) -1);
6600 				assert(m1->vmp_busy);
6601 				assert(!m1->vmp_wanted);
6602 				assert(!m1->vmp_laundry);
6603 			} else {
6604 				/*
6605 				 * try to relocate/steal the page
6606 				 */
6607 				if (abort_run == TRUE) {
6608 					continue;
6609 				}
6610 
6611 				assert(m1->vmp_q_state != VM_PAGE_NOT_ON_Q);
6612 
6613 				vm_object_t object = VM_PAGE_OBJECT(m1);
6614 
6615 				if (object != locked_object) {
6616 					if (locked_object) {
6617 						vm_object_unlock(locked_object);
6618 						locked_object = VM_OBJECT_NULL;
6619 					}
6620 					if (vm_object_lock_try(object)) {
6621 						locked_object = object;
6622 					} else {
6623 						/* object must be locked to relocate its pages */
6624 						tmp_start_idx = cur_idx;
6625 						abort_run = TRUE;
6626 						continue;
6627 					}
6628 				}
6629 
6630 				kr = vm_page_relocate(m1, &compressed_pages);
6631 				if (kr != KERN_SUCCESS) {
6632 					if (locked_object) {
6633 						vm_object_unlock(locked_object);
6634 						locked_object = VM_OBJECT_NULL;
6635 					}
6636 					tmp_start_idx = cur_idx;
6637 					abort_run = TRUE;
6638 					continue;
6639 				}
6640 
6641 				stolen_pages++;
6642 			}
6643 
6644 			/* m1 is ours at this point ... */
6645 
6646 			if (m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR) {
6647 				/*
6648 				 * The Q state is preserved on m1 because vm_page_queues_remove doesn't
6649 				 * change it for pages marked as used-by-compressor.
6650 				 */
6651 				vm_page_assign_special_state(m1, VM_PAGE_SPECIAL_Q_BG);
6652 			}
6653 			VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6654 			m1->vmp_snext = m;
6655 			m = m1;
6656 		}
6657 
6658 		if (locked_object) {
6659 			vm_object_unlock(locked_object);
6660 			locked_object = VM_OBJECT_NULL;
6661 		}
6662 
6663 		if (abort_run == TRUE) {
6664 			/*
6665 			 * want the index of the last
6666 			 * page in this run that was
6667 			 * successfully 'stolen', so back
6668 			 * it up 1 for the auto-decrement on use
6669 			 * and 1 more to bump back over this page
6670 			 */
6671 			page_idx = tmp_start_idx + 2;
6672 			if (page_idx >= vm_pages_count) {
6673 				if (wrapped) {
6674 					if (m != VM_PAGE_NULL) {
6675 						vm_page_unlock_queues();
6676 						vm_page_free_list(m, FALSE);
6677 						vm_page_lock_queues();
6678 						m = VM_PAGE_NULL;
6679 					}
6680 					dumped_run++;
6681 					goto done_scanning;
6682 				}
6683 				page_idx = last_idx = 0;
6684 				wrapped = TRUE;
6685 			}
6686 			abort_run = FALSE;
6687 
6688 			/*
6689 			 * We didn't find a contiguous range but we didn't
6690 			 * start from the very first page.
6691 			 * Start again from the very first page.
6692 			 */
6693 			RESET_STATE_OF_RUN();
6694 
6695 			if (flags & KMA_LOMEM) {
6696 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = page_idx;
6697 			} else {
6698 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx;
6699 			}
6700 
6701 			last_idx = page_idx;
6702 
6703 			if (m != VM_PAGE_NULL) {
6704 				vm_page_unlock_queues();
6705 				vm_page_free_list(m, FALSE);
6706 				vm_page_lock_queues();
6707 				m = VM_PAGE_NULL;
6708 			}
6709 			dumped_run++;
6710 
6711 			vm_free_page_lock();
6712 			/*
6713 			 * reset our free page limit since we
6714 			 * dropped the lock protecting the vm_page_free_queue
6715 			 */
6716 			free_available = vm_page_free_count - vm_page_free_reserved;
6717 			goto retry;
6718 		}
6719 
6720 		for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
6721 			assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6722 			assert(m1->vmp_wire_count == 0);
6723 
6724 			if (wire == TRUE) {
6725 				m1->vmp_wire_count++;
6726 				m1->vmp_q_state = VM_PAGE_IS_WIRED;
6727 			} else {
6728 				m1->vmp_gobbled = TRUE;
6729 			}
6730 		}
6731 		if (wire == FALSE) {
6732 			vm_page_gobble_count += npages;
6733 		}
6734 
6735 		/*
6736 		 * gobbled pages are also counted as wired pages
6737 		 */
6738 		vm_page_wire_count += npages;
6739 
6740 		assert(vm_page_verify_contiguous(m, npages));
6741 	}
6742 done_scanning:
6743 	PAGE_REPLACEMENT_ALLOWED(FALSE);
6744 
6745 	vm_page_unlock_queues();
6746 
6747 #if DEBUG
6748 	clock_get_system_microtime(&tv_end_sec, &tv_end_usec);
6749 
6750 	tv_end_sec -= tv_start_sec;
6751 	if (tv_end_usec < tv_start_usec) {
6752 		tv_end_sec--;
6753 		tv_end_usec += 1000000;
6754 	}
6755 	tv_end_usec -= tv_start_usec;
6756 	if (tv_end_usec >= 1000000) {
6757 		tv_end_sec++;
6758 		tv_end_sec -= 1000000;
6759 	}
6760 	if (vm_page_find_contig_debug) {
6761 		printf("%s(num=%d,low=%d): found %d pages at 0x%llx in %ld.%06ds...  started at %d...  scanned %d pages...  yielded %d times...  dumped run %d times... stole %d pages... stole %d compressed pages\n",
6762 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6763 		    (long)tv_end_sec, tv_end_usec, orig_last_idx,
6764 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages);
6765 	}
6766 
6767 #endif
6768 #if MACH_ASSERT
6769 	vm_page_verify_free_lists();
6770 #endif
6771 	if (m == NULL && zone_gc_called < 2) {
6772 		printf("%s(num=%d,low=%d): found %d pages at 0x%llx...scanned %d pages...  yielded %d times...  dumped run %d times... stole %d pages... stole %d compressed pages... wired count is %d\n",
6773 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6774 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages, vm_page_wire_count);
6775 
6776 		if (consider_buffer_cache_collect != NULL) {
6777 			(void)(*consider_buffer_cache_collect)(1);
6778 		}
6779 
6780 		zone_gc(zone_gc_called ? ZONE_GC_DRAIN : ZONE_GC_TRIM);
6781 
6782 		zone_gc_called++;
6783 
6784 		printf("vm_page_find_contiguous: zone_gc called... wired count is %d\n", vm_page_wire_count);
6785 		goto full_scan_again;
6786 	}
6787 
6788 	return m;
6789 }
6790 
6791 /*
6792  *	Allocate a list of contiguous, wired pages.
6793  */
6794 kern_return_t
cpm_allocate(vm_size_t size,vm_page_t * list,ppnum_t max_pnum,ppnum_t pnum_mask,boolean_t wire,int flags)6795 cpm_allocate(
6796 	vm_size_t       size,
6797 	vm_page_t       *list,
6798 	ppnum_t         max_pnum,
6799 	ppnum_t         pnum_mask,
6800 	boolean_t       wire,
6801 	int             flags)
6802 {
6803 	vm_page_t               pages;
6804 	unsigned int            npages;
6805 
6806 	if (size % PAGE_SIZE != 0) {
6807 		return KERN_INVALID_ARGUMENT;
6808 	}
6809 
6810 	npages = (unsigned int) (size / PAGE_SIZE);
6811 	if (npages != size / PAGE_SIZE) {
6812 		/* 32-bit overflow */
6813 		return KERN_INVALID_ARGUMENT;
6814 	}
6815 
6816 	/*
6817 	 *	Obtain a pointer to a subset of the free
6818 	 *	list large enough to satisfy the request;
6819 	 *	the region will be physically contiguous.
6820 	 */
6821 	pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags);
6822 
6823 	if (pages == VM_PAGE_NULL) {
6824 		return KERN_NO_SPACE;
6825 	}
6826 	/*
6827 	 * determine need for wakeups
6828 	 */
6829 	if (vm_page_free_count < vm_page_free_min) {
6830 		vm_free_page_lock();
6831 		if (vm_pageout_running == FALSE) {
6832 			vm_free_page_unlock();
6833 			thread_wakeup((event_t) &vm_page_free_wanted);
6834 		} else {
6835 			vm_free_page_unlock();
6836 		}
6837 	}
6838 
6839 	VM_CHECK_MEMORYSTATUS;
6840 
6841 	/*
6842 	 *	The CPM pages should now be available and
6843 	 *	ordered by ascending physical address.
6844 	 */
6845 	assert(vm_page_verify_contiguous(pages, npages));
6846 
6847 	if (flags & KMA_ZERO) {
6848 		for (vm_page_t m = pages; m; m = NEXT_PAGE(m)) {
6849 			vm_page_zero_fill(m);
6850 		}
6851 	}
6852 
6853 	*list = pages;
6854 	return KERN_SUCCESS;
6855 }
6856 
6857 
6858 unsigned int vm_max_delayed_work_limit = DEFAULT_DELAYED_WORK_LIMIT;
6859 
6860 /*
6861  * when working on a 'run' of pages, it is necessary to hold
6862  * the vm_page_queue_lock (a hot global lock) for certain operations
6863  * on the page... however, the majority of the work can be done
6864  * while merely holding the object lock... in fact there are certain
6865  * collections of pages that don't require any work brokered by the
6866  * vm_page_queue_lock... to mitigate the time spent behind the global
6867  * lock, go to a 2 pass algorithm... collect pages up to DELAYED_WORK_LIMIT
6868  * while doing all of the work that doesn't require the vm_page_queue_lock...
6869  * then call vm_page_do_delayed_work to acquire the vm_page_queue_lock and do the
6870  * necessary work for each page... we will grab the busy bit on the page
6871  * if it's not already held so that vm_page_do_delayed_work can drop the object lock
6872  * if it can't immediately take the vm_page_queue_lock in order to compete
6873  * for the locks in the same order that vm_pageout_scan takes them.
6874  * the operation names are modeled after the names of the routines that
6875  * need to be called in order to make the changes very obvious in the
6876  * original loop
6877  */
6878 
6879 void
vm_page_do_delayed_work(vm_object_t object,vm_tag_t tag,struct vm_page_delayed_work * dwp,int dw_count)6880 vm_page_do_delayed_work(
6881 	vm_object_t     object,
6882 	vm_tag_t        tag,
6883 	struct vm_page_delayed_work *dwp,
6884 	int             dw_count)
6885 {
6886 	int             j;
6887 	vm_page_t       m;
6888 	vm_page_t       local_free_q = VM_PAGE_NULL;
6889 
6890 	/*
6891 	 * pageout_scan takes the vm_page_lock_queues first
6892 	 * then tries for the object lock... to avoid what
6893 	 * is effectively a lock inversion, we'll go to the
6894 	 * trouble of taking them in that same order... otherwise
6895 	 * if this object contains the majority of the pages resident
6896 	 * in the UBC (or a small set of large objects actively being
6897 	 * worked on contain the majority of the pages), we could
6898 	 * cause the pageout_scan thread to 'starve' in its attempt
6899 	 * to find pages to move to the free queue, since it has to
6900 	 * successfully acquire the object lock of any candidate page
6901 	 * before it can steal/clean it.
6902 	 */
6903 	if (!vm_page_trylock_queues()) {
6904 		vm_object_unlock(object);
6905 
6906 		/*
6907 		 * "Turnstile enabled vm_pageout_scan" can be runnable
6908 		 * for a very long time without getting on a core.
6909 		 * If this is a higher priority thread it could be
6910 		 * waiting here for a very long time respecting the fact
6911 		 * that pageout_scan would like its object after VPS does
6912 		 * a mutex_pause(0).
6913 		 * So we cap the number of yields in the vm_object_lock_avoid()
6914 		 * case to a single mutex_pause(0) which will give vm_pageout_scan
6915 		 * 10us to run and grab the object if needed.
6916 		 */
6917 		vm_page_lock_queues();
6918 
6919 		for (j = 0;; j++) {
6920 			if ((!vm_object_lock_avoid(object) ||
6921 			    (vps_dynamic_priority_enabled && (j > 0))) &&
6922 			    _vm_object_lock_try(object)) {
6923 				break;
6924 			}
6925 			vm_page_unlock_queues();
6926 			mutex_pause(j);
6927 			vm_page_lock_queues();
6928 		}
6929 	}
6930 	for (j = 0; j < dw_count; j++, dwp++) {
6931 		m = dwp->dw_m;
6932 
6933 		if (dwp->dw_mask & DW_vm_pageout_throttle_up) {
6934 			vm_pageout_throttle_up(m);
6935 		}
6936 #if CONFIG_PHANTOM_CACHE
6937 		if (dwp->dw_mask & DW_vm_phantom_cache_update) {
6938 			vm_phantom_cache_update(m);
6939 		}
6940 #endif
6941 		if (dwp->dw_mask & DW_vm_page_wire) {
6942 			vm_page_wire(m, tag, FALSE);
6943 		} else if (dwp->dw_mask & DW_vm_page_unwire) {
6944 			boolean_t       queueit;
6945 
6946 			queueit = (dwp->dw_mask & (DW_vm_page_free | DW_vm_page_deactivate_internal)) ? FALSE : TRUE;
6947 
6948 			vm_page_unwire(m, queueit);
6949 		}
6950 		if (dwp->dw_mask & DW_vm_page_free) {
6951 			vm_page_free_prepare_queues(m);
6952 
6953 			assert(m->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0);
6954 			/*
6955 			 * Add this page to our list of reclaimed pages,
6956 			 * to be freed later.
6957 			 */
6958 			m->vmp_snext = local_free_q;
6959 			local_free_q = m;
6960 		} else {
6961 			if (dwp->dw_mask & DW_vm_page_deactivate_internal) {
6962 				vm_page_deactivate_internal(m, FALSE);
6963 			} else if (dwp->dw_mask & DW_vm_page_activate) {
6964 				if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6965 					vm_page_activate(m);
6966 				}
6967 			} else if (dwp->dw_mask & DW_vm_page_speculate) {
6968 				vm_page_speculate(m, TRUE);
6969 			} else if (dwp->dw_mask & DW_enqueue_cleaned) {
6970 				/*
6971 				 * if we didn't hold the object lock and did this,
6972 				 * we might disconnect the page, then someone might
6973 				 * soft fault it back in, then we would put it on the
6974 				 * cleaned queue, and so we would have a referenced (maybe even dirty)
6975 				 * page on that queue, which we don't want
6976 				 */
6977 				int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
6978 
6979 				if ((refmod_state & VM_MEM_REFERENCED)) {
6980 					/*
6981 					 * this page has been touched since it got cleaned; let's activate it
6982 					 * if it hasn't already been
6983 					 */
6984 					VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
6985 					VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1);
6986 
6987 					if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6988 						vm_page_activate(m);
6989 					}
6990 				} else {
6991 					m->vmp_reference = FALSE;
6992 					vm_page_enqueue_cleaned(m);
6993 				}
6994 			} else if (dwp->dw_mask & DW_vm_page_lru) {
6995 				vm_page_lru(m);
6996 			} else if (dwp->dw_mask & DW_VM_PAGE_QUEUES_REMOVE) {
6997 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
6998 					vm_page_queues_remove(m, TRUE);
6999 				}
7000 			}
7001 			if (dwp->dw_mask & DW_set_reference) {
7002 				m->vmp_reference = TRUE;
7003 			} else if (dwp->dw_mask & DW_clear_reference) {
7004 				m->vmp_reference = FALSE;
7005 			}
7006 
7007 			if (dwp->dw_mask & DW_move_page) {
7008 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
7009 					vm_page_queues_remove(m, FALSE);
7010 
7011 					assert(!is_kernel_object(VM_PAGE_OBJECT(m)));
7012 
7013 					vm_page_enqueue_inactive(m, FALSE);
7014 				}
7015 			}
7016 			if (dwp->dw_mask & DW_clear_busy) {
7017 				m->vmp_busy = FALSE;
7018 			}
7019 
7020 			if (dwp->dw_mask & DW_PAGE_WAKEUP) {
7021 				vm_page_wakeup(object, m);
7022 			}
7023 		}
7024 	}
7025 	vm_page_unlock_queues();
7026 
7027 	if (local_free_q) {
7028 		vm_page_free_list(local_free_q, TRUE);
7029 	}
7030 
7031 	VM_CHECK_MEMORYSTATUS;
7032 }
7033 
7034 __abortlike
7035 static void
__vm_page_alloc_list_failed_panic(vm_size_t page_count,kma_flags_t flags,kern_return_t kr)7036 __vm_page_alloc_list_failed_panic(
7037 	vm_size_t       page_count,
7038 	kma_flags_t     flags,
7039 	kern_return_t   kr)
7040 {
7041 	panic("vm_page_alloc_list(%zd, 0x%x) failed unexpectedly with %d",
7042 	    (size_t)page_count, flags, kr);
7043 }
7044 
7045 kern_return_t
vm_page_alloc_list(vm_size_t page_count,kma_flags_t flags,vm_page_t * list)7046 vm_page_alloc_list(
7047 	vm_size_t   page_count,
7048 	kma_flags_t flags,
7049 	vm_page_t  *list)
7050 {
7051 	vm_page_t       page_list = VM_PAGE_NULL;
7052 	vm_page_t       mem;
7053 	kern_return_t   kr = KERN_SUCCESS;
7054 	int             page_grab_count = 0;
7055 #if DEVELOPMENT || DEBUG
7056 	task_t          task;
7057 #endif /* DEVELOPMENT || DEBUG */
7058 
7059 	for (vm_size_t i = 0; i < page_count; i++) {
7060 		for (;;) {
7061 			if (flags & KMA_LOMEM) {
7062 				mem = vm_page_grablo();
7063 			} else {
7064 				uint_t options = VM_PAGE_GRAB_OPTIONS_NONE;
7065 				mem = vm_page_grab_options(options);
7066 			}
7067 
7068 			if (mem != VM_PAGE_NULL) {
7069 				break;
7070 			}
7071 
7072 			if (flags & KMA_NOPAGEWAIT) {
7073 				kr = KERN_RESOURCE_SHORTAGE;
7074 				goto out;
7075 			}
7076 			if ((flags & KMA_LOMEM) && (vm_lopage_needed == TRUE)) {
7077 				kr = KERN_RESOURCE_SHORTAGE;
7078 				goto out;
7079 			}
7080 
7081 			/* VM privileged threads should have waited in vm_page_grab() and not get here. */
7082 			assert(!(current_thread()->options & TH_OPT_VMPRIV));
7083 
7084 			if ((flags & KMA_NOFAIL) == 0 && ptoa_64(page_count) > max_mem / 4) {
7085 				uint64_t unavailable = ptoa_64(vm_page_wire_count + vm_page_free_target);
7086 				if (unavailable > max_mem || ptoa_64(page_count) > (max_mem - unavailable)) {
7087 					kr = KERN_RESOURCE_SHORTAGE;
7088 					goto out;
7089 				}
7090 			}
7091 			VM_PAGE_WAIT();
7092 		}
7093 
7094 		page_grab_count++;
7095 		mem->vmp_snext = page_list;
7096 		page_list = mem;
7097 	}
7098 
7099 	if ((KMA_ZERO | KMA_NOENCRYPT) & flags) {
7100 		for (mem = page_list; mem; mem = mem->vmp_snext) {
7101 			vm_page_zero_fill(mem);
7102 		}
7103 	}
7104 
7105 out:
7106 #if DEBUG || DEVELOPMENT
7107 	task = current_task_early();
7108 	if (task != NULL) {
7109 		ledger_credit(task->ledger, task_ledgers.pages_grabbed_kern, page_grab_count);
7110 	}
7111 #endif
7112 
7113 	if (kr == KERN_SUCCESS) {
7114 		*list = page_list;
7115 	} else if (flags & KMA_NOFAIL) {
7116 		__vm_page_alloc_list_failed_panic(page_count, flags, kr);
7117 	} else {
7118 		vm_page_free_list(page_list, FALSE);
7119 	}
7120 
7121 	return kr;
7122 }
7123 
7124 void
vm_page_set_offset(vm_page_t page,vm_object_offset_t offset)7125 vm_page_set_offset(vm_page_t page, vm_object_offset_t offset)
7126 {
7127 	page->vmp_offset = offset;
7128 }
7129 
7130 vm_page_t
vm_page_get_next(vm_page_t page)7131 vm_page_get_next(vm_page_t page)
7132 {
7133 	return page->vmp_snext;
7134 }
7135 
7136 vm_object_offset_t
vm_page_get_offset(vm_page_t page)7137 vm_page_get_offset(vm_page_t page)
7138 {
7139 	return page->vmp_offset;
7140 }
7141 
7142 ppnum_t
vm_page_get_phys_page(vm_page_t page)7143 vm_page_get_phys_page(vm_page_t page)
7144 {
7145 	return VM_PAGE_GET_PHYS_PAGE(page);
7146 }
7147 
7148 
7149 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
7150 
7151 #if HIBERNATION
7152 
7153 static vm_page_t hibernate_gobble_queue;
7154 
7155 static int  hibernate_drain_pageout_queue(struct vm_pageout_queue *);
7156 static int  hibernate_flush_dirty_pages(int);
7157 static int  hibernate_flush_queue(vm_page_queue_head_t *, int);
7158 
7159 void hibernate_flush_wait(void);
7160 void hibernate_mark_in_progress(void);
7161 void hibernate_clear_in_progress(void);
7162 
7163 void            hibernate_free_range(int, int);
7164 void            hibernate_hash_insert_page(vm_page_t);
7165 uint32_t        hibernate_mark_as_unneeded(addr64_t, addr64_t, hibernate_page_list_t *, hibernate_page_list_t *);
7166 uint32_t        hibernate_teardown_vm_structs(hibernate_page_list_t *, hibernate_page_list_t *);
7167 ppnum_t         hibernate_lookup_paddr(unsigned int);
7168 
7169 struct hibernate_statistics {
7170 	int hibernate_considered;
7171 	int hibernate_reentered_on_q;
7172 	int hibernate_found_dirty;
7173 	int hibernate_skipped_cleaning;
7174 	int hibernate_skipped_transient;
7175 	int hibernate_skipped_precious;
7176 	int hibernate_skipped_external;
7177 	int hibernate_queue_nolock;
7178 	int hibernate_queue_paused;
7179 	int hibernate_throttled;
7180 	int hibernate_throttle_timeout;
7181 	int hibernate_drained;
7182 	int hibernate_drain_timeout;
7183 	int cd_lock_failed;
7184 	int cd_found_precious;
7185 	int cd_found_wired;
7186 	int cd_found_busy;
7187 	int cd_found_unusual;
7188 	int cd_found_cleaning;
7189 	int cd_found_laundry;
7190 	int cd_found_dirty;
7191 	int cd_found_xpmapped;
7192 	int cd_skipped_xpmapped;
7193 	int cd_local_free;
7194 	int cd_total_free;
7195 	int cd_vm_page_wire_count;
7196 	int cd_vm_struct_pages_unneeded;
7197 	int cd_pages;
7198 	int cd_discarded;
7199 	int cd_count_wire;
7200 } hibernate_stats;
7201 
7202 #if CONFIG_SPTM
7203 /**
7204  * On SPTM-based systems don't save any executable pages into the hibernation
7205  * image. The SPTM has stronger guarantees around not allowing write access to
7206  * the executable pages than on older systems, which prevents XNU from being
7207  * able to restore any pages mapped as executable.
7208  */
7209 #define HIBERNATE_XPMAPPED_LIMIT        0ULL
7210 #else /* CONFIG_SPTM */
7211 /*
7212  * clamp the number of 'xpmapped' pages we'll sweep into the hibernation image
7213  * so that we don't overrun the estimated image size, which would
7214  * result in a hibernation failure.
7215  *
7216  * We use a size value instead of pages because we don't want to take up more space
7217  * on disk if the system has a 16K page size vs 4K. Also, we are not guaranteed
7218  * to have that additional space available.
7219  *
7220  * Since this was set at 40000 pages on X86 we are going to use 160MB as our
7221  * xpmapped size.
7222  */
7223 #define HIBERNATE_XPMAPPED_LIMIT        ((160 * 1024 * 1024ULL) / PAGE_SIZE)
7224 #endif /* CONFIG_SPTM */
7225 
7226 static int
hibernate_drain_pageout_queue(struct vm_pageout_queue * q)7227 hibernate_drain_pageout_queue(struct vm_pageout_queue *q)
7228 {
7229 	wait_result_t   wait_result;
7230 
7231 	vm_page_lock_queues();
7232 
7233 	while (!vm_page_queue_empty(&q->pgo_pending)) {
7234 		q->pgo_draining = TRUE;
7235 
7236 		assert_wait_timeout((event_t) (&q->pgo_laundry + 1), THREAD_INTERRUPTIBLE, 5000, 1000 * NSEC_PER_USEC);
7237 
7238 		vm_page_unlock_queues();
7239 
7240 		wait_result = thread_block(THREAD_CONTINUE_NULL);
7241 
7242 		if (wait_result == THREAD_TIMED_OUT && !vm_page_queue_empty(&q->pgo_pending)) {
7243 			hibernate_stats.hibernate_drain_timeout++;
7244 
7245 			if (q == &vm_pageout_queue_external) {
7246 				return 0;
7247 			}
7248 
7249 			return 1;
7250 		}
7251 		vm_page_lock_queues();
7252 
7253 		hibernate_stats.hibernate_drained++;
7254 	}
7255 	vm_page_unlock_queues();
7256 
7257 	return 0;
7258 }
7259 
7260 
7261 boolean_t hibernate_skip_external = FALSE;
7262 
7263 static int
hibernate_flush_queue(vm_page_queue_head_t * q,int qcount)7264 hibernate_flush_queue(vm_page_queue_head_t *q, int qcount)
7265 {
7266 	vm_page_t       m;
7267 	vm_object_t     l_object = NULL;
7268 	vm_object_t     m_object = NULL;
7269 	int             refmod_state = 0;
7270 	int             try_failed_count = 0;
7271 	int             retval = 0;
7272 	int             current_run = 0;
7273 	struct  vm_pageout_queue *iq;
7274 	struct  vm_pageout_queue *eq;
7275 	struct  vm_pageout_queue *tq;
7276 
7277 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START,
7278 	    VM_KERNEL_UNSLIDE_OR_PERM(q), qcount);
7279 
7280 	iq = &vm_pageout_queue_internal;
7281 	eq = &vm_pageout_queue_external;
7282 
7283 	vm_page_lock_queues();
7284 
7285 	while (qcount && !vm_page_queue_empty(q)) {
7286 		if (current_run++ == 1000) {
7287 			if (hibernate_should_abort()) {
7288 				retval = 1;
7289 				break;
7290 			}
7291 			current_run = 0;
7292 		}
7293 
7294 		m = (vm_page_t) vm_page_queue_first(q);
7295 		m_object = VM_PAGE_OBJECT(m);
7296 
7297 		/*
7298 		 * check to see if we currently are working
7299 		 * with the same object... if so, we've
7300 		 * already got the lock
7301 		 */
7302 		if (m_object != l_object) {
7303 			/*
7304 			 * the object associated with candidate page is
7305 			 * different from the one we were just working
7306 			 * with... dump the lock if we still own it
7307 			 */
7308 			if (l_object != NULL) {
7309 				vm_object_unlock(l_object);
7310 				l_object = NULL;
7311 			}
7312 			/*
7313 			 * Try to lock object; since we've alread got the
7314 			 * page queues lock, we can only 'try' for this one.
7315 			 * if the 'try' fails, we need to do a mutex_pause
7316 			 * to allow the owner of the object lock a chance to
7317 			 * run...
7318 			 */
7319 			if (!vm_object_lock_try_scan(m_object)) {
7320 				if (try_failed_count > 20) {
7321 					hibernate_stats.hibernate_queue_nolock++;
7322 
7323 					goto reenter_pg_on_q;
7324 				}
7325 
7326 				vm_page_unlock_queues();
7327 				mutex_pause(try_failed_count++);
7328 				vm_page_lock_queues();
7329 
7330 				hibernate_stats.hibernate_queue_paused++;
7331 				continue;
7332 			} else {
7333 				l_object = m_object;
7334 			}
7335 		}
7336 		if (!m_object->alive || m->vmp_cleaning || m->vmp_laundry || m->vmp_busy || m->vmp_absent || VMP_ERROR_GET(m)) {
7337 			/*
7338 			 * page is not to be cleaned
7339 			 * put it back on the head of its queue
7340 			 */
7341 			if (m->vmp_cleaning) {
7342 				hibernate_stats.hibernate_skipped_cleaning++;
7343 			} else {
7344 				hibernate_stats.hibernate_skipped_transient++;
7345 			}
7346 
7347 			goto reenter_pg_on_q;
7348 		}
7349 		if (m_object->vo_copy == VM_OBJECT_NULL) {
7350 			if (m_object->purgable == VM_PURGABLE_VOLATILE || m_object->purgable == VM_PURGABLE_EMPTY) {
7351 				/*
7352 				 * let the normal hibernate image path
7353 				 * deal with these
7354 				 */
7355 				goto reenter_pg_on_q;
7356 			}
7357 		}
7358 		if (!m->vmp_dirty && m->vmp_pmapped) {
7359 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7360 
7361 			if ((refmod_state & VM_MEM_MODIFIED)) {
7362 				SET_PAGE_DIRTY(m, FALSE);
7363 			}
7364 		} else {
7365 			refmod_state = 0;
7366 		}
7367 
7368 		if (!m->vmp_dirty) {
7369 			/*
7370 			 * page is not to be cleaned
7371 			 * put it back on the head of its queue
7372 			 */
7373 			if (m->vmp_precious) {
7374 				hibernate_stats.hibernate_skipped_precious++;
7375 			}
7376 
7377 			goto reenter_pg_on_q;
7378 		}
7379 
7380 		if (hibernate_skip_external == TRUE && !m_object->internal) {
7381 			hibernate_stats.hibernate_skipped_external++;
7382 
7383 			goto reenter_pg_on_q;
7384 		}
7385 		tq = NULL;
7386 
7387 		if (m_object->internal) {
7388 			if (VM_PAGE_Q_THROTTLED(iq)) {
7389 				tq = iq;
7390 			}
7391 		} else if (VM_PAGE_Q_THROTTLED(eq)) {
7392 			tq = eq;
7393 		}
7394 
7395 		if (tq != NULL) {
7396 			wait_result_t   wait_result;
7397 			int             wait_count = 5;
7398 
7399 			if (l_object != NULL) {
7400 				vm_object_unlock(l_object);
7401 				l_object = NULL;
7402 			}
7403 
7404 			while (retval == 0) {
7405 				tq->pgo_throttled = TRUE;
7406 
7407 				assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000 * NSEC_PER_USEC);
7408 
7409 				vm_page_unlock_queues();
7410 
7411 				wait_result = thread_block(THREAD_CONTINUE_NULL);
7412 
7413 				vm_page_lock_queues();
7414 
7415 				if (wait_result != THREAD_TIMED_OUT) {
7416 					break;
7417 				}
7418 				if (!VM_PAGE_Q_THROTTLED(tq)) {
7419 					break;
7420 				}
7421 
7422 				if (hibernate_should_abort()) {
7423 					retval = 1;
7424 				}
7425 
7426 				if (--wait_count == 0) {
7427 					hibernate_stats.hibernate_throttle_timeout++;
7428 
7429 					if (tq == eq) {
7430 						hibernate_skip_external = TRUE;
7431 						break;
7432 					}
7433 					retval = 1;
7434 				}
7435 			}
7436 			if (retval) {
7437 				break;
7438 			}
7439 
7440 			hibernate_stats.hibernate_throttled++;
7441 
7442 			continue;
7443 		}
7444 		/*
7445 		 * we've already factored out pages in the laundry which
7446 		 * means this page can't be on the pageout queue so it's
7447 		 * safe to do the vm_page_queues_remove
7448 		 */
7449 		vm_page_queues_remove(m, TRUE);
7450 
7451 		if (m_object->internal == TRUE) {
7452 			pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(m), PMAP_OPTIONS_COMPRESSOR, NULL);
7453 		}
7454 
7455 		vm_pageout_cluster(m);
7456 
7457 		hibernate_stats.hibernate_found_dirty++;
7458 
7459 		goto next_pg;
7460 
7461 reenter_pg_on_q:
7462 		vm_page_queue_remove(q, m, vmp_pageq);
7463 		vm_page_queue_enter(q, m, vmp_pageq);
7464 
7465 		hibernate_stats.hibernate_reentered_on_q++;
7466 next_pg:
7467 		hibernate_stats.hibernate_considered++;
7468 
7469 		qcount--;
7470 		try_failed_count = 0;
7471 	}
7472 	if (l_object != NULL) {
7473 		vm_object_unlock(l_object);
7474 		l_object = NULL;
7475 	}
7476 
7477 	vm_page_unlock_queues();
7478 
7479 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_END, hibernate_stats.hibernate_found_dirty, retval, 0, 0, 0);
7480 
7481 	return retval;
7482 }
7483 
7484 
7485 static int
hibernate_flush_dirty_pages(int pass)7486 hibernate_flush_dirty_pages(int pass)
7487 {
7488 	struct vm_speculative_age_q     *aq;
7489 	uint32_t        i;
7490 
7491 	if (vm_page_local_q) {
7492 		zpercpu_foreach_cpu(lid) {
7493 			vm_page_reactivate_local(lid, TRUE, FALSE);
7494 		}
7495 	}
7496 
7497 	for (i = 0; i <= vm_page_max_speculative_age_q; i++) {
7498 		int             qcount;
7499 		vm_page_t       m;
7500 
7501 		aq = &vm_page_queue_speculative[i];
7502 
7503 		if (vm_page_queue_empty(&aq->age_q)) {
7504 			continue;
7505 		}
7506 		qcount = 0;
7507 
7508 		vm_page_lockspin_queues();
7509 
7510 		vm_page_queue_iterate(&aq->age_q, m, vmp_pageq) {
7511 			qcount++;
7512 		}
7513 		vm_page_unlock_queues();
7514 
7515 		if (qcount) {
7516 			if (hibernate_flush_queue(&aq->age_q, qcount)) {
7517 				return 1;
7518 			}
7519 		}
7520 	}
7521 	if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_page_anonymous_count - vm_page_cleaned_count)) {
7522 		return 1;
7523 	}
7524 	/* XXX FBDP TODO: flush secluded queue */
7525 	if (hibernate_flush_queue(&vm_page_queue_anonymous, vm_page_anonymous_count)) {
7526 		return 1;
7527 	}
7528 	if (hibernate_flush_queue(&vm_page_queue_cleaned, vm_page_cleaned_count)) {
7529 		return 1;
7530 	}
7531 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7532 		return 1;
7533 	}
7534 
7535 	if (pass == 1) {
7536 		vm_compressor_record_warmup_start();
7537 	}
7538 
7539 	if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) {
7540 		if (pass == 1) {
7541 			vm_compressor_record_warmup_end();
7542 		}
7543 		return 1;
7544 	}
7545 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7546 		if (pass == 1) {
7547 			vm_compressor_record_warmup_end();
7548 		}
7549 		return 1;
7550 	}
7551 	if (pass == 1) {
7552 		vm_compressor_record_warmup_end();
7553 	}
7554 
7555 	if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external)) {
7556 		return 1;
7557 	}
7558 
7559 	return 0;
7560 }
7561 
7562 
7563 void
hibernate_reset_stats()7564 hibernate_reset_stats()
7565 {
7566 	bzero(&hibernate_stats, sizeof(struct hibernate_statistics));
7567 }
7568 
7569 
7570 int
hibernate_flush_memory()7571 hibernate_flush_memory()
7572 {
7573 	int     retval;
7574 
7575 	assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
7576 
7577 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0);
7578 
7579 	hibernate_cleaning_in_progress = TRUE;
7580 	hibernate_skip_external = FALSE;
7581 
7582 	if ((retval = hibernate_flush_dirty_pages(1)) == 0) {
7583 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_START, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7584 
7585 		vm_compressor_flush();
7586 
7587 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7588 
7589 		if (consider_buffer_cache_collect != NULL) {
7590 			unsigned int orig_wire_count;
7591 
7592 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, 0, 0, 0, 0, 0);
7593 			orig_wire_count = vm_page_wire_count;
7594 
7595 			(void)(*consider_buffer_cache_collect)(1);
7596 			zone_gc(ZONE_GC_DRAIN);
7597 
7598 			HIBLOG("hibernate_flush_memory: buffer_cache_gc freed up %d wired pages\n", orig_wire_count - vm_page_wire_count);
7599 
7600 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0);
7601 		}
7602 	}
7603 	hibernate_cleaning_in_progress = FALSE;
7604 
7605 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0);
7606 
7607 	if (retval) {
7608 		HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n", VM_PAGE_COMPRESSOR_COUNT);
7609 	}
7610 
7611 
7612 	HIBPRINT("hibernate_flush_memory() considered(%d) reentered_on_q(%d) found_dirty(%d)\n",
7613 	    hibernate_stats.hibernate_considered,
7614 	    hibernate_stats.hibernate_reentered_on_q,
7615 	    hibernate_stats.hibernate_found_dirty);
7616 	HIBPRINT("   skipped_cleaning(%d) skipped_transient(%d) skipped_precious(%d) skipped_external(%d) queue_nolock(%d)\n",
7617 	    hibernate_stats.hibernate_skipped_cleaning,
7618 	    hibernate_stats.hibernate_skipped_transient,
7619 	    hibernate_stats.hibernate_skipped_precious,
7620 	    hibernate_stats.hibernate_skipped_external,
7621 	    hibernate_stats.hibernate_queue_nolock);
7622 	HIBPRINT("   queue_paused(%d) throttled(%d) throttle_timeout(%d) drained(%d) drain_timeout(%d)\n",
7623 	    hibernate_stats.hibernate_queue_paused,
7624 	    hibernate_stats.hibernate_throttled,
7625 	    hibernate_stats.hibernate_throttle_timeout,
7626 	    hibernate_stats.hibernate_drained,
7627 	    hibernate_stats.hibernate_drain_timeout);
7628 
7629 	return retval;
7630 }
7631 
7632 
7633 static void
hibernate_page_list_zero(hibernate_page_list_t * list)7634 hibernate_page_list_zero(hibernate_page_list_t *list)
7635 {
7636 	uint32_t             bank;
7637 	hibernate_bitmap_t * bitmap;
7638 
7639 	bitmap = &list->bank_bitmap[0];
7640 	for (bank = 0; bank < list->bank_count; bank++) {
7641 		uint32_t last_bit;
7642 
7643 		bzero((void *) &bitmap->bitmap[0], bitmap->bitmapwords << 2);
7644 		// set out-of-bound bits at end of bitmap.
7645 		last_bit = ((bitmap->last_page - bitmap->first_page + 1) & 31);
7646 		if (last_bit) {
7647 			bitmap->bitmap[bitmap->bitmapwords - 1] = (0xFFFFFFFF >> last_bit);
7648 		}
7649 
7650 		bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords];
7651 	}
7652 }
7653 
7654 void
hibernate_free_gobble_pages(void)7655 hibernate_free_gobble_pages(void)
7656 {
7657 	vm_page_t m, next;
7658 	uint32_t  count = 0;
7659 
7660 	m = (vm_page_t) hibernate_gobble_queue;
7661 	while (m) {
7662 		next = m->vmp_snext;
7663 		vm_page_free(m);
7664 		count++;
7665 		m = next;
7666 	}
7667 	hibernate_gobble_queue = VM_PAGE_NULL;
7668 
7669 	if (count) {
7670 		HIBLOG("Freed %d pages\n", count);
7671 	}
7672 }
7673 
7674 static boolean_t
hibernate_consider_discard(vm_page_t m,boolean_t preflight)7675 hibernate_consider_discard(vm_page_t m, boolean_t preflight)
7676 {
7677 	vm_object_t object = NULL;
7678 	int                  refmod_state;
7679 	boolean_t            discard = FALSE;
7680 
7681 	do{
7682 		if (m->vmp_private) {
7683 			panic("hibernate_consider_discard: private");
7684 		}
7685 
7686 		object = VM_PAGE_OBJECT(m);
7687 
7688 		if (!vm_object_lock_try(object)) {
7689 			object = NULL;
7690 			if (!preflight) {
7691 				hibernate_stats.cd_lock_failed++;
7692 			}
7693 			break;
7694 		}
7695 		if (VM_PAGE_WIRED(m)) {
7696 			if (!preflight) {
7697 				hibernate_stats.cd_found_wired++;
7698 			}
7699 			break;
7700 		}
7701 		if (m->vmp_precious) {
7702 			if (!preflight) {
7703 				hibernate_stats.cd_found_precious++;
7704 			}
7705 			break;
7706 		}
7707 		if (m->vmp_busy || !object->alive) {
7708 			/*
7709 			 *	Somebody is playing with this page.
7710 			 */
7711 			if (!preflight) {
7712 				hibernate_stats.cd_found_busy++;
7713 			}
7714 			break;
7715 		}
7716 		if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7717 			/*
7718 			 * If it's unusual in anyway, ignore it
7719 			 */
7720 			if (!preflight) {
7721 				hibernate_stats.cd_found_unusual++;
7722 			}
7723 			break;
7724 		}
7725 		if (m->vmp_cleaning) {
7726 			if (!preflight) {
7727 				hibernate_stats.cd_found_cleaning++;
7728 			}
7729 			break;
7730 		}
7731 		if (m->vmp_laundry) {
7732 			if (!preflight) {
7733 				hibernate_stats.cd_found_laundry++;
7734 			}
7735 			break;
7736 		}
7737 		if (!m->vmp_dirty) {
7738 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7739 
7740 			if (refmod_state & VM_MEM_REFERENCED) {
7741 				m->vmp_reference = TRUE;
7742 			}
7743 			if (refmod_state & VM_MEM_MODIFIED) {
7744 				SET_PAGE_DIRTY(m, FALSE);
7745 			}
7746 		}
7747 
7748 		/*
7749 		 * If it's clean or purgeable we can discard the page on wakeup.
7750 		 */
7751 		discard = (!m->vmp_dirty)
7752 		    || (VM_PURGABLE_VOLATILE == object->purgable)
7753 		    || (VM_PURGABLE_EMPTY == object->purgable);
7754 
7755 
7756 		if (discard == FALSE) {
7757 			if (!preflight) {
7758 				hibernate_stats.cd_found_dirty++;
7759 			}
7760 		} else if (m->vmp_xpmapped && m->vmp_reference && !object->internal) {
7761 			if (hibernate_stats.cd_found_xpmapped < HIBERNATE_XPMAPPED_LIMIT) {
7762 				if (!preflight) {
7763 					hibernate_stats.cd_found_xpmapped++;
7764 				}
7765 				discard = FALSE;
7766 			} else {
7767 				if (!preflight) {
7768 					hibernate_stats.cd_skipped_xpmapped++;
7769 				}
7770 			}
7771 		}
7772 	}while (FALSE);
7773 
7774 	if (object) {
7775 		vm_object_unlock(object);
7776 	}
7777 
7778 	return discard;
7779 }
7780 
7781 
7782 static void
hibernate_discard_page(vm_page_t m)7783 hibernate_discard_page(vm_page_t m)
7784 {
7785 	vm_object_t m_object;
7786 
7787 	if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7788 		/*
7789 		 * If it's unusual in anyway, ignore
7790 		 */
7791 		return;
7792 	}
7793 
7794 	m_object = VM_PAGE_OBJECT(m);
7795 
7796 #if MACH_ASSERT || DEBUG
7797 	if (!vm_object_lock_try(m_object)) {
7798 		panic("hibernate_discard_page(%p) !vm_object_lock_try", m);
7799 	}
7800 #else
7801 	/* No need to lock page queue for token delete, hibernate_vm_unlock()
7802 	 *  makes sure these locks are uncontended before sleep */
7803 #endif /* MACH_ASSERT || DEBUG */
7804 
7805 	if (m->vmp_pmapped == TRUE) {
7806 		__unused int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
7807 	}
7808 
7809 	if (m->vmp_laundry) {
7810 		panic("hibernate_discard_page(%p) laundry", m);
7811 	}
7812 	if (m->vmp_private) {
7813 		panic("hibernate_discard_page(%p) private", m);
7814 	}
7815 	if (m->vmp_fictitious) {
7816 		panic("hibernate_discard_page(%p) fictitious", m);
7817 	}
7818 
7819 	if (VM_PURGABLE_VOLATILE == m_object->purgable) {
7820 		/* object should be on a queue */
7821 		assert((m_object->objq.next != NULL) && (m_object->objq.prev != NULL));
7822 		purgeable_q_t old_queue = vm_purgeable_object_remove(m_object);
7823 		assert(old_queue);
7824 		if (m_object->purgeable_when_ripe) {
7825 			vm_purgeable_token_delete_first(old_queue);
7826 		}
7827 		vm_object_lock_assert_exclusive(m_object);
7828 		VM_OBJECT_SET_PURGABLE(m_object, VM_PURGABLE_EMPTY);
7829 
7830 		/*
7831 		 * Purgeable ledgers:  pages of VOLATILE and EMPTY objects are
7832 		 * accounted in the "volatile" ledger, so no change here.
7833 		 * We have to update vm_page_purgeable_count, though, since we're
7834 		 * effectively purging this object.
7835 		 */
7836 		unsigned int delta;
7837 		assert(m_object->resident_page_count >= m_object->wired_page_count);
7838 		delta = (m_object->resident_page_count - m_object->wired_page_count);
7839 		assert(vm_page_purgeable_count >= delta);
7840 		assert(delta > 0);
7841 		OSAddAtomic(-delta, (SInt32 *)&vm_page_purgeable_count);
7842 	}
7843 
7844 	vm_page_free(m);
7845 
7846 #if MACH_ASSERT || DEBUG
7847 	vm_object_unlock(m_object);
7848 #endif  /* MACH_ASSERT || DEBUG */
7849 }
7850 
7851 /*
7852  *  Grab locks for hibernate_page_list_setall()
7853  */
7854 void
hibernate_vm_lock_queues(void)7855 hibernate_vm_lock_queues(void)
7856 {
7857 	vm_object_lock(compressor_object);
7858 	vm_page_lock_queues();
7859 	vm_free_page_lock();
7860 	lck_mtx_lock(&vm_purgeable_queue_lock);
7861 
7862 	if (vm_page_local_q) {
7863 		zpercpu_foreach(lq, vm_page_local_q) {
7864 			VPL_LOCK(&lq->vpl_lock);
7865 		}
7866 	}
7867 }
7868 
7869 void
hibernate_vm_unlock_queues(void)7870 hibernate_vm_unlock_queues(void)
7871 {
7872 	if (vm_page_local_q) {
7873 		zpercpu_foreach(lq, vm_page_local_q) {
7874 			VPL_UNLOCK(&lq->vpl_lock);
7875 		}
7876 	}
7877 	lck_mtx_unlock(&vm_purgeable_queue_lock);
7878 	vm_free_page_unlock();
7879 	vm_page_unlock_queues();
7880 	vm_object_unlock(compressor_object);
7881 }
7882 
7883 #if CONFIG_SPTM
7884 static bool
hibernate_sptm_should_force_page_to_wired_pagelist(vm_page_t vmp)7885 hibernate_sptm_should_force_page_to_wired_pagelist(vm_page_t vmp)
7886 {
7887 	const sptm_paddr_t paddr = ptoa_64(VM_PAGE_GET_PHYS_PAGE(vmp));
7888 	const sptm_frame_type_t frame_type = sptm_get_frame_type(paddr);
7889 	const vm_object_t vmp_objp = VM_PAGE_OBJECT(vmp);
7890 
7891 	return frame_type == XNU_USER_JIT || frame_type == XNU_USER_DEBUG ||
7892 	       (frame_type == XNU_USER_EXEC && vmp_objp->internal == TRUE);
7893 }
7894 #endif
7895 
7896 /*
7897  *  Bits zero in the bitmaps => page needs to be saved. All pages default to be saved,
7898  *  pages known to VM to not need saving are subtracted.
7899  *  Wired pages to be saved are present in page_list_wired, pageable in page_list.
7900  */
7901 
7902 void
hibernate_page_list_setall(hibernate_page_list_t * page_list,hibernate_page_list_t * page_list_wired,hibernate_page_list_t * page_list_pal,boolean_t preflight,boolean_t will_discard,uint32_t * pagesOut)7903 hibernate_page_list_setall(hibernate_page_list_t * page_list,
7904     hibernate_page_list_t * page_list_wired,
7905     hibernate_page_list_t * page_list_pal,
7906     boolean_t preflight,
7907     boolean_t will_discard,
7908     uint32_t * pagesOut)
7909 {
7910 	uint64_t start, end, nsec;
7911 	vm_page_t m;
7912 	vm_page_t next;
7913 	uint32_t pages = page_list->page_count;
7914 	uint32_t count_anonymous = 0, count_throttled = 0, count_compressor = 0;
7915 	uint32_t count_inactive = 0, count_active = 0, count_speculative = 0, count_cleaned = 0;
7916 	uint32_t count_wire = pages;
7917 	uint32_t count_discard_active    = 0;
7918 	uint32_t count_discard_inactive  = 0;
7919 	uint32_t count_retired = 0;
7920 	uint32_t count_discard_cleaned   = 0;
7921 	uint32_t count_discard_purgeable = 0;
7922 	uint32_t count_discard_speculative = 0;
7923 	uint32_t count_discard_vm_struct_pages = 0;
7924 	uint32_t i;
7925 	uint32_t             bank;
7926 	hibernate_bitmap_t * bitmap;
7927 	hibernate_bitmap_t * bitmap_wired;
7928 	boolean_t                    discard_all;
7929 	boolean_t            discard = FALSE;
7930 
7931 	HIBLOG("hibernate_page_list_setall(preflight %d) start\n", preflight);
7932 
7933 	if (preflight) {
7934 		page_list       = NULL;
7935 		page_list_wired = NULL;
7936 		page_list_pal   = NULL;
7937 		discard_all     = FALSE;
7938 	} else {
7939 		discard_all     = will_discard;
7940 	}
7941 
7942 #if MACH_ASSERT || DEBUG
7943 	if (!preflight) {
7944 		assert(hibernate_vm_locks_are_safe());
7945 		vm_page_lock_queues();
7946 		if (vm_page_local_q) {
7947 			zpercpu_foreach(lq, vm_page_local_q) {
7948 				VPL_LOCK(&lq->vpl_lock);
7949 			}
7950 		}
7951 	}
7952 #endif  /* MACH_ASSERT || DEBUG */
7953 
7954 
7955 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0);
7956 
7957 	clock_get_uptime(&start);
7958 
7959 	if (!preflight) {
7960 		hibernate_page_list_zero(page_list);
7961 		hibernate_page_list_zero(page_list_wired);
7962 		hibernate_page_list_zero(page_list_pal);
7963 
7964 		hibernate_stats.cd_vm_page_wire_count = vm_page_wire_count;
7965 		hibernate_stats.cd_pages = pages;
7966 	}
7967 
7968 	if (vm_page_local_q) {
7969 		zpercpu_foreach_cpu(lid) {
7970 			vm_page_reactivate_local(lid, TRUE, !preflight);
7971 		}
7972 	}
7973 
7974 	if (preflight) {
7975 		vm_object_lock(compressor_object);
7976 		vm_page_lock_queues();
7977 		vm_free_page_lock();
7978 	}
7979 
7980 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
7981 
7982 	hibernation_vmqueues_inspection = TRUE;
7983 
7984 	m = (vm_page_t) hibernate_gobble_queue;
7985 	while (m) {
7986 		pages--;
7987 		count_wire--;
7988 		if (!preflight) {
7989 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7990 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7991 		}
7992 		m = m->vmp_snext;
7993 	}
7994 
7995 	if (!preflight) {
7996 		percpu_foreach(free_pages_head, free_pages) {
7997 			for (m = *free_pages_head; m; m = m->vmp_snext) {
7998 				assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
7999 
8000 				pages--;
8001 				count_wire--;
8002 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8003 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8004 
8005 				hibernate_stats.cd_local_free++;
8006 				hibernate_stats.cd_total_free++;
8007 			}
8008 		}
8009 	}
8010 
8011 	for (i = 0; i < vm_colors; i++) {
8012 		vm_page_queue_iterate(&vm_page_queue_free[i].qhead, m, vmp_pageq) {
8013 			assert(m->vmp_q_state == VM_PAGE_ON_FREE_Q);
8014 
8015 			pages--;
8016 			count_wire--;
8017 			if (!preflight) {
8018 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8019 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8020 
8021 				hibernate_stats.cd_total_free++;
8022 			}
8023 		}
8024 	}
8025 
8026 	vm_page_queue_iterate(&vm_lopage_queue_free, m, vmp_pageq) {
8027 		assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
8028 
8029 		pages--;
8030 		count_wire--;
8031 		if (!preflight) {
8032 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8033 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8034 
8035 			hibernate_stats.cd_total_free++;
8036 		}
8037 	}
8038 
8039 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
8040 	while (m && !vm_page_queue_end(&vm_page_queue_throttled, (vm_page_queue_entry_t)m)) {
8041 		assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
8042 
8043 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8044 		discard = FALSE;
8045 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
8046 		    && hibernate_consider_discard(m, preflight)) {
8047 			if (!preflight) {
8048 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8049 			}
8050 			count_discard_inactive++;
8051 			discard = discard_all;
8052 		} else {
8053 			count_throttled++;
8054 		}
8055 		count_wire--;
8056 		if (!preflight) {
8057 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8058 		}
8059 
8060 		if (discard) {
8061 			hibernate_discard_page(m);
8062 		}
8063 		m = next;
8064 	}
8065 
8066 	m = (vm_page_t)vm_page_queue_first(&vm_page_queue_anonymous);
8067 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
8068 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
8069 		bool force_to_wired_list = false;       /* Default to NOT forcing page into the wired page list */
8070 #if CONFIG_SPTM
8071 		force_to_wired_list = hibernate_sptm_should_force_page_to_wired_pagelist(m);
8072 #endif
8073 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8074 		discard = FALSE;
8075 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
8076 		    hibernate_consider_discard(m, preflight)) {
8077 			if (!preflight) {
8078 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8079 			}
8080 			if (m->vmp_dirty) {
8081 				count_discard_purgeable++;
8082 			} else {
8083 				count_discard_inactive++;
8084 			}
8085 			discard = discard_all;
8086 		} else {
8087 			/*
8088 			 * If the page must be force-added to the wired page list, prevent it from appearing
8089 			 * in the unwired page list.
8090 			 */
8091 			if (force_to_wired_list) {
8092 				if (!preflight) {
8093 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8094 				}
8095 			} else {
8096 				count_anonymous++;
8097 			}
8098 		}
8099 		/*
8100 		 * If the page is NOT being forced into the wired page list, remove it from the
8101 		 * wired page list here.
8102 		 */
8103 		if (!force_to_wired_list) {
8104 			count_wire--;
8105 			if (!preflight) {
8106 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8107 			}
8108 		}
8109 		if (discard) {
8110 			hibernate_discard_page(m);
8111 		}
8112 		m = next;
8113 	}
8114 
8115 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
8116 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
8117 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
8118 
8119 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8120 		discard = FALSE;
8121 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
8122 		    hibernate_consider_discard(m, preflight)) {
8123 			if (!preflight) {
8124 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8125 			}
8126 			if (m->vmp_dirty) {
8127 				count_discard_purgeable++;
8128 			} else {
8129 				count_discard_cleaned++;
8130 			}
8131 			discard = discard_all;
8132 		} else {
8133 			count_cleaned++;
8134 		}
8135 		count_wire--;
8136 		if (!preflight) {
8137 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8138 		}
8139 		if (discard) {
8140 			hibernate_discard_page(m);
8141 		}
8142 		m = next;
8143 	}
8144 
8145 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
8146 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
8147 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
8148 		bool force_to_wired_list = false;       /* Default to NOT forcing page into the wired page list */
8149 #if CONFIG_SPTM
8150 		force_to_wired_list = hibernate_sptm_should_force_page_to_wired_pagelist(m);
8151 #endif
8152 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8153 		discard = FALSE;
8154 		if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) &&
8155 		    hibernate_consider_discard(m, preflight)) {
8156 			if (!preflight) {
8157 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8158 			}
8159 			if (m->vmp_dirty) {
8160 				count_discard_purgeable++;
8161 			} else {
8162 				count_discard_active++;
8163 			}
8164 			discard = discard_all;
8165 		} else {
8166 			/*
8167 			 * If the page must be force-added to the wired page list, prevent it from appearing
8168 			 * in the unwired page list.
8169 			 */
8170 			if (force_to_wired_list) {
8171 				if (!preflight) {
8172 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8173 				}
8174 			} else {
8175 				count_active++;
8176 			}
8177 		}
8178 		/*
8179 		 * If the page is NOT being forced into the wired page list, remove it from the
8180 		 * wired page list here.
8181 		 */
8182 		if (!force_to_wired_list) {
8183 			count_wire--;
8184 			if (!preflight) {
8185 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8186 			}
8187 		}
8188 		if (discard) {
8189 			hibernate_discard_page(m);
8190 		}
8191 		m = next;
8192 	}
8193 
8194 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
8195 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
8196 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
8197 		bool force_to_wired_list = false;        /* Default to NOT forcing page into the wired page list */
8198 #if CONFIG_SPTM
8199 		force_to_wired_list = hibernate_sptm_should_force_page_to_wired_pagelist(m);
8200 #endif
8201 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8202 		discard = FALSE;
8203 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
8204 		    hibernate_consider_discard(m, preflight)) {
8205 			if (!preflight) {
8206 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8207 			}
8208 			if (m->vmp_dirty) {
8209 				count_discard_purgeable++;
8210 			} else {
8211 				count_discard_inactive++;
8212 			}
8213 			discard = discard_all;
8214 		} else {
8215 			/*
8216 			 * If the page must be force-added to the wired page list, prevent it from appearing
8217 			 * in the unwired page list.
8218 			 */
8219 			if (force_to_wired_list) {
8220 				if (!preflight) {
8221 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8222 				}
8223 			} else {
8224 				count_inactive++;
8225 			}
8226 		}
8227 		/*
8228 		 * If the page is NOT being forced into the wired page list, remove it from the
8229 		 * wired page list here.
8230 		 */
8231 		if (!force_to_wired_list) {
8232 			count_wire--;
8233 			if (!preflight) {
8234 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8235 			}
8236 		}
8237 		if (discard) {
8238 			hibernate_discard_page(m);
8239 		}
8240 		m = next;
8241 	}
8242 	/* XXX FBDP TODO: secluded queue */
8243 
8244 	for (i = 0; i <= vm_page_max_speculative_age_q; i++) {
8245 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
8246 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
8247 			assertf(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q,
8248 			    "Bad page: %p (0x%x:0x%x) on queue %d has state: %d (Discard: %d, Preflight: %d)",
8249 			    m, m->vmp_pageq.next, m->vmp_pageq.prev, i, m->vmp_q_state, discard, preflight);
8250 
8251 			next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8252 			discard = FALSE;
8253 			if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
8254 			    hibernate_consider_discard(m, preflight)) {
8255 				if (!preflight) {
8256 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8257 				}
8258 				count_discard_speculative++;
8259 				discard = discard_all;
8260 			} else {
8261 				count_speculative++;
8262 			}
8263 			count_wire--;
8264 			if (!preflight) {
8265 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8266 			}
8267 			if (discard) {
8268 				hibernate_discard_page(m);
8269 			}
8270 			m = next;
8271 		}
8272 	}
8273 
8274 	vm_page_queue_iterate(&compressor_object->memq, m, vmp_listq) {
8275 		assert(m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR);
8276 
8277 		count_compressor++;
8278 		count_wire--;
8279 		if (!preflight) {
8280 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
8281 		}
8282 	}
8283 
8284 
8285 	if (preflight == FALSE && discard_all == TRUE) {
8286 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START);
8287 
8288 		HIBLOG("hibernate_teardown started\n");
8289 		count_discard_vm_struct_pages = hibernate_teardown_vm_structs(page_list, page_list_wired);
8290 		HIBLOG("hibernate_teardown completed - discarded %d\n", count_discard_vm_struct_pages);
8291 
8292 		pages -= count_discard_vm_struct_pages;
8293 		count_wire -= count_discard_vm_struct_pages;
8294 
8295 		hibernate_stats.cd_vm_struct_pages_unneeded = count_discard_vm_struct_pages;
8296 
8297 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_END);
8298 	}
8299 
8300 	if (!preflight) {
8301 		// pull wired from hibernate_bitmap
8302 		bitmap = &page_list->bank_bitmap[0];
8303 		bitmap_wired = &page_list_wired->bank_bitmap[0];
8304 		for (bank = 0; bank < page_list->bank_count; bank++) {
8305 			for (i = 0; i < bitmap->bitmapwords; i++) {
8306 				bitmap->bitmap[i] = bitmap->bitmap[i] | ~bitmap_wired->bitmap[i];
8307 			}
8308 			bitmap = (hibernate_bitmap_t *)&bitmap->bitmap[bitmap->bitmapwords];
8309 			bitmap_wired = (hibernate_bitmap_t *) &bitmap_wired->bitmap[bitmap_wired->bitmapwords];
8310 		}
8311 	}
8312 
8313 	// machine dependent adjustments
8314 	hibernate_page_list_setall_machine(page_list, page_list_wired, preflight, &pages);
8315 
8316 	if (!preflight) {
8317 		hibernate_stats.cd_count_wire = count_wire;
8318 		hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable +
8319 		    count_discard_speculative + count_discard_cleaned + count_discard_vm_struct_pages;
8320 	}
8321 
8322 	clock_get_uptime(&end);
8323 	absolutetime_to_nanoseconds(end - start, &nsec);
8324 	HIBLOG("hibernate_page_list_setall time: %qd ms\n", nsec / 1000000ULL);
8325 
8326 	HIBLOG("pages %d, wire %d, act %d, inact %d, cleaned %d spec %d, zf %d, throt %d, compr %d, xpmapped %d\n  %s discard act %d inact %d purgeable %d spec %d cleaned %d retired %d\n",
8327 	    pages, count_wire, count_active, count_inactive, count_cleaned, count_speculative, count_anonymous, count_throttled, count_compressor, hibernate_stats.cd_found_xpmapped,
8328 	    discard_all ? "did" : "could",
8329 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned, count_retired);
8330 
8331 	if (hibernate_stats.cd_skipped_xpmapped) {
8332 		HIBLOG("WARNING: hibernate_page_list_setall skipped %d xpmapped pages\n", hibernate_stats.cd_skipped_xpmapped);
8333 	}
8334 
8335 	*pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative - count_discard_cleaned - count_retired;
8336 
8337 	if (preflight && will_discard) {
8338 		*pagesOut -= count_compressor + count_throttled + count_anonymous + count_inactive + count_cleaned + count_speculative + count_active;
8339 		/*
8340 		 * We try to keep max HIBERNATE_XPMAPPED_LIMIT pages around in the hibernation image
8341 		 * even if these are clean and so we need to size the hibernation image accordingly.
8342 		 *
8343 		 * NB: We have to assume all HIBERNATE_XPMAPPED_LIMIT pages might show up because 'dirty'
8344 		 * xpmapped pages aren't distinguishable from other 'dirty' pages in preflight. So we might
8345 		 * only see part of the xpmapped pages if we look at 'cd_found_xpmapped' which solely tracks
8346 		 * clean xpmapped pages.
8347 		 *
8348 		 * Since these pages are all cleaned by the time we are in the post-preflight phase, we might
8349 		 * see a much larger number in 'cd_found_xpmapped' now than we did in the preflight phase
8350 		 */
8351 		*pagesOut +=  HIBERNATE_XPMAPPED_LIMIT;
8352 	}
8353 
8354 	hibernation_vmqueues_inspection = FALSE;
8355 
8356 #if MACH_ASSERT || DEBUG
8357 	if (!preflight) {
8358 		if (vm_page_local_q) {
8359 			zpercpu_foreach(lq, vm_page_local_q) {
8360 				VPL_UNLOCK(&lq->vpl_lock);
8361 			}
8362 		}
8363 		vm_page_unlock_queues();
8364 	}
8365 #endif  /* MACH_ASSERT || DEBUG */
8366 
8367 	if (preflight) {
8368 		vm_free_page_unlock();
8369 		vm_page_unlock_queues();
8370 		vm_object_unlock(compressor_object);
8371 	}
8372 
8373 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_END, count_wire, *pagesOut, 0, 0, 0);
8374 }
8375 
8376 void
hibernate_page_list_discard(hibernate_page_list_t * page_list)8377 hibernate_page_list_discard(hibernate_page_list_t * page_list)
8378 {
8379 	uint64_t  start, end, nsec;
8380 	vm_page_t m;
8381 	vm_page_t next;
8382 	uint32_t  i;
8383 	uint32_t  count_discard_active    = 0;
8384 	uint32_t  count_discard_inactive  = 0;
8385 	uint32_t  count_discard_purgeable = 0;
8386 	uint32_t  count_discard_cleaned   = 0;
8387 	uint32_t  count_discard_speculative = 0;
8388 
8389 
8390 #if MACH_ASSERT || DEBUG
8391 	vm_page_lock_queues();
8392 	if (vm_page_local_q) {
8393 		zpercpu_foreach(lq, vm_page_local_q) {
8394 			VPL_LOCK(&lq->vpl_lock);
8395 		}
8396 	}
8397 #endif  /* MACH_ASSERT || DEBUG */
8398 
8399 	clock_get_uptime(&start);
8400 
8401 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_anonymous);
8402 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
8403 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
8404 
8405 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8406 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8407 			if (m->vmp_dirty) {
8408 				count_discard_purgeable++;
8409 			} else {
8410 				count_discard_inactive++;
8411 			}
8412 			hibernate_discard_page(m);
8413 		}
8414 		m = next;
8415 	}
8416 
8417 	for (i = 0; i <= vm_page_max_speculative_age_q; i++) {
8418 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
8419 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
8420 			assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q);
8421 
8422 			next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8423 			if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8424 				count_discard_speculative++;
8425 				hibernate_discard_page(m);
8426 			}
8427 			m = next;
8428 		}
8429 	}
8430 
8431 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
8432 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
8433 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
8434 
8435 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8436 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8437 			if (m->vmp_dirty) {
8438 				count_discard_purgeable++;
8439 			} else {
8440 				count_discard_inactive++;
8441 			}
8442 			hibernate_discard_page(m);
8443 		}
8444 		m = next;
8445 	}
8446 	/* XXX FBDP TODO: secluded queue */
8447 
8448 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
8449 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
8450 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
8451 
8452 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8453 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8454 			if (m->vmp_dirty) {
8455 				count_discard_purgeable++;
8456 			} else {
8457 				count_discard_active++;
8458 			}
8459 			hibernate_discard_page(m);
8460 		}
8461 		m = next;
8462 	}
8463 
8464 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
8465 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
8466 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
8467 
8468 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8469 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8470 			if (m->vmp_dirty) {
8471 				count_discard_purgeable++;
8472 			} else {
8473 				count_discard_cleaned++;
8474 			}
8475 			hibernate_discard_page(m);
8476 		}
8477 		m = next;
8478 	}
8479 
8480 #if MACH_ASSERT || DEBUG
8481 	if (vm_page_local_q) {
8482 		zpercpu_foreach(lq, vm_page_local_q) {
8483 			VPL_UNLOCK(&lq->vpl_lock);
8484 		}
8485 	}
8486 	vm_page_unlock_queues();
8487 #endif  /* MACH_ASSERT || DEBUG */
8488 
8489 	clock_get_uptime(&end);
8490 	absolutetime_to_nanoseconds(end - start, &nsec);
8491 	HIBLOG("hibernate_page_list_discard time: %qd ms, discarded act %d inact %d purgeable %d spec %d cleaned %d\n",
8492 	    nsec / 1000000ULL,
8493 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned);
8494 }
8495 
8496 boolean_t       hibernate_paddr_map_inited = FALSE;
8497 unsigned int    hibernate_teardown_last_valid_compact_indx = -1;
8498 vm_page_t       hibernate_rebuild_hash_list = NULL;
8499 
8500 unsigned int    hibernate_teardown_found_tabled_pages = 0;
8501 unsigned int    hibernate_teardown_found_created_pages = 0;
8502 unsigned int    hibernate_teardown_found_free_pages = 0;
8503 unsigned int    hibernate_teardown_vm_page_free_count;
8504 
8505 
8506 struct ppnum_mapping {
8507 	struct ppnum_mapping    *ppnm_next;
8508 	ppnum_t                 ppnm_base_paddr;
8509 	unsigned int            ppnm_sindx;
8510 	unsigned int            ppnm_eindx;
8511 };
8512 
8513 struct ppnum_mapping    *ppnm_head;
8514 struct ppnum_mapping    *ppnm_last_found = NULL;
8515 
8516 
8517 void
hibernate_create_paddr_map(void)8518 hibernate_create_paddr_map(void)
8519 {
8520 	unsigned int    i;
8521 	ppnum_t         next_ppnum_in_run = 0;
8522 	struct ppnum_mapping *ppnm = NULL;
8523 
8524 	if (hibernate_paddr_map_inited == FALSE) {
8525 		for (i = 0; i < vm_pages_count; i++) {
8526 			if (ppnm) {
8527 				ppnm->ppnm_eindx = i;
8528 			}
8529 
8530 			if (ppnm == NULL || VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) != next_ppnum_in_run) {
8531 				ppnm = zalloc_permanent_type(struct ppnum_mapping);
8532 
8533 				ppnm->ppnm_next = ppnm_head;
8534 				ppnm_head = ppnm;
8535 
8536 				ppnm->ppnm_sindx = i;
8537 				ppnm->ppnm_base_paddr = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]);
8538 			}
8539 			next_ppnum_in_run = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) + 1;
8540 		}
8541 		ppnm->ppnm_eindx = vm_pages_count;
8542 
8543 		hibernate_paddr_map_inited = TRUE;
8544 	}
8545 }
8546 
8547 ppnum_t
hibernate_lookup_paddr(unsigned int indx)8548 hibernate_lookup_paddr(unsigned int indx)
8549 {
8550 	struct ppnum_mapping *ppnm = NULL;
8551 
8552 	ppnm = ppnm_last_found;
8553 
8554 	if (ppnm) {
8555 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8556 			goto done;
8557 		}
8558 	}
8559 	for (ppnm = ppnm_head; ppnm; ppnm = ppnm->ppnm_next) {
8560 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8561 			ppnm_last_found = ppnm;
8562 			break;
8563 		}
8564 	}
8565 	if (ppnm == NULL) {
8566 		panic("hibernate_lookup_paddr of %d failed", indx);
8567 	}
8568 done:
8569 	return ppnm->ppnm_base_paddr + (indx - ppnm->ppnm_sindx);
8570 }
8571 
8572 
8573 uint32_t
hibernate_mark_as_unneeded(addr64_t saddr,addr64_t eaddr,hibernate_page_list_t * page_list,hibernate_page_list_t * page_list_wired)8574 hibernate_mark_as_unneeded(addr64_t saddr, addr64_t eaddr, hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8575 {
8576 	addr64_t        saddr_aligned;
8577 	addr64_t        eaddr_aligned;
8578 	addr64_t        addr;
8579 	ppnum_t         paddr;
8580 	unsigned int    mark_as_unneeded_pages = 0;
8581 
8582 	saddr_aligned = (saddr + PAGE_MASK_64) & ~PAGE_MASK_64;
8583 	eaddr_aligned = eaddr & ~PAGE_MASK_64;
8584 
8585 	for (addr = saddr_aligned; addr < eaddr_aligned; addr += PAGE_SIZE_64) {
8586 		paddr = pmap_find_phys(kernel_pmap, addr);
8587 
8588 		assert(paddr);
8589 
8590 		hibernate_page_bitset(page_list, TRUE, paddr);
8591 		hibernate_page_bitset(page_list_wired, TRUE, paddr);
8592 
8593 		mark_as_unneeded_pages++;
8594 	}
8595 	return mark_as_unneeded_pages;
8596 }
8597 
8598 
8599 void
hibernate_hash_insert_page(vm_page_t mem)8600 hibernate_hash_insert_page(vm_page_t mem)
8601 {
8602 	vm_page_bucket_t *bucket;
8603 	int             hash_id;
8604 	vm_object_t     m_object;
8605 
8606 	m_object = VM_PAGE_OBJECT(mem);
8607 
8608 	assert(mem->vmp_hashed);
8609 	assert(m_object);
8610 	assert(mem->vmp_offset != (vm_object_offset_t) -1);
8611 
8612 	/*
8613 	 *	Insert it into the object_object/offset hash table
8614 	 */
8615 	hash_id = vm_page_hash(m_object, mem->vmp_offset);
8616 	bucket = &vm_page_buckets[hash_id];
8617 
8618 	mem->vmp_next_m = bucket->page_list;
8619 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
8620 }
8621 
8622 
8623 void
hibernate_free_range(int sindx,int eindx)8624 hibernate_free_range(int sindx, int eindx)
8625 {
8626 	vm_page_t       mem;
8627 	unsigned int    color;
8628 
8629 	while (sindx < eindx) {
8630 		mem = &vm_pages[sindx];
8631 
8632 		vm_page_init(mem, hibernate_lookup_paddr(sindx), FALSE);
8633 
8634 		mem->vmp_lopage = FALSE;
8635 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8636 
8637 		color = VM_PAGE_GET_COLOR(mem);
8638 #if defined(__x86_64__)
8639 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
8640 #else
8641 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8642 #endif
8643 		vm_page_free_count++;
8644 
8645 		sindx++;
8646 	}
8647 }
8648 
8649 void
hibernate_rebuild_vm_structs(void)8650 hibernate_rebuild_vm_structs(void)
8651 {
8652 	int             i, cindx, sindx, eindx;
8653 	vm_page_t       mem, tmem, mem_next;
8654 	AbsoluteTime    startTime, endTime;
8655 	uint64_t        nsec;
8656 
8657 	if (hibernate_rebuild_needed == FALSE) {
8658 		return;
8659 	}
8660 
8661 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START);
8662 	HIBLOG("hibernate_rebuild started\n");
8663 
8664 	clock_get_uptime(&startTime);
8665 
8666 	pal_hib_rebuild_pmap_structs();
8667 
8668 	bzero(&vm_page_buckets[0], vm_page_bucket_count * sizeof(vm_page_bucket_t));
8669 	eindx = vm_pages_count;
8670 
8671 	/*
8672 	 * Mark all the vm_pages[] that have not been initialized yet as being
8673 	 * transient. This is needed to ensure that buddy page search is corrrect.
8674 	 * Without this random data in these vm_pages[] can trip the buddy search
8675 	 */
8676 	for (i = hibernate_teardown_last_valid_compact_indx + 1; i < eindx; ++i) {
8677 		vm_pages[i].vmp_q_state = VM_PAGE_NOT_ON_Q;
8678 	}
8679 
8680 	for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) {
8681 		mem = &vm_pages[cindx];
8682 		assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
8683 		/*
8684 		 * hibernate_teardown_vm_structs leaves the location where
8685 		 * this vm_page_t must be located in "next".
8686 		 */
8687 		tmem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8688 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
8689 
8690 		sindx = (int)(tmem - &vm_pages[0]);
8691 
8692 		if (mem != tmem) {
8693 			/*
8694 			 * this vm_page_t was moved by hibernate_teardown_vm_structs,
8695 			 * so move it back to its real location
8696 			 */
8697 			*tmem = *mem;
8698 			mem = tmem;
8699 		}
8700 		if (mem->vmp_hashed) {
8701 			hibernate_hash_insert_page(mem);
8702 		}
8703 		/*
8704 		 * the 'hole' between this vm_page_t and the previous
8705 		 * vm_page_t we moved needs to be initialized as
8706 		 * a range of free vm_page_t's
8707 		 */
8708 		hibernate_free_range(sindx + 1, eindx);
8709 
8710 		eindx = sindx;
8711 	}
8712 	if (sindx) {
8713 		hibernate_free_range(0, sindx);
8714 	}
8715 
8716 	assert(vm_page_free_count == hibernate_teardown_vm_page_free_count);
8717 
8718 	/*
8719 	 * process the list of vm_page_t's that were entered in the hash,
8720 	 * but were not located in the vm_pages arrary... these are
8721 	 * vm_page_t's that were created on the fly (i.e. fictitious)
8722 	 */
8723 	for (mem = hibernate_rebuild_hash_list; mem; mem = mem_next) {
8724 		mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8725 
8726 		mem->vmp_next_m = 0;
8727 		hibernate_hash_insert_page(mem);
8728 	}
8729 	hibernate_rebuild_hash_list = NULL;
8730 
8731 	clock_get_uptime(&endTime);
8732 	SUB_ABSOLUTETIME(&endTime, &startTime);
8733 	absolutetime_to_nanoseconds(endTime, &nsec);
8734 
8735 	HIBLOG("hibernate_rebuild completed - took %qd msecs\n", nsec / 1000000ULL);
8736 
8737 	hibernate_rebuild_needed = FALSE;
8738 
8739 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END);
8740 }
8741 
8742 uint32_t
hibernate_teardown_vm_structs(hibernate_page_list_t * page_list,hibernate_page_list_t * page_list_wired)8743 hibernate_teardown_vm_structs(hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8744 {
8745 	unsigned int    i;
8746 	unsigned int    compact_target_indx;
8747 	vm_page_t       mem, mem_next;
8748 	vm_page_bucket_t *bucket;
8749 	unsigned int    mark_as_unneeded_pages = 0;
8750 	unsigned int    unneeded_vm_page_bucket_pages = 0;
8751 	unsigned int    unneeded_vm_pages_pages = 0;
8752 	unsigned int    unneeded_pmap_pages = 0;
8753 	addr64_t        start_of_unneeded = 0;
8754 	addr64_t        end_of_unneeded = 0;
8755 
8756 
8757 	if (hibernate_should_abort()) {
8758 		return 0;
8759 	}
8760 
8761 	hibernate_rebuild_needed = TRUE;
8762 
8763 	HIBLOG("hibernate_teardown: wired_pages %d, free_pages %d, active_pages %d, inactive_pages %d, speculative_pages %d, cleaned_pages %d, compressor_pages %d\n",
8764 	    vm_page_wire_count, vm_page_free_count, vm_page_active_count, vm_page_inactive_count, vm_page_speculative_count,
8765 	    vm_page_cleaned_count, compressor_object->resident_page_count);
8766 
8767 	for (i = 0; i < vm_page_bucket_count; i++) {
8768 		bucket = &vm_page_buckets[i];
8769 
8770 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); mem != VM_PAGE_NULL; mem = mem_next) {
8771 			assert(mem->vmp_hashed);
8772 
8773 			mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8774 
8775 			if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) {
8776 				mem->vmp_next_m = VM_PAGE_PACK_PTR(hibernate_rebuild_hash_list);
8777 				hibernate_rebuild_hash_list = mem;
8778 			}
8779 		}
8780 	}
8781 	unneeded_vm_page_bucket_pages = hibernate_mark_as_unneeded((addr64_t)&vm_page_buckets[0], (addr64_t)&vm_page_buckets[vm_page_bucket_count], page_list, page_list_wired);
8782 	mark_as_unneeded_pages += unneeded_vm_page_bucket_pages;
8783 
8784 	hibernate_teardown_vm_page_free_count = vm_page_free_count;
8785 
8786 	compact_target_indx = 0;
8787 
8788 	for (i = 0; i < vm_pages_count; i++) {
8789 		mem = &vm_pages[i];
8790 
8791 		if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
8792 			unsigned int color;
8793 
8794 			assert(mem->vmp_busy);
8795 			assert(!mem->vmp_lopage);
8796 
8797 			color = VM_PAGE_GET_COLOR(mem);
8798 
8799 			vm_page_queue_remove(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8800 
8801 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
8802 
8803 			vm_page_free_count--;
8804 
8805 			hibernate_teardown_found_free_pages++;
8806 
8807 			if (vm_pages[compact_target_indx].vmp_q_state != VM_PAGE_ON_FREE_Q) {
8808 				compact_target_indx = i;
8809 			}
8810 		} else {
8811 			/*
8812 			 * record this vm_page_t's original location
8813 			 * we need this even if it doesn't get moved
8814 			 * as an indicator to the rebuild function that
8815 			 * we don't have to move it
8816 			 */
8817 			mem->vmp_next_m = VM_PAGE_PACK_PTR(mem);
8818 
8819 			if (vm_pages[compact_target_indx].vmp_q_state == VM_PAGE_ON_FREE_Q) {
8820 				/*
8821 				 * we've got a hole to fill, so
8822 				 * move this vm_page_t to it's new home
8823 				 */
8824 				vm_pages[compact_target_indx] = *mem;
8825 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8826 
8827 				hibernate_teardown_last_valid_compact_indx = compact_target_indx;
8828 				compact_target_indx++;
8829 			} else {
8830 				hibernate_teardown_last_valid_compact_indx = i;
8831 			}
8832 		}
8833 	}
8834 	unneeded_vm_pages_pages = hibernate_mark_as_unneeded((addr64_t)&vm_pages[hibernate_teardown_last_valid_compact_indx + 1],
8835 	    (addr64_t)&vm_pages[vm_pages_count - 1], page_list, page_list_wired);
8836 	mark_as_unneeded_pages += unneeded_vm_pages_pages;
8837 
8838 	pal_hib_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded);
8839 
8840 	if (start_of_unneeded) {
8841 		unneeded_pmap_pages = hibernate_mark_as_unneeded(start_of_unneeded, end_of_unneeded, page_list, page_list_wired);
8842 		mark_as_unneeded_pages += unneeded_pmap_pages;
8843 	}
8844 	HIBLOG("hibernate_teardown: mark_as_unneeded_pages %d, %d, %d\n", unneeded_vm_page_bucket_pages, unneeded_vm_pages_pages, unneeded_pmap_pages);
8845 
8846 	return mark_as_unneeded_pages;
8847 }
8848 
8849 
8850 #endif /* HIBERNATION */
8851 
8852 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
8853 
8854 #include <mach_vm_debug.h>
8855 #if     MACH_VM_DEBUG
8856 
8857 #include <mach_debug/hash_info.h>
8858 #include <vm/vm_debug_internal.h>
8859 
8860 /*
8861  *	Routine:	vm_page_info
8862  *	Purpose:
8863  *		Return information about the global VP table.
8864  *		Fills the buffer with as much information as possible
8865  *		and returns the desired size of the buffer.
8866  *	Conditions:
8867  *		Nothing locked.  The caller should provide
8868  *		possibly-pageable memory.
8869  */
8870 
8871 unsigned int
vm_page_info(hash_info_bucket_t * info,unsigned int count)8872 vm_page_info(
8873 	hash_info_bucket_t *info,
8874 	unsigned int count)
8875 {
8876 	unsigned int i;
8877 	lck_spin_t      *bucket_lock;
8878 
8879 	if (vm_page_bucket_count < count) {
8880 		count = vm_page_bucket_count;
8881 	}
8882 
8883 	for (i = 0; i < count; i++) {
8884 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
8885 		unsigned int bucket_count = 0;
8886 		vm_page_t m;
8887 
8888 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8889 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8890 
8891 		for (m = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8892 		    m != VM_PAGE_NULL;
8893 		    m = (vm_page_t)(VM_PAGE_UNPACK_PTR(m->vmp_next_m))) {
8894 			bucket_count++;
8895 		}
8896 
8897 		lck_spin_unlock(bucket_lock);
8898 
8899 		/* don't touch pageable memory while holding locks */
8900 		info[i].hib_count = bucket_count;
8901 	}
8902 
8903 	return vm_page_bucket_count;
8904 }
8905 #endif  /* MACH_VM_DEBUG */
8906 
8907 #if VM_PAGE_BUCKETS_CHECK
8908 void
vm_page_buckets_check(void)8909 vm_page_buckets_check(void)
8910 {
8911 	unsigned int i;
8912 	vm_page_t p;
8913 	unsigned int p_hash;
8914 	vm_page_bucket_t *bucket;
8915 	lck_spin_t      *bucket_lock;
8916 
8917 	if (!vm_page_buckets_check_ready) {
8918 		return;
8919 	}
8920 
8921 #if HIBERNATION
8922 	if (hibernate_rebuild_needed ||
8923 	    hibernate_rebuild_hash_list) {
8924 		panic("BUCKET_CHECK: hibernation in progress: "
8925 		    "rebuild_needed=%d rebuild_hash_list=%p\n",
8926 		    hibernate_rebuild_needed,
8927 		    hibernate_rebuild_hash_list);
8928 	}
8929 #endif /* HIBERNATION */
8930 
8931 #if VM_PAGE_FAKE_BUCKETS
8932 	char *cp;
8933 	for (cp = (char *) vm_page_fake_buckets_start;
8934 	    cp < (char *) vm_page_fake_buckets_end;
8935 	    cp++) {
8936 		if (*cp != 0x5a) {
8937 			panic("BUCKET_CHECK: corruption at %p in fake buckets "
8938 			    "[0x%llx:0x%llx]\n",
8939 			    cp,
8940 			    (uint64_t) vm_page_fake_buckets_start,
8941 			    (uint64_t) vm_page_fake_buckets_end);
8942 		}
8943 	}
8944 #endif /* VM_PAGE_FAKE_BUCKETS */
8945 
8946 	for (i = 0; i < vm_page_bucket_count; i++) {
8947 		vm_object_t     p_object;
8948 
8949 		bucket = &vm_page_buckets[i];
8950 		if (!bucket->page_list) {
8951 			continue;
8952 		}
8953 
8954 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8955 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8956 		p = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8957 
8958 		while (p != VM_PAGE_NULL) {
8959 			p_object = VM_PAGE_OBJECT(p);
8960 
8961 			if (!p->vmp_hashed) {
8962 				panic("BUCKET_CHECK: page %p (%p,0x%llx) "
8963 				    "hash %d in bucket %d at %p "
8964 				    "is not hashed\n",
8965 				    p, p_object, p->vmp_offset,
8966 				    p_hash, i, bucket);
8967 			}
8968 			p_hash = vm_page_hash(p_object, p->vmp_offset);
8969 			if (p_hash != i) {
8970 				panic("BUCKET_CHECK: corruption in bucket %d "
8971 				    "at %p: page %p object %p offset 0x%llx "
8972 				    "hash %d\n",
8973 				    i, bucket, p, p_object, p->vmp_offset,
8974 				    p_hash);
8975 			}
8976 			p = (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m));
8977 		}
8978 		lck_spin_unlock(bucket_lock);
8979 	}
8980 
8981 //	printf("BUCKET_CHECK: checked buckets\n");
8982 }
8983 #endif /* VM_PAGE_BUCKETS_CHECK */
8984 
8985 /*
8986  * 'vm_fault_enter' will place newly created pages (zero-fill and COW) onto the
8987  * local queues if they exist... its the only spot in the system where we add pages
8988  * to those queues...  once on those queues, those pages can only move to one of the
8989  * global page queues or the free queues... they NEVER move from local q to local q.
8990  * the 'local' state is stable when vm_page_queues_remove is called since we're behind
8991  * the global vm_page_queue_lock at this point...  we still need to take the local lock
8992  * in case this operation is being run on a different CPU then the local queue's identity,
8993  * but we don't have to worry about the page moving to a global queue or becoming wired
8994  * while we're grabbing the local lock since those operations would require the global
8995  * vm_page_queue_lock to be held, and we already own it.
8996  *
8997  * this is why its safe to utilze the wire_count field in the vm_page_t as the local_id...
8998  * 'wired' and local are ALWAYS mutually exclusive conditions.
8999  */
9000 
9001 void
vm_page_queues_remove(vm_page_t mem,boolean_t remove_from_specialq)9002 vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_specialq)
9003 {
9004 	boolean_t       was_pageable = TRUE;
9005 	vm_object_t     m_object;
9006 
9007 	m_object = VM_PAGE_OBJECT(mem);
9008 
9009 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
9010 
9011 	if (mem->vmp_q_state == VM_PAGE_NOT_ON_Q) {
9012 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
9013 		if (remove_from_specialq == TRUE) {
9014 			vm_page_remove_from_specialq(mem);
9015 		}
9016 		/*if (mem->vmp_on_specialq != VM_PAGE_SPECIAL_Q_EMPTY) {
9017 		 *       assert(mem->vmp_specialq.next != 0);
9018 		 *       assert(mem->vmp_specialq.prev != 0);
9019 		 *  } else {*/
9020 		if (mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
9021 			assert(mem->vmp_specialq.next == 0);
9022 			assert(mem->vmp_specialq.prev == 0);
9023 		}
9024 		return;
9025 	}
9026 
9027 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
9028 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
9029 		assert(mem->vmp_specialq.next == 0 &&
9030 		    mem->vmp_specialq.prev == 0 &&
9031 		    mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
9032 		return;
9033 	}
9034 	if (mem->vmp_q_state == VM_PAGE_IS_WIRED) {
9035 		/*
9036 		 * might put these guys on a list for debugging purposes
9037 		 * if we do, we'll need to remove this assert
9038 		 */
9039 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
9040 		assert(mem->vmp_specialq.next == 0 &&
9041 		    mem->vmp_specialq.prev == 0);
9042 		/*
9043 		 * Recall that vmp_on_specialq also means a request to put
9044 		 * it on the special Q. So we don't want to reset that bit
9045 		 * just because a wiring request came in. We might want to
9046 		 * put it on the special queue post-unwiring.
9047 		 *
9048 		 * &&
9049 		 * mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
9050 		 */
9051 		return;
9052 	}
9053 
9054 	assert(m_object != compressor_object);
9055 	assert(!is_kernel_object(m_object));
9056 	assert(!mem->vmp_fictitious);
9057 
9058 	switch (mem->vmp_q_state) {
9059 	case VM_PAGE_ON_ACTIVE_LOCAL_Q:
9060 	{
9061 		struct vpl      *lq;
9062 
9063 		lq = zpercpu_get_cpu(vm_page_local_q, mem->vmp_local_id);
9064 		VPL_LOCK(&lq->vpl_lock);
9065 		vm_page_queue_remove(&lq->vpl_queue, mem, vmp_pageq);
9066 		mem->vmp_local_id = 0;
9067 		lq->vpl_count--;
9068 		if (m_object->internal) {
9069 			lq->vpl_internal_count--;
9070 		} else {
9071 			lq->vpl_external_count--;
9072 		}
9073 		VPL_UNLOCK(&lq->vpl_lock);
9074 		was_pageable = FALSE;
9075 		break;
9076 	}
9077 	case VM_PAGE_ON_ACTIVE_Q:
9078 	{
9079 		vm_page_queue_remove(&vm_page_queue_active, mem, vmp_pageq);
9080 		vm_page_active_count--;
9081 		break;
9082 	}
9083 
9084 	case VM_PAGE_ON_INACTIVE_INTERNAL_Q:
9085 	{
9086 		assert(m_object->internal == TRUE);
9087 
9088 		vm_page_inactive_count--;
9089 		vm_page_queue_remove(&vm_page_queue_anonymous, mem, vmp_pageq);
9090 		vm_page_anonymous_count--;
9091 
9092 		vm_purgeable_q_advance_all();
9093 		vm_page_balance_inactive(3);
9094 		break;
9095 	}
9096 
9097 	case VM_PAGE_ON_INACTIVE_EXTERNAL_Q:
9098 	{
9099 		assert(m_object->internal == FALSE);
9100 
9101 		vm_page_inactive_count--;
9102 		vm_page_queue_remove(&vm_page_queue_inactive, mem, vmp_pageq);
9103 		vm_purgeable_q_advance_all();
9104 		vm_page_balance_inactive(3);
9105 		break;
9106 	}
9107 
9108 	case VM_PAGE_ON_INACTIVE_CLEANED_Q:
9109 	{
9110 		assert(m_object->internal == FALSE);
9111 
9112 		vm_page_inactive_count--;
9113 		vm_page_queue_remove(&vm_page_queue_cleaned, mem, vmp_pageq);
9114 		vm_page_cleaned_count--;
9115 		vm_page_balance_inactive(3);
9116 		break;
9117 	}
9118 
9119 	case VM_PAGE_ON_THROTTLED_Q:
9120 	{
9121 		assert(m_object->internal == TRUE);
9122 
9123 		vm_page_queue_remove(&vm_page_queue_throttled, mem, vmp_pageq);
9124 		vm_page_throttled_count--;
9125 		was_pageable = FALSE;
9126 		break;
9127 	}
9128 
9129 	case VM_PAGE_ON_SPECULATIVE_Q:
9130 	{
9131 		assert(m_object->internal == FALSE);
9132 
9133 		vm_page_remque(&mem->vmp_pageq);
9134 		vm_page_speculative_count--;
9135 		vm_page_balance_inactive(3);
9136 		break;
9137 	}
9138 
9139 #if CONFIG_SECLUDED_MEMORY
9140 	case VM_PAGE_ON_SECLUDED_Q:
9141 	{
9142 		vm_page_queue_remove(&vm_page_queue_secluded, mem, vmp_pageq);
9143 		vm_page_secluded_count--;
9144 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
9145 		if (m_object == VM_OBJECT_NULL) {
9146 			vm_page_secluded_count_free--;
9147 			was_pageable = FALSE;
9148 		} else {
9149 			assert(!m_object->internal);
9150 			vm_page_secluded_count_inuse--;
9151 			was_pageable = FALSE;
9152 //			was_pageable = TRUE;
9153 		}
9154 		break;
9155 	}
9156 #endif /* CONFIG_SECLUDED_MEMORY */
9157 
9158 	default:
9159 	{
9160 		/*
9161 		 *	if (mem->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)
9162 		 *              NOTE: vm_page_queues_remove does not deal with removing pages from the pageout queue...
9163 		 *              the caller is responsible for determing if the page is on that queue, and if so, must
9164 		 *              either first remove it (it needs both the page queues lock and the object lock to do
9165 		 *              this via vm_pageout_steal_laundry), or avoid the call to vm_page_queues_remove
9166 		 *
9167 		 *	we also don't expect to encounter VM_PAGE_ON_FREE_Q, VM_PAGE_ON_FREE_LOCAL_Q, VM_PAGE_ON_FREE_LOPAGE_Q
9168 		 *	or any of the undefined states
9169 		 */
9170 		panic("vm_page_queues_remove - bad page q_state (%p, %d)", mem, mem->vmp_q_state);
9171 		break;
9172 	}
9173 	}
9174 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
9175 	mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
9176 
9177 	if (remove_from_specialq == TRUE) {
9178 		vm_page_remove_from_specialq(mem);
9179 	}
9180 	if (was_pageable) {
9181 		if (m_object->internal) {
9182 			vm_page_pageable_internal_count--;
9183 		} else {
9184 			vm_page_pageable_external_count--;
9185 		}
9186 	}
9187 }
9188 
9189 void
vm_page_remove_internal(vm_page_t page)9190 vm_page_remove_internal(vm_page_t page)
9191 {
9192 	vm_object_t __object = VM_PAGE_OBJECT(page);
9193 	if (page == __object->memq_hint) {
9194 		vm_page_t       __new_hint;
9195 		vm_page_queue_entry_t   __qe;
9196 		__qe = (vm_page_queue_entry_t)vm_page_queue_next(&page->vmp_listq);
9197 		if (vm_page_queue_end(&__object->memq, __qe)) {
9198 			__qe = (vm_page_queue_entry_t)vm_page_queue_prev(&page->vmp_listq);
9199 			if (vm_page_queue_end(&__object->memq, __qe)) {
9200 				__qe = NULL;
9201 			}
9202 		}
9203 		__new_hint = (vm_page_t)((uintptr_t) __qe);
9204 		__object->memq_hint = __new_hint;
9205 	}
9206 	vm_page_queue_remove(&__object->memq, page, vmp_listq);
9207 #if CONFIG_SECLUDED_MEMORY
9208 	if (__object->eligible_for_secluded) {
9209 		vm_page_secluded.eligible_for_secluded--;
9210 	}
9211 #endif /* CONFIG_SECLUDED_MEMORY */
9212 }
9213 
9214 void
vm_page_enqueue_inactive(vm_page_t mem,boolean_t first)9215 vm_page_enqueue_inactive(vm_page_t mem, boolean_t first)
9216 {
9217 	vm_object_t     m_object;
9218 
9219 	m_object = VM_PAGE_OBJECT(mem);
9220 
9221 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
9222 	assert(!mem->vmp_fictitious);
9223 	assert(!mem->vmp_laundry);
9224 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
9225 	vm_page_check_pageable_safe(mem);
9226 
9227 	if (m_object->internal) {
9228 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_INTERNAL_Q;
9229 
9230 		if (first == TRUE) {
9231 			vm_page_queue_enter_first(&vm_page_queue_anonymous, mem, vmp_pageq);
9232 		} else {
9233 			vm_page_queue_enter(&vm_page_queue_anonymous, mem, vmp_pageq);
9234 		}
9235 
9236 		vm_page_anonymous_count++;
9237 		vm_page_pageable_internal_count++;
9238 	} else {
9239 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_EXTERNAL_Q;
9240 
9241 		if (first == TRUE) {
9242 			vm_page_queue_enter_first(&vm_page_queue_inactive, mem, vmp_pageq);
9243 		} else {
9244 			vm_page_queue_enter(&vm_page_queue_inactive, mem, vmp_pageq);
9245 		}
9246 
9247 		vm_page_pageable_external_count++;
9248 	}
9249 	vm_page_inactive_count++;
9250 	token_new_pagecount++;
9251 
9252 	vm_page_add_to_specialq(mem, FALSE);
9253 }
9254 
9255 void
vm_page_enqueue_active(vm_page_t mem,boolean_t first)9256 vm_page_enqueue_active(vm_page_t mem, boolean_t first)
9257 {
9258 	vm_object_t     m_object;
9259 
9260 	m_object = VM_PAGE_OBJECT(mem);
9261 
9262 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
9263 	assert(!mem->vmp_fictitious);
9264 	assert(!mem->vmp_laundry);
9265 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
9266 	vm_page_check_pageable_safe(mem);
9267 
9268 	mem->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
9269 	if (first == TRUE) {
9270 		vm_page_queue_enter_first(&vm_page_queue_active, mem, vmp_pageq);
9271 	} else {
9272 		vm_page_queue_enter(&vm_page_queue_active, mem, vmp_pageq);
9273 	}
9274 	vm_page_active_count++;
9275 
9276 	if (m_object->internal) {
9277 		vm_page_pageable_internal_count++;
9278 	} else {
9279 		vm_page_pageable_external_count++;
9280 	}
9281 
9282 	vm_page_add_to_specialq(mem, FALSE);
9283 	vm_page_balance_inactive(3);
9284 }
9285 
9286 /*
9287  * Pages from special kernel objects shouldn't
9288  * be placed on pageable queues.
9289  */
9290 void
vm_page_check_pageable_safe(vm_page_t page)9291 vm_page_check_pageable_safe(vm_page_t page)
9292 {
9293 	vm_object_t     page_object;
9294 
9295 	page_object = VM_PAGE_OBJECT(page);
9296 
9297 	if (is_kernel_object(page_object)) {
9298 		panic("vm_page_check_pageable_safe: trying to add page"
9299 		    "from a kernel object to pageable queue");
9300 	}
9301 
9302 	if (page_object == compressor_object) {
9303 		panic("vm_page_check_pageable_safe: trying to add page"
9304 		    "from compressor object (%p) to pageable queue", compressor_object);
9305 	}
9306 }
9307 
9308 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
9309 * wired page diagnose
9310 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
9311 
9312 #include <libkern/OSKextLibPrivate.h>
9313 
9314 #define KA_SIZE(namelen, subtotalscount)        \
9315 	(sizeof(struct vm_allocation_site) + (namelen) + 1 + ((subtotalscount) * sizeof(struct vm_allocation_total)))
9316 
9317 #define KA_NAME(alloc)  \
9318 	((char *)(&(alloc)->subtotals[(alloc->subtotalscount)]))
9319 
9320 #define KA_NAME_LEN(alloc)      \
9321     (VM_TAG_NAME_LEN_MAX & (alloc->flags >> VM_TAG_NAME_LEN_SHIFT))
9322 
9323 vm_tag_t
vm_tag_bt(void)9324 vm_tag_bt(void)
9325 {
9326 	uintptr_t* frameptr;
9327 	uintptr_t* frameptr_next;
9328 	uintptr_t retaddr;
9329 	uintptr_t kstackb, kstackt;
9330 	const vm_allocation_site_t * site;
9331 	thread_t cthread;
9332 	kern_allocation_name_t name;
9333 
9334 	cthread = current_thread();
9335 	if (__improbable(cthread == NULL)) {
9336 		return VM_KERN_MEMORY_OSFMK;
9337 	}
9338 
9339 	if ((name = thread_get_kernel_state(cthread)->allocation_name)) {
9340 		if (!name->tag) {
9341 			vm_tag_alloc(name);
9342 		}
9343 		return name->tag;
9344 	}
9345 
9346 	kstackb = cthread->kernel_stack;
9347 	kstackt = kstackb + kernel_stack_size;
9348 
9349 	/* Load stack frame pointer (EBP on x86) into frameptr */
9350 	frameptr = __builtin_frame_address(0);
9351 	site = NULL;
9352 	while (frameptr != NULL) {
9353 		/* Verify thread stack bounds */
9354 		if (((uintptr_t)(frameptr + 2) > kstackt) || ((uintptr_t)frameptr < kstackb)) {
9355 			break;
9356 		}
9357 
9358 		/* Next frame pointer is pointed to by the previous one */
9359 		frameptr_next = (uintptr_t*) *frameptr;
9360 #if defined(HAS_APPLE_PAC)
9361 		frameptr_next = ptrauth_strip(frameptr_next, ptrauth_key_frame_pointer);
9362 #endif
9363 
9364 		/* Pull return address from one spot above the frame pointer */
9365 		retaddr = *(frameptr + 1);
9366 
9367 #if defined(HAS_APPLE_PAC)
9368 		retaddr = (uintptr_t) ptrauth_strip((void *)retaddr, ptrauth_key_return_address);
9369 #endif
9370 
9371 		if (((retaddr < vm_kernel_builtinkmod_text_end) && (retaddr >= vm_kernel_builtinkmod_text))
9372 		    || (retaddr < vm_kernel_stext) || (retaddr > vm_kernel_top)) {
9373 			site = OSKextGetAllocationSiteForCaller(retaddr);
9374 			break;
9375 		}
9376 		frameptr = frameptr_next;
9377 	}
9378 
9379 	if (site) {
9380 		return site->tag;
9381 	}
9382 
9383 #if MACH_ASSERT
9384 	/*
9385 	 * Kernel tests appear here as unrecognized call sites and would get
9386 	 * no memory tag. Give them a default tag to prevent panics later.
9387 	 */
9388 	if (thread_get_test_option(test_option_vm_prevent_wire_tag_panic)) {
9389 		return VM_KERN_MEMORY_OSFMK;
9390 	}
9391 #endif
9392 
9393 	return VM_KERN_MEMORY_NONE;
9394 }
9395 
9396 static uint64_t free_tag_bits[VM_MAX_TAG_VALUE / 64];
9397 
9398 void
vm_tag_alloc_locked(vm_allocation_site_t * site,vm_allocation_site_t ** releasesiteP)9399 vm_tag_alloc_locked(vm_allocation_site_t * site, vm_allocation_site_t ** releasesiteP)
9400 {
9401 	vm_tag_t tag;
9402 	uint64_t avail;
9403 	uint32_t idx;
9404 	vm_allocation_site_t * prev;
9405 
9406 	if (site->tag) {
9407 		return;
9408 	}
9409 
9410 	idx = 0;
9411 	while (TRUE) {
9412 		avail = free_tag_bits[idx];
9413 		if (avail) {
9414 			tag = (vm_tag_t)__builtin_clzll(avail);
9415 			avail &= ~(1ULL << (63 - tag));
9416 			free_tag_bits[idx] = avail;
9417 			tag += (idx << 6);
9418 			break;
9419 		}
9420 		idx++;
9421 		if (idx >= ARRAY_COUNT(free_tag_bits)) {
9422 			for (idx = 0; idx < ARRAY_COUNT(vm_allocation_sites); idx++) {
9423 				prev = vm_allocation_sites[idx];
9424 				if (!prev) {
9425 					continue;
9426 				}
9427 				if (!KA_NAME_LEN(prev)) {
9428 					continue;
9429 				}
9430 				if (!prev->tag) {
9431 					continue;
9432 				}
9433 				if (prev->total) {
9434 					continue;
9435 				}
9436 				if (1 != prev->refcount) {
9437 					continue;
9438 				}
9439 
9440 				assert(idx == prev->tag);
9441 				tag = (vm_tag_t)idx;
9442 				prev->tag = VM_KERN_MEMORY_NONE;
9443 				*releasesiteP = prev;
9444 				break;
9445 			}
9446 			if (idx >= ARRAY_COUNT(vm_allocation_sites)) {
9447 				tag = VM_KERN_MEMORY_ANY;
9448 			}
9449 			break;
9450 		}
9451 	}
9452 	site->tag = tag;
9453 
9454 	OSAddAtomic16(1, &site->refcount);
9455 
9456 	if (VM_KERN_MEMORY_ANY != tag) {
9457 		vm_allocation_sites[tag] = site;
9458 	}
9459 
9460 	if (tag > vm_allocation_tag_highest) {
9461 		vm_allocation_tag_highest = tag;
9462 	}
9463 }
9464 
9465 static void
vm_tag_free_locked(vm_tag_t tag)9466 vm_tag_free_locked(vm_tag_t tag)
9467 {
9468 	uint64_t avail;
9469 	uint32_t idx;
9470 	uint64_t bit;
9471 
9472 	if (VM_KERN_MEMORY_ANY == tag) {
9473 		return;
9474 	}
9475 
9476 	idx = (tag >> 6);
9477 	avail = free_tag_bits[idx];
9478 	tag &= 63;
9479 	bit = (1ULL << (63 - tag));
9480 	assert(!(avail & bit));
9481 	free_tag_bits[idx] = (avail | bit);
9482 }
9483 
9484 static void
vm_tag_init(void)9485 vm_tag_init(void)
9486 {
9487 	vm_tag_t tag;
9488 	for (tag = VM_KERN_MEMORY_FIRST_DYNAMIC; tag < VM_KERN_MEMORY_ANY; tag++) {
9489 		vm_tag_free_locked(tag);
9490 	}
9491 
9492 	for (tag = VM_KERN_MEMORY_ANY + 1; tag < VM_MAX_TAG_VALUE; tag++) {
9493 		vm_tag_free_locked(tag);
9494 	}
9495 }
9496 
9497 vm_tag_t
vm_tag_alloc(vm_allocation_site_t * site)9498 vm_tag_alloc(vm_allocation_site_t * site)
9499 {
9500 	vm_allocation_site_t * releasesite;
9501 
9502 	if (!site->tag) {
9503 		releasesite = NULL;
9504 		lck_ticket_lock(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
9505 		vm_tag_alloc_locked(site, &releasesite);
9506 		lck_ticket_unlock(&vm_allocation_sites_lock);
9507 		if (releasesite) {
9508 			kern_allocation_name_release(releasesite);
9509 		}
9510 	}
9511 
9512 	return site->tag;
9513 }
9514 
9515 #if VM_BTLOG_TAGS
9516 #define VM_KERN_MEMORY_STR_MAX_LEN (32)
9517 TUNABLE_STR(vmtaglog, VM_KERN_MEMORY_STR_MAX_LEN, "vmtaglog", "");
9518 #define VM_TAG_BTLOG_SIZE (16u << 10)
9519 
9520 btlog_t vmtaglog_btlog;
9521 vm_tag_t vmtaglog_tag;
9522 
9523 static void
vm_tag_log(vm_object_t object,int64_t delta,void * fp)9524 vm_tag_log(vm_object_t object, int64_t delta, void *fp)
9525 {
9526 	if (is_kernel_object(object)) {
9527 		/* kernel object backtraces are tracked in vm entries */
9528 		return;
9529 	}
9530 	if (delta > 0) {
9531 		btref_t ref = btref_get(fp, BTREF_GET_NOWAIT);
9532 		btlog_record(vmtaglog_btlog, object, 0, ref);
9533 	} else if (object->wired_page_count == 0) {
9534 		btlog_erase(vmtaglog_btlog, object);
9535 	}
9536 }
9537 
9538 #ifndef ARRAY_SIZE
9539 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
9540 #endif /* ARRAY_SIZE */
9541 #define VM_KERN_MEMORY_ELEM(name) [VM_KERN_MEMORY_##name] = #name
9542 const char *vm_kern_memory_strs[] = {
9543 	VM_KERN_MEMORY_ELEM(OSFMK),
9544 	VM_KERN_MEMORY_ELEM(BSD),
9545 	VM_KERN_MEMORY_ELEM(IOKIT),
9546 	VM_KERN_MEMORY_ELEM(LIBKERN),
9547 	VM_KERN_MEMORY_ELEM(OSKEXT),
9548 	VM_KERN_MEMORY_ELEM(KEXT),
9549 	VM_KERN_MEMORY_ELEM(IPC),
9550 	VM_KERN_MEMORY_ELEM(STACK),
9551 	VM_KERN_MEMORY_ELEM(CPU),
9552 	VM_KERN_MEMORY_ELEM(PMAP),
9553 	VM_KERN_MEMORY_ELEM(PTE),
9554 	VM_KERN_MEMORY_ELEM(ZONE),
9555 	VM_KERN_MEMORY_ELEM(KALLOC),
9556 	VM_KERN_MEMORY_ELEM(COMPRESSOR),
9557 	VM_KERN_MEMORY_ELEM(COMPRESSED_DATA),
9558 	VM_KERN_MEMORY_ELEM(PHANTOM_CACHE),
9559 	VM_KERN_MEMORY_ELEM(WAITQ),
9560 	VM_KERN_MEMORY_ELEM(DIAG),
9561 	VM_KERN_MEMORY_ELEM(LOG),
9562 	VM_KERN_MEMORY_ELEM(FILE),
9563 	VM_KERN_MEMORY_ELEM(MBUF),
9564 	VM_KERN_MEMORY_ELEM(UBC),
9565 	VM_KERN_MEMORY_ELEM(SECURITY),
9566 	VM_KERN_MEMORY_ELEM(MLOCK),
9567 	VM_KERN_MEMORY_ELEM(REASON),
9568 	VM_KERN_MEMORY_ELEM(SKYWALK),
9569 	VM_KERN_MEMORY_ELEM(LTABLE),
9570 	VM_KERN_MEMORY_ELEM(HV),
9571 	VM_KERN_MEMORY_ELEM(KALLOC_DATA),
9572 	VM_KERN_MEMORY_ELEM(RETIRED),
9573 	VM_KERN_MEMORY_ELEM(KALLOC_TYPE),
9574 	VM_KERN_MEMORY_ELEM(TRIAGE),
9575 	VM_KERN_MEMORY_ELEM(RECOUNT),
9576 };
9577 
9578 static vm_tag_t
vm_tag_str_to_idx(char tagstr[VM_KERN_MEMORY_STR_MAX_LEN])9579 vm_tag_str_to_idx(char tagstr[VM_KERN_MEMORY_STR_MAX_LEN])
9580 {
9581 	for (vm_tag_t i = VM_KERN_MEMORY_OSFMK; i < ARRAY_SIZE(vm_kern_memory_strs); i++) {
9582 		if (!strncmp(vm_kern_memory_strs[i], tagstr, VM_KERN_MEMORY_STR_MAX_LEN)) {
9583 			return i;
9584 		}
9585 	}
9586 
9587 	printf("Unable to find vm tag %s for btlog\n", tagstr);
9588 	return VM_KERN_MEMORY_NONE;
9589 }
9590 
9591 __startup_func
9592 static void
vm_btlog_init(void)9593 vm_btlog_init(void)
9594 {
9595 	vmtaglog_tag = vm_tag_str_to_idx(vmtaglog);
9596 
9597 	if (vmtaglog_tag != VM_KERN_MEMORY_NONE) {
9598 		vmtaglog_btlog = btlog_create(BTLOG_HASH, VM_TAG_BTLOG_SIZE, 0);
9599 	}
9600 }
9601 STARTUP(ZALLOC, STARTUP_RANK_FIRST, vm_btlog_init);
9602 #endif /* VM_BTLOG_TAGS */
9603 
9604 void
vm_tag_update_size(vm_tag_t tag,int64_t delta,vm_object_t object)9605 vm_tag_update_size(vm_tag_t tag, int64_t delta, vm_object_t object)
9606 {
9607 	assert(VM_KERN_MEMORY_NONE != tag && tag < VM_MAX_TAG_VALUE);
9608 
9609 	kern_allocation_update_size(vm_allocation_sites[tag], delta, object);
9610 }
9611 
9612 uint64_t
vm_tag_get_size(vm_tag_t tag)9613 vm_tag_get_size(vm_tag_t tag)
9614 {
9615 	vm_allocation_site_t *allocation;
9616 
9617 	assert(VM_KERN_MEMORY_NONE != tag && tag < VM_MAX_TAG_VALUE);
9618 
9619 	allocation = vm_allocation_sites[tag];
9620 	return allocation ? os_atomic_load(&allocation->total, relaxed) : 0;
9621 }
9622 
9623 void
kern_allocation_update_size(kern_allocation_name_t allocation,int64_t delta,__unused vm_object_t object)9624 kern_allocation_update_size(kern_allocation_name_t allocation, int64_t delta, __unused vm_object_t object)
9625 {
9626 	uint64_t value;
9627 
9628 	value = os_atomic_add(&allocation->total, delta, relaxed);
9629 	if (delta < 0) {
9630 		assertf(value + (uint64_t)-delta > value,
9631 		    "tag %d, site %p", allocation->tag, allocation);
9632 	}
9633 
9634 #if DEBUG || DEVELOPMENT
9635 	if (value > allocation->peak) {
9636 		os_atomic_max(&allocation->peak, value, relaxed);
9637 	}
9638 #endif /* DEBUG || DEVELOPMENT */
9639 
9640 	if (value == (uint64_t)delta && !allocation->tag) {
9641 		vm_tag_alloc(allocation);
9642 	}
9643 
9644 #if VM_BTLOG_TAGS
9645 	if (vmtaglog_tag && (allocation->tag == vmtaglog_tag) && object) {
9646 		vm_tag_log(object, delta, __builtin_frame_address(0));
9647 	}
9648 #endif /* VM_BTLOG_TAGS */
9649 }
9650 
9651 #if VM_TAG_SIZECLASSES
9652 
9653 void
vm_allocation_zones_init(void)9654 vm_allocation_zones_init(void)
9655 {
9656 	vm_offset_t   addr;
9657 	vm_size_t     size;
9658 
9659 	const vm_tag_t early_tags[] = {
9660 		VM_KERN_MEMORY_DIAG,
9661 		VM_KERN_MEMORY_KALLOC,
9662 		VM_KERN_MEMORY_KALLOC_DATA,
9663 		VM_KERN_MEMORY_KALLOC_TYPE,
9664 		VM_KERN_MEMORY_LIBKERN,
9665 		VM_KERN_MEMORY_OSFMK,
9666 		VM_KERN_MEMORY_RECOUNT,
9667 	};
9668 
9669 	size = VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *)
9670 	    + ARRAY_COUNT(early_tags) * VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9671 
9672 	kmem_alloc(kernel_map, &addr, round_page(size),
9673 	    KMA_NOFAIL | KMA_KOBJECT | KMA_ZERO | KMA_PERMANENT,
9674 	    VM_KERN_MEMORY_DIAG);
9675 
9676 	vm_allocation_zone_totals = (vm_allocation_zone_total_t **) addr;
9677 	addr += VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *);
9678 
9679 	// prepopulate early tag ranges so allocations
9680 	// in vm_tag_update_zone_size() and early boot won't recurse
9681 	for (size_t i = 0; i < ARRAY_COUNT(early_tags); i++) {
9682 		vm_allocation_zone_totals[early_tags[i]] = (vm_allocation_zone_total_t *)addr;
9683 		addr += VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9684 	}
9685 }
9686 
9687 __attribute__((noinline))
9688 static vm_tag_t
vm_tag_zone_stats_alloc(vm_tag_t tag,zalloc_flags_t flags)9689 vm_tag_zone_stats_alloc(vm_tag_t tag, zalloc_flags_t flags)
9690 {
9691 	vm_allocation_zone_total_t *stats;
9692 	vm_size_t size = sizeof(*stats) * VM_TAG_SIZECLASSES;
9693 
9694 	flags = Z_VM_TAG(Z_ZERO | flags, VM_KERN_MEMORY_DIAG);
9695 	stats = kalloc_data(size, flags);
9696 	if (!stats) {
9697 		return VM_KERN_MEMORY_NONE;
9698 	}
9699 	if (!os_atomic_cmpxchg(&vm_allocation_zone_totals[tag], NULL, stats, release)) {
9700 		kfree_data(stats, size);
9701 	}
9702 	return tag;
9703 }
9704 
9705 vm_tag_t
vm_tag_will_update_zone(vm_tag_t tag,uint32_t zflags)9706 vm_tag_will_update_zone(vm_tag_t tag, uint32_t zflags)
9707 {
9708 	assert(VM_KERN_MEMORY_NONE != tag);
9709 	assert(tag < VM_MAX_TAG_VALUE);
9710 
9711 	if (__probable(vm_allocation_zone_totals[tag])) {
9712 		return tag;
9713 	}
9714 	return vm_tag_zone_stats_alloc(tag, zflags);
9715 }
9716 
9717 void
vm_tag_update_zone_size(vm_tag_t tag,uint32_t zidx,long delta)9718 vm_tag_update_zone_size(vm_tag_t tag, uint32_t zidx, long delta)
9719 {
9720 	vm_allocation_zone_total_t *stats;
9721 	vm_size_t value;
9722 
9723 	assert(VM_KERN_MEMORY_NONE != tag);
9724 	assert(tag < VM_MAX_TAG_VALUE);
9725 
9726 	if (zidx >= VM_TAG_SIZECLASSES) {
9727 		return;
9728 	}
9729 
9730 	stats = vm_allocation_zone_totals[tag];
9731 	assert(stats);
9732 	stats += zidx;
9733 
9734 	value = os_atomic_add(&stats->vazt_total, delta, relaxed);
9735 	if (delta < 0) {
9736 		assertf((long)value >= 0, "zidx %d, tag %d, %p", zidx, tag, stats);
9737 		return;
9738 	} else if (os_atomic_load(&stats->vazt_peak, relaxed) < value) {
9739 		os_atomic_max(&stats->vazt_peak, value, relaxed);
9740 	}
9741 }
9742 
9743 #endif /* VM_TAG_SIZECLASSES */
9744 
9745 void
kern_allocation_update_subtotal(kern_allocation_name_t allocation,vm_tag_t subtag,int64_t delta)9746 kern_allocation_update_subtotal(kern_allocation_name_t allocation, vm_tag_t subtag, int64_t delta)
9747 {
9748 	kern_allocation_name_t other;
9749 	struct vm_allocation_total * total;
9750 	uint32_t subidx;
9751 
9752 	assert(VM_KERN_MEMORY_NONE != subtag);
9753 	lck_ticket_lock(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
9754 	for (subidx = 0; subidx < allocation->subtotalscount; subidx++) {
9755 		total = &allocation->subtotals[subidx];
9756 		if (subtag == total->tag) {
9757 			break;
9758 		}
9759 	}
9760 	if (subidx >= allocation->subtotalscount) {
9761 		for (subidx = 0; subidx < allocation->subtotalscount; subidx++) {
9762 			total = &allocation->subtotals[subidx];
9763 			if ((VM_KERN_MEMORY_NONE == total->tag)
9764 			    || !total->total) {
9765 				total->tag = (vm_tag_t)subtag;
9766 				break;
9767 			}
9768 		}
9769 	}
9770 	assert(subidx < allocation->subtotalscount);
9771 	if (subidx >= allocation->subtotalscount) {
9772 		lck_ticket_unlock(&vm_allocation_sites_lock);
9773 		return;
9774 	}
9775 	if (delta < 0) {
9776 		assertf(total->total >= ((uint64_t)-delta), "name %p", allocation);
9777 	}
9778 	OSAddAtomic64(delta, &total->total);
9779 	lck_ticket_unlock(&vm_allocation_sites_lock);
9780 
9781 	other = vm_allocation_sites[subtag];
9782 	assert(other);
9783 	if (delta < 0) {
9784 		assertf(other->mapped >= ((uint64_t)-delta), "other %p", other);
9785 	}
9786 	OSAddAtomic64(delta, &other->mapped);
9787 }
9788 
9789 const char *
kern_allocation_get_name(kern_allocation_name_t allocation)9790 kern_allocation_get_name(kern_allocation_name_t allocation)
9791 {
9792 	return KA_NAME(allocation);
9793 }
9794 
9795 kern_allocation_name_t
kern_allocation_name_allocate(const char * name,uint16_t subtotalscount)9796 kern_allocation_name_allocate(const char * name, uint16_t subtotalscount)
9797 {
9798 	kern_allocation_name_t allocation;
9799 	uint16_t namelen;
9800 
9801 	namelen = (uint16_t)strnlen(name, MACH_MEMORY_INFO_NAME_MAX_LEN - 1);
9802 
9803 	allocation = kalloc_data(KA_SIZE(namelen, subtotalscount), Z_WAITOK | Z_ZERO);
9804 	allocation->refcount       = 1;
9805 	allocation->subtotalscount = subtotalscount;
9806 	allocation->flags          = (uint16_t)(namelen << VM_TAG_NAME_LEN_SHIFT);
9807 	strlcpy(KA_NAME(allocation), name, namelen + 1);
9808 
9809 	vm_tag_alloc(allocation);
9810 	return allocation;
9811 }
9812 
9813 void
kern_allocation_name_release(kern_allocation_name_t allocation)9814 kern_allocation_name_release(kern_allocation_name_t allocation)
9815 {
9816 	assert(allocation->refcount > 0);
9817 	if (1 == OSAddAtomic16(-1, &allocation->refcount)) {
9818 		kfree_data(allocation,
9819 		    KA_SIZE(KA_NAME_LEN(allocation), allocation->subtotalscount));
9820 	}
9821 }
9822 
9823 #if !VM_TAG_ACTIVE_UPDATE
9824 static void
vm_page_count_object(mach_memory_info_t * info,unsigned int __unused num_info,vm_object_t object)9825 vm_page_count_object(mach_memory_info_t * info, unsigned int __unused num_info, vm_object_t object)
9826 {
9827 	if (!object->wired_page_count) {
9828 		return;
9829 	}
9830 	if (!is_kernel_object(object)) {
9831 		assert(object->wire_tag < num_info);
9832 		info[object->wire_tag].size += ptoa_64(object->wired_page_count);
9833 	}
9834 }
9835 
9836 typedef void (*vm_page_iterate_proc)(mach_memory_info_t * info,
9837     unsigned int num_info, vm_object_t object);
9838 
9839 static void
vm_page_iterate_purgeable_objects(mach_memory_info_t * info,unsigned int num_info,vm_page_iterate_proc proc,purgeable_q_t queue,int group)9840 vm_page_iterate_purgeable_objects(mach_memory_info_t * info, unsigned int num_info,
9841     vm_page_iterate_proc proc, purgeable_q_t queue,
9842     int group)
9843 {
9844 	vm_object_t object;
9845 
9846 	for (object = (vm_object_t) queue_first(&queue->objq[group]);
9847 	    !queue_end(&queue->objq[group], (queue_entry_t) object);
9848 	    object = (vm_object_t) queue_next(&object->objq)) {
9849 		proc(info, num_info, object);
9850 	}
9851 }
9852 
9853 static void
vm_page_iterate_objects(mach_memory_info_t * info,unsigned int num_info,vm_page_iterate_proc proc)9854 vm_page_iterate_objects(mach_memory_info_t * info, unsigned int num_info,
9855     vm_page_iterate_proc proc)
9856 {
9857 	vm_object_t     object;
9858 
9859 	lck_spin_lock_grp(&vm_objects_wired_lock, &vm_page_lck_grp_bucket);
9860 	queue_iterate(&vm_objects_wired,
9861 	    object,
9862 	    vm_object_t,
9863 	    wired_objq)
9864 	{
9865 		proc(info, num_info, object);
9866 	}
9867 	lck_spin_unlock(&vm_objects_wired_lock);
9868 }
9869 #endif /* ! VM_TAG_ACTIVE_UPDATE */
9870 
9871 static uint64_t
process_account(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes,boolean_t iterated,bool redact_info __unused)9872 process_account(mach_memory_info_t * info, unsigned int num_info,
9873     uint64_t zones_collectable_bytes, boolean_t iterated, bool redact_info __unused)
9874 {
9875 	size_t                 namelen;
9876 	unsigned int           idx, count, nextinfo;
9877 	vm_allocation_site_t * site;
9878 	lck_ticket_lock(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
9879 
9880 	for (idx = 0; idx <= vm_allocation_tag_highest; idx++) {
9881 		site = vm_allocation_sites[idx];
9882 		if (!site) {
9883 			continue;
9884 		}
9885 		info[idx].mapped = site->mapped;
9886 		info[idx].tag    = site->tag;
9887 		if (!iterated) {
9888 			info[idx].size = site->total;
9889 #if DEBUG || DEVELOPMENT
9890 			info[idx].peak = site->peak;
9891 #endif /* DEBUG || DEVELOPMENT */
9892 		} else {
9893 			if (!site->subtotalscount && (site->total != info[idx].size)) {
9894 				printf("tag mismatch[%d] 0x%qx, iter 0x%qx\n", idx, site->total, info[idx].size);
9895 				info[idx].size = site->total;
9896 			}
9897 		}
9898 		info[idx].flags |= VM_KERN_SITE_WIRED;
9899 		if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC) {
9900 			info[idx].site   = idx;
9901 			info[idx].flags |= VM_KERN_SITE_TAG;
9902 			if (VM_KERN_MEMORY_ZONE == idx) {
9903 				info[idx].flags |= VM_KERN_SITE_HIDE;
9904 				info[idx].flags &= ~VM_KERN_SITE_WIRED;
9905 				info[idx].collectable_bytes = zones_collectable_bytes;
9906 			}
9907 		} else if ((namelen = (VM_TAG_NAME_LEN_MAX & (site->flags >> VM_TAG_NAME_LEN_SHIFT)))) {
9908 			info[idx].site   = 0;
9909 			info[idx].flags |= VM_KERN_SITE_NAMED;
9910 			if (namelen > sizeof(info[idx].name)) {
9911 				namelen = sizeof(info[idx].name);
9912 			}
9913 			strncpy(&info[idx].name[0], KA_NAME(site), namelen);
9914 		} else if (VM_TAG_KMOD & site->flags) {
9915 			info[idx].site   = OSKextGetKmodIDForSite(site, NULL, 0);
9916 			info[idx].flags |= VM_KERN_SITE_KMOD;
9917 		} else {
9918 			info[idx].site   = VM_KERNEL_UNSLIDE(site);
9919 			info[idx].flags |= VM_KERN_SITE_KERNEL;
9920 		}
9921 	}
9922 
9923 	nextinfo = (vm_allocation_tag_highest + 1);
9924 	count    = nextinfo;
9925 	if (count >= num_info) {
9926 		count = num_info;
9927 	}
9928 
9929 	for (idx = 0; idx < count; idx++) {
9930 		site = vm_allocation_sites[idx];
9931 		if (!site) {
9932 			continue;
9933 		}
9934 #if VM_TAG_SIZECLASSES
9935 		vm_allocation_zone_total_t * zone;
9936 		unsigned int                 zidx;
9937 
9938 		if (!redact_info
9939 		    && vm_allocation_zone_totals
9940 		    && (zone = vm_allocation_zone_totals[idx])
9941 		    && (nextinfo < num_info)) {
9942 			for (zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
9943 				if (!zone[zidx].vazt_peak) {
9944 					continue;
9945 				}
9946 				info[nextinfo]        = info[idx];
9947 				info[nextinfo].zone   = zone_index_from_tag_index(zidx);
9948 				info[nextinfo].flags  &= ~VM_KERN_SITE_WIRED;
9949 				info[nextinfo].flags  |= VM_KERN_SITE_ZONE;
9950 				info[nextinfo].flags  |= VM_KERN_SITE_KALLOC;
9951 				info[nextinfo].size   = zone[zidx].vazt_total;
9952 				info[nextinfo].peak   = zone[zidx].vazt_peak;
9953 				info[nextinfo].mapped = 0;
9954 				nextinfo++;
9955 			}
9956 		}
9957 #endif /* VM_TAG_SIZECLASSES */
9958 		if (site->subtotalscount) {
9959 			uint64_t mapped, mapcost, take;
9960 			uint32_t sub;
9961 			vm_tag_t alloctag;
9962 
9963 			info[idx].size = site->total;
9964 			mapped = info[idx].size;
9965 			info[idx].mapped = mapped;
9966 			mapcost = 0;
9967 			for (sub = 0; sub < site->subtotalscount; sub++) {
9968 				alloctag = site->subtotals[sub].tag;
9969 				assert(alloctag < num_info);
9970 				if (info[alloctag].name[0]) {
9971 					continue;
9972 				}
9973 				take = site->subtotals[sub].total;
9974 				if (take > info[alloctag].size) {
9975 					take = info[alloctag].size;
9976 				}
9977 				if (take > mapped) {
9978 					take = mapped;
9979 				}
9980 				info[alloctag].mapped  -= take;
9981 				info[alloctag].size    -= take;
9982 				mapped                 -= take;
9983 				mapcost                += take;
9984 			}
9985 			info[idx].size = mapcost;
9986 		}
9987 	}
9988 	lck_ticket_unlock(&vm_allocation_sites_lock);
9989 
9990 	return 0;
9991 }
9992 
9993 uint32_t
vm_page_diagnose_estimate(void)9994 vm_page_diagnose_estimate(void)
9995 {
9996 	vm_allocation_site_t * site;
9997 	uint32_t               count = zone_view_count;
9998 	uint32_t               idx;
9999 
10000 	lck_ticket_lock(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
10001 	for (idx = 0; idx < VM_MAX_TAG_VALUE; idx++) {
10002 		site = vm_allocation_sites[idx];
10003 		if (!site) {
10004 			continue;
10005 		}
10006 		count++;
10007 #if VM_TAG_SIZECLASSES
10008 		if (vm_allocation_zone_totals) {
10009 			vm_allocation_zone_total_t * zone;
10010 			zone = vm_allocation_zone_totals[idx];
10011 			if (!zone) {
10012 				continue;
10013 			}
10014 			for (uint32_t zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
10015 				count += (zone[zidx].vazt_peak != 0);
10016 			}
10017 		}
10018 #endif
10019 	}
10020 	lck_ticket_unlock(&vm_allocation_sites_lock);
10021 
10022 	/* some slop for new tags created */
10023 	count += 8;
10024 	count += VM_KERN_COUNTER_COUNT;
10025 
10026 	return count;
10027 }
10028 
10029 static void
vm_page_diagnose_zone_stats(mach_memory_info_t * info,zone_stats_t zstats,bool percpu)10030 vm_page_diagnose_zone_stats(mach_memory_info_t *info, zone_stats_t zstats,
10031     bool percpu)
10032 {
10033 	zpercpu_foreach(zs, zstats) {
10034 		info->size += zs->zs_mem_allocated - zs->zs_mem_freed;
10035 	}
10036 	if (percpu) {
10037 		info->size *= zpercpu_count();
10038 	}
10039 	info->flags |= VM_KERN_SITE_NAMED | VM_KERN_SITE_ZONE_VIEW;
10040 }
10041 
10042 static void
vm_page_add_info(mach_memory_info_t * info,zone_stats_t stats,bool per_cpu,const char * parent_heap_name,const char * parent_zone_name,const char * view_name)10043 vm_page_add_info(
10044 	mach_memory_info_t     *info,
10045 	zone_stats_t            stats,
10046 	bool                    per_cpu,
10047 	const char             *parent_heap_name,
10048 	const char             *parent_zone_name,
10049 	const char             *view_name)
10050 {
10051 	vm_page_diagnose_zone_stats(info, stats, per_cpu);
10052 	snprintf(info->name, sizeof(info->name),
10053 	    "%s%s[%s]", parent_heap_name, parent_zone_name, view_name);
10054 }
10055 
10056 static void
vm_page_diagnose_zone(mach_memory_info_t * info,zone_t z)10057 vm_page_diagnose_zone(mach_memory_info_t *info, zone_t z)
10058 {
10059 	vm_page_add_info(info, z->z_stats, z->z_percpu, zone_heap_name(z),
10060 	    z->z_name, "raw");
10061 }
10062 
10063 static void
vm_page_add_view(mach_memory_info_t * info,zone_stats_t stats,const char * parent_heap_name,const char * parent_zone_name,const char * view_name)10064 vm_page_add_view(
10065 	mach_memory_info_t     *info,
10066 	zone_stats_t            stats,
10067 	const char             *parent_heap_name,
10068 	const char             *parent_zone_name,
10069 	const char             *view_name)
10070 {
10071 	vm_page_add_info(info, stats, false, parent_heap_name, parent_zone_name,
10072 	    view_name);
10073 }
10074 
10075 static uint32_t
vm_page_diagnose_heap_views(mach_memory_info_t * info,kalloc_heap_t kh,const char * parent_heap_name,const char * parent_zone_name)10076 vm_page_diagnose_heap_views(
10077 	mach_memory_info_t     *info,
10078 	kalloc_heap_t           kh,
10079 	const char             *parent_heap_name,
10080 	const char             *parent_zone_name)
10081 {
10082 	uint32_t i = 0;
10083 
10084 	while (kh) {
10085 		vm_page_add_view(info + i, kh->kh_stats, parent_heap_name,
10086 		    parent_zone_name, kh->kh_name);
10087 		kh = kh->kh_views;
10088 		i++;
10089 	}
10090 	return i;
10091 }
10092 
10093 static uint32_t
vm_page_diagnose_heap(mach_memory_info_t * info,kalloc_heap_t kheap)10094 vm_page_diagnose_heap(mach_memory_info_t *info, kalloc_heap_t kheap)
10095 {
10096 	uint32_t i = 0;
10097 
10098 	for (; i < KHEAP_NUM_ZONES; i++) {
10099 		vm_page_diagnose_zone(info + i, zone_by_id(kheap->kh_zstart + i));
10100 	}
10101 
10102 	i += vm_page_diagnose_heap_views(info + i, kheap->kh_views, kheap->kh_name,
10103 	    NULL);
10104 	return i;
10105 }
10106 
10107 static int
vm_page_diagnose_kt_heaps(mach_memory_info_t * info)10108 vm_page_diagnose_kt_heaps(mach_memory_info_t *info)
10109 {
10110 	uint32_t idx = 0;
10111 	vm_page_add_view(info + idx, KHEAP_KT_VAR->kh_stats, KHEAP_KT_VAR->kh_name,
10112 	    "", "raw");
10113 	idx++;
10114 
10115 	for (uint32_t i = 0; i < KT_VAR_MAX_HEAPS; i++) {
10116 		struct kheap_info heap = kalloc_type_heap_array[i];
10117 		char heap_num_tmp[MAX_ZONE_NAME] = "";
10118 		const char *heap_num;
10119 
10120 		snprintf(&heap_num_tmp[0], MAX_ZONE_NAME, "%u", i);
10121 		heap_num = &heap_num_tmp[0];
10122 
10123 		for (kalloc_type_var_view_t ktv = heap.kt_views; ktv;
10124 		    ktv = (kalloc_type_var_view_t) ktv->kt_next) {
10125 			if (ktv->kt_stats && ktv->kt_stats != KHEAP_KT_VAR->kh_stats) {
10126 				vm_page_add_view(info + idx, ktv->kt_stats, KHEAP_KT_VAR->kh_name,
10127 				    heap_num, ktv->kt_name);
10128 				idx++;
10129 			}
10130 		}
10131 
10132 		idx += vm_page_diagnose_heap_views(info + idx, heap.kh_views,
10133 		    KHEAP_KT_VAR->kh_name, heap_num);
10134 	}
10135 
10136 	return idx;
10137 }
10138 
10139 kern_return_t
vm_page_diagnose(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes,bool redact_info)10140 vm_page_diagnose(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes, bool redact_info)
10141 {
10142 	uint64_t                 wired_size;
10143 	uint64_t                 wired_managed_size;
10144 	uint64_t                 wired_reserved_size;
10145 	boolean_t                iterate;
10146 	mach_memory_info_t     * counts;
10147 	uint32_t                 i;
10148 
10149 	bzero(info, num_info * sizeof(mach_memory_info_t));
10150 
10151 	if (!vm_page_wire_count_initial) {
10152 		return KERN_ABORTED;
10153 	}
10154 
10155 #if !XNU_TARGET_OS_OSX
10156 	wired_size          = ptoa_64(vm_page_wire_count);
10157 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count);
10158 #else /* !XNU_TARGET_OS_OSX */
10159 	wired_size          = ptoa_64(vm_page_wire_count + vm_lopage_free_count + vm_page_throttled_count);
10160 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count + vm_page_throttled_count);
10161 #endif /* !XNU_TARGET_OS_OSX */
10162 	wired_managed_size  = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial);
10163 
10164 	wired_size += booter_size;
10165 
10166 	assert(num_info >= VM_KERN_COUNTER_COUNT);
10167 	num_info -= VM_KERN_COUNTER_COUNT;
10168 	counts = &info[num_info];
10169 
10170 #define SET_COUNT(xcount, xsize, xflags)                        \
10171     counts[xcount].tag   = VM_MAX_TAG_VALUE + xcount;   \
10172     counts[xcount].site  = (xcount);                            \
10173     counts[xcount].size  = (xsize);                                 \
10174     counts[xcount].mapped  = (xsize);                           \
10175     counts[xcount].flags = VM_KERN_SITE_COUNTER | xflags;
10176 
10177 	SET_COUNT(VM_KERN_COUNT_MANAGED, ptoa_64(vm_page_pages), 0);
10178 	SET_COUNT(VM_KERN_COUNT_WIRED, wired_size, 0);
10179 	SET_COUNT(VM_KERN_COUNT_WIRED_MANAGED, wired_managed_size, 0);
10180 	SET_COUNT(VM_KERN_COUNT_RESERVED, wired_reserved_size, VM_KERN_SITE_WIRED);
10181 	SET_COUNT(VM_KERN_COUNT_STOLEN, ptoa_64(vm_page_stolen_count), VM_KERN_SITE_WIRED);
10182 	SET_COUNT(VM_KERN_COUNT_LOPAGE, ptoa_64(vm_lopage_free_count), VM_KERN_SITE_WIRED);
10183 	SET_COUNT(VM_KERN_COUNT_WIRED_BOOT, ptoa_64(vm_page_wire_count_on_boot), 0);
10184 	SET_COUNT(VM_KERN_COUNT_BOOT_STOLEN, booter_size, VM_KERN_SITE_WIRED);
10185 	SET_COUNT(VM_KERN_COUNT_WIRED_STATIC_KERNELCACHE, ptoa_64(vm_page_kernelcache_count), 0);
10186 #if CONFIG_SPTM
10187 	SET_COUNT(VM_KERN_COUNT_EXCLAVES_CARVEOUT, SPTMArgs->sk_carveout_size, 0);
10188 #endif
10189 
10190 #define SET_MAP(xcount, xsize, xfree, xlargest) \
10191     counts[xcount].site    = (xcount);                  \
10192     counts[xcount].size    = (xsize);                   \
10193     counts[xcount].mapped  = (xsize);                   \
10194     counts[xcount].free    = (xfree);                   \
10195     counts[xcount].largest = (xlargest);                \
10196     counts[xcount].flags   = VM_KERN_SITE_COUNTER;
10197 
10198 	vm_map_size_t map_size, map_free, map_largest;
10199 
10200 	vm_map_sizes(kernel_map, &map_size, &map_free, &map_largest);
10201 	SET_MAP(VM_KERN_COUNT_MAP_KERNEL, map_size, map_free, map_largest);
10202 
10203 	zone_map_sizes(&map_size, &map_free, &map_largest);
10204 	SET_MAP(VM_KERN_COUNT_MAP_ZONE, map_size, map_free, map_largest);
10205 
10206 	assert(num_info >= zone_view_count);
10207 	num_info -= zone_view_count;
10208 	counts = &info[num_info];
10209 	i = 0;
10210 
10211 	if (!redact_info) {
10212 		if (KHEAP_DATA_BUFFERS->kh_heap_id == KHEAP_ID_DATA_BUFFERS) {
10213 			i += vm_page_diagnose_heap(counts + i, KHEAP_DATA_BUFFERS);
10214 		}
10215 		if (KHEAP_KT_VAR->kh_heap_id == KHEAP_ID_KT_VAR) {
10216 			i += vm_page_diagnose_kt_heaps(counts + i);
10217 		}
10218 		assert(i <= zone_view_count);
10219 
10220 		zone_index_foreach(zidx) {
10221 			zone_t z = &zone_array[zidx];
10222 			zone_security_flags_t zsflags = zone_security_array[zidx];
10223 			zone_view_t zv = z->z_views;
10224 
10225 			if (zv == NULL) {
10226 				continue;
10227 			}
10228 
10229 			zone_stats_t zv_stats_head = z->z_stats;
10230 			bool has_raw_view = false;
10231 
10232 			for (; zv; zv = zv->zv_next) {
10233 				/*
10234 				 * kalloc_types that allocate from the same zone are linked
10235 				 * as views. Only print the ones that have their own stats.
10236 				 */
10237 				if (zv->zv_stats == zv_stats_head) {
10238 					continue;
10239 				}
10240 				has_raw_view = true;
10241 				vm_page_diagnose_zone_stats(counts + i, zv->zv_stats,
10242 				    z->z_percpu);
10243 				snprintf(counts[i].name, sizeof(counts[i].name), "%s%s[%s]",
10244 				    zone_heap_name(z), z->z_name, zv->zv_name);
10245 				i++;
10246 				assert(i <= zone_view_count);
10247 			}
10248 
10249 			/*
10250 			 * Print raw views for non kalloc or kalloc_type zones
10251 			 */
10252 			bool kalloc_type = zsflags.z_kalloc_type;
10253 			if ((zsflags.z_kheap_id == KHEAP_ID_NONE && !kalloc_type) ||
10254 			    (kalloc_type && has_raw_view)) {
10255 				vm_page_diagnose_zone(counts + i, z);
10256 				i++;
10257 				assert(i <= zone_view_count);
10258 			}
10259 		}
10260 	}
10261 
10262 	iterate = !VM_TAG_ACTIVE_UPDATE;
10263 	if (iterate) {
10264 		enum                       { kMaxKernelDepth = 1 };
10265 		vm_map_t                     maps[kMaxKernelDepth];
10266 		vm_map_entry_t               entries[kMaxKernelDepth];
10267 		vm_map_t                     map;
10268 		vm_map_entry_t               entry;
10269 		vm_object_offset_t           offset;
10270 		vm_page_t                    page;
10271 		int                          stackIdx, count;
10272 
10273 #if !VM_TAG_ACTIVE_UPDATE
10274 		vm_page_iterate_objects(info, num_info, &vm_page_count_object);
10275 #endif /* ! VM_TAG_ACTIVE_UPDATE */
10276 
10277 		map = kernel_map;
10278 		stackIdx = 0;
10279 		while (map) {
10280 			vm_map_lock(map);
10281 			for (entry = map->hdr.links.next; map; entry = entry->vme_next) {
10282 				if (entry->is_sub_map) {
10283 					assert(stackIdx < kMaxKernelDepth);
10284 					maps[stackIdx] = map;
10285 					entries[stackIdx] = entry;
10286 					stackIdx++;
10287 					map = VME_SUBMAP(entry);
10288 					entry = NULL;
10289 					break;
10290 				}
10291 				if (is_kernel_object(VME_OBJECT(entry))) {
10292 					count = 0;
10293 					vm_object_lock(VME_OBJECT(entry));
10294 					for (offset = entry->vme_start; offset < entry->vme_end; offset += page_size) {
10295 						page = vm_page_lookup(VME_OBJECT(entry), offset);
10296 						if (page && VM_PAGE_WIRED(page)) {
10297 							count++;
10298 						}
10299 					}
10300 					vm_object_unlock(VME_OBJECT(entry));
10301 
10302 					if (count) {
10303 						assert(VME_ALIAS(entry) != VM_KERN_MEMORY_NONE);
10304 						assert(VME_ALIAS(entry) < num_info);
10305 						info[VME_ALIAS(entry)].size += ptoa_64(count);
10306 					}
10307 				}
10308 				while (map && (entry == vm_map_last_entry(map))) {
10309 					vm_map_unlock(map);
10310 					if (!stackIdx) {
10311 						map = NULL;
10312 					} else {
10313 						--stackIdx;
10314 						map = maps[stackIdx];
10315 						entry = entries[stackIdx];
10316 					}
10317 				}
10318 			}
10319 		}
10320 	}
10321 
10322 	process_account(info, num_info, zones_collectable_bytes, iterate, redact_info);
10323 
10324 	return KERN_SUCCESS;
10325 }
10326 
10327 #if DEBUG || DEVELOPMENT
10328 
10329 kern_return_t
vm_kern_allocation_info(uintptr_t addr,vm_size_t * size,vm_tag_t * tag,vm_size_t * zone_size)10330 vm_kern_allocation_info(uintptr_t addr, vm_size_t * size, vm_tag_t * tag, vm_size_t * zone_size)
10331 {
10332 	kern_return_t  ret;
10333 	vm_size_t      zsize;
10334 	vm_map_t       map;
10335 	vm_map_entry_t entry;
10336 
10337 	zsize = zone_element_info((void *) addr, tag);
10338 	if (zsize) {
10339 		*zone_size = *size = zsize;
10340 		return KERN_SUCCESS;
10341 	}
10342 
10343 	*zone_size = 0;
10344 	ret = KERN_INVALID_ADDRESS;
10345 	for (map = kernel_map; map;) {
10346 		vm_map_lock(map);
10347 		if (!vm_map_lookup_entry_allow_pgz(map, addr, &entry)) {
10348 			break;
10349 		}
10350 		if (entry->is_sub_map) {
10351 			if (map != kernel_map) {
10352 				break;
10353 			}
10354 			map = VME_SUBMAP(entry);
10355 			continue;
10356 		}
10357 		if (entry->vme_start != addr) {
10358 			break;
10359 		}
10360 		*tag = (vm_tag_t)VME_ALIAS(entry);
10361 		*size = (entry->vme_end - addr);
10362 		ret = KERN_SUCCESS;
10363 		break;
10364 	}
10365 	if (map != kernel_map) {
10366 		vm_map_unlock(map);
10367 	}
10368 	vm_map_unlock(kernel_map);
10369 
10370 	return ret;
10371 }
10372 
10373 #endif /* DEBUG || DEVELOPMENT */
10374 
10375 uint32_t
vm_tag_get_kext(vm_tag_t tag,char * name,vm_size_t namelen)10376 vm_tag_get_kext(vm_tag_t tag, char * name, vm_size_t namelen)
10377 {
10378 	vm_allocation_site_t * site;
10379 	uint32_t               kmodId;
10380 
10381 	kmodId = 0;
10382 	lck_ticket_lock(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
10383 	if ((site = vm_allocation_sites[tag])) {
10384 		if (VM_TAG_KMOD & site->flags) {
10385 			kmodId = OSKextGetKmodIDForSite(site, name, namelen);
10386 		}
10387 	}
10388 	lck_ticket_unlock(&vm_allocation_sites_lock);
10389 
10390 	return kmodId;
10391 }
10392 
10393 
10394 #if CONFIG_SECLUDED_MEMORY
10395 /*
10396  * Note that there's no locking around other accesses to vm_page_secluded_target.
10397  * That should be OK, since these are the only place where it can be changed after
10398  * initialization. Other users (like vm_pageout) may see the wrong value briefly,
10399  * but will eventually get the correct value. This brief mismatch is OK as pageout
10400  * and page freeing will auto-adjust the vm_page_secluded_count to match the target
10401  * over time.
10402  */
10403 unsigned int vm_page_secluded_suppress_cnt = 0;
10404 unsigned int vm_page_secluded_save_target;
10405 
10406 LCK_GRP_DECLARE(secluded_suppress_slock_grp, "secluded_suppress_slock");
10407 LCK_SPIN_DECLARE(secluded_suppress_slock, &secluded_suppress_slock_grp);
10408 
10409 void
start_secluded_suppression(task_t task)10410 start_secluded_suppression(task_t task)
10411 {
10412 	if (task->task_suppressed_secluded) {
10413 		return;
10414 	}
10415 	lck_spin_lock(&secluded_suppress_slock);
10416 	if (!task->task_suppressed_secluded && vm_page_secluded_suppress_cnt++ == 0) {
10417 		task->task_suppressed_secluded = TRUE;
10418 		vm_page_secluded_save_target = vm_page_secluded_target;
10419 		vm_page_secluded_target = 0;
10420 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
10421 	}
10422 	lck_spin_unlock(&secluded_suppress_slock);
10423 }
10424 
10425 void
stop_secluded_suppression(task_t task)10426 stop_secluded_suppression(task_t task)
10427 {
10428 	lck_spin_lock(&secluded_suppress_slock);
10429 	if (task->task_suppressed_secluded && --vm_page_secluded_suppress_cnt == 0) {
10430 		task->task_suppressed_secluded = FALSE;
10431 		vm_page_secluded_target = vm_page_secluded_save_target;
10432 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
10433 	}
10434 	lck_spin_unlock(&secluded_suppress_slock);
10435 }
10436 
10437 #endif /* CONFIG_SECLUDED_MEMORY */
10438 
10439 /*
10440  * Move the list of retired pages on the vm_page_queue_retired to
10441  * their final resting place on retired_pages_object.
10442  */
10443 void
vm_retire_boot_pages(void)10444 vm_retire_boot_pages(void)
10445 {
10446 }
10447 
10448 /*
10449  * This holds the reported physical address if an ECC error leads to a panic.
10450  * SMC will store it in PMU SRAM under the 'sECC' key.
10451  */
10452 uint64_t ecc_panic_physical_address = 0;
10453 
10454 
10455 boolean_t
vm_page_created(vm_page_t page)10456 vm_page_created(vm_page_t page)
10457 {
10458 	return (page < &vm_pages[0]) || (page >= &vm_pages[vm_pages_count]);
10459 }
10460