xref: /xnu-8792.41.9/osfmk/vm/vm_resident.c (revision 5c2921b07a2480ab43ec66f5b9e41cb872bc554f)
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.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_pageout.h>
88 #include <vm/vm_kern.h>                 /* kmem_alloc() */
89 #include <kern/misc_protos.h>
90 #include <mach_debug/zone_info.h>
91 #include <vm/cpm.h>
92 #include <pexpert/pexpert.h>
93 #include <pexpert/device_tree.h>
94 #include <san/kasan.h>
95 
96 #include <vm/vm_protos.h>
97 #include <vm/memory_object.h>
98 #include <vm/vm_purgeable_internal.h>
99 #include <vm/vm_compressor.h>
100 #if defined (__x86_64__)
101 #include <i386/misc_protos.h>
102 #endif
103 
104 #if CONFIG_PHANTOM_CACHE
105 #include <vm/vm_phantom_cache.h>
106 #endif
107 
108 #if HIBERNATION
109 #include <IOKit/IOHibernatePrivate.h>
110 #include <machine/pal_hibernate.h>
111 #endif /* HIBERNATION */
112 
113 #include <sys/kdebug.h>
114 
115 #if defined(HAS_APPLE_PAC)
116 #include <ptrauth.h>
117 #endif
118 #if defined(__arm64__)
119 #include <arm/cpu_internal.h>
120 #endif /* defined(__arm64__) */
121 
122 #if MACH_ASSERT
123 
124 #define ASSERT_PMAP_FREE(mem) pmap_assert_free(VM_PAGE_GET_PHYS_PAGE(mem))
125 
126 #else /* MACH_ASSERT */
127 
128 #define ASSERT_PMAP_FREE(mem) /* nothing */
129 
130 #endif /* MACH_ASSERT */
131 
132 extern boolean_t vm_pageout_running;
133 extern thread_t  vm_pageout_scan_thread;
134 extern boolean_t vps_dynamic_priority_enabled;
135 
136 char    vm_page_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
137 char    vm_page_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
138 char    vm_page_non_speculative_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
139 char    vm_page_active_or_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE];
140 
141 #if CONFIG_SECLUDED_MEMORY
142 struct vm_page_secluded_data vm_page_secluded;
143 #endif /* CONFIG_SECLUDED_MEMORY */
144 
145 #if DEVELOPMENT || DEBUG
146 extern struct memory_object_pager_ops shared_region_pager_ops;
147 unsigned int shared_region_pagers_resident_count = 0;
148 unsigned int shared_region_pagers_resident_peak = 0;
149 #endif /* DEVELOPMENT || DEBUG */
150 
151 
152 
153 int             PERCPU_DATA(start_color);
154 vm_page_t       PERCPU_DATA(free_pages);
155 boolean_t       hibernate_cleaning_in_progress = FALSE;
156 boolean_t       vm_page_free_verify = TRUE;
157 
158 uint32_t        vm_lopage_free_count = 0;
159 uint32_t        vm_lopage_free_limit = 0;
160 uint32_t        vm_lopage_lowater    = 0;
161 boolean_t       vm_lopage_refill = FALSE;
162 boolean_t       vm_lopage_needed = FALSE;
163 
164 int             speculative_age_index = 0;
165 int             speculative_steal_index = 0;
166 struct vm_speculative_age_q vm_page_queue_speculative[VM_PAGE_MAX_SPECULATIVE_AGE_Q + 1];
167 
168 boolean_t       hibernation_vmqueues_inspection = FALSE; /* Tracks if the hibernation code is looking at the VM queues.
169                                                           * Updated and checked behind the vm_page_queues_lock. */
170 
171 static void             vm_page_free_prepare(vm_page_t  page);
172 static vm_page_t        vm_page_grab_fictitious_common(ppnum_t, boolean_t);
173 
174 static void vm_tag_init(void);
175 
176 /* for debugging purposes */
177 SECURITY_READ_ONLY_EARLY(uint32_t) vm_packed_from_vm_pages_array_mask =
178     VM_PAGE_PACKED_FROM_ARRAY;
179 SECURITY_READ_ONLY_EARLY(vm_packing_params_t) vm_page_packing_params =
180     VM_PACKING_PARAMS(VM_PAGE_PACKED_PTR);
181 
182 /*
183  *	Associated with page of user-allocatable memory is a
184  *	page structure.
185  */
186 
187 /*
188  *	These variables record the values returned by vm_page_bootstrap,
189  *	for debugging purposes.  The implementation of pmap_steal_memory
190  *	and pmap_startup here also uses them internally.
191  */
192 
193 vm_offset_t virtual_space_start;
194 vm_offset_t virtual_space_end;
195 uint32_t        vm_page_pages;
196 
197 /*
198  *	The vm_page_lookup() routine, which provides for fast
199  *	(virtual memory object, offset) to page lookup, employs
200  *	the following hash table.  The vm_page_{insert,remove}
201  *	routines install and remove associations in the table.
202  *	[This table is often called the virtual-to-physical,
203  *	or VP, table.]
204  */
205 typedef struct {
206 	vm_page_packed_t page_list;
207 #if     MACH_PAGE_HASH_STATS
208 	int             cur_count;              /* current count */
209 	int             hi_count;               /* high water mark */
210 #endif /* MACH_PAGE_HASH_STATS */
211 } vm_page_bucket_t;
212 
213 
214 #define BUCKETS_PER_LOCK        16
215 
216 SECURITY_READ_ONLY_LATE(vm_page_bucket_t *) vm_page_buckets;                /* Array of buckets */
217 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_bucket_count = 0;       /* How big is array? */
218 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_hash_mask;              /* Mask for hash function */
219 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_hash_shift;             /* Shift for hash function */
220 SECURITY_READ_ONLY_LATE(uint32_t)           vm_page_bucket_hash;            /* Basic bucket hash */
221 SECURITY_READ_ONLY_LATE(unsigned int)       vm_page_bucket_lock_count = 0;  /* How big is array of locks? */
222 
223 #ifndef VM_TAG_ACTIVE_UPDATE
224 #error VM_TAG_ACTIVE_UPDATE
225 #endif
226 #ifndef VM_TAG_SIZECLASSES
227 #error VM_TAG_SIZECLASSES
228 #endif
229 
230 /* for debugging */
231 SECURITY_READ_ONLY_LATE(bool) vm_tag_active_update = VM_TAG_ACTIVE_UPDATE;
232 SECURITY_READ_ONLY_LATE(lck_spin_t *) vm_page_bucket_locks;
233 
234 vm_allocation_site_t            vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC + 1];
235 vm_allocation_site_t *          vm_allocation_sites[VM_MAX_TAG_VALUE];
236 #if VM_TAG_SIZECLASSES
237 static vm_allocation_zone_total_t **vm_allocation_zone_totals;
238 #endif /* VM_TAG_SIZECLASSES */
239 
240 vm_tag_t vm_allocation_tag_highest;
241 
242 #if VM_PAGE_BUCKETS_CHECK
243 boolean_t vm_page_buckets_check_ready = FALSE;
244 #if VM_PAGE_FAKE_BUCKETS
245 vm_page_bucket_t *vm_page_fake_buckets; /* decoy buckets */
246 vm_map_offset_t vm_page_fake_buckets_start, vm_page_fake_buckets_end;
247 #endif /* VM_PAGE_FAKE_BUCKETS */
248 #endif /* VM_PAGE_BUCKETS_CHECK */
249 
250 #if     MACH_PAGE_HASH_STATS
251 /* This routine is only for debug.  It is intended to be called by
252  * hand by a developer using a kernel debugger.  This routine prints
253  * out vm_page_hash table statistics to the kernel debug console.
254  */
255 void
hash_debug(void)256 hash_debug(void)
257 {
258 	int     i;
259 	int     numbuckets = 0;
260 	int     highsum = 0;
261 	int     maxdepth = 0;
262 
263 	for (i = 0; i < vm_page_bucket_count; i++) {
264 		if (vm_page_buckets[i].hi_count) {
265 			numbuckets++;
266 			highsum += vm_page_buckets[i].hi_count;
267 			if (vm_page_buckets[i].hi_count > maxdepth) {
268 				maxdepth = vm_page_buckets[i].hi_count;
269 			}
270 		}
271 	}
272 	printf("Total number of buckets: %d\n", vm_page_bucket_count);
273 	printf("Number used buckets:     %d = %d%%\n",
274 	    numbuckets, 100 * numbuckets / vm_page_bucket_count);
275 	printf("Number unused buckets:   %d = %d%%\n",
276 	    vm_page_bucket_count - numbuckets,
277 	    100 * (vm_page_bucket_count - numbuckets) / vm_page_bucket_count);
278 	printf("Sum of bucket max depth: %d\n", highsum);
279 	printf("Average bucket depth:    %d.%2d\n",
280 	    highsum / vm_page_bucket_count,
281 	    highsum % vm_page_bucket_count);
282 	printf("Maximum bucket depth:    %d\n", maxdepth);
283 }
284 #endif /* MACH_PAGE_HASH_STATS */
285 
286 /*
287  *	The virtual page size is currently implemented as a runtime
288  *	variable, but is constant once initialized using vm_set_page_size.
289  *	This initialization must be done in the machine-dependent
290  *	bootstrap sequence, before calling other machine-independent
291  *	initializations.
292  *
293  *	All references to the virtual page size outside this
294  *	module must use the PAGE_SIZE, PAGE_MASK and PAGE_SHIFT
295  *	constants.
296  */
297 #if defined(__arm64__)
298 vm_size_t       page_size;
299 vm_size_t       page_mask;
300 int             page_shift;
301 #else
302 vm_size_t       page_size  = PAGE_SIZE;
303 vm_size_t       page_mask  = PAGE_MASK;
304 int             page_shift = PAGE_SHIFT;
305 #endif
306 
307 SECURITY_READ_ONLY_LATE(vm_page_t) vm_pages = VM_PAGE_NULL;
308 SECURITY_READ_ONLY_LATE(vm_page_t) vm_page_array_beginning_addr;
309 vm_page_t                          vm_page_array_ending_addr;
310 
311 unsigned int    vm_pages_count = 0;
312 
313 /*
314  *	Resident pages that represent real memory
315  *	are allocated from a set of free lists,
316  *	one per color.
317  */
318 unsigned int    vm_colors;
319 unsigned int    vm_color_mask;                  /* mask is == (vm_colors-1) */
320 unsigned int    vm_cache_geometry_colors = 0;   /* set by hw dependent code during startup */
321 unsigned int    vm_free_magazine_refill_limit = 0;
322 
323 
324 struct vm_page_queue_free_head {
325 	vm_page_queue_head_t    qhead;
326 } VM_PAGE_PACKED_ALIGNED;
327 
328 struct vm_page_queue_free_head  vm_page_queue_free[MAX_COLORS];
329 
330 
331 unsigned int    vm_page_free_wanted;
332 unsigned int    vm_page_free_wanted_privileged;
333 #if CONFIG_SECLUDED_MEMORY
334 unsigned int    vm_page_free_wanted_secluded;
335 #endif /* CONFIG_SECLUDED_MEMORY */
336 unsigned int    vm_page_free_count;
337 
338 /*
339  *	Occasionally, the virtual memory system uses
340  *	resident page structures that do not refer to
341  *	real pages, for example to leave a page with
342  *	important state information in the VP table.
343  *
344  *	These page structures are allocated the way
345  *	most other kernel structures are.
346  */
347 SECURITY_READ_ONLY_LATE(zone_t) vm_page_zone;
348 vm_locks_array_t vm_page_locks;
349 
350 LCK_ATTR_DECLARE(vm_page_lck_attr, 0, 0);
351 LCK_GRP_DECLARE(vm_page_lck_grp_free, "vm_page_free");
352 LCK_GRP_DECLARE(vm_page_lck_grp_queue, "vm_page_queue");
353 LCK_GRP_DECLARE(vm_page_lck_grp_local, "vm_page_queue_local");
354 LCK_GRP_DECLARE(vm_page_lck_grp_purge, "vm_page_purge");
355 LCK_GRP_DECLARE(vm_page_lck_grp_alloc, "vm_page_alloc");
356 LCK_GRP_DECLARE(vm_page_lck_grp_bucket, "vm_page_bucket");
357 LCK_SPIN_DECLARE_ATTR(vm_objects_wired_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr);
358 LCK_TICKET_DECLARE(vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
359 
360 unsigned int    vm_page_local_q_soft_limit = 250;
361 unsigned int    vm_page_local_q_hard_limit = 500;
362 struct vpl     *__zpercpu vm_page_local_q;
363 
364 /* N.B. Guard and fictitious pages must not
365  * be assigned a zero phys_page value.
366  */
367 /*
368  *	Fictitious pages don't have a physical address,
369  *	but we must initialize phys_page to something.
370  *	For debugging, this should be a strange value
371  *	that the pmap module can recognize in assertions.
372  */
373 const ppnum_t vm_page_fictitious_addr = (ppnum_t) -1;
374 
375 /*
376  *	Guard pages are not accessible so they don't
377  *      need a physical address, but we need to enter
378  *	one in the pmap.
379  *	Let's make it recognizable and make sure that
380  *	we don't use a real physical page with that
381  *	physical address.
382  */
383 const ppnum_t vm_page_guard_addr = (ppnum_t) -2;
384 
385 /*
386  *	Resident page structures are also chained on
387  *	queues that are used by the page replacement
388  *	system (pageout daemon).  These queues are
389  *	defined here, but are shared by the pageout
390  *	module.  The inactive queue is broken into
391  *	file backed and anonymous for convenience as the
392  *	pageout daemon often assignes a higher
393  *	importance to anonymous pages (less likely to pick)
394  */
395 vm_page_queue_head_t    vm_page_queue_active VM_PAGE_PACKED_ALIGNED;
396 vm_page_queue_head_t    vm_page_queue_inactive VM_PAGE_PACKED_ALIGNED;
397 #if CONFIG_SECLUDED_MEMORY
398 vm_page_queue_head_t    vm_page_queue_secluded VM_PAGE_PACKED_ALIGNED;
399 #endif /* CONFIG_SECLUDED_MEMORY */
400 vm_page_queue_head_t    vm_page_queue_anonymous VM_PAGE_PACKED_ALIGNED;  /* inactive memory queue for anonymous pages */
401 vm_page_queue_head_t    vm_page_queue_throttled VM_PAGE_PACKED_ALIGNED;
402 
403 queue_head_t    vm_objects_wired;
404 
405 void vm_update_darkwake_mode(boolean_t);
406 
407 vm_page_queue_head_t    vm_page_queue_donate VM_PAGE_PACKED_ALIGNED;
408 uint32_t        vm_page_donate_mode;
409 uint32_t        vm_page_donate_target, vm_page_donate_target_high, vm_page_donate_target_low;
410 uint32_t        vm_page_donate_count;
411 bool            vm_page_donate_queue_ripe;
412 
413 
414 vm_page_queue_head_t    vm_page_queue_background VM_PAGE_PACKED_ALIGNED;
415 uint32_t        vm_page_background_target;
416 uint32_t        vm_page_background_target_snapshot;
417 uint32_t        vm_page_background_count;
418 uint64_t        vm_page_background_promoted_count;
419 
420 uint32_t        vm_page_background_internal_count;
421 uint32_t        vm_page_background_external_count;
422 
423 uint32_t        vm_page_background_mode;
424 uint32_t        vm_page_background_exclude_external;
425 
426 unsigned int    vm_page_active_count;
427 unsigned int    vm_page_inactive_count;
428 unsigned int    vm_page_kernelcache_count;
429 #if CONFIG_SECLUDED_MEMORY
430 unsigned int    vm_page_secluded_count;
431 unsigned int    vm_page_secluded_count_free;
432 unsigned int    vm_page_secluded_count_inuse;
433 unsigned int    vm_page_secluded_count_over_target;
434 #endif /* CONFIG_SECLUDED_MEMORY */
435 unsigned int    vm_page_anonymous_count;
436 unsigned int    vm_page_throttled_count;
437 unsigned int    vm_page_speculative_count;
438 
439 unsigned int    vm_page_wire_count;
440 unsigned int    vm_page_wire_count_on_boot = 0;
441 unsigned int    vm_page_stolen_count = 0;
442 unsigned int    vm_page_wire_count_initial;
443 unsigned int    vm_page_gobble_count = 0;
444 unsigned int    vm_page_kern_lpage_count = 0;
445 
446 uint64_t        booter_size;  /* external so it can be found in core dumps */
447 
448 #define VM_PAGE_WIRE_COUNT_WARNING      0
449 #define VM_PAGE_GOBBLE_COUNT_WARNING    0
450 
451 unsigned int    vm_page_purgeable_count = 0; /* # of pages purgeable now */
452 unsigned int    vm_page_purgeable_wired_count = 0; /* # of purgeable pages that are wired now */
453 uint64_t        vm_page_purged_count = 0;    /* total count of purged pages */
454 
455 unsigned int    vm_page_xpmapped_external_count = 0;
456 unsigned int    vm_page_external_count = 0;
457 unsigned int    vm_page_internal_count = 0;
458 unsigned int    vm_page_pageable_external_count = 0;
459 unsigned int    vm_page_pageable_internal_count = 0;
460 
461 #if DEVELOPMENT || DEBUG
462 unsigned int    vm_page_speculative_recreated = 0;
463 unsigned int    vm_page_speculative_created = 0;
464 unsigned int    vm_page_speculative_used = 0;
465 #endif
466 
467 vm_page_queue_head_t    vm_page_queue_cleaned VM_PAGE_PACKED_ALIGNED;
468 
469 unsigned int    vm_page_cleaned_count = 0;
470 
471 uint64_t        max_valid_dma_address = 0xffffffffffffffffULL;
472 ppnum_t         max_valid_low_ppnum = PPNUM_MAX;
473 
474 
475 /*
476  *	Several page replacement parameters are also
477  *	shared with this module, so that page allocation
478  *	(done here in vm_page_alloc) can trigger the
479  *	pageout daemon.
480  */
481 unsigned int    vm_page_free_target = 0;
482 unsigned int    vm_page_free_min = 0;
483 unsigned int    vm_page_throttle_limit = 0;
484 unsigned int    vm_page_inactive_target = 0;
485 #if CONFIG_SECLUDED_MEMORY
486 unsigned int    vm_page_secluded_target = 0;
487 #endif /* CONFIG_SECLUDED_MEMORY */
488 unsigned int    vm_page_anonymous_min = 0;
489 unsigned int    vm_page_free_reserved = 0;
490 
491 
492 /*
493  *	The VM system has a couple of heuristics for deciding
494  *	that pages are "uninteresting" and should be placed
495  *	on the inactive queue as likely candidates for replacement.
496  *	These variables let the heuristics be controlled at run-time
497  *	to make experimentation easier.
498  */
499 
500 boolean_t vm_page_deactivate_hint = TRUE;
501 
502 struct vm_page_stats_reusable vm_page_stats_reusable;
503 
504 /*
505  *	vm_set_page_size:
506  *
507  *	Sets the page size, perhaps based upon the memory
508  *	size.  Must be called before any use of page-size
509  *	dependent functions.
510  *
511  *	Sets page_shift and page_mask from page_size.
512  */
513 void
vm_set_page_size(void)514 vm_set_page_size(void)
515 {
516 	page_size  = PAGE_SIZE;
517 	page_mask  = PAGE_MASK;
518 	page_shift = PAGE_SHIFT;
519 
520 	if ((page_mask & page_size) != 0) {
521 		panic("vm_set_page_size: page size not a power of two");
522 	}
523 
524 	for (page_shift = 0;; page_shift++) {
525 		if ((1U << page_shift) == page_size) {
526 			break;
527 		}
528 	}
529 }
530 
531 #if defined (__x86_64__)
532 
533 #define MAX_CLUMP_SIZE      16
534 #define DEFAULT_CLUMP_SIZE  4
535 
536 unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold;
537 
538 #if DEVELOPMENT || DEBUG
539 unsigned long vm_clump_stats[MAX_CLUMP_SIZE + 1];
540 unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes;
541 
542 static inline void
vm_clump_update_stats(unsigned int c)543 vm_clump_update_stats(unsigned int c)
544 {
545 	assert(c <= vm_clump_size);
546 	if (c > 0 && c <= vm_clump_size) {
547 		vm_clump_stats[c] += c;
548 	}
549 	vm_clump_allocs += c;
550 }
551 #endif  /*  if DEVELOPMENT || DEBUG */
552 
553 /* Called once to setup the VM clump knobs */
554 static void
vm_page_setup_clump(void)555 vm_page_setup_clump( void )
556 {
557 	unsigned int override, n;
558 
559 	vm_clump_size = DEFAULT_CLUMP_SIZE;
560 	if (PE_parse_boot_argn("clump_size", &override, sizeof(override))) {
561 		vm_clump_size = override;
562 	}
563 
564 	if (vm_clump_size > MAX_CLUMP_SIZE) {
565 		panic("vm_page_setup_clump:: clump_size is too large!");
566 	}
567 	if (vm_clump_size < 1) {
568 		panic("vm_page_setup_clump:: clump_size must be >= 1");
569 	}
570 	if ((vm_clump_size & (vm_clump_size - 1)) != 0) {
571 		panic("vm_page_setup_clump:: clump_size must be a power of 2");
572 	}
573 
574 	vm_clump_promote_threshold = vm_clump_size;
575 	vm_clump_mask = vm_clump_size - 1;
576 	for (vm_clump_shift = 0, n = vm_clump_size; n > 1; n >>= 1, vm_clump_shift++) {
577 		;
578 	}
579 
580 #if DEVELOPMENT || DEBUG
581 	bzero(vm_clump_stats, sizeof(vm_clump_stats));
582 	vm_clump_allocs = vm_clump_inserts = vm_clump_inrange = vm_clump_promotes = 0;
583 #endif  /*  if DEVELOPMENT || DEBUG */
584 }
585 
586 #endif  /* #if defined (__x86_64__) */
587 
588 #define COLOR_GROUPS_TO_STEAL   4
589 
590 /* Called once during statup, once the cache geometry is known.
591  */
592 static void
vm_page_set_colors(void)593 vm_page_set_colors( void )
594 {
595 	unsigned int    n, override;
596 
597 #if defined (__x86_64__)
598 	/* adjust #colors because we need to color outside the clump boundary */
599 	vm_cache_geometry_colors >>= vm_clump_shift;
600 #endif
601 	if (PE_parse_boot_argn("colors", &override, sizeof(override))) {                /* colors specified as a boot-arg? */
602 		n = override;
603 	} else if (vm_cache_geometry_colors) {                  /* do we know what the cache geometry is? */
604 		n = vm_cache_geometry_colors;
605 	} else {
606 		n = DEFAULT_COLORS;                             /* use default if all else fails */
607 	}
608 	if (n == 0) {
609 		n = 1;
610 	}
611 	if (n > MAX_COLORS) {
612 		n = MAX_COLORS;
613 	}
614 
615 	/* the count must be a power of 2  */
616 	if ((n & (n - 1)) != 0) {
617 		n = DEFAULT_COLORS;                             /* use default if all else fails */
618 	}
619 	vm_colors = n;
620 	vm_color_mask = n - 1;
621 
622 	vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL;
623 
624 #if defined (__x86_64__)
625 	/* adjust for reduction in colors due to clumping and multiple cores */
626 	if (real_ncpus) {
627 		vm_free_magazine_refill_limit *= (vm_clump_size * real_ncpus);
628 	}
629 #endif
630 }
631 
632 /*
633  * During single threaded early boot we don't initialize all pages.
634  * This avoids some delay during boot. They'll be initialized and
635  * added to the free list as needed or after we are multithreaded by
636  * what becomes the pageout thread.
637  */
638 static boolean_t fill = FALSE;
639 static unsigned int fillval;
640 uint_t vm_delayed_count = 0;    /* when non-zero, indicates we may have more pages to init */
641 ppnum_t delay_above_pnum = PPNUM_MAX;
642 
643 /*
644  * For x86 first 8 Gig initializes quickly and gives us lots of lowmem + mem above to start off with.
645  * If ARM ever uses delayed page initialization, this value may need to be quite different.
646  */
647 #define DEFAULT_DELAY_ABOVE_PHYS_GB (8)
648 
649 /*
650  * When we have to dip into more delayed pages due to low memory, free up
651  * a large chunk to get things back to normal. This avoids contention on the
652  * delayed code allocating page by page.
653  */
654 #define VM_DELAY_PAGE_CHUNK ((1024 * 1024 * 1024) / PAGE_SIZE)
655 
656 /*
657  * Get and initialize the next delayed page.
658  */
659 static vm_page_t
vm_get_delayed_page(int grab_options)660 vm_get_delayed_page(int grab_options)
661 {
662 	vm_page_t p;
663 	ppnum_t   pnum;
664 
665 	/*
666 	 * Get a new page if we have one.
667 	 */
668 	vm_free_page_lock();
669 	if (vm_delayed_count == 0) {
670 		vm_free_page_unlock();
671 		return NULL;
672 	}
673 
674 	if (!pmap_next_page(&pnum)) {
675 		vm_delayed_count = 0;
676 		vm_free_page_unlock();
677 		return NULL;
678 	}
679 
680 
681 	assert(vm_delayed_count > 0);
682 	--vm_delayed_count;
683 
684 #if defined(__x86_64__)
685 	/* x86 cluster code requires increasing phys_page in vm_pages[] */
686 	if (vm_pages_count > 0) {
687 		assert(pnum > vm_pages[vm_pages_count - 1].vmp_phys_page);
688 	}
689 #endif
690 	p = &vm_pages[vm_pages_count];
691 	assert(p < vm_page_array_ending_addr);
692 	vm_page_init(p, pnum, FALSE);
693 	++vm_pages_count;
694 	++vm_page_pages;
695 	vm_free_page_unlock();
696 
697 	/*
698 	 * These pages were initially counted as wired, undo that now.
699 	 */
700 	if (grab_options & VM_PAGE_GRAB_Q_LOCK_HELD) {
701 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
702 	} else {
703 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
704 		vm_page_lockspin_queues();
705 	}
706 	--vm_page_wire_count;
707 	--vm_page_wire_count_initial;
708 	if (vm_page_wire_count_on_boot != 0) {
709 		--vm_page_wire_count_on_boot;
710 	}
711 	if (!(grab_options & VM_PAGE_GRAB_Q_LOCK_HELD)) {
712 		vm_page_unlock_queues();
713 	}
714 
715 
716 	if (fill) {
717 		fillPage(pnum, fillval);
718 	}
719 	return p;
720 }
721 
722 static void vm_page_module_init_delayed(void);
723 
724 /*
725  * Free all remaining delayed pages to the free lists.
726  */
727 void
vm_free_delayed_pages(void)728 vm_free_delayed_pages(void)
729 {
730 	vm_page_t   p;
731 	vm_page_t   list = NULL;
732 	uint_t      cnt = 0;
733 	vm_offset_t start_free_va;
734 	int64_t     free_size;
735 
736 	while ((p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE)) != NULL) {
737 		if (vm_himemory_mode) {
738 			vm_page_release(p, FALSE);
739 		} else {
740 			p->vmp_snext = list;
741 			list = p;
742 		}
743 		++cnt;
744 	}
745 
746 	/*
747 	 * Free the pages in reverse order if not himemory mode.
748 	 * Hence the low memory pages will be first on free lists. (LIFO)
749 	 */
750 	while (list != NULL) {
751 		p = list;
752 		list = p->vmp_snext;
753 		p->vmp_snext = NULL;
754 		vm_page_release(p, FALSE);
755 	}
756 #if DEVELOPMENT || DEBUG
757 	kprintf("vm_free_delayed_pages: initialized %d free pages\n", cnt);
758 #endif
759 
760 	/*
761 	 * Free up any unused full pages at the end of the vm_pages[] array
762 	 */
763 	start_free_va = round_page((vm_offset_t)&vm_pages[vm_pages_count]);
764 
765 #if defined(__x86_64__)
766 	/*
767 	 * Since x86 might have used large pages for vm_pages[], we can't
768 	 * free starting in the middle of a partially used large page.
769 	 */
770 	if (pmap_query_pagesize(kernel_pmap, start_free_va) == I386_LPGBYTES) {
771 		start_free_va = ((start_free_va + I386_LPGMASK) & ~I386_LPGMASK);
772 	}
773 #endif
774 	if (start_free_va < (vm_offset_t)vm_page_array_ending_addr) {
775 		free_size = trunc_page((vm_offset_t)vm_page_array_ending_addr - start_free_va);
776 		if (free_size > 0) {
777 			ml_static_mfree(start_free_va, (vm_offset_t)free_size);
778 			vm_page_array_ending_addr = (void *)start_free_va;
779 
780 			/*
781 			 * Note there's no locking here, as only this thread will ever change this value.
782 			 * The reader, vm_page_diagnose, doesn't grab any locks for the counts it looks at.
783 			 */
784 			vm_page_stolen_count -= (free_size >> PAGE_SHIFT);
785 
786 #if DEVELOPMENT || DEBUG
787 			kprintf("Freeing final unused %ld bytes from vm_pages[] at 0x%lx\n",
788 			    (long)free_size, (long)start_free_va);
789 #endif
790 		}
791 	}
792 
793 
794 	/*
795 	 * now we can create the VM page array zone
796 	 */
797 	vm_page_module_init_delayed();
798 }
799 
800 /*
801  * Try and free up enough delayed pages to match a contig memory allocation.
802  */
803 static void
vm_free_delayed_pages_contig(uint_t npages,ppnum_t max_pnum,ppnum_t pnum_mask)804 vm_free_delayed_pages_contig(
805 	uint_t    npages,
806 	ppnum_t   max_pnum,
807 	ppnum_t   pnum_mask)
808 {
809 	vm_page_t p;
810 	ppnum_t   pnum;
811 	uint_t    cnt = 0;
812 
813 	/*
814 	 * Treat 0 as the absolute max page number.
815 	 */
816 	if (max_pnum == 0) {
817 		max_pnum = PPNUM_MAX;
818 	}
819 
820 	/*
821 	 * Free till we get a properly aligned start page
822 	 */
823 	for (;;) {
824 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
825 		if (p == NULL) {
826 			return;
827 		}
828 		pnum = VM_PAGE_GET_PHYS_PAGE(p);
829 		vm_page_release(p, FALSE);
830 		if (pnum >= max_pnum) {
831 			return;
832 		}
833 		if ((pnum & pnum_mask) == 0) {
834 			break;
835 		}
836 	}
837 
838 	/*
839 	 * Having a healthy pool of free pages will help performance. We don't
840 	 * want to fall back to the delayed code for every page allocation.
841 	 */
842 	if (vm_page_free_count < VM_DELAY_PAGE_CHUNK) {
843 		npages += VM_DELAY_PAGE_CHUNK;
844 	}
845 
846 	/*
847 	 * Now free up the pages
848 	 */
849 	for (cnt = 1; cnt < npages; ++cnt) {
850 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
851 		if (p == NULL) {
852 			return;
853 		}
854 		vm_page_release(p, FALSE);
855 	}
856 }
857 
858 #define ROUNDUP_NEXTP2(X) (1U << (32 - __builtin_clz((X) - 1)))
859 
860 void
vm_page_init_local_q(unsigned int num_cpus)861 vm_page_init_local_q(unsigned int num_cpus)
862 {
863 	struct vpl *t_local_q;
864 
865 	/*
866 	 * no point in this for a uni-processor system
867 	 */
868 	if (num_cpus >= 2) {
869 		ml_cpu_info_t cpu_info;
870 
871 		/*
872 		 * Force the allocation alignment to a cacheline,
873 		 * because the `vpl` struct has a lock and will be taken
874 		 * cross CPU so we want to isolate the rest of the per-CPU
875 		 * data to avoid false sharing due to this lock being taken.
876 		 */
877 
878 		ml_cpu_get_info(&cpu_info);
879 
880 		t_local_q = zalloc_percpu_permanent(sizeof(struct vpl),
881 		    cpu_info.cache_line_size - 1);
882 
883 		zpercpu_foreach(lq, t_local_q) {
884 			VPL_LOCK_INIT(lq, &vm_page_lck_grp_local, &vm_page_lck_attr);
885 			vm_page_queue_init(&lq->vpl_queue);
886 		}
887 
888 		/* make the initialization visible to all cores */
889 		os_atomic_store(&vm_page_local_q, t_local_q, release);
890 	}
891 }
892 
893 /*
894  * vm_init_before_launchd
895  *
896  * This should be called right before launchd is loaded.
897  */
898 void
vm_init_before_launchd()899 vm_init_before_launchd()
900 {
901 	vm_page_lockspin_queues();
902 	vm_page_wire_count_on_boot = vm_page_wire_count;
903 	vm_page_unlock_queues();
904 }
905 
906 
907 /*
908  *	vm_page_bootstrap:
909  *
910  *	Initializes the resident memory module.
911  *
912  *	Allocates memory for the page cells, and
913  *	for the object/offset-to-page hash table headers.
914  *	Each page cell is initialized and placed on the free list.
915  *	Returns the range of available kernel virtual memory.
916  */
917 __startup_func
918 void
vm_page_bootstrap(vm_offset_t * startp,vm_offset_t * endp)919 vm_page_bootstrap(
920 	vm_offset_t             *startp,
921 	vm_offset_t             *endp)
922 {
923 	unsigned int            i;
924 	unsigned int            log1;
925 	unsigned int            log2;
926 	unsigned int            size;
927 
928 	/*
929 	 *	Initialize the page queues.
930 	 */
931 
932 	lck_mtx_init(&vm_page_queue_free_lock, &vm_page_lck_grp_free, &vm_page_lck_attr);
933 	lck_mtx_init(&vm_page_queue_lock, &vm_page_lck_grp_queue, &vm_page_lck_attr);
934 	lck_mtx_init(&vm_purgeable_queue_lock, &vm_page_lck_grp_purge, &vm_page_lck_attr);
935 
936 	for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) {
937 		int group;
938 
939 		purgeable_queues[i].token_q_head = 0;
940 		purgeable_queues[i].token_q_tail = 0;
941 		for (group = 0; group < NUM_VOLATILE_GROUPS; group++) {
942 			queue_init(&purgeable_queues[i].objq[group]);
943 		}
944 
945 		purgeable_queues[i].type = i;
946 		purgeable_queues[i].new_pages = 0;
947 #if MACH_ASSERT
948 		purgeable_queues[i].debug_count_tokens = 0;
949 		purgeable_queues[i].debug_count_objects = 0;
950 #endif
951 	}
952 	;
953 	purgeable_nonvolatile_count = 0;
954 	queue_init(&purgeable_nonvolatile_queue);
955 
956 	for (i = 0; i < MAX_COLORS; i++) {
957 		vm_page_queue_init(&vm_page_queue_free[i].qhead);
958 	}
959 
960 	vm_page_queue_init(&vm_lopage_queue_free);
961 	vm_page_queue_init(&vm_page_queue_active);
962 	vm_page_queue_init(&vm_page_queue_inactive);
963 #if CONFIG_SECLUDED_MEMORY
964 	vm_page_queue_init(&vm_page_queue_secluded);
965 #endif /* CONFIG_SECLUDED_MEMORY */
966 	vm_page_queue_init(&vm_page_queue_cleaned);
967 	vm_page_queue_init(&vm_page_queue_throttled);
968 	vm_page_queue_init(&vm_page_queue_anonymous);
969 	queue_init(&vm_objects_wired);
970 
971 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
972 		vm_page_queue_init(&vm_page_queue_speculative[i].age_q);
973 
974 		vm_page_queue_speculative[i].age_ts.tv_sec = 0;
975 		vm_page_queue_speculative[i].age_ts.tv_nsec = 0;
976 	}
977 
978 	vm_page_queue_init(&vm_page_queue_donate);
979 	vm_page_queue_init(&vm_page_queue_background);
980 
981 	vm_page_background_count = 0;
982 	vm_page_background_internal_count = 0;
983 	vm_page_background_external_count = 0;
984 	vm_page_background_promoted_count = 0;
985 
986 	vm_page_background_target = (unsigned int)(atop_64(max_mem) / 25);
987 
988 	if (vm_page_background_target > VM_PAGE_BACKGROUND_TARGET_MAX) {
989 		vm_page_background_target = VM_PAGE_BACKGROUND_TARGET_MAX;
990 	}
991 
992 #if    defined(__LP64__)
993 	vm_page_background_mode = VM_PAGE_BG_ENABLED;
994 	vm_page_donate_mode = VM_PAGE_DONATE_ENABLED;
995 #else
996 	vm_page_background_mode = VM_PAGE_BG_DISABLED;
997 	vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
998 #endif
999 	vm_page_background_exclude_external = 0;
1000 
1001 	PE_parse_boot_argn("vm_page_bg_mode", &vm_page_background_mode, sizeof(vm_page_background_mode));
1002 	PE_parse_boot_argn("vm_page_bg_exclude_external", &vm_page_background_exclude_external, sizeof(vm_page_background_exclude_external));
1003 	PE_parse_boot_argn("vm_page_bg_target", &vm_page_background_target, sizeof(vm_page_background_target));
1004 
1005 	if (vm_page_background_mode != VM_PAGE_BG_DISABLED && vm_page_background_mode != VM_PAGE_BG_ENABLED) {
1006 		vm_page_background_mode = VM_PAGE_BG_DISABLED;
1007 	}
1008 
1009 	PE_parse_boot_argn("vm_page_donate_mode", &vm_page_donate_mode, sizeof(vm_page_donate_mode));
1010 	if (vm_page_donate_mode != VM_PAGE_DONATE_DISABLED && vm_page_donate_mode != VM_PAGE_DONATE_ENABLED) {
1011 		vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
1012 	}
1013 
1014 	vm_page_donate_target_high = VM_PAGE_DONATE_TARGET_HIGHWATER;
1015 	vm_page_donate_target_low = VM_PAGE_DONATE_TARGET_LOWWATER;
1016 	vm_page_donate_target = vm_page_donate_target_high;
1017 	vm_page_donate_count = 0;
1018 
1019 	vm_page_free_wanted = 0;
1020 	vm_page_free_wanted_privileged = 0;
1021 #if CONFIG_SECLUDED_MEMORY
1022 	vm_page_free_wanted_secluded = 0;
1023 #endif /* CONFIG_SECLUDED_MEMORY */
1024 
1025 #if defined (__x86_64__)
1026 	/* this must be called before vm_page_set_colors() */
1027 	vm_page_setup_clump();
1028 #endif
1029 
1030 	vm_page_set_colors();
1031 
1032 	bzero(vm_page_inactive_states, sizeof(vm_page_inactive_states));
1033 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1034 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1035 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1036 
1037 	bzero(vm_page_pageable_states, sizeof(vm_page_pageable_states));
1038 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1039 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1040 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1041 	vm_page_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1042 	vm_page_pageable_states[VM_PAGE_ON_SPECULATIVE_Q] = 1;
1043 	vm_page_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1044 #if CONFIG_SECLUDED_MEMORY
1045 	vm_page_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1046 #endif /* CONFIG_SECLUDED_MEMORY */
1047 
1048 	bzero(vm_page_non_speculative_pageable_states, sizeof(vm_page_non_speculative_pageable_states));
1049 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1050 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1051 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1052 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1053 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1054 #if CONFIG_SECLUDED_MEMORY
1055 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1056 #endif /* CONFIG_SECLUDED_MEMORY */
1057 
1058 	bzero(vm_page_active_or_inactive_states, sizeof(vm_page_active_or_inactive_states));
1059 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1060 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1061 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1062 	vm_page_active_or_inactive_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1063 #if CONFIG_SECLUDED_MEMORY
1064 	vm_page_active_or_inactive_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1065 #endif /* CONFIG_SECLUDED_MEMORY */
1066 
1067 	for (vm_tag_t t = 0; t < VM_KERN_MEMORY_FIRST_DYNAMIC; t++) {
1068 		vm_allocation_sites_static[t].refcount = 2;
1069 		vm_allocation_sites_static[t].tag = t;
1070 		vm_allocation_sites[t] = &vm_allocation_sites_static[t];
1071 	}
1072 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].refcount = 2;
1073 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].tag = VM_KERN_MEMORY_ANY;
1074 	vm_allocation_sites[VM_KERN_MEMORY_ANY] = &vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC];
1075 
1076 	/*
1077 	 *	Steal memory for the map and zone subsystems.
1078 	 */
1079 	kernel_startup_initialize_upto(STARTUP_SUB_PMAP_STEAL);
1080 
1081 	/*
1082 	 *	Allocate (and initialize) the virtual-to-physical
1083 	 *	table hash buckets.
1084 	 *
1085 	 *	The number of buckets should be a power of two to
1086 	 *	get a good hash function.  The following computation
1087 	 *	chooses the first power of two that is greater
1088 	 *	than the number of physical pages in the system.
1089 	 */
1090 
1091 	if (vm_page_bucket_count == 0) {
1092 		unsigned int npages = pmap_free_pages();
1093 
1094 		vm_page_bucket_count = 1;
1095 		while (vm_page_bucket_count < npages) {
1096 			vm_page_bucket_count <<= 1;
1097 		}
1098 	}
1099 	vm_page_bucket_lock_count = (vm_page_bucket_count + BUCKETS_PER_LOCK - 1) / BUCKETS_PER_LOCK;
1100 
1101 	vm_page_hash_mask = vm_page_bucket_count - 1;
1102 
1103 	/*
1104 	 *	Calculate object shift value for hashing algorithm:
1105 	 *		O = log2(sizeof(struct vm_object))
1106 	 *		B = log2(vm_page_bucket_count)
1107 	 *	        hash shifts the object left by
1108 	 *		B/2 - O
1109 	 */
1110 	size = vm_page_bucket_count;
1111 	for (log1 = 0; size > 1; log1++) {
1112 		size /= 2;
1113 	}
1114 	size = sizeof(struct vm_object);
1115 	for (log2 = 0; size > 1; log2++) {
1116 		size /= 2;
1117 	}
1118 	vm_page_hash_shift = log1 / 2 - log2 + 1;
1119 
1120 	vm_page_bucket_hash = 1 << ((log1 + 1) >> 1);           /* Get (ceiling of sqrt of table size) */
1121 	vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2);          /* Get (ceiling of quadroot of table size) */
1122 	vm_page_bucket_hash |= 1;                                                       /* Set bit and add 1 - always must be 1 to insure unique series */
1123 
1124 	if (vm_page_hash_mask & vm_page_bucket_count) {
1125 		printf("vm_page_bootstrap: WARNING -- strange page hash\n");
1126 	}
1127 
1128 #if VM_PAGE_BUCKETS_CHECK
1129 #if VM_PAGE_FAKE_BUCKETS
1130 	/*
1131 	 * Allocate a decoy set of page buckets, to detect
1132 	 * any stomping there.
1133 	 */
1134 	vm_page_fake_buckets = (vm_page_bucket_t *)
1135 	    pmap_steal_memory(vm_page_bucket_count *
1136 	    sizeof(vm_page_bucket_t));
1137 	vm_page_fake_buckets_start = (vm_map_offset_t) vm_page_fake_buckets;
1138 	vm_page_fake_buckets_end =
1139 	    vm_map_round_page((vm_page_fake_buckets_start +
1140 	    (vm_page_bucket_count *
1141 	    sizeof(vm_page_bucket_t))),
1142 	    PAGE_MASK);
1143 	char *cp;
1144 	for (cp = (char *)vm_page_fake_buckets_start;
1145 	    cp < (char *)vm_page_fake_buckets_end;
1146 	    cp++) {
1147 		*cp = 0x5a;
1148 	}
1149 #endif /* VM_PAGE_FAKE_BUCKETS */
1150 #endif /* VM_PAGE_BUCKETS_CHECK */
1151 
1152 	kernel_debug_string_early("vm_page_buckets");
1153 	vm_page_buckets = (vm_page_bucket_t *)
1154 	    pmap_steal_memory(vm_page_bucket_count *
1155 	    sizeof(vm_page_bucket_t));
1156 
1157 	kernel_debug_string_early("vm_page_bucket_locks");
1158 	vm_page_bucket_locks = (lck_spin_t *)
1159 	    pmap_steal_memory(vm_page_bucket_lock_count *
1160 	    sizeof(lck_spin_t));
1161 
1162 	for (i = 0; i < vm_page_bucket_count; i++) {
1163 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
1164 
1165 		bucket->page_list = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
1166 #if     MACH_PAGE_HASH_STATS
1167 		bucket->cur_count = 0;
1168 		bucket->hi_count = 0;
1169 #endif /* MACH_PAGE_HASH_STATS */
1170 	}
1171 
1172 	for (i = 0; i < vm_page_bucket_lock_count; i++) {
1173 		lck_spin_init(&vm_page_bucket_locks[i], &vm_page_lck_grp_bucket, &vm_page_lck_attr);
1174 	}
1175 
1176 	vm_tag_init();
1177 
1178 #if VM_PAGE_BUCKETS_CHECK
1179 	vm_page_buckets_check_ready = TRUE;
1180 #endif /* VM_PAGE_BUCKETS_CHECK */
1181 
1182 	/*
1183 	 *	Machine-dependent code allocates the resident page table.
1184 	 *	It uses vm_page_init to initialize the page frames.
1185 	 *	The code also returns to us the virtual space available
1186 	 *	to the kernel.  We don't trust the pmap module
1187 	 *	to get the alignment right.
1188 	 */
1189 
1190 	kernel_debug_string_early("pmap_startup");
1191 	pmap_startup(&virtual_space_start, &virtual_space_end);
1192 	virtual_space_start = round_page(virtual_space_start);
1193 	virtual_space_end = trunc_page(virtual_space_end);
1194 
1195 	*startp = virtual_space_start;
1196 	*endp = virtual_space_end;
1197 
1198 	/*
1199 	 *	Compute the initial "wire" count.
1200 	 *	Up until now, the pages which have been set aside are not under
1201 	 *	the VM system's control, so although they aren't explicitly
1202 	 *	wired, they nonetheless can't be moved. At this moment,
1203 	 *	all VM managed pages are "free", courtesy of pmap_startup.
1204 	 */
1205 	assert((unsigned int) atop_64(max_mem) == atop_64(max_mem));
1206 	vm_page_wire_count = ((unsigned int) atop_64(max_mem)) -
1207 	    vm_page_free_count - vm_lopage_free_count;
1208 #if CONFIG_SECLUDED_MEMORY
1209 	vm_page_wire_count -= vm_page_secluded_count;
1210 #endif
1211 	vm_page_wire_count_initial = vm_page_wire_count;
1212 
1213 	/* capture this for later use */
1214 	booter_size = ml_get_booter_memory_size();
1215 
1216 	printf("vm_page_bootstrap: %d free pages, %d wired pages, (up to %d of which are delayed free)\n",
1217 	    vm_page_free_count, vm_page_wire_count, vm_delayed_count);
1218 
1219 	kernel_debug_string_early("vm_page_bootstrap complete");
1220 }
1221 
1222 #ifndef MACHINE_PAGES
1223 /*
1224  * This is the early boot time allocator for data structures needed to bootstrap the VM system.
1225  * On x86 it will allocate large pages if size is sufficiently large. We don't need to do this
1226  * on ARM yet, due to the combination of a large base page size and smaller RAM devices.
1227  */
1228 static void *
pmap_steal_memory_internal(vm_size_t size,boolean_t might_free)1229 pmap_steal_memory_internal(
1230 	vm_size_t size,
1231 	boolean_t might_free)
1232 {
1233 	kern_return_t kr;
1234 	vm_offset_t addr;
1235 	vm_offset_t map_addr;
1236 	ppnum_t phys_page;
1237 
1238 	/*
1239 	 * Size needs to be aligned to word size.
1240 	 */
1241 	size = (size + sizeof(void *) - 1) & ~(sizeof(void *) - 1);
1242 
1243 	/*
1244 	 * On the first call, get the initial values for virtual address space
1245 	 * and page align them.
1246 	 */
1247 	if (virtual_space_start == virtual_space_end) {
1248 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
1249 		virtual_space_start = round_page(virtual_space_start);
1250 		virtual_space_end = trunc_page(virtual_space_end);
1251 
1252 #if defined(__x86_64__)
1253 		/*
1254 		 * Release remaining unused section of preallocated KVA and the 4K page tables
1255 		 * that map it. This makes the VA available for large page mappings.
1256 		 */
1257 		Idle_PTs_release(virtual_space_start, virtual_space_end);
1258 #endif
1259 	}
1260 
1261 	/*
1262 	 * Allocate the virtual space for this request. On x86, we'll align to a large page
1263 	 * address if the size is big enough to back with at least 1 large page.
1264 	 */
1265 #if defined(__x86_64__)
1266 	if (size >= I386_LPGBYTES) {
1267 		virtual_space_start = ((virtual_space_start + I386_LPGMASK) & ~I386_LPGMASK);
1268 	}
1269 #endif
1270 	addr = virtual_space_start;
1271 	virtual_space_start += size;
1272 
1273 	//kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)size);	/* (TEST/DEBUG) */
1274 
1275 	/*
1276 	 * Allocate and map physical pages to back the new virtual space.
1277 	 */
1278 	map_addr = round_page(addr);
1279 	while (map_addr < addr + size) {
1280 #if defined(__x86_64__)
1281 		/*
1282 		 * Back with a large page if properly aligned on x86
1283 		 */
1284 		if ((map_addr & I386_LPGMASK) == 0 &&
1285 		    map_addr + I386_LPGBYTES <= addr + size &&
1286 		    pmap_pre_expand_large(kernel_pmap, map_addr) == KERN_SUCCESS &&
1287 		    pmap_next_page_large(&phys_page) == KERN_SUCCESS) {
1288 			kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1289 			    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1290 			    VM_WIMG_USE_DEFAULT | VM_MEM_SUPERPAGE, FALSE);
1291 
1292 			if (kr != KERN_SUCCESS) {
1293 				panic("pmap_steal_memory: pmap_enter() large failed, new_addr=%#lx, phys_page=%u",
1294 				    (unsigned long)map_addr, phys_page);
1295 			}
1296 			map_addr += I386_LPGBYTES;
1297 			vm_page_wire_count += I386_LPGBYTES >> PAGE_SHIFT;
1298 			vm_page_stolen_count += I386_LPGBYTES >> PAGE_SHIFT;
1299 			vm_page_kern_lpage_count++;
1300 			continue;
1301 		}
1302 #endif
1303 
1304 		if (!pmap_next_page_hi(&phys_page, might_free)) {
1305 			panic("pmap_steal_memory() size: 0x%llx", (uint64_t)size);
1306 		}
1307 
1308 #if defined(__x86_64__)
1309 		pmap_pre_expand(kernel_pmap, map_addr);
1310 #endif
1311 
1312 		kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1313 		    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1314 		    VM_WIMG_USE_DEFAULT, FALSE);
1315 
1316 		if (kr != KERN_SUCCESS) {
1317 			panic("pmap_steal_memory() pmap_enter failed, map_addr=%#lx, phys_page=%u",
1318 			    (unsigned long)map_addr, phys_page);
1319 		}
1320 		map_addr += PAGE_SIZE;
1321 
1322 		/*
1323 		 * Account for newly stolen memory
1324 		 */
1325 		vm_page_wire_count++;
1326 		vm_page_stolen_count++;
1327 	}
1328 
1329 #if defined(__x86_64__)
1330 	/*
1331 	 * The call with might_free is currently the last use of pmap_steal_memory*().
1332 	 * Notify the pmap layer to record which high pages were allocated so far.
1333 	 */
1334 	if (might_free) {
1335 		pmap_hi_pages_done();
1336 	}
1337 #endif
1338 #if KASAN
1339 	kasan_notify_address(round_page(addr), size);
1340 #endif
1341 	return (void *) addr;
1342 }
1343 
1344 void *
pmap_steal_memory(vm_size_t size)1345 pmap_steal_memory(
1346 	vm_size_t size)
1347 {
1348 	return pmap_steal_memory_internal(size, FALSE);
1349 }
1350 
1351 void *
pmap_steal_freeable_memory(vm_size_t size)1352 pmap_steal_freeable_memory(
1353 	vm_size_t size)
1354 {
1355 	return pmap_steal_memory_internal(size, TRUE);
1356 }
1357 
1358 
1359 #if CONFIG_SECLUDED_MEMORY
1360 /* boot-args to control secluded memory */
1361 unsigned int secluded_mem_mb = 0;       /* # of MBs of RAM to seclude */
1362 int secluded_for_iokit = 1;             /* IOKit can use secluded memory */
1363 int secluded_for_apps = 1;              /* apps can use secluded memory */
1364 int secluded_for_filecache = 2;         /* filecache can use seclude memory */
1365 #if 11
1366 int secluded_for_fbdp = 0;
1367 #endif
1368 uint64_t secluded_shutoff_trigger = 0;
1369 uint64_t secluded_shutoff_headroom = 150 * 1024 * 1024; /* original value from N56 */
1370 #endif /* CONFIG_SECLUDED_MEMORY */
1371 
1372 
1373 #if defined(__arm64__)
1374 extern void patch_low_glo_vm_page_info(void *, void *, uint32_t);
1375 unsigned int vm_first_phys_ppnum = 0;
1376 #endif
1377 
1378 void vm_page_release_startup(vm_page_t mem);
1379 void
pmap_startup(vm_offset_t * startp,vm_offset_t * endp)1380 pmap_startup(
1381 	vm_offset_t     *startp,
1382 	vm_offset_t     *endp)
1383 {
1384 	unsigned int    i, npages;
1385 	ppnum_t         phys_page;
1386 	uint64_t        mem_sz;
1387 	uint64_t        start_ns;
1388 	uint64_t        now_ns;
1389 	uint_t          low_page_count = 0;
1390 
1391 #if    defined(__LP64__)
1392 	/*
1393 	 * make sure we are aligned on a 64 byte boundary
1394 	 * for VM_PAGE_PACK_PTR (it clips off the low-order
1395 	 * 6 bits of the pointer)
1396 	 */
1397 	if (virtual_space_start != virtual_space_end) {
1398 		virtual_space_start = round_page(virtual_space_start);
1399 	}
1400 #endif
1401 
1402 	/*
1403 	 * We calculate how many page frames we will have
1404 	 * and then allocate the page structures in one chunk.
1405 	 *
1406 	 * Note that the calculation here doesn't take into account
1407 	 * the memory needed to map what's being allocated, i.e. the page
1408 	 * table entries. So the actual number of pages we get will be
1409 	 * less than this. To do someday: include that in the computation.
1410 	 *
1411 	 * Also for ARM, we don't use the count of free_pages, but rather the
1412 	 * range from last page to first page (ignore holes due to retired pages).
1413 	 */
1414 #if defined(__arm64__)
1415 	mem_sz = pmap_free_pages_span() * (uint64_t)PAGE_SIZE;
1416 #else /* defined(__arm64__) */
1417 	mem_sz = pmap_free_pages() * (uint64_t)PAGE_SIZE;
1418 #endif /* defined(__arm64__) */
1419 	mem_sz += round_page(virtual_space_start) - virtual_space_start;        /* Account for any slop */
1420 	npages = (uint_t)(mem_sz / (PAGE_SIZE + sizeof(*vm_pages)));    /* scaled to include the vm_page_ts */
1421 
1422 
1423 	vm_pages = (vm_page_t) pmap_steal_freeable_memory(npages * sizeof *vm_pages);
1424 
1425 	/*
1426 	 * Check if we want to initialize pages to a known value
1427 	 */
1428 	if (PE_parse_boot_argn("fill", &fillval, sizeof(fillval))) {
1429 		fill = TRUE;
1430 	}
1431 #if     DEBUG
1432 	/* This slows down booting the DEBUG kernel, particularly on
1433 	 * large memory systems, but is worthwhile in deterministically
1434 	 * trapping uninitialized memory usage.
1435 	 */
1436 	if (!fill) {
1437 		fill = TRUE;
1438 		fillval = 0xDEB8F177;
1439 	}
1440 #endif
1441 	if (fill) {
1442 		kprintf("Filling vm_pages with pattern: 0x%x\n", fillval);
1443 	}
1444 
1445 #if CONFIG_SECLUDED_MEMORY
1446 	/*
1447 	 * Figure out how much secluded memory to have before we start
1448 	 * release pages to free lists.
1449 	 * The default, if specified nowhere else, is no secluded mem.
1450 	 */
1451 	secluded_mem_mb = 0;
1452 	if (max_mem > 1 * 1024 * 1024 * 1024) {
1453 		/* default to 90MB for devices with > 1GB of RAM */
1454 		secluded_mem_mb = 90;
1455 	}
1456 	/* override with value from device tree, if provided */
1457 	PE_get_default("kern.secluded_mem_mb",
1458 	    &secluded_mem_mb, sizeof(secluded_mem_mb));
1459 	/* override with value from boot-args, if provided */
1460 	PE_parse_boot_argn("secluded_mem_mb",
1461 	    &secluded_mem_mb,
1462 	    sizeof(secluded_mem_mb));
1463 
1464 	vm_page_secluded_target = (unsigned int)
1465 	    ((secluded_mem_mb * 1024ULL * 1024ULL) / PAGE_SIZE);
1466 	PE_parse_boot_argn("secluded_for_iokit",
1467 	    &secluded_for_iokit,
1468 	    sizeof(secluded_for_iokit));
1469 	PE_parse_boot_argn("secluded_for_apps",
1470 	    &secluded_for_apps,
1471 	    sizeof(secluded_for_apps));
1472 	PE_parse_boot_argn("secluded_for_filecache",
1473 	    &secluded_for_filecache,
1474 	    sizeof(secluded_for_filecache));
1475 #if 11
1476 	PE_parse_boot_argn("secluded_for_fbdp",
1477 	    &secluded_for_fbdp,
1478 	    sizeof(secluded_for_fbdp));
1479 #endif
1480 
1481 	/*
1482 	 * Allow a really large app to effectively use secluded memory until it exits.
1483 	 */
1484 	if (vm_page_secluded_target != 0) {
1485 		/*
1486 		 * Get an amount from boot-args, else use 1/2 of max_mem.
1487 		 * 1/2 max_mem was chosen from a Peace daemon tentpole test which
1488 		 * used munch to induce jetsam thrashing of false idle daemons on N56.
1489 		 */
1490 		int secluded_shutoff_mb;
1491 		if (PE_parse_boot_argn("secluded_shutoff_mb", &secluded_shutoff_mb,
1492 		    sizeof(secluded_shutoff_mb))) {
1493 			secluded_shutoff_trigger = (uint64_t)secluded_shutoff_mb * 1024 * 1024;
1494 		} else {
1495 			secluded_shutoff_trigger = max_mem / 2;
1496 		}
1497 
1498 		/* ensure the headroom value is sensible and avoid underflows */
1499 		assert(secluded_shutoff_trigger == 0 || secluded_shutoff_trigger > secluded_shutoff_headroom);
1500 	}
1501 
1502 #endif /* CONFIG_SECLUDED_MEMORY */
1503 
1504 #if defined(__x86_64__)
1505 
1506 	/*
1507 	 * Decide how much memory we delay freeing at boot time.
1508 	 */
1509 	uint32_t delay_above_gb;
1510 	if (!PE_parse_boot_argn("delay_above_gb", &delay_above_gb, sizeof(delay_above_gb))) {
1511 		delay_above_gb = DEFAULT_DELAY_ABOVE_PHYS_GB;
1512 	}
1513 
1514 	if (delay_above_gb == 0) {
1515 		delay_above_pnum = PPNUM_MAX;
1516 	} else {
1517 		delay_above_pnum = delay_above_gb * (1024 * 1024 * 1024 / PAGE_SIZE);
1518 	}
1519 
1520 	/* make sure we have sane breathing room: 1G above low memory */
1521 	if (delay_above_pnum <= max_valid_low_ppnum) {
1522 		delay_above_pnum = max_valid_low_ppnum + ((1024 * 1024 * 1024) >> PAGE_SHIFT);
1523 	}
1524 
1525 	if (delay_above_pnum < PPNUM_MAX) {
1526 		printf("pmap_startup() delaying init/free of page nums > 0x%x\n", delay_above_pnum);
1527 	}
1528 
1529 #endif /* defined(__x86_64__) */
1530 
1531 	/*
1532 	 * Initialize and release the page frames.
1533 	 */
1534 	kernel_debug_string_early("page_frame_init");
1535 
1536 	vm_page_array_beginning_addr = &vm_pages[0];
1537 	vm_page_array_ending_addr = &vm_pages[npages];  /* used by ptr packing/unpacking code */
1538 #if VM_PAGE_PACKED_FROM_ARRAY
1539 	if (npages >= VM_PAGE_PACKED_FROM_ARRAY) {
1540 		panic("pmap_startup(): too many pages to support vm_page packing");
1541 	}
1542 #endif
1543 
1544 	vm_delayed_count = 0;
1545 
1546 	absolutetime_to_nanoseconds(mach_absolute_time(), &start_ns);
1547 	vm_pages_count = 0;
1548 	for (i = 0; i < npages; i++) {
1549 		/* Did we run out of pages? */
1550 		if (!pmap_next_page(&phys_page)) {
1551 			break;
1552 		}
1553 
1554 		if (phys_page < max_valid_low_ppnum) {
1555 			++low_page_count;
1556 		}
1557 
1558 		/* Are we at high enough pages to delay the rest? */
1559 		if (low_page_count > vm_lopage_free_limit && phys_page > delay_above_pnum) {
1560 			vm_delayed_count = pmap_free_pages();
1561 			break;
1562 		}
1563 
1564 #if defined(__arm64__)
1565 		if (i == 0) {
1566 			vm_first_phys_ppnum = phys_page;
1567 			patch_low_glo_vm_page_info((void *)vm_page_array_beginning_addr,
1568 			    (void *)vm_page_array_ending_addr, vm_first_phys_ppnum);
1569 		}
1570 #endif /* defined(__arm64__) */
1571 
1572 #if defined(__x86_64__)
1573 		/* The x86 clump freeing code requires increasing ppn's to work correctly */
1574 		if (i > 0) {
1575 			assert(phys_page > vm_pages[i - 1].vmp_phys_page);
1576 		}
1577 #endif
1578 		++vm_pages_count;
1579 		vm_page_init(&vm_pages[i], phys_page, FALSE);
1580 		if (fill) {
1581 			fillPage(phys_page, fillval);
1582 		}
1583 		if (vm_himemory_mode) {
1584 			vm_page_release_startup(&vm_pages[i]);
1585 		}
1586 	}
1587 	vm_page_pages = vm_pages_count; /* used to report to user space */
1588 
1589 	if (!vm_himemory_mode) {
1590 		do {
1591 			if (!VMP_ERROR_GET(&vm_pages[--i])) {               /* skip retired pages */
1592 				vm_page_release_startup(&vm_pages[i]);
1593 			}
1594 		} while (i != 0);
1595 	}
1596 
1597 	absolutetime_to_nanoseconds(mach_absolute_time(), &now_ns);
1598 	printf("pmap_startup() init/release time: %lld microsec\n", (now_ns - start_ns) / NSEC_PER_USEC);
1599 	printf("pmap_startup() delayed init/release of %d pages\n", vm_delayed_count);
1600 
1601 #if defined(__LP64__)
1602 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[0]))) != &vm_pages[0]) {
1603 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[0] - %p", (void *)&vm_pages[0]);
1604 	}
1605 
1606 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[vm_pages_count - 1]))) != &vm_pages[vm_pages_count - 1]) {
1607 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[vm_pages_count-1] - %p", (void *)&vm_pages[vm_pages_count - 1]);
1608 	}
1609 #endif
1610 
1611 	VM_CHECK_MEMORYSTATUS;
1612 
1613 	/*
1614 	 * We have to re-align virtual_space_start,
1615 	 * because pmap_steal_memory has been using it.
1616 	 */
1617 	virtual_space_start = round_page(virtual_space_start);
1618 	*startp = virtual_space_start;
1619 	*endp = virtual_space_end;
1620 }
1621 #endif  /* MACHINE_PAGES */
1622 
1623 /*
1624  * Create the zone that represents the vm_pages[] array. Nothing ever allocates
1625  * or frees to this zone. It's just here for reporting purposes via zprint command.
1626  * This needs to be done after all initially delayed pages are put on the free lists.
1627  */
1628 static void
vm_page_module_init_delayed(void)1629 vm_page_module_init_delayed(void)
1630 {
1631 	(void)zone_create_ext("vm pages array", sizeof(struct vm_page),
1632 	    ZC_NONE, ZONE_ID_VM_PAGES, ^(zone_t z) {
1633 		uint64_t vm_page_zone_pages, vm_page_array_zone_data_size;
1634 
1635 		zone_set_exhaustible(z, 0);
1636 		/*
1637 		 * Reflect size and usage information for vm_pages[].
1638 		 */
1639 
1640 		z->z_elems_avail = (uint32_t)(vm_page_array_ending_addr - vm_pages);
1641 		z->z_elems_free = z->z_elems_avail - vm_pages_count;
1642 		zpercpu_get_cpu(z->z_stats, 0)->zs_mem_allocated =
1643 		vm_pages_count * sizeof(struct vm_page);
1644 		vm_page_array_zone_data_size = (uint64_t)vm_page_array_ending_addr - (uint64_t)vm_pages;
1645 		vm_page_zone_pages = atop(round_page((vm_offset_t)vm_page_array_zone_data_size));
1646 		z->z_wired_cur += vm_page_zone_pages;
1647 		z->z_wired_hwm = z->z_wired_cur;
1648 		z->z_va_cur = z->z_wired_cur;
1649 		/* since zone accounts for these, take them out of stolen */
1650 		VM_PAGE_MOVE_STOLEN(vm_page_zone_pages);
1651 	});
1652 }
1653 
1654 /*
1655  * Create the vm_pages zone. This is used for the vm_page structures for the pages
1656  * that are scavanged from other boot time usages by ml_static_mfree(). As such,
1657  * this needs to happen in early VM bootstrap.
1658  */
1659 
1660 __startup_func
1661 static void
vm_page_module_init(void)1662 vm_page_module_init(void)
1663 {
1664 	vm_size_t vm_page_with_ppnum_size;
1665 
1666 	/*
1667 	 * Since the pointers to elements in this zone will be packed, they
1668 	 * must have appropriate size. Not strictly what sizeof() reports.
1669 	 */
1670 	vm_page_with_ppnum_size =
1671 	    (sizeof(struct vm_page_with_ppnum) + (VM_PAGE_PACKED_PTR_ALIGNMENT - 1)) &
1672 	    ~(VM_PAGE_PACKED_PTR_ALIGNMENT - 1);
1673 
1674 	vm_page_zone = zone_create_ext("vm pages", vm_page_with_ppnum_size,
1675 	    ZC_ALIGNMENT_REQUIRED | ZC_VM_LP64 | ZC_NOTBITAG,
1676 	    ZONE_ID_ANY, ^(zone_t z) {
1677 		/*
1678 		 * The number "10" is a small number that is larger than the number
1679 		 * of fictitious pages that any single caller will attempt to allocate
1680 		 * without blocking.
1681 		 *
1682 		 * The largest such number at the moment is kmem_alloc()
1683 		 * when 2 guard pages are asked. 10 is simply a somewhat larger number,
1684 		 * taking into account the 50% hysteresis the zone allocator uses.
1685 		 *
1686 		 * Note: this works at all because the zone allocator
1687 		 *       doesn't ever allocate fictitious pages.
1688 		 */
1689 		zone_raise_reserve(z, 10);
1690 	});
1691 }
1692 STARTUP(ZALLOC, STARTUP_RANK_SECOND, vm_page_module_init);
1693 
1694 /*
1695  *	Routine:	vm_page_create
1696  *	Purpose:
1697  *		After the VM system is up, machine-dependent code
1698  *		may stumble across more physical memory.  For example,
1699  *		memory that it was reserving for a frame buffer.
1700  *		vm_page_create turns this memory into available pages.
1701  */
1702 
1703 void
vm_page_create(ppnum_t start,ppnum_t end)1704 vm_page_create(
1705 	ppnum_t start,
1706 	ppnum_t end)
1707 {
1708 	ppnum_t         phys_page;
1709 	vm_page_t       m;
1710 
1711 	for (phys_page = start;
1712 	    phys_page < end;
1713 	    phys_page++) {
1714 		m = vm_page_grab_fictitious_common(phys_page, TRUE);
1715 		m->vmp_fictitious = FALSE;
1716 		pmap_clear_noencrypt(phys_page);
1717 
1718 
1719 		vm_free_page_lock();
1720 		vm_page_pages++;
1721 		vm_free_page_unlock();
1722 		vm_page_release(m, FALSE);
1723 	}
1724 }
1725 
1726 
1727 /*
1728  *	vm_page_hash:
1729  *
1730  *	Distributes the object/offset key pair among hash buckets.
1731  *
1732  *	NOTE:	The bucket count must be a power of 2
1733  */
1734 #define vm_page_hash(object, offset) (\
1735 	( (natural_t)((uintptr_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
1736 	 & vm_page_hash_mask)
1737 
1738 
1739 /*
1740  *	vm_page_insert:		[ internal use only ]
1741  *
1742  *	Inserts the given mem entry into the object/object-page
1743  *	table and object list.
1744  *
1745  *	The object must be locked.
1746  */
1747 void
vm_page_insert(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)1748 vm_page_insert(
1749 	vm_page_t               mem,
1750 	vm_object_t             object,
1751 	vm_object_offset_t      offset)
1752 {
1753 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, FALSE, FALSE, NULL);
1754 }
1755 
1756 void
vm_page_insert_wired(vm_page_t mem,vm_object_t object,vm_object_offset_t offset,vm_tag_t tag)1757 vm_page_insert_wired(
1758 	vm_page_t               mem,
1759 	vm_object_t             object,
1760 	vm_object_offset_t      offset,
1761 	vm_tag_t                tag)
1762 {
1763 	vm_page_insert_internal(mem, object, offset, tag, FALSE, TRUE, FALSE, FALSE, NULL);
1764 }
1765 
1766 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)1767 vm_page_insert_internal(
1768 	vm_page_t               mem,
1769 	vm_object_t             object,
1770 	vm_object_offset_t      offset,
1771 	vm_tag_t                tag,
1772 	boolean_t               queues_lock_held,
1773 	boolean_t               insert_in_hash,
1774 	boolean_t               batch_pmap_op,
1775 	boolean_t               batch_accounting,
1776 	uint64_t                *delayed_ledger_update)
1777 {
1778 	vm_page_bucket_t        *bucket;
1779 	lck_spin_t              *bucket_lock;
1780 	int                     hash_id;
1781 	task_t                  owner;
1782 	int                     ledger_idx_volatile;
1783 	int                     ledger_idx_nonvolatile;
1784 	int                     ledger_idx_volatile_compressed;
1785 	int                     ledger_idx_nonvolatile_compressed;
1786 	boolean_t               do_footprint;
1787 
1788 #if 0
1789 	/*
1790 	 * we may not hold the page queue lock
1791 	 * so this check isn't safe to make
1792 	 */
1793 	VM_PAGE_CHECK(mem);
1794 #endif
1795 
1796 	assertf(page_aligned(offset), "0x%llx\n", offset);
1797 
1798 	assert(!VM_PAGE_WIRED(mem) || mem->vmp_private || mem->vmp_fictitious || (tag != VM_KERN_MEMORY_NONE));
1799 
1800 	vm_object_lock_assert_exclusive(object);
1801 	LCK_MTX_ASSERT(&vm_page_queue_lock,
1802 	    queues_lock_held ? LCK_MTX_ASSERT_OWNED
1803 	    : LCK_MTX_ASSERT_NOTOWNED);
1804 
1805 	if (queues_lock_held == FALSE) {
1806 		assert(!VM_PAGE_PAGEABLE(mem));
1807 	}
1808 
1809 	if (insert_in_hash == TRUE) {
1810 #if DEBUG || VM_PAGE_BUCKETS_CHECK
1811 		if (mem->vmp_tabled || mem->vmp_object) {
1812 			panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
1813 			    "already in (obj=%p,off=0x%llx)",
1814 			    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
1815 		}
1816 #endif
1817 		if (object->internal && (offset >= object->vo_size)) {
1818 			panic("vm_page_insert_internal: (page=%p,obj=%p,off=0x%llx,size=0x%llx) inserted at offset past object bounds",
1819 			    mem, object, offset, object->vo_size);
1820 		}
1821 
1822 		assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
1823 
1824 		/*
1825 		 *	Record the object/offset pair in this page
1826 		 */
1827 
1828 		mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
1829 		mem->vmp_offset = offset;
1830 
1831 #if CONFIG_SECLUDED_MEMORY
1832 		if (object->eligible_for_secluded) {
1833 			vm_page_secluded.eligible_for_secluded++;
1834 		}
1835 #endif /* CONFIG_SECLUDED_MEMORY */
1836 
1837 		/*
1838 		 *	Insert it into the object_object/offset hash table
1839 		 */
1840 		hash_id = vm_page_hash(object, offset);
1841 		bucket = &vm_page_buckets[hash_id];
1842 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
1843 
1844 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
1845 
1846 		mem->vmp_next_m = bucket->page_list;
1847 		bucket->page_list = VM_PAGE_PACK_PTR(mem);
1848 		assert(mem == (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)));
1849 
1850 #if     MACH_PAGE_HASH_STATS
1851 		if (++bucket->cur_count > bucket->hi_count) {
1852 			bucket->hi_count = bucket->cur_count;
1853 		}
1854 #endif /* MACH_PAGE_HASH_STATS */
1855 		mem->vmp_hashed = TRUE;
1856 		lck_spin_unlock(bucket_lock);
1857 	}
1858 
1859 	{
1860 		unsigned int    cache_attr;
1861 
1862 		cache_attr = object->wimg_bits & VM_WIMG_MASK;
1863 
1864 		if (cache_attr != VM_WIMG_USE_DEFAULT) {
1865 			PMAP_SET_CACHE_ATTR(mem, object, cache_attr, batch_pmap_op);
1866 		}
1867 	}
1868 	/*
1869 	 *	Now link into the object's list of backed pages.
1870 	 */
1871 	vm_page_queue_enter(&object->memq, mem, vmp_listq);
1872 	object->memq_hint = mem;
1873 	mem->vmp_tabled = TRUE;
1874 
1875 	/*
1876 	 *	Show that the object has one more resident page.
1877 	 */
1878 
1879 	object->resident_page_count++;
1880 	if (VM_PAGE_WIRED(mem)) {
1881 		assert(mem->vmp_wire_count > 0);
1882 		VM_OBJECT_WIRED_PAGE_UPDATE_START(object);
1883 		VM_OBJECT_WIRED_PAGE_ADD(object, mem);
1884 		VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag);
1885 	}
1886 	assert(object->resident_page_count >= object->wired_page_count);
1887 
1888 #if DEVELOPMENT || DEBUG
1889 	if (object->object_is_shared_cache &&
1890 	    object->pager != NULL &&
1891 	    object->pager->mo_pager_ops == &shared_region_pager_ops) {
1892 		int new, old;
1893 		assert(!object->internal);
1894 		new = OSAddAtomic(+1, &shared_region_pagers_resident_count);
1895 		do {
1896 			old = shared_region_pagers_resident_peak;
1897 		} while (old < new &&
1898 		    !OSCompareAndSwap(old, new, &shared_region_pagers_resident_peak));
1899 	}
1900 #endif /* DEVELOPMENT || DEBUG */
1901 
1902 	if (batch_accounting == FALSE) {
1903 		if (object->internal) {
1904 			OSAddAtomic(1, &vm_page_internal_count);
1905 		} else {
1906 			OSAddAtomic(1, &vm_page_external_count);
1907 		}
1908 	}
1909 
1910 	/*
1911 	 * It wouldn't make sense to insert a "reusable" page in
1912 	 * an object (the page would have been marked "reusable" only
1913 	 * at the time of a madvise(MADV_FREE_REUSABLE) if it was already
1914 	 * in the object at that time).
1915 	 * But a page could be inserted in a "all_reusable" object, if
1916 	 * something faults it in (a vm_read() from another task or a
1917 	 * "use-after-free" issue in user space, for example).  It can
1918 	 * also happen if we're relocating a page from that object to
1919 	 * a different physical page during a physically-contiguous
1920 	 * allocation.
1921 	 */
1922 	assert(!mem->vmp_reusable);
1923 	if (object->all_reusable) {
1924 		OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count);
1925 	}
1926 
1927 	if (object->purgable == VM_PURGABLE_DENY &&
1928 	    !object->vo_ledger_tag) {
1929 		owner = TASK_NULL;
1930 	} else {
1931 		owner = VM_OBJECT_OWNER(object);
1932 		vm_object_ledger_tag_ledgers(object,
1933 		    &ledger_idx_volatile,
1934 		    &ledger_idx_nonvolatile,
1935 		    &ledger_idx_volatile_compressed,
1936 		    &ledger_idx_nonvolatile_compressed,
1937 		    &do_footprint);
1938 	}
1939 	if (owner &&
1940 	    (object->purgable == VM_PURGABLE_NONVOLATILE ||
1941 	    object->purgable == VM_PURGABLE_DENY ||
1942 	    VM_PAGE_WIRED(mem))) {
1943 		if (delayed_ledger_update) {
1944 			*delayed_ledger_update += PAGE_SIZE;
1945 		} else {
1946 			/* more non-volatile bytes */
1947 			ledger_credit(owner->ledger,
1948 			    ledger_idx_nonvolatile,
1949 			    PAGE_SIZE);
1950 			if (do_footprint) {
1951 				/* more footprint */
1952 				ledger_credit(owner->ledger,
1953 				    task_ledgers.phys_footprint,
1954 				    PAGE_SIZE);
1955 			}
1956 		}
1957 	} else if (owner &&
1958 	    (object->purgable == VM_PURGABLE_VOLATILE ||
1959 	    object->purgable == VM_PURGABLE_EMPTY)) {
1960 		assert(!VM_PAGE_WIRED(mem));
1961 		/* more volatile bytes */
1962 		ledger_credit(owner->ledger,
1963 		    ledger_idx_volatile,
1964 		    PAGE_SIZE);
1965 	}
1966 
1967 	if (object->purgable == VM_PURGABLE_VOLATILE) {
1968 		if (VM_PAGE_WIRED(mem)) {
1969 			OSAddAtomic(+1, &vm_page_purgeable_wired_count);
1970 		} else {
1971 			OSAddAtomic(+1, &vm_page_purgeable_count);
1972 		}
1973 	} else if (object->purgable == VM_PURGABLE_EMPTY &&
1974 	    mem->vmp_q_state == VM_PAGE_ON_THROTTLED_Q) {
1975 		/*
1976 		 * This page belongs to a purged VM object but hasn't
1977 		 * been purged (because it was "busy").
1978 		 * It's in the "throttled" queue and hence not
1979 		 * visible to vm_pageout_scan().  Move it to a pageable
1980 		 * queue, so that it can eventually be reclaimed, instead
1981 		 * of lingering in the "empty" object.
1982 		 */
1983 		if (queues_lock_held == FALSE) {
1984 			vm_page_lockspin_queues();
1985 		}
1986 		vm_page_deactivate(mem);
1987 		if (queues_lock_held == FALSE) {
1988 			vm_page_unlock_queues();
1989 		}
1990 	}
1991 
1992 #if VM_OBJECT_TRACKING_OP_MODIFIED
1993 	if (vm_object_tracking_btlog &&
1994 	    object->internal &&
1995 	    object->resident_page_count == 0 &&
1996 	    object->pager == NULL &&
1997 	    object->shadow != NULL &&
1998 	    object->shadow->copy == object) {
1999 		btlog_record(vm_object_tracking_btlog, object,
2000 		    VM_OBJECT_TRACKING_OP_MODIFIED,
2001 		    btref_get(__builtin_frame_address(0), 0));
2002 	}
2003 #endif /* VM_OBJECT_TRACKING_OP_MODIFIED */
2004 }
2005 
2006 /*
2007  *	vm_page_replace:
2008  *
2009  *	Exactly like vm_page_insert, except that we first
2010  *	remove any existing page at the given offset in object.
2011  *
2012  *	The object must be locked.
2013  */
2014 void
vm_page_replace(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)2015 vm_page_replace(
2016 	vm_page_t               mem,
2017 	vm_object_t             object,
2018 	vm_object_offset_t      offset)
2019 {
2020 	vm_page_bucket_t *bucket;
2021 	vm_page_t        found_m = VM_PAGE_NULL;
2022 	lck_spin_t      *bucket_lock;
2023 	int             hash_id;
2024 
2025 #if 0
2026 	/*
2027 	 * we don't hold the page queue lock
2028 	 * so this check isn't safe to make
2029 	 */
2030 	VM_PAGE_CHECK(mem);
2031 #endif
2032 	vm_object_lock_assert_exclusive(object);
2033 #if DEBUG || VM_PAGE_BUCKETS_CHECK
2034 	if (mem->vmp_tabled || mem->vmp_object) {
2035 		panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) "
2036 		    "already in (obj=%p,off=0x%llx)",
2037 		    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
2038 	}
2039 #endif
2040 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2041 
2042 	assert(!VM_PAGE_PAGEABLE(mem));
2043 
2044 	/*
2045 	 *	Record the object/offset pair in this page
2046 	 */
2047 	mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
2048 	mem->vmp_offset = offset;
2049 
2050 	/*
2051 	 *	Insert it into the object_object/offset hash table,
2052 	 *	replacing any page that might have been there.
2053 	 */
2054 
2055 	hash_id = vm_page_hash(object, offset);
2056 	bucket = &vm_page_buckets[hash_id];
2057 	bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2058 
2059 	lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2060 
2061 	if (bucket->page_list) {
2062 		vm_page_packed_t *mp = &bucket->page_list;
2063 		vm_page_t m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp));
2064 
2065 		do {
2066 			/*
2067 			 * compare packed object pointers
2068 			 */
2069 			if (m->vmp_object == mem->vmp_object && m->vmp_offset == offset) {
2070 				/*
2071 				 * Remove old page from hash list
2072 				 */
2073 				*mp = m->vmp_next_m;
2074 				m->vmp_hashed = FALSE;
2075 				m->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2076 
2077 				found_m = m;
2078 				break;
2079 			}
2080 			mp = &m->vmp_next_m;
2081 		} while ((m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp))));
2082 
2083 		mem->vmp_next_m = bucket->page_list;
2084 	} else {
2085 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2086 	}
2087 	/*
2088 	 * insert new page at head of hash list
2089 	 */
2090 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
2091 	mem->vmp_hashed = TRUE;
2092 
2093 	lck_spin_unlock(bucket_lock);
2094 
2095 	if (found_m) {
2096 		/*
2097 		 * there was already a page at the specified
2098 		 * offset for this object... remove it from
2099 		 * the object and free it back to the free list
2100 		 */
2101 		vm_page_free_unlocked(found_m, FALSE);
2102 	}
2103 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL);
2104 }
2105 
2106 /*
2107  *	vm_page_remove:		[ internal use only ]
2108  *
2109  *	Removes the given mem entry from the object/offset-page
2110  *	table and the object page list.
2111  *
2112  *	The object must be locked.
2113  */
2114 
2115 void
vm_page_remove(vm_page_t mem,boolean_t remove_from_hash)2116 vm_page_remove(
2117 	vm_page_t       mem,
2118 	boolean_t       remove_from_hash)
2119 {
2120 	vm_page_bucket_t *bucket;
2121 	vm_page_t       this;
2122 	lck_spin_t      *bucket_lock;
2123 	int             hash_id;
2124 	task_t          owner;
2125 	vm_object_t     m_object;
2126 	int             ledger_idx_volatile;
2127 	int             ledger_idx_nonvolatile;
2128 	int             ledger_idx_volatile_compressed;
2129 	int             ledger_idx_nonvolatile_compressed;
2130 	int             do_footprint;
2131 
2132 	m_object = VM_PAGE_OBJECT(mem);
2133 
2134 	vm_object_lock_assert_exclusive(m_object);
2135 	assert(mem->vmp_tabled);
2136 	assert(!mem->vmp_cleaning);
2137 	assert(!mem->vmp_laundry);
2138 
2139 	if (VM_PAGE_PAGEABLE(mem)) {
2140 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2141 	}
2142 #if 0
2143 	/*
2144 	 * we don't hold the page queue lock
2145 	 * so this check isn't safe to make
2146 	 */
2147 	VM_PAGE_CHECK(mem);
2148 #endif
2149 	if (remove_from_hash == TRUE) {
2150 		/*
2151 		 *	Remove from the object_object/offset hash table
2152 		 */
2153 		hash_id = vm_page_hash(m_object, mem->vmp_offset);
2154 		bucket = &vm_page_buckets[hash_id];
2155 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2156 
2157 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2158 
2159 		if ((this = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))) == mem) {
2160 			/* optimize for common case */
2161 
2162 			bucket->page_list = mem->vmp_next_m;
2163 		} else {
2164 			vm_page_packed_t        *prev;
2165 
2166 			for (prev = &this->vmp_next_m;
2167 			    (this = (vm_page_t)(VM_PAGE_UNPACK_PTR(*prev))) != mem;
2168 			    prev = &this->vmp_next_m) {
2169 				continue;
2170 			}
2171 			*prev = this->vmp_next_m;
2172 		}
2173 #if     MACH_PAGE_HASH_STATS
2174 		bucket->cur_count--;
2175 #endif /* MACH_PAGE_HASH_STATS */
2176 		mem->vmp_hashed = FALSE;
2177 		this->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2178 		lck_spin_unlock(bucket_lock);
2179 	}
2180 	/*
2181 	 *	Now remove from the object's list of backed pages.
2182 	 */
2183 
2184 	vm_page_remove_internal(mem);
2185 
2186 	/*
2187 	 *	And show that the object has one fewer resident
2188 	 *	page.
2189 	 */
2190 
2191 	assert(m_object->resident_page_count > 0);
2192 	m_object->resident_page_count--;
2193 
2194 #if DEVELOPMENT || DEBUG
2195 	if (m_object->object_is_shared_cache &&
2196 	    m_object->pager != NULL &&
2197 	    m_object->pager->mo_pager_ops == &shared_region_pager_ops) {
2198 		assert(!m_object->internal);
2199 		OSAddAtomic(-1, &shared_region_pagers_resident_count);
2200 	}
2201 #endif /* DEVELOPMENT || DEBUG */
2202 
2203 	if (m_object->internal) {
2204 #if DEBUG
2205 		assert(vm_page_internal_count);
2206 #endif /* DEBUG */
2207 
2208 		OSAddAtomic(-1, &vm_page_internal_count);
2209 	} else {
2210 		assert(vm_page_external_count);
2211 		OSAddAtomic(-1, &vm_page_external_count);
2212 
2213 		if (mem->vmp_xpmapped) {
2214 			assert(vm_page_xpmapped_external_count);
2215 			OSAddAtomic(-1, &vm_page_xpmapped_external_count);
2216 		}
2217 	}
2218 	if (!m_object->internal &&
2219 	    m_object->cached_list.next &&
2220 	    m_object->cached_list.prev) {
2221 		if (m_object->resident_page_count == 0) {
2222 			vm_object_cache_remove(m_object);
2223 		}
2224 	}
2225 
2226 	if (VM_PAGE_WIRED(mem)) {
2227 		assert(mem->vmp_wire_count > 0);
2228 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
2229 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
2230 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
2231 	}
2232 	assert(m_object->resident_page_count >=
2233 	    m_object->wired_page_count);
2234 	if (mem->vmp_reusable) {
2235 		assert(m_object->reusable_page_count > 0);
2236 		m_object->reusable_page_count--;
2237 		assert(m_object->reusable_page_count <=
2238 		    m_object->resident_page_count);
2239 		mem->vmp_reusable = FALSE;
2240 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2241 		vm_page_stats_reusable.reused_remove++;
2242 	} else if (m_object->all_reusable) {
2243 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2244 		vm_page_stats_reusable.reused_remove++;
2245 	}
2246 
2247 	if (m_object->purgable == VM_PURGABLE_DENY &&
2248 	    !m_object->vo_ledger_tag) {
2249 		owner = TASK_NULL;
2250 	} else {
2251 		owner = VM_OBJECT_OWNER(m_object);
2252 		vm_object_ledger_tag_ledgers(m_object,
2253 		    &ledger_idx_volatile,
2254 		    &ledger_idx_nonvolatile,
2255 		    &ledger_idx_volatile_compressed,
2256 		    &ledger_idx_nonvolatile_compressed,
2257 		    &do_footprint);
2258 	}
2259 	if (owner &&
2260 	    (m_object->purgable == VM_PURGABLE_NONVOLATILE ||
2261 	    m_object->purgable == VM_PURGABLE_DENY ||
2262 	    VM_PAGE_WIRED(mem))) {
2263 		/* less non-volatile bytes */
2264 		ledger_debit(owner->ledger,
2265 		    ledger_idx_nonvolatile,
2266 		    PAGE_SIZE);
2267 		if (do_footprint) {
2268 			/* less footprint */
2269 			ledger_debit(owner->ledger,
2270 			    task_ledgers.phys_footprint,
2271 			    PAGE_SIZE);
2272 		}
2273 	} else if (owner &&
2274 	    (m_object->purgable == VM_PURGABLE_VOLATILE ||
2275 	    m_object->purgable == VM_PURGABLE_EMPTY)) {
2276 		assert(!VM_PAGE_WIRED(mem));
2277 		/* less volatile bytes */
2278 		ledger_debit(owner->ledger,
2279 		    ledger_idx_volatile,
2280 		    PAGE_SIZE);
2281 	}
2282 	if (m_object->purgable == VM_PURGABLE_VOLATILE) {
2283 		if (VM_PAGE_WIRED(mem)) {
2284 			assert(vm_page_purgeable_wired_count > 0);
2285 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
2286 		} else {
2287 			assert(vm_page_purgeable_count > 0);
2288 			OSAddAtomic(-1, &vm_page_purgeable_count);
2289 		}
2290 	}
2291 
2292 	if (m_object->set_cache_attr == TRUE) {
2293 		pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(mem), 0);
2294 	}
2295 
2296 	mem->vmp_tabled = FALSE;
2297 	mem->vmp_object = 0;
2298 	mem->vmp_offset = (vm_object_offset_t) -1;
2299 }
2300 
2301 
2302 /*
2303  *	vm_page_lookup:
2304  *
2305  *	Returns the page associated with the object/offset
2306  *	pair specified; if none is found, VM_PAGE_NULL is returned.
2307  *
2308  *	The object must be locked.  No side effects.
2309  */
2310 
2311 #define VM_PAGE_HASH_LOOKUP_THRESHOLD   10
2312 
2313 #if DEBUG_VM_PAGE_LOOKUP
2314 
2315 struct {
2316 	uint64_t        vpl_total;
2317 	uint64_t        vpl_empty_obj;
2318 	uint64_t        vpl_bucket_NULL;
2319 	uint64_t        vpl_hit_hint;
2320 	uint64_t        vpl_hit_hint_next;
2321 	uint64_t        vpl_hit_hint_prev;
2322 	uint64_t        vpl_fast;
2323 	uint64_t        vpl_slow;
2324 	uint64_t        vpl_hit;
2325 	uint64_t        vpl_miss;
2326 
2327 	uint64_t        vpl_fast_elapsed;
2328 	uint64_t        vpl_slow_elapsed;
2329 } vm_page_lookup_stats __attribute__((aligned(8)));
2330 
2331 #endif
2332 
2333 #define KDP_VM_PAGE_WALK_MAX    1000
2334 
2335 vm_page_t
kdp_vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2336 kdp_vm_page_lookup(
2337 	vm_object_t             object,
2338 	vm_object_offset_t      offset)
2339 {
2340 	vm_page_t cur_page;
2341 	int num_traversed = 0;
2342 
2343 	if (not_in_kdp) {
2344 		panic("panic: kdp_vm_page_lookup done outside of kernel debugger");
2345 	}
2346 
2347 	vm_page_queue_iterate(&object->memq, cur_page, vmp_listq) {
2348 		if (cur_page->vmp_offset == offset) {
2349 			return cur_page;
2350 		}
2351 		num_traversed++;
2352 
2353 		if (num_traversed >= KDP_VM_PAGE_WALK_MAX) {
2354 			return VM_PAGE_NULL;
2355 		}
2356 	}
2357 
2358 	return VM_PAGE_NULL;
2359 }
2360 
2361 vm_page_t
vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2362 vm_page_lookup(
2363 	vm_object_t             object,
2364 	vm_object_offset_t      offset)
2365 {
2366 	vm_page_t       mem;
2367 	vm_page_bucket_t *bucket;
2368 	vm_page_queue_entry_t   qe;
2369 	lck_spin_t      *bucket_lock = NULL;
2370 	int             hash_id;
2371 #if DEBUG_VM_PAGE_LOOKUP
2372 	uint64_t        start, elapsed;
2373 
2374 	OSAddAtomic64(1, &vm_page_lookup_stats.vpl_total);
2375 #endif
2376 
2377 #if CONFIG_KERNEL_TBI
2378 	if (VM_KERNEL_ADDRESS(offset)) {
2379 		offset = VM_KERNEL_STRIP_UPTR(offset);
2380 	}
2381 #endif /* CONFIG_KERNEL_TBI */
2382 
2383 	vm_object_lock_assert_held(object);
2384 	assertf(page_aligned(offset), "offset 0x%llx\n", offset);
2385 
2386 	if (object->resident_page_count == 0) {
2387 #if DEBUG_VM_PAGE_LOOKUP
2388 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_empty_obj);
2389 #endif
2390 		return VM_PAGE_NULL;
2391 	}
2392 
2393 	mem = object->memq_hint;
2394 
2395 	if (mem != VM_PAGE_NULL) {
2396 		assert(VM_PAGE_OBJECT(mem) == object);
2397 
2398 		if (mem->vmp_offset == offset) {
2399 #if DEBUG_VM_PAGE_LOOKUP
2400 			OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint);
2401 #endif
2402 			return mem;
2403 		}
2404 		qe = (vm_page_queue_entry_t)vm_page_queue_next(&mem->vmp_listq);
2405 
2406 		if (!vm_page_queue_end(&object->memq, qe)) {
2407 			vm_page_t       next_page;
2408 
2409 			next_page = (vm_page_t)((uintptr_t)qe);
2410 			assert(VM_PAGE_OBJECT(next_page) == object);
2411 
2412 			if (next_page->vmp_offset == offset) {
2413 				object->memq_hint = next_page; /* new hint */
2414 #if DEBUG_VM_PAGE_LOOKUP
2415 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next);
2416 #endif
2417 				return next_page;
2418 			}
2419 		}
2420 		qe = (vm_page_queue_entry_t)vm_page_queue_prev(&mem->vmp_listq);
2421 
2422 		if (!vm_page_queue_end(&object->memq, qe)) {
2423 			vm_page_t prev_page;
2424 
2425 			prev_page = (vm_page_t)((uintptr_t)qe);
2426 			assert(VM_PAGE_OBJECT(prev_page) == object);
2427 
2428 			if (prev_page->vmp_offset == offset) {
2429 				object->memq_hint = prev_page; /* new hint */
2430 #if DEBUG_VM_PAGE_LOOKUP
2431 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev);
2432 #endif
2433 				return prev_page;
2434 			}
2435 		}
2436 	}
2437 	/*
2438 	 * Search the hash table for this object/offset pair
2439 	 */
2440 	hash_id = vm_page_hash(object, offset);
2441 	bucket = &vm_page_buckets[hash_id];
2442 
2443 	/*
2444 	 * since we hold the object lock, we are guaranteed that no
2445 	 * new pages can be inserted into this object... this in turn
2446 	 * guarantess that the page we're looking for can't exist
2447 	 * if the bucket it hashes to is currently NULL even when looked
2448 	 * at outside the scope of the hash bucket lock... this is a
2449 	 * really cheap optimiztion to avoid taking the lock
2450 	 */
2451 	if (!bucket->page_list) {
2452 #if DEBUG_VM_PAGE_LOOKUP
2453 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL);
2454 #endif
2455 		return VM_PAGE_NULL;
2456 	}
2457 
2458 #if DEBUG_VM_PAGE_LOOKUP
2459 	start = mach_absolute_time();
2460 #endif
2461 	if (object->resident_page_count <= VM_PAGE_HASH_LOOKUP_THRESHOLD) {
2462 		/*
2463 		 * on average, it's roughly 3 times faster to run a short memq list
2464 		 * than to take the spin lock and go through the hash list
2465 		 */
2466 		mem = (vm_page_t)vm_page_queue_first(&object->memq);
2467 
2468 		while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2469 			if (mem->vmp_offset == offset) {
2470 				break;
2471 			}
2472 
2473 			mem = (vm_page_t)vm_page_queue_next(&mem->vmp_listq);
2474 		}
2475 		if (vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2476 			mem = NULL;
2477 		}
2478 	} else {
2479 		vm_page_object_t        packed_object;
2480 
2481 		packed_object = VM_PAGE_PACK_OBJECT(object);
2482 
2483 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2484 
2485 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2486 
2487 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
2488 		    mem != VM_PAGE_NULL;
2489 		    mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m))) {
2490 #if 0
2491 			/*
2492 			 * we don't hold the page queue lock
2493 			 * so this check isn't safe to make
2494 			 */
2495 			VM_PAGE_CHECK(mem);
2496 #endif
2497 			if ((mem->vmp_object == packed_object) && (mem->vmp_offset == offset)) {
2498 				break;
2499 			}
2500 		}
2501 		lck_spin_unlock(bucket_lock);
2502 	}
2503 
2504 #if DEBUG_VM_PAGE_LOOKUP
2505 	elapsed = mach_absolute_time() - start;
2506 
2507 	if (bucket_lock) {
2508 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_slow);
2509 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_slow_elapsed);
2510 	} else {
2511 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_fast);
2512 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_fast_elapsed);
2513 	}
2514 	if (mem != VM_PAGE_NULL) {
2515 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit);
2516 	} else {
2517 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss);
2518 	}
2519 #endif
2520 	if (mem != VM_PAGE_NULL) {
2521 		assert(VM_PAGE_OBJECT(mem) == object);
2522 
2523 		object->memq_hint = mem;
2524 	}
2525 	return mem;
2526 }
2527 
2528 
2529 /*
2530  *	vm_page_rename:
2531  *
2532  *	Move the given memory entry from its
2533  *	current object to the specified target object/offset.
2534  *
2535  *	The object must be locked.
2536  */
2537 void
vm_page_rename(vm_page_t mem,vm_object_t new_object,vm_object_offset_t new_offset)2538 vm_page_rename(
2539 	vm_page_t               mem,
2540 	vm_object_t             new_object,
2541 	vm_object_offset_t      new_offset)
2542 {
2543 	boolean_t       internal_to_external, external_to_internal;
2544 	vm_tag_t        tag;
2545 	vm_object_t     m_object;
2546 
2547 	m_object = VM_PAGE_OBJECT(mem);
2548 
2549 	assert(m_object != new_object);
2550 	assert(m_object);
2551 
2552 	/*
2553 	 *	Changes to mem->vmp_object require the page lock because
2554 	 *	the pageout daemon uses that lock to get the object.
2555 	 */
2556 	vm_page_lockspin_queues();
2557 
2558 	internal_to_external = FALSE;
2559 	external_to_internal = FALSE;
2560 
2561 	if (mem->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q) {
2562 		/*
2563 		 * it's much easier to get the vm_page_pageable_xxx accounting correct
2564 		 * if we first move the page to the active queue... it's going to end
2565 		 * up there anyway, and we don't do vm_page_rename's frequently enough
2566 		 * for this to matter.
2567 		 */
2568 		vm_page_queues_remove(mem, FALSE);
2569 		vm_page_activate(mem);
2570 	}
2571 	if (VM_PAGE_PAGEABLE(mem)) {
2572 		if (m_object->internal && !new_object->internal) {
2573 			internal_to_external = TRUE;
2574 		}
2575 		if (!m_object->internal && new_object->internal) {
2576 			external_to_internal = TRUE;
2577 		}
2578 	}
2579 
2580 	tag = m_object->wire_tag;
2581 	vm_page_remove(mem, TRUE);
2582 	vm_page_insert_internal(mem, new_object, new_offset, tag, TRUE, TRUE, FALSE, FALSE, NULL);
2583 
2584 	if (internal_to_external) {
2585 		vm_page_pageable_internal_count--;
2586 		vm_page_pageable_external_count++;
2587 	} else if (external_to_internal) {
2588 		vm_page_pageable_external_count--;
2589 		vm_page_pageable_internal_count++;
2590 	}
2591 
2592 	vm_page_unlock_queues();
2593 }
2594 
2595 /*
2596  *	vm_page_init:
2597  *
2598  *	Initialize the fields in a new page.
2599  *	This takes a structure with random values and initializes it
2600  *	so that it can be given to vm_page_release or vm_page_insert.
2601  */
2602 void
vm_page_init(vm_page_t mem,ppnum_t phys_page,boolean_t lopage)2603 vm_page_init(
2604 	vm_page_t mem,
2605 	ppnum_t   phys_page,
2606 	boolean_t lopage)
2607 {
2608 	uint_t    i;
2609 	uintptr_t *p;
2610 
2611 	assert(phys_page);
2612 
2613 #if DEBUG
2614 	if ((phys_page != vm_page_fictitious_addr) && (phys_page != vm_page_guard_addr)) {
2615 		if (!(pmap_valid_page(phys_page))) {
2616 			panic("vm_page_init: non-DRAM phys_page 0x%x", phys_page);
2617 		}
2618 	}
2619 #endif /* DEBUG */
2620 
2621 	/*
2622 	 * Initialize the fields of the vm_page. If adding any new fields to vm_page,
2623 	 * try to use initial values which match 0. This minimizes the number of writes
2624 	 * needed for boot-time initialization.
2625 	 *
2626 	 * Kernel bzero() isn't an inline yet, so do it by hand for performance.
2627 	 */
2628 	assert(VM_PAGE_NOT_ON_Q == 0);
2629 	assert(sizeof(*mem) % sizeof(uintptr_t) == 0);
2630 	for (p = (uintptr_t *)(void *)mem, i = sizeof(*mem) / sizeof(uintptr_t); i != 0; --i) {
2631 		*p++ = 0;
2632 	}
2633 	mem->vmp_offset = (vm_object_offset_t)-1;
2634 	mem->vmp_busy = TRUE;
2635 	mem->vmp_lopage = lopage;
2636 
2637 	VM_PAGE_SET_PHYS_PAGE(mem, phys_page);
2638 #if 0
2639 	/*
2640 	 * we're leaving this turned off for now... currently pages
2641 	 * come off the free list and are either immediately dirtied/referenced
2642 	 * due to zero-fill or COW faults, or are used to read or write files...
2643 	 * in the file I/O case, the UPL mechanism takes care of clearing
2644 	 * the state of the HW ref/mod bits in a somewhat fragile way.
2645 	 * Since we may change the way this works in the future (to toughen it up),
2646 	 * I'm leaving this as a reminder of where these bits could get cleared
2647 	 */
2648 
2649 	/*
2650 	 * make sure both the h/w referenced and modified bits are
2651 	 * clear at this point... we are especially dependent on
2652 	 * not finding a 'stale' h/w modified in a number of spots
2653 	 * once this page goes back into use
2654 	 */
2655 	pmap_clear_refmod(phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
2656 #endif
2657 }
2658 
2659 /*
2660  *	vm_page_grab_fictitious:
2661  *
2662  *	Remove a fictitious page from the free list.
2663  *	Returns VM_PAGE_NULL if there are no free pages.
2664  */
2665 
2666 static vm_page_t
vm_page_grab_fictitious_common(ppnum_t phys_addr,boolean_t canwait)2667 vm_page_grab_fictitious_common(ppnum_t phys_addr, boolean_t canwait)
2668 {
2669 	vm_page_t m;
2670 
2671 	m = zalloc_flags(vm_page_zone, canwait ? Z_WAITOK : Z_NOWAIT);
2672 	if (m) {
2673 		vm_page_init(m, phys_addr, FALSE);
2674 		m->vmp_fictitious = TRUE;
2675 	}
2676 	return m;
2677 }
2678 
2679 vm_page_t
vm_page_grab_fictitious(boolean_t canwait)2680 vm_page_grab_fictitious(boolean_t canwait)
2681 {
2682 	return vm_page_grab_fictitious_common(vm_page_fictitious_addr, canwait);
2683 }
2684 
2685 int vm_guard_count;
2686 
2687 
2688 vm_page_t
vm_page_grab_guard(boolean_t canwait)2689 vm_page_grab_guard(boolean_t canwait)
2690 {
2691 	vm_page_t page;
2692 	page = vm_page_grab_fictitious_common(vm_page_guard_addr, canwait);
2693 	if (page) {
2694 		OSAddAtomic(1, &vm_guard_count);
2695 	}
2696 	return page;
2697 }
2698 
2699 
2700 /*
2701  *	vm_page_release_fictitious:
2702  *
2703  *	Release a fictitious page to the zone pool
2704  */
2705 void
vm_page_release_fictitious(vm_page_t m)2706 vm_page_release_fictitious(
2707 	vm_page_t m)
2708 {
2709 	assert((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || (m->vmp_q_state == VM_PAGE_IS_WIRED));
2710 	assert(m->vmp_fictitious);
2711 	assert(VM_PAGE_GET_PHYS_PAGE(m) == vm_page_fictitious_addr ||
2712 	    VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr);
2713 
2714 
2715 	if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) {
2716 		OSAddAtomic(-1, &vm_guard_count);
2717 	}
2718 
2719 	zfree(vm_page_zone, m);
2720 }
2721 
2722 /*
2723  *	vm_pool_low():
2724  *
2725  *	Return true if it is not likely that a non-vm_privileged thread
2726  *	can get memory without blocking.  Advisory only, since the
2727  *	situation may change under us.
2728  */
2729 bool
vm_pool_low(void)2730 vm_pool_low(void)
2731 {
2732 	/* No locking, at worst we will fib. */
2733 	return vm_page_free_count <= vm_page_free_reserved;
2734 }
2735 
2736 boolean_t vm_darkwake_mode = FALSE;
2737 
2738 /*
2739  * vm_update_darkwake_mode():
2740  *
2741  * Tells the VM that the system is in / out of darkwake.
2742  *
2743  * Today, the VM only lowers/raises the background queue target
2744  * so as to favor consuming more/less background pages when
2745  * darwake is ON/OFF.
2746  *
2747  * We might need to do more things in the future.
2748  */
2749 
2750 void
vm_update_darkwake_mode(boolean_t darkwake_mode)2751 vm_update_darkwake_mode(boolean_t darkwake_mode)
2752 {
2753 #if XNU_TARGET_OS_OSX && defined(__arm64__)
2754 #pragma unused(darkwake_mode)
2755 	assert(vm_darkwake_mode == FALSE);
2756 	/*
2757 	 * Darkwake mode isn't supported for AS macOS.
2758 	 */
2759 	return;
2760 #else /* XNU_TARGET_OS_OSX && __arm64__ */
2761 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2762 
2763 	vm_page_lockspin_queues();
2764 
2765 	if (vm_darkwake_mode == darkwake_mode) {
2766 		/*
2767 		 * No change.
2768 		 */
2769 		vm_page_unlock_queues();
2770 		return;
2771 	}
2772 
2773 	vm_darkwake_mode = darkwake_mode;
2774 
2775 	if (vm_darkwake_mode == TRUE) {
2776 		/* save background target to restore later */
2777 		vm_page_background_target_snapshot = vm_page_background_target;
2778 
2779 		/* target is set to 0...no protection for background pages */
2780 		vm_page_background_target = 0;
2781 	} else if (vm_darkwake_mode == FALSE) {
2782 		if (vm_page_background_target_snapshot) {
2783 			vm_page_background_target = vm_page_background_target_snapshot;
2784 		}
2785 	}
2786 	vm_page_unlock_queues();
2787 #endif
2788 }
2789 
2790 void
vm_page_update_special_state(vm_page_t mem)2791 vm_page_update_special_state(vm_page_t mem)
2792 {
2793 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR || mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
2794 		return;
2795 	}
2796 
2797 	int mode = mem->vmp_on_specialq;
2798 
2799 	switch (mode) {
2800 	case VM_PAGE_SPECIAL_Q_BG:
2801 	{
2802 		task_t  my_task = current_task_early();
2803 
2804 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2805 			return;
2806 		}
2807 
2808 		if (my_task) {
2809 			if (task_get_darkwake_mode(my_task)) {
2810 				return;
2811 			}
2812 		}
2813 
2814 		if (my_task) {
2815 			if (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG)) {
2816 				return;
2817 			}
2818 		}
2819 		vm_page_lockspin_queues();
2820 
2821 		vm_page_background_promoted_count++;
2822 
2823 		vm_page_remove_from_specialq(mem);
2824 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2825 
2826 		vm_page_unlock_queues();
2827 		break;
2828 	}
2829 
2830 	case VM_PAGE_SPECIAL_Q_DONATE:
2831 	{
2832 		task_t  my_task = current_task_early();
2833 
2834 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2835 			return;
2836 		}
2837 
2838 		if (my_task->donates_own_pages == false) {
2839 			vm_page_lockspin_queues();
2840 
2841 			vm_page_remove_from_specialq(mem);
2842 			mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2843 
2844 			vm_page_unlock_queues();
2845 		}
2846 		break;
2847 	}
2848 
2849 	default:
2850 	{
2851 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2852 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2853 		break;
2854 	}
2855 	}
2856 }
2857 
2858 
2859 void
vm_page_assign_special_state(vm_page_t mem,int mode)2860 vm_page_assign_special_state(vm_page_t mem, int mode)
2861 {
2862 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
2863 		return;
2864 	}
2865 
2866 	switch (mode) {
2867 	case VM_PAGE_SPECIAL_Q_BG:
2868 	{
2869 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2870 			return;
2871 		}
2872 
2873 		task_t  my_task = current_task_early();
2874 
2875 		if (my_task) {
2876 			if (task_get_darkwake_mode(my_task)) {
2877 				mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
2878 				return;
2879 			}
2880 		}
2881 
2882 		if (my_task) {
2883 			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);
2884 		}
2885 		break;
2886 	}
2887 
2888 	case VM_PAGE_SPECIAL_Q_DONATE:
2889 	{
2890 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2891 			return;
2892 		}
2893 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
2894 		break;
2895 	}
2896 
2897 	default:
2898 		break;
2899 	}
2900 }
2901 
2902 
2903 void
vm_page_remove_from_specialq(vm_page_t mem)2904 vm_page_remove_from_specialq(
2905 	vm_page_t       mem)
2906 {
2907 	vm_object_t     m_object;
2908 	unsigned short  mode;
2909 
2910 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2911 
2912 	mode = mem->vmp_on_specialq;
2913 
2914 	switch (mode) {
2915 	case VM_PAGE_SPECIAL_Q_BG:
2916 	{
2917 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2918 			vm_page_queue_remove(&vm_page_queue_background, mem, vmp_specialq);
2919 
2920 			mem->vmp_specialq.next = 0;
2921 			mem->vmp_specialq.prev = 0;
2922 
2923 			vm_page_background_count--;
2924 
2925 			m_object = VM_PAGE_OBJECT(mem);
2926 
2927 			if (m_object->internal) {
2928 				vm_page_background_internal_count--;
2929 			} else {
2930 				vm_page_background_external_count--;
2931 			}
2932 		}
2933 		break;
2934 	}
2935 
2936 	case VM_PAGE_SPECIAL_Q_DONATE:
2937 	{
2938 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2939 			vm_page_queue_remove((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
2940 			mem->vmp_specialq.next = 0;
2941 			mem->vmp_specialq.prev = 0;
2942 			vm_page_donate_count--;
2943 			if (vm_page_donate_queue_ripe && (vm_page_donate_count < vm_page_donate_target)) {
2944 				assert(vm_page_donate_target == vm_page_donate_target_low);
2945 				vm_page_donate_target = vm_page_donate_target_high;
2946 				vm_page_donate_queue_ripe = false;
2947 			}
2948 		}
2949 
2950 		break;
2951 	}
2952 
2953 	default:
2954 	{
2955 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2956 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2957 		break;
2958 	}
2959 	}
2960 }
2961 
2962 
2963 void
vm_page_add_to_specialq(vm_page_t mem,boolean_t first)2964 vm_page_add_to_specialq(
2965 	vm_page_t       mem,
2966 	boolean_t       first)
2967 {
2968 	vm_object_t     m_object;
2969 
2970 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2971 
2972 	if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2973 		return;
2974 	}
2975 
2976 	int mode = mem->vmp_on_specialq;
2977 
2978 	switch (mode) {
2979 	case VM_PAGE_SPECIAL_Q_BG:
2980 	{
2981 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2982 			return;
2983 		}
2984 
2985 		m_object = VM_PAGE_OBJECT(mem);
2986 
2987 		if (vm_page_background_exclude_external && !m_object->internal) {
2988 			return;
2989 		}
2990 
2991 		if (first == TRUE) {
2992 			vm_page_queue_enter_first(&vm_page_queue_background, mem, vmp_specialq);
2993 		} else {
2994 			vm_page_queue_enter(&vm_page_queue_background, mem, vmp_specialq);
2995 		}
2996 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
2997 
2998 		vm_page_background_count++;
2999 
3000 		if (m_object->internal) {
3001 			vm_page_background_internal_count++;
3002 		} else {
3003 			vm_page_background_external_count++;
3004 		}
3005 		break;
3006 	}
3007 
3008 	case VM_PAGE_SPECIAL_Q_DONATE:
3009 	{
3010 		if (first == TRUE) {
3011 			vm_page_queue_enter_first((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3012 		} else {
3013 			vm_page_queue_enter((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3014 		}
3015 		vm_page_donate_count++;
3016 		if (!vm_page_donate_queue_ripe && (vm_page_donate_count > vm_page_donate_target)) {
3017 			assert(vm_page_donate_target == vm_page_donate_target_high);
3018 			vm_page_donate_target = vm_page_donate_target_low;
3019 			vm_page_donate_queue_ripe = true;
3020 		}
3021 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
3022 		break;
3023 	}
3024 
3025 	default:
3026 		break;
3027 	}
3028 }
3029 
3030 /*
3031  * This can be switched to FALSE to help debug drivers
3032  * that are having problems with memory > 4G.
3033  */
3034 boolean_t       vm_himemory_mode = TRUE;
3035 
3036 /*
3037  * this interface exists to support hardware controllers
3038  * incapable of generating DMAs with more than 32 bits
3039  * of address on platforms with physical memory > 4G...
3040  */
3041 unsigned int    vm_lopages_allocated_q = 0;
3042 unsigned int    vm_lopages_allocated_cpm_success = 0;
3043 unsigned int    vm_lopages_allocated_cpm_failed = 0;
3044 vm_page_queue_head_t    vm_lopage_queue_free VM_PAGE_PACKED_ALIGNED;
3045 
3046 vm_page_t
vm_page_grablo(void)3047 vm_page_grablo(void)
3048 {
3049 	vm_page_t       mem;
3050 
3051 	if (vm_lopage_needed == FALSE) {
3052 		return vm_page_grab();
3053 	}
3054 
3055 	vm_free_page_lock_spin();
3056 
3057 	if (!vm_page_queue_empty(&vm_lopage_queue_free)) {
3058 		vm_page_queue_remove_first(&vm_lopage_queue_free, mem, vmp_pageq);
3059 		assert(vm_lopage_free_count);
3060 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
3061 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3062 
3063 		vm_lopage_free_count--;
3064 		vm_lopages_allocated_q++;
3065 
3066 		if (vm_lopage_free_count < vm_lopage_lowater) {
3067 			vm_lopage_refill = TRUE;
3068 		}
3069 
3070 		vm_free_page_unlock();
3071 
3072 		if (current_task()->donates_own_pages) {
3073 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3074 		} else {
3075 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3076 		}
3077 	} else {
3078 		vm_free_page_unlock();
3079 
3080 		if (cpm_allocate(PAGE_SIZE, &mem, atop(PPNUM_MAX), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) {
3081 			vm_free_page_lock_spin();
3082 			vm_lopages_allocated_cpm_failed++;
3083 			vm_free_page_unlock();
3084 
3085 			return VM_PAGE_NULL;
3086 		}
3087 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3088 
3089 		mem->vmp_busy = TRUE;
3090 
3091 		vm_page_lockspin_queues();
3092 
3093 		mem->vmp_gobbled = FALSE;
3094 		vm_page_gobble_count--;
3095 		vm_page_wire_count--;
3096 
3097 		vm_lopages_allocated_cpm_success++;
3098 		vm_page_unlock_queues();
3099 	}
3100 	assert(mem->vmp_busy);
3101 	assert(!mem->vmp_pmapped);
3102 	assert(!mem->vmp_wpmapped);
3103 	assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3104 
3105 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3106 
3107 	counter_inc(&vm_page_grab_count);
3108 	VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, 0, 1, 0, 0);
3109 
3110 	return mem;
3111 }
3112 
3113 /*
3114  *	vm_page_grab:
3115  *
3116  *	first try to grab a page from the per-cpu free list...
3117  *	this must be done while pre-emption is disabled... if
3118  *      a page is available, we're done...
3119  *	if no page is available, grab the vm_page_queue_free_lock
3120  *	and see if current number of free pages would allow us
3121  *      to grab at least 1... if not, return VM_PAGE_NULL as before...
3122  *	if there are pages available, disable preemption and
3123  *      recheck the state of the per-cpu free list... we could
3124  *	have been preempted and moved to a different cpu, or
3125  *      some other thread could have re-filled it... if still
3126  *	empty, figure out how many pages we can steal from the
3127  *	global free queue and move to the per-cpu queue...
3128  *	return 1 of these pages when done... only wakeup the
3129  *      pageout_scan thread if we moved pages from the global
3130  *	list... no need for the wakeup if we've satisfied the
3131  *	request from the per-cpu queue.
3132  */
3133 
3134 #if CONFIG_SECLUDED_MEMORY
3135 vm_page_t vm_page_grab_secluded(void);
3136 #endif /* CONFIG_SECLUDED_MEMORY */
3137 
3138 static inline void
3139 vm_page_grab_diags(void);
3140 
3141 vm_page_t
vm_page_grab(void)3142 vm_page_grab(void)
3143 {
3144 	return vm_page_grab_options(VM_PAGE_GRAB_OPTIONS_NONE);
3145 }
3146 
3147 #if HIBERNATION
3148 boolean_t       hibernate_rebuild_needed = FALSE;
3149 #endif /* HIBERNATION */
3150 
3151 vm_page_t
vm_page_grab_options(int grab_options)3152 vm_page_grab_options(
3153 	int grab_options)
3154 {
3155 	vm_page_t       mem;
3156 
3157 restart:
3158 	disable_preemption();
3159 
3160 	if ((mem = *PERCPU_GET(free_pages))) {
3161 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
3162 
3163 #if HIBERNATION
3164 		if (hibernate_rebuild_needed) {
3165 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3166 		}
3167 #endif /* HIBERNATION */
3168 
3169 		vm_page_grab_diags();
3170 
3171 		vm_offset_t pcpu_base = current_percpu_base();
3172 		counter_inc_preemption_disabled(&vm_page_grab_count);
3173 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = mem->vmp_snext;
3174 		VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3175 
3176 		VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3177 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3178 		enable_preemption();
3179 
3180 		assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3181 		assert(mem->vmp_tabled == FALSE);
3182 		assert(mem->vmp_object == 0);
3183 		assert(!mem->vmp_laundry);
3184 		ASSERT_PMAP_FREE(mem);
3185 		assert(mem->vmp_busy);
3186 		assert(!mem->vmp_pmapped);
3187 		assert(!mem->vmp_wpmapped);
3188 		assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3189 
3190 		task_t  cur_task = current_task_early();
3191 		if (cur_task && cur_task != kernel_task) {
3192 			if (cur_task->donates_own_pages) {
3193 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3194 			} else {
3195 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3196 			}
3197 		}
3198 		return mem;
3199 	}
3200 	enable_preemption();
3201 
3202 
3203 	/*
3204 	 *	Optionally produce warnings if the wire or gobble
3205 	 *	counts exceed some threshold.
3206 	 */
3207 #if VM_PAGE_WIRE_COUNT_WARNING
3208 	if (vm_page_wire_count >= VM_PAGE_WIRE_COUNT_WARNING) {
3209 		printf("mk: vm_page_grab(): high wired page count of %d\n",
3210 		    vm_page_wire_count);
3211 	}
3212 #endif
3213 #if VM_PAGE_GOBBLE_COUNT_WARNING
3214 	if (vm_page_gobble_count >= VM_PAGE_GOBBLE_COUNT_WARNING) {
3215 		printf("mk: vm_page_grab(): high gobbled page count of %d\n",
3216 		    vm_page_gobble_count);
3217 	}
3218 #endif
3219 
3220 	/*
3221 	 * If free count is low and we have delayed pages from early boot,
3222 	 * get one of those instead.
3223 	 */
3224 	if (__improbable(vm_delayed_count > 0 &&
3225 	    vm_page_free_count <= vm_page_free_target &&
3226 	    (mem = vm_get_delayed_page(grab_options)) != NULL)) {
3227 		return mem;
3228 	}
3229 
3230 	vm_free_page_lock_spin();
3231 
3232 	/*
3233 	 *	Only let privileged threads (involved in pageout)
3234 	 *	dip into the reserved pool.
3235 	 */
3236 	if ((vm_page_free_count < vm_page_free_reserved) &&
3237 	    !(current_thread()->options & TH_OPT_VMPRIV)) {
3238 		/* no page for us in the free queue... */
3239 		vm_free_page_unlock();
3240 		mem = VM_PAGE_NULL;
3241 
3242 #if CONFIG_SECLUDED_MEMORY
3243 		/* ... but can we try and grab from the secluded queue? */
3244 		if (vm_page_secluded_count > 0 &&
3245 		    ((grab_options & VM_PAGE_GRAB_SECLUDED) ||
3246 		    task_can_use_secluded_mem(current_task(), TRUE))) {
3247 			mem = vm_page_grab_secluded();
3248 			if (grab_options & VM_PAGE_GRAB_SECLUDED) {
3249 				vm_page_secluded.grab_for_iokit++;
3250 				if (mem) {
3251 					vm_page_secluded.grab_for_iokit_success++;
3252 				}
3253 			}
3254 			if (mem) {
3255 				VM_CHECK_MEMORYSTATUS;
3256 
3257 				vm_page_grab_diags();
3258 				counter_inc(&vm_page_grab_count);
3259 				VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3260 
3261 
3262 				return mem;
3263 			}
3264 		}
3265 #else /* CONFIG_SECLUDED_MEMORY */
3266 		(void) grab_options;
3267 #endif /* CONFIG_SECLUDED_MEMORY */
3268 	} else {
3269 		vm_page_t        head;
3270 		vm_page_t        tail;
3271 		unsigned int     pages_to_steal;
3272 		unsigned int     color;
3273 		unsigned int clump_end, sub_count;
3274 
3275 		while (vm_page_free_count == 0) {
3276 			vm_free_page_unlock();
3277 			/*
3278 			 * must be a privileged thread to be
3279 			 * in this state since a non-privileged
3280 			 * thread would have bailed if we were
3281 			 * under the vm_page_free_reserved mark
3282 			 */
3283 			VM_PAGE_WAIT();
3284 			vm_free_page_lock_spin();
3285 		}
3286 
3287 		/*
3288 		 * Need to repopulate the per-CPU free list from the global free list.
3289 		 * Note we don't do any processing of pending retirement pages here.
3290 		 * That'll happen in the code above when the page comes off the per-CPU list.
3291 		 */
3292 		disable_preemption();
3293 
3294 		/*
3295 		 * If we got preempted the cache might now have pages.
3296 		 */
3297 		if ((mem = *PERCPU_GET(free_pages))) {
3298 			vm_free_page_unlock();
3299 			enable_preemption();
3300 			goto restart;
3301 		}
3302 
3303 		if (vm_page_free_count <= vm_page_free_reserved) {
3304 			pages_to_steal = 1;
3305 		} else {
3306 			if (vm_free_magazine_refill_limit <= (vm_page_free_count - vm_page_free_reserved)) {
3307 				pages_to_steal = vm_free_magazine_refill_limit;
3308 			} else {
3309 				pages_to_steal = (vm_page_free_count - vm_page_free_reserved);
3310 			}
3311 		}
3312 		color = *PERCPU_GET(start_color);
3313 		head = tail = NULL;
3314 
3315 		vm_page_free_count -= pages_to_steal;
3316 		clump_end = sub_count = 0;
3317 
3318 		while (pages_to_steal--) {
3319 			while (vm_page_queue_empty(&vm_page_queue_free[color].qhead)) {
3320 				color = (color + 1) & vm_color_mask;
3321 			}
3322 #if defined(__x86_64__)
3323 			vm_page_queue_remove_first_with_clump(&vm_page_queue_free[color].qhead,
3324 			    mem, clump_end);
3325 #else
3326 			vm_page_queue_remove_first(&vm_page_queue_free[color].qhead,
3327 			    mem, vmp_pageq);
3328 #endif
3329 
3330 			assert(mem->vmp_q_state == VM_PAGE_ON_FREE_Q);
3331 
3332 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3333 
3334 #if defined(__arm64__)
3335 			color = (color + 1) & vm_color_mask;
3336 #else
3337 
3338 #if DEVELOPMENT || DEBUG
3339 
3340 			sub_count++;
3341 			if (clump_end) {
3342 				vm_clump_update_stats(sub_count);
3343 				sub_count = 0;
3344 				color = (color + 1) & vm_color_mask;
3345 			}
3346 #else
3347 			if (clump_end) {
3348 				color = (color + 1) & vm_color_mask;
3349 			}
3350 
3351 #endif /* if DEVELOPMENT || DEBUG */
3352 
3353 #endif  /* if defined(__arm64__) */
3354 
3355 			if (head == NULL) {
3356 				head = mem;
3357 			} else {
3358 				tail->vmp_snext = mem;
3359 			}
3360 			tail = mem;
3361 
3362 			assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3363 			assert(mem->vmp_tabled == FALSE);
3364 			assert(mem->vmp_object == 0);
3365 			assert(!mem->vmp_laundry);
3366 
3367 			mem->vmp_q_state = VM_PAGE_ON_FREE_LOCAL_Q;
3368 
3369 			ASSERT_PMAP_FREE(mem);
3370 			assert(mem->vmp_busy);
3371 			assert(!mem->vmp_pmapped);
3372 			assert(!mem->vmp_wpmapped);
3373 			assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3374 		}
3375 #if defined (__x86_64__) && (DEVELOPMENT || DEBUG)
3376 		vm_clump_update_stats(sub_count);
3377 #endif
3378 
3379 #if HIBERNATION
3380 		if (hibernate_rebuild_needed) {
3381 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3382 		}
3383 #endif /* HIBERNATION */
3384 		vm_offset_t pcpu_base = current_percpu_base();
3385 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = head;
3386 		*PERCPU_GET_WITH_BASE(pcpu_base, start_color) = color;
3387 
3388 		vm_free_page_unlock();
3389 		enable_preemption();
3390 		goto restart;
3391 	}
3392 
3393 	/*
3394 	 *	Decide if we should poke the pageout daemon.
3395 	 *	We do this if the free count is less than the low
3396 	 *	water mark. VM Pageout Scan will keep running till
3397 	 *	the free_count > free_target (& hence above free_min).
3398 	 *	This wakeup is to catch the possibility of the counts
3399 	 *	dropping between VM Pageout Scan parking and this check.
3400 	 *
3401 	 *	We don't have the counts locked ... if they change a little,
3402 	 *	it doesn't really matter.
3403 	 */
3404 	if (vm_page_free_count < vm_page_free_min) {
3405 		vm_free_page_lock();
3406 		if (vm_pageout_running == FALSE) {
3407 			vm_free_page_unlock();
3408 			thread_wakeup((event_t) &vm_page_free_wanted);
3409 		} else {
3410 			vm_free_page_unlock();
3411 		}
3412 	}
3413 
3414 	VM_CHECK_MEMORYSTATUS;
3415 
3416 	if (mem) {
3417 //		dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 4);	/* (TEST/DEBUG) */
3418 
3419 		task_t  cur_task = current_task_early();
3420 		if (cur_task && cur_task != kernel_task) {
3421 			if (cur_task->donates_own_pages) {
3422 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3423 			} else {
3424 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3425 			}
3426 		}
3427 	}
3428 	return mem;
3429 }
3430 
3431 #if CONFIG_SECLUDED_MEMORY
3432 vm_page_t
vm_page_grab_secluded(void)3433 vm_page_grab_secluded(void)
3434 {
3435 	vm_page_t       mem;
3436 	vm_object_t     object;
3437 	int             refmod_state;
3438 
3439 	if (vm_page_secluded_count == 0) {
3440 		/* no secluded pages to grab... */
3441 		return VM_PAGE_NULL;
3442 	}
3443 
3444 	/* secluded queue is protected by the VM page queue lock */
3445 	vm_page_lock_queues();
3446 
3447 	if (vm_page_secluded_count == 0) {
3448 		/* no secluded pages to grab... */
3449 		vm_page_unlock_queues();
3450 		return VM_PAGE_NULL;
3451 	}
3452 
3453 #if 00
3454 	/* can we grab from the secluded queue? */
3455 	if (vm_page_secluded_count > vm_page_secluded_target ||
3456 	    (vm_page_secluded_count > 0 &&
3457 	    task_can_use_secluded_mem(current_task(), TRUE))) {
3458 		/* OK */
3459 	} else {
3460 		/* can't grab from secluded queue... */
3461 		vm_page_unlock_queues();
3462 		return VM_PAGE_NULL;
3463 	}
3464 #endif
3465 
3466 	/* we can grab a page from secluded queue! */
3467 	assert((vm_page_secluded_count_free +
3468 	    vm_page_secluded_count_inuse) ==
3469 	    vm_page_secluded_count);
3470 	if (current_task()->task_can_use_secluded_mem) {
3471 		assert(num_tasks_can_use_secluded_mem > 0);
3472 	}
3473 	assert(!vm_page_queue_empty(&vm_page_queue_secluded));
3474 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3475 	mem = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3476 	assert(mem->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3477 	vm_page_queues_remove(mem, TRUE);
3478 
3479 	object = VM_PAGE_OBJECT(mem);
3480 
3481 	assert(!mem->vmp_fictitious);
3482 	assert(!VM_PAGE_WIRED(mem));
3483 	if (object == VM_OBJECT_NULL) {
3484 		/* free for grab! */
3485 		vm_page_unlock_queues();
3486 		vm_page_secluded.grab_success_free++;
3487 
3488 		assert(mem->vmp_busy);
3489 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3490 		assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3491 		assert(mem->vmp_pageq.next == 0);
3492 		assert(mem->vmp_pageq.prev == 0);
3493 		assert(mem->vmp_listq.next == 0);
3494 		assert(mem->vmp_listq.prev == 0);
3495 		assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3496 		assert(mem->vmp_specialq.next == 0);
3497 		assert(mem->vmp_specialq.prev == 0);
3498 		return mem;
3499 	}
3500 
3501 	assert(!object->internal);
3502 //	vm_page_pageable_external_count--;
3503 
3504 	if (!vm_object_lock_try(object)) {
3505 //		printf("SECLUDED: page %p: object %p locked\n", mem, object);
3506 		vm_page_secluded.grab_failure_locked++;
3507 reactivate_secluded_page:
3508 		vm_page_activate(mem);
3509 		vm_page_unlock_queues();
3510 		return VM_PAGE_NULL;
3511 	}
3512 	if (mem->vmp_busy ||
3513 	    mem->vmp_cleaning ||
3514 	    mem->vmp_laundry) {
3515 		/* can't steal page in this state... */
3516 		vm_object_unlock(object);
3517 		vm_page_secluded.grab_failure_state++;
3518 		goto reactivate_secluded_page;
3519 	}
3520 
3521 	mem->vmp_busy = TRUE;
3522 	refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem));
3523 	if (refmod_state & VM_MEM_REFERENCED) {
3524 		mem->vmp_reference = TRUE;
3525 	}
3526 	if (refmod_state & VM_MEM_MODIFIED) {
3527 		SET_PAGE_DIRTY(mem, FALSE);
3528 	}
3529 	if (mem->vmp_dirty || mem->vmp_precious) {
3530 		/* can't grab a dirty page; re-activate */
3531 //		printf("SECLUDED: dirty page %p\n", mem);
3532 		PAGE_WAKEUP_DONE(mem);
3533 		vm_page_secluded.grab_failure_dirty++;
3534 		vm_object_unlock(object);
3535 		goto reactivate_secluded_page;
3536 	}
3537 	if (mem->vmp_reference) {
3538 		/* it's been used but we do need to grab a page... */
3539 	}
3540 
3541 	vm_page_unlock_queues();
3542 
3543 
3544 	/* finish what vm_page_free() would have done... */
3545 	vm_page_free_prepare_object(mem, TRUE);
3546 	vm_object_unlock(object);
3547 	object = VM_OBJECT_NULL;
3548 	if (vm_page_free_verify) {
3549 		ASSERT_PMAP_FREE(mem);
3550 	}
3551 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3552 	vm_page_secluded.grab_success_other++;
3553 
3554 	assert(mem->vmp_busy);
3555 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3556 	assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3557 	assert(mem->vmp_pageq.next == 0);
3558 	assert(mem->vmp_pageq.prev == 0);
3559 	assert(mem->vmp_listq.next == 0);
3560 	assert(mem->vmp_listq.prev == 0);
3561 	assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3562 	assert(mem->vmp_specialq.next == 0);
3563 	assert(mem->vmp_specialq.prev == 0);
3564 
3565 	return mem;
3566 }
3567 
3568 uint64_t
vm_page_secluded_drain(void)3569 vm_page_secluded_drain(void)
3570 {
3571 	vm_page_t local_freeq;
3572 	int local_freed;
3573 	uint64_t num_reclaimed;
3574 	unsigned int saved_secluded_count, saved_secluded_target;
3575 
3576 	num_reclaimed = 0;
3577 	local_freeq = NULL;
3578 	local_freed = 0;
3579 
3580 	vm_page_lock_queues();
3581 
3582 	saved_secluded_count = vm_page_secluded_count;
3583 	saved_secluded_target = vm_page_secluded_target;
3584 	vm_page_secluded_target = 0;
3585 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3586 	while (vm_page_secluded_count) {
3587 		vm_page_t secluded_page;
3588 
3589 		assert((vm_page_secluded_count_free +
3590 		    vm_page_secluded_count_inuse) ==
3591 		    vm_page_secluded_count);
3592 		secluded_page = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3593 		assert(secluded_page->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3594 
3595 		vm_page_queues_remove(secluded_page, FALSE);
3596 		assert(!secluded_page->vmp_fictitious);
3597 		assert(!VM_PAGE_WIRED(secluded_page));
3598 
3599 		if (secluded_page->vmp_object == 0) {
3600 			/* transfer to free queue */
3601 			assert(secluded_page->vmp_busy);
3602 			secluded_page->vmp_snext = local_freeq;
3603 			local_freeq = secluded_page;
3604 			local_freed += 1;
3605 		} else {
3606 			/* transfer to head of active queue */
3607 			vm_page_enqueue_active(secluded_page, FALSE);
3608 			secluded_page = VM_PAGE_NULL;
3609 		}
3610 		num_reclaimed++;
3611 	}
3612 	vm_page_secluded_target = saved_secluded_target;
3613 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3614 
3615 //	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);
3616 
3617 	vm_page_unlock_queues();
3618 
3619 	if (local_freed) {
3620 		vm_page_free_list(local_freeq, TRUE);
3621 		local_freeq = NULL;
3622 		local_freed = 0;
3623 	}
3624 
3625 	return num_reclaimed;
3626 }
3627 #endif /* CONFIG_SECLUDED_MEMORY */
3628 
3629 
3630 static inline void
vm_page_grab_diags()3631 vm_page_grab_diags()
3632 {
3633 #if DEVELOPMENT || DEBUG
3634 	task_t task = current_task_early();
3635 	if (task == NULL) {
3636 		return;
3637 	}
3638 
3639 	ledger_credit(task->ledger, task_ledgers.pages_grabbed, 1);
3640 #endif /* DEVELOPMENT || DEBUG */
3641 }
3642 
3643 /*
3644  *	vm_page_release:
3645  *
3646  *	Return a page to the free list.
3647  */
3648 
3649 void
vm_page_release(vm_page_t mem,boolean_t page_queues_locked)3650 vm_page_release(
3651 	vm_page_t       mem,
3652 	boolean_t       page_queues_locked)
3653 {
3654 	unsigned int    color;
3655 	int     need_wakeup = 0;
3656 	int     need_priv_wakeup = 0;
3657 #if CONFIG_SECLUDED_MEMORY
3658 	int     need_secluded_wakeup = 0;
3659 #endif /* CONFIG_SECLUDED_MEMORY */
3660 	event_t wakeup_event = NULL;
3661 
3662 	if (page_queues_locked) {
3663 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3664 	} else {
3665 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
3666 	}
3667 
3668 	assert(!mem->vmp_private && !mem->vmp_fictitious);
3669 	if (vm_page_free_verify) {
3670 		ASSERT_PMAP_FREE(mem);
3671 	}
3672 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 5);	/* (TEST/DEBUG) */
3673 
3674 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3675 
3676 	vm_free_page_lock_spin();
3677 
3678 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3679 	assert(mem->vmp_busy);
3680 	assert(!mem->vmp_laundry);
3681 	assert(mem->vmp_object == 0);
3682 	assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
3683 	assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3684 	assert(mem->vmp_specialq.next == 0 && mem->vmp_specialq.prev == 0);
3685 
3686 	/* Clear any specialQ hints before releasing page to the free pool*/
3687 	mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
3688 
3689 	if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
3690 	    vm_lopage_free_count < vm_lopage_free_limit &&
3691 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3692 		/*
3693 		 * this exists to support hardware controllers
3694 		 * incapable of generating DMAs with more than 32 bits
3695 		 * of address on platforms with physical memory > 4G...
3696 		 */
3697 		vm_page_queue_enter_first(&vm_lopage_queue_free, mem, vmp_pageq);
3698 		vm_lopage_free_count++;
3699 
3700 		if (vm_lopage_free_count >= vm_lopage_free_limit) {
3701 			vm_lopage_refill = FALSE;
3702 		}
3703 
3704 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3705 		mem->vmp_lopage = TRUE;
3706 #if CONFIG_SECLUDED_MEMORY
3707 	} else if (vm_page_free_count > vm_page_free_reserved &&
3708 	    vm_page_secluded_count < vm_page_secluded_target &&
3709 	    num_tasks_can_use_secluded_mem == 0) {
3710 		/*
3711 		 * XXX FBDP TODO: also avoid refilling secluded queue
3712 		 * when some IOKit objects are already grabbing from it...
3713 		 */
3714 		if (!page_queues_locked) {
3715 			if (!vm_page_trylock_queues()) {
3716 				/* take locks in right order */
3717 				vm_free_page_unlock();
3718 				vm_page_lock_queues();
3719 				vm_free_page_lock_spin();
3720 			}
3721 		}
3722 		mem->vmp_lopage = FALSE;
3723 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3724 		vm_page_queue_enter_first(&vm_page_queue_secluded, mem, vmp_pageq);
3725 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3726 		vm_page_secluded_count++;
3727 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3728 		vm_page_secluded_count_free++;
3729 		if (!page_queues_locked) {
3730 			vm_page_unlock_queues();
3731 		}
3732 		LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_OWNED);
3733 		if (vm_page_free_wanted_secluded > 0) {
3734 			vm_page_free_wanted_secluded--;
3735 			need_secluded_wakeup = 1;
3736 		}
3737 #endif /* CONFIG_SECLUDED_MEMORY */
3738 	} else {
3739 		mem->vmp_lopage = FALSE;
3740 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3741 
3742 		color = VM_PAGE_GET_COLOR(mem);
3743 #if defined(__x86_64__)
3744 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
3745 #else
3746 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
3747 #endif
3748 		vm_page_free_count++;
3749 		/*
3750 		 *	Check if we should wake up someone waiting for page.
3751 		 *	But don't bother waking them unless they can allocate.
3752 		 *
3753 		 *	We wakeup only one thread, to prevent starvation.
3754 		 *	Because the scheduling system handles wait queues FIFO,
3755 		 *	if we wakeup all waiting threads, one greedy thread
3756 		 *	can starve multiple niceguy threads.  When the threads
3757 		 *	all wakeup, the greedy threads runs first, grabs the page,
3758 		 *	and waits for another page.  It will be the first to run
3759 		 *	when the next page is freed.
3760 		 *
3761 		 *	However, there is a slight danger here.
3762 		 *	The thread we wake might not use the free page.
3763 		 *	Then the other threads could wait indefinitely
3764 		 *	while the page goes unused.  To forestall this,
3765 		 *	the pageout daemon will keep making free pages
3766 		 *	as long as vm_page_free_wanted is non-zero.
3767 		 */
3768 
3769 		assert(vm_page_free_count > 0);
3770 		if (vm_page_free_wanted_privileged > 0) {
3771 			vm_page_free_wanted_privileged--;
3772 			need_priv_wakeup = 1;
3773 #if CONFIG_SECLUDED_MEMORY
3774 		} else if (vm_page_free_wanted_secluded > 0 &&
3775 		    vm_page_free_count > vm_page_free_reserved) {
3776 			vm_page_free_wanted_secluded--;
3777 			need_secluded_wakeup = 1;
3778 #endif /* CONFIG_SECLUDED_MEMORY */
3779 		} else if (vm_page_free_wanted > 0 &&
3780 		    vm_page_free_count > vm_page_free_reserved) {
3781 			vm_page_free_wanted--;
3782 			need_wakeup = 1;
3783 		}
3784 	}
3785 	vm_pageout_vminfo.vm_page_pages_freed++;
3786 
3787 	vm_free_page_unlock();
3788 
3789 	VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, 1, 0, 0, 0);
3790 
3791 	if (need_priv_wakeup) {
3792 		wakeup_event = &vm_page_free_wanted_privileged;
3793 	}
3794 #if CONFIG_SECLUDED_MEMORY
3795 	else if (need_secluded_wakeup) {
3796 		wakeup_event = &vm_page_free_wanted_secluded;
3797 	}
3798 #endif /* CONFIG_SECLUDED_MEMORY */
3799 	else if (need_wakeup) {
3800 		wakeup_event = &vm_page_free_count;
3801 	}
3802 
3803 	if (wakeup_event) {
3804 		if (vps_dynamic_priority_enabled == TRUE) {
3805 			thread_t thread_woken = NULL;
3806 			wakeup_one_with_inheritor((event_t) wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
3807 			/*
3808 			 * (80947592) if this is the last reference on this
3809 			 * thread, calling thread_deallocate() here
3810 			 * might take the tasks_threads_lock,
3811 			 * sadly thread_create_internal is doing several
3812 			 * allocations under this lock, which can result in
3813 			 * deadlocks with the pageout scan daemon.
3814 			 *
3815 			 * FIXME: we should disallow allocations under the
3816 			 * task_thread_locks, but that is a larger fix to make.
3817 			 */
3818 			thread_deallocate_safe(thread_woken);
3819 		} else {
3820 			thread_wakeup_one((event_t) wakeup_event);
3821 		}
3822 	}
3823 
3824 	VM_CHECK_MEMORYSTATUS;
3825 }
3826 
3827 /*
3828  * This version of vm_page_release() is used only at startup
3829  * when we are single-threaded and pages are being released
3830  * for the first time. Hence, no locking or unnecessary checks are made.
3831  * Note: VM_CHECK_MEMORYSTATUS invoked by the caller.
3832  */
3833 void
vm_page_release_startup(vm_page_t mem)3834 vm_page_release_startup(
3835 	vm_page_t       mem)
3836 {
3837 	vm_page_queue_t queue_free;
3838 
3839 	if (vm_lopage_free_count < vm_lopage_free_limit &&
3840 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3841 		mem->vmp_lopage = TRUE;
3842 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3843 		vm_lopage_free_count++;
3844 		queue_free = &vm_lopage_queue_free;
3845 #if CONFIG_SECLUDED_MEMORY
3846 	} else if (vm_page_secluded_count < vm_page_secluded_target) {
3847 		mem->vmp_lopage = FALSE;
3848 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3849 		vm_page_secluded_count++;
3850 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3851 		vm_page_secluded_count_free++;
3852 		queue_free = &vm_page_queue_secluded;
3853 #endif /* CONFIG_SECLUDED_MEMORY */
3854 	} else {
3855 		mem->vmp_lopage = FALSE;
3856 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3857 		vm_page_free_count++;
3858 		queue_free = &vm_page_queue_free[VM_PAGE_GET_COLOR(mem)].qhead;
3859 	}
3860 	if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
3861 #if defined(__x86_64__)
3862 		vm_page_queue_enter_clump(queue_free, mem);
3863 #else
3864 		vm_page_queue_enter(queue_free, mem, vmp_pageq);
3865 #endif
3866 	} else {
3867 		vm_page_queue_enter_first(queue_free, mem, vmp_pageq);
3868 	}
3869 }
3870 
3871 /*
3872  *	vm_page_wait:
3873  *
3874  *	Wait for a page to become available.
3875  *	If there are plenty of free pages, then we don't sleep.
3876  *
3877  *	Returns:
3878  *		TRUE:  There may be another page, try again
3879  *		FALSE: We were interrupted out of our wait, don't try again
3880  */
3881 
3882 boolean_t
vm_page_wait(int interruptible)3883 vm_page_wait(
3884 	int     interruptible )
3885 {
3886 	/*
3887 	 *	We can't use vm_page_free_reserved to make this
3888 	 *	determination.  Consider: some thread might
3889 	 *	need to allocate two pages.  The first allocation
3890 	 *	succeeds, the second fails.  After the first page is freed,
3891 	 *	a call to vm_page_wait must really block.
3892 	 */
3893 	kern_return_t   wait_result;
3894 	int             need_wakeup = 0;
3895 	int             is_privileged = current_thread()->options & TH_OPT_VMPRIV;
3896 	event_t         wait_event = NULL;
3897 
3898 	vm_free_page_lock_spin();
3899 
3900 	if (is_privileged && vm_page_free_count) {
3901 		vm_free_page_unlock();
3902 		return TRUE;
3903 	}
3904 
3905 	if (vm_page_free_count >= vm_page_free_target) {
3906 		vm_free_page_unlock();
3907 		return TRUE;
3908 	}
3909 
3910 	if (is_privileged) {
3911 		if (vm_page_free_wanted_privileged++ == 0) {
3912 			need_wakeup = 1;
3913 		}
3914 		wait_event = (event_t)&vm_page_free_wanted_privileged;
3915 #if CONFIG_SECLUDED_MEMORY
3916 	} else if (secluded_for_apps &&
3917 	    task_can_use_secluded_mem(current_task(), FALSE)) {
3918 #if 00
3919 		/* XXX FBDP: need pageq lock for this... */
3920 		/* XXX FBDP: might wait even if pages available, */
3921 		/* XXX FBDP: hopefully not for too long... */
3922 		if (vm_page_secluded_count > 0) {
3923 			vm_free_page_unlock();
3924 			return TRUE;
3925 		}
3926 #endif
3927 		if (vm_page_free_wanted_secluded++ == 0) {
3928 			need_wakeup = 1;
3929 		}
3930 		wait_event = (event_t)&vm_page_free_wanted_secluded;
3931 #endif /* CONFIG_SECLUDED_MEMORY */
3932 	} else {
3933 		if (vm_page_free_wanted++ == 0) {
3934 			need_wakeup = 1;
3935 		}
3936 		wait_event = (event_t)&vm_page_free_count;
3937 	}
3938 
3939 	/*
3940 	 * We don't do a vm_pageout_scan wakeup if we already have
3941 	 * some waiters because vm_pageout_scan checks for waiters
3942 	 * before it returns and does so behind the vm_page_queue_free_lock,
3943 	 * which we own when we bump the waiter counts.
3944 	 */
3945 
3946 	if (vps_dynamic_priority_enabled == TRUE) {
3947 		/*
3948 		 * We are waking up vm_pageout_scan here. If it needs
3949 		 * the vm_page_queue_free_lock before we unlock it
3950 		 * we'll end up just blocking and incur an extra
3951 		 * context switch. Could be a perf. issue.
3952 		 */
3953 
3954 		if (need_wakeup) {
3955 			thread_wakeup((event_t)&vm_page_free_wanted);
3956 		}
3957 
3958 		/*
3959 		 * LD: This event is going to get recorded every time because
3960 		 * we don't get back THREAD_WAITING from lck_mtx_sleep_with_inheritor.
3961 		 * We just block in that routine.
3962 		 */
3963 		VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
3964 		    vm_page_free_wanted_privileged,
3965 		    vm_page_free_wanted,
3966 #if CONFIG_SECLUDED_MEMORY
3967 		    vm_page_free_wanted_secluded,
3968 #else /* CONFIG_SECLUDED_MEMORY */
3969 		    0,
3970 #endif /* CONFIG_SECLUDED_MEMORY */
3971 		    0);
3972 		wait_result =  lck_mtx_sleep_with_inheritor(&vm_page_queue_free_lock,
3973 		    LCK_SLEEP_UNLOCK,
3974 		    wait_event,
3975 		    vm_pageout_scan_thread,
3976 		    interruptible,
3977 		    0);
3978 	} else {
3979 		wait_result = assert_wait(wait_event, interruptible);
3980 
3981 		vm_free_page_unlock();
3982 
3983 		if (need_wakeup) {
3984 			thread_wakeup((event_t)&vm_page_free_wanted);
3985 		}
3986 
3987 		if (wait_result == THREAD_WAITING) {
3988 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
3989 			    vm_page_free_wanted_privileged,
3990 			    vm_page_free_wanted,
3991 #if CONFIG_SECLUDED_MEMORY
3992 			    vm_page_free_wanted_secluded,
3993 #else /* CONFIG_SECLUDED_MEMORY */
3994 			    0,
3995 #endif /* CONFIG_SECLUDED_MEMORY */
3996 			    0);
3997 			wait_result = thread_block(THREAD_CONTINUE_NULL);
3998 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block,
3999 			    VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0);
4000 		}
4001 	}
4002 
4003 	return (wait_result == THREAD_AWAKENED) || (wait_result == THREAD_NOT_WAITING);
4004 }
4005 
4006 /*
4007  *	vm_page_alloc:
4008  *
4009  *	Allocate and return a memory cell associated
4010  *	with this VM object/offset pair.
4011  *
4012  *	Object must be locked.
4013  */
4014 
4015 vm_page_t
vm_page_alloc(vm_object_t object,vm_object_offset_t offset)4016 vm_page_alloc(
4017 	vm_object_t             object,
4018 	vm_object_offset_t      offset)
4019 {
4020 	vm_page_t       mem;
4021 	int             grab_options;
4022 
4023 	vm_object_lock_assert_exclusive(object);
4024 	grab_options = 0;
4025 #if CONFIG_SECLUDED_MEMORY
4026 	if (object->can_grab_secluded) {
4027 		grab_options |= VM_PAGE_GRAB_SECLUDED;
4028 	}
4029 #endif /* CONFIG_SECLUDED_MEMORY */
4030 	mem = vm_page_grab_options(grab_options);
4031 	if (mem == VM_PAGE_NULL) {
4032 		return VM_PAGE_NULL;
4033 	}
4034 
4035 	vm_page_insert(mem, object, offset);
4036 
4037 	return mem;
4038 }
4039 
4040 /*
4041  *	vm_page_free_prepare:
4042  *
4043  *	Removes page from any queue it may be on
4044  *	and disassociates it from its VM object.
4045  *
4046  *	Object and page queues must be locked prior to entry.
4047  */
4048 static void
vm_page_free_prepare(vm_page_t mem)4049 vm_page_free_prepare(
4050 	vm_page_t       mem)
4051 {
4052 	vm_page_free_prepare_queues(mem);
4053 	vm_page_free_prepare_object(mem, TRUE);
4054 }
4055 
4056 
4057 void
vm_page_free_prepare_queues(vm_page_t mem)4058 vm_page_free_prepare_queues(
4059 	vm_page_t       mem)
4060 {
4061 	vm_object_t     m_object;
4062 
4063 	VM_PAGE_CHECK(mem);
4064 
4065 	assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
4066 	assert(!mem->vmp_cleaning);
4067 	m_object = VM_PAGE_OBJECT(mem);
4068 
4069 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4070 	if (m_object) {
4071 		vm_object_lock_assert_exclusive(m_object);
4072 	}
4073 	if (mem->vmp_laundry) {
4074 		/*
4075 		 * We may have to free a page while it's being laundered
4076 		 * if we lost its pager (due to a forced unmount, for example).
4077 		 * We need to call vm_pageout_steal_laundry() before removing
4078 		 * the page from its VM object, so that we can remove it
4079 		 * from its pageout queue and adjust the laundry accounting
4080 		 */
4081 		vm_pageout_steal_laundry(mem, TRUE);
4082 	}
4083 
4084 	vm_page_queues_remove(mem, TRUE);
4085 
4086 	if (VM_PAGE_WIRED(mem)) {
4087 		assert(mem->vmp_wire_count > 0);
4088 
4089 		if (m_object) {
4090 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4091 			VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4092 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4093 
4094 			assert(m_object->resident_page_count >=
4095 			    m_object->wired_page_count);
4096 
4097 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4098 				OSAddAtomic(+1, &vm_page_purgeable_count);
4099 				assert(vm_page_purgeable_wired_count > 0);
4100 				OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4101 			}
4102 			if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4103 			    m_object->purgable == VM_PURGABLE_EMPTY) &&
4104 			    m_object->vo_owner != TASK_NULL) {
4105 				task_t          owner;
4106 				int             ledger_idx_volatile;
4107 				int             ledger_idx_nonvolatile;
4108 				int             ledger_idx_volatile_compressed;
4109 				int             ledger_idx_nonvolatile_compressed;
4110 				boolean_t       do_footprint;
4111 
4112 				owner = VM_OBJECT_OWNER(m_object);
4113 				vm_object_ledger_tag_ledgers(
4114 					m_object,
4115 					&ledger_idx_volatile,
4116 					&ledger_idx_nonvolatile,
4117 					&ledger_idx_volatile_compressed,
4118 					&ledger_idx_nonvolatile_compressed,
4119 					&do_footprint);
4120 				/*
4121 				 * While wired, this page was accounted
4122 				 * as "non-volatile" but it should now
4123 				 * be accounted as "volatile".
4124 				 */
4125 				/* one less "non-volatile"... */
4126 				ledger_debit(owner->ledger,
4127 				    ledger_idx_nonvolatile,
4128 				    PAGE_SIZE);
4129 				if (do_footprint) {
4130 					/* ... and "phys_footprint" */
4131 					ledger_debit(owner->ledger,
4132 					    task_ledgers.phys_footprint,
4133 					    PAGE_SIZE);
4134 				}
4135 				/* one more "volatile" */
4136 				ledger_credit(owner->ledger,
4137 				    ledger_idx_volatile,
4138 				    PAGE_SIZE);
4139 			}
4140 		}
4141 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4142 			vm_page_wire_count--;
4143 		}
4144 
4145 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4146 		mem->vmp_wire_count = 0;
4147 		assert(!mem->vmp_gobbled);
4148 	} else if (mem->vmp_gobbled) {
4149 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4150 			vm_page_wire_count--;
4151 		}
4152 		vm_page_gobble_count--;
4153 	}
4154 }
4155 
4156 
4157 void
vm_page_free_prepare_object(vm_page_t mem,boolean_t remove_from_hash)4158 vm_page_free_prepare_object(
4159 	vm_page_t       mem,
4160 	boolean_t       remove_from_hash)
4161 {
4162 	if (mem->vmp_tabled) {
4163 		vm_page_remove(mem, remove_from_hash);  /* clears tabled, object, offset */
4164 	}
4165 	PAGE_WAKEUP(mem);               /* clears wanted */
4166 
4167 	if (mem->vmp_private) {
4168 		mem->vmp_private = FALSE;
4169 		mem->vmp_fictitious = TRUE;
4170 		VM_PAGE_SET_PHYS_PAGE(mem, vm_page_fictitious_addr);
4171 	}
4172 	if (!mem->vmp_fictitious) {
4173 		assert(mem->vmp_pageq.next == 0);
4174 		assert(mem->vmp_pageq.prev == 0);
4175 		assert(mem->vmp_listq.next == 0);
4176 		assert(mem->vmp_listq.prev == 0);
4177 		assert(mem->vmp_specialq.next == 0);
4178 		assert(mem->vmp_specialq.prev == 0);
4179 		assert(mem->vmp_next_m == 0);
4180 		ASSERT_PMAP_FREE(mem);
4181 		{
4182 			vm_page_init(mem, VM_PAGE_GET_PHYS_PAGE(mem), mem->vmp_lopage);
4183 		}
4184 	}
4185 }
4186 
4187 
4188 /*
4189  *	vm_page_free:
4190  *
4191  *	Returns the given page to the free list,
4192  *	disassociating it with any VM object.
4193  *
4194  *	Object and page queues must be locked prior to entry.
4195  */
4196 void
vm_page_free(vm_page_t mem)4197 vm_page_free(
4198 	vm_page_t       mem)
4199 {
4200 	vm_page_free_prepare(mem);
4201 
4202 	if (mem->vmp_fictitious) {
4203 		vm_page_release_fictitious(mem);
4204 	} else {
4205 		vm_page_release(mem, TRUE);  /* page queues are locked */
4206 	}
4207 }
4208 
4209 
4210 void
vm_page_free_unlocked(vm_page_t mem,boolean_t remove_from_hash)4211 vm_page_free_unlocked(
4212 	vm_page_t       mem,
4213 	boolean_t       remove_from_hash)
4214 {
4215 	vm_page_lockspin_queues();
4216 	vm_page_free_prepare_queues(mem);
4217 	vm_page_unlock_queues();
4218 
4219 	vm_page_free_prepare_object(mem, remove_from_hash);
4220 
4221 	if (mem->vmp_fictitious) {
4222 		vm_page_release_fictitious(mem);
4223 	} else {
4224 		vm_page_release(mem, FALSE); /* page queues are not locked */
4225 	}
4226 }
4227 
4228 
4229 /*
4230  * Free a list of pages.  The list can be up to several hundred pages,
4231  * as blocked up by vm_pageout_scan().
4232  * The big win is not having to take the free list lock once
4233  * per page.
4234  *
4235  * The VM page queues lock (vm_page_queue_lock) should NOT be held.
4236  * The VM page free queues lock (vm_page_queue_free_lock) should NOT be held.
4237  */
4238 void
vm_page_free_list(vm_page_t freeq,boolean_t prepare_object)4239 vm_page_free_list(
4240 	vm_page_t       freeq,
4241 	boolean_t       prepare_object)
4242 {
4243 	vm_page_t       mem;
4244 	vm_page_t       nxt;
4245 	vm_page_t       local_freeq;
4246 	int             pg_count;
4247 
4248 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
4249 	LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_NOTOWNED);
4250 
4251 	while (freeq) {
4252 		pg_count = 0;
4253 		local_freeq = VM_PAGE_NULL;
4254 		mem = freeq;
4255 
4256 		/*
4257 		 * break up the processing into smaller chunks so
4258 		 * that we can 'pipeline' the pages onto the
4259 		 * free list w/o introducing too much
4260 		 * contention on the global free queue lock
4261 		 */
4262 		while (mem && pg_count < 64) {
4263 			assert((mem->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
4264 			    (mem->vmp_q_state == VM_PAGE_IS_WIRED));
4265 			assert(mem->vmp_specialq.next == 0 &&
4266 			    mem->vmp_specialq.prev == 0);
4267 			/*
4268 			 * &&
4269 			 *   mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
4270 			 */
4271 			nxt = mem->vmp_snext;
4272 			mem->vmp_snext = NULL;
4273 			assert(mem->vmp_pageq.prev == 0);
4274 
4275 			if (vm_page_free_verify && !mem->vmp_fictitious && !mem->vmp_private) {
4276 				ASSERT_PMAP_FREE(mem);
4277 			}
4278 			if (prepare_object == TRUE) {
4279 				vm_page_free_prepare_object(mem, TRUE);
4280 			}
4281 
4282 			if (!mem->vmp_fictitious) {
4283 				assert(mem->vmp_busy);
4284 
4285 				if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
4286 				    vm_lopage_free_count < vm_lopage_free_limit &&
4287 				    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
4288 					vm_page_release(mem, FALSE); /* page queues are not locked */
4289 #if CONFIG_SECLUDED_MEMORY
4290 				} else if (vm_page_secluded_count < vm_page_secluded_target &&
4291 				    num_tasks_can_use_secluded_mem == 0) {
4292 					vm_page_release(mem,
4293 					    FALSE);             /* page queues are not locked */
4294 #endif /* CONFIG_SECLUDED_MEMORY */
4295 				} else {
4296 					/*
4297 					 * IMPORTANT: we can't set the page "free" here
4298 					 * because that would make the page eligible for
4299 					 * a physically-contiguous allocation (see
4300 					 * vm_page_find_contiguous()) right away (we don't
4301 					 * hold the vm_page_queue_free lock).  That would
4302 					 * cause trouble because the page is not actually
4303 					 * in the free queue yet...
4304 					 */
4305 					mem->vmp_snext = local_freeq;
4306 					local_freeq = mem;
4307 					pg_count++;
4308 
4309 					pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
4310 				}
4311 			} else {
4312 				assert(VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_fictitious_addr ||
4313 				    VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_guard_addr);
4314 				vm_page_release_fictitious(mem);
4315 			}
4316 			mem = nxt;
4317 		}
4318 		freeq = mem;
4319 
4320 		if ((mem = local_freeq)) {
4321 			unsigned int    avail_free_count;
4322 			unsigned int    need_wakeup = 0;
4323 			unsigned int    need_priv_wakeup = 0;
4324 #if CONFIG_SECLUDED_MEMORY
4325 			unsigned int    need_wakeup_secluded = 0;
4326 #endif /* CONFIG_SECLUDED_MEMORY */
4327 			event_t         priv_wakeup_event, secluded_wakeup_event, normal_wakeup_event;
4328 			boolean_t       priv_wakeup_all, secluded_wakeup_all, normal_wakeup_all;
4329 
4330 			vm_free_page_lock_spin();
4331 
4332 			while (mem) {
4333 				int     color;
4334 
4335 				nxt = mem->vmp_snext;
4336 
4337 				assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
4338 				assert(mem->vmp_busy);
4339 				mem->vmp_lopage = FALSE;
4340 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
4341 
4342 				color = VM_PAGE_GET_COLOR(mem);
4343 #if defined(__x86_64__)
4344 				vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
4345 #else
4346 				vm_page_queue_enter(&vm_page_queue_free[color].qhead,
4347 				    mem, vmp_pageq);
4348 #endif
4349 				mem = nxt;
4350 			}
4351 			vm_pageout_vminfo.vm_page_pages_freed += pg_count;
4352 			vm_page_free_count += pg_count;
4353 			avail_free_count = vm_page_free_count;
4354 
4355 			VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, pg_count, 0, 0, 0);
4356 
4357 			if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) {
4358 				if (avail_free_count < vm_page_free_wanted_privileged) {
4359 					need_priv_wakeup = avail_free_count;
4360 					vm_page_free_wanted_privileged -= avail_free_count;
4361 					avail_free_count = 0;
4362 				} else {
4363 					need_priv_wakeup = vm_page_free_wanted_privileged;
4364 					avail_free_count -= vm_page_free_wanted_privileged;
4365 					vm_page_free_wanted_privileged = 0;
4366 				}
4367 			}
4368 #if CONFIG_SECLUDED_MEMORY
4369 			if (vm_page_free_wanted_secluded > 0 &&
4370 			    avail_free_count > vm_page_free_reserved) {
4371 				unsigned int available_pages;
4372 				available_pages = (avail_free_count -
4373 				    vm_page_free_reserved);
4374 				if (available_pages <
4375 				    vm_page_free_wanted_secluded) {
4376 					need_wakeup_secluded = available_pages;
4377 					vm_page_free_wanted_secluded -=
4378 					    available_pages;
4379 					avail_free_count -= available_pages;
4380 				} else {
4381 					need_wakeup_secluded =
4382 					    vm_page_free_wanted_secluded;
4383 					avail_free_count -=
4384 					    vm_page_free_wanted_secluded;
4385 					vm_page_free_wanted_secluded = 0;
4386 				}
4387 			}
4388 #endif /* CONFIG_SECLUDED_MEMORY */
4389 			if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) {
4390 				unsigned int  available_pages;
4391 
4392 				available_pages = avail_free_count - vm_page_free_reserved;
4393 
4394 				if (available_pages >= vm_page_free_wanted) {
4395 					need_wakeup = vm_page_free_wanted;
4396 					vm_page_free_wanted = 0;
4397 				} else {
4398 					need_wakeup = available_pages;
4399 					vm_page_free_wanted -= available_pages;
4400 				}
4401 			}
4402 			vm_free_page_unlock();
4403 
4404 			priv_wakeup_event = NULL;
4405 			secluded_wakeup_event = NULL;
4406 			normal_wakeup_event = NULL;
4407 
4408 			priv_wakeup_all = FALSE;
4409 			secluded_wakeup_all = FALSE;
4410 			normal_wakeup_all = FALSE;
4411 
4412 
4413 			if (need_priv_wakeup != 0) {
4414 				/*
4415 				 * There shouldn't be that many VM-privileged threads,
4416 				 * so let's wake them all up, even if we don't quite
4417 				 * have enough pages to satisfy them all.
4418 				 */
4419 				priv_wakeup_event = (event_t)&vm_page_free_wanted_privileged;
4420 				priv_wakeup_all = TRUE;
4421 			}
4422 #if CONFIG_SECLUDED_MEMORY
4423 			if (need_wakeup_secluded != 0 &&
4424 			    vm_page_free_wanted_secluded == 0) {
4425 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4426 				secluded_wakeup_all = TRUE;
4427 				need_wakeup_secluded = 0;
4428 			} else {
4429 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4430 			}
4431 #endif /* CONFIG_SECLUDED_MEMORY */
4432 			if (need_wakeup != 0 && vm_page_free_wanted == 0) {
4433 				/*
4434 				 * We don't expect to have any more waiters
4435 				 * after this, so let's wake them all up at
4436 				 * once.
4437 				 */
4438 				normal_wakeup_event = (event_t) &vm_page_free_count;
4439 				normal_wakeup_all = TRUE;
4440 				need_wakeup = 0;
4441 			} else {
4442 				normal_wakeup_event = (event_t) &vm_page_free_count;
4443 			}
4444 
4445 			if (priv_wakeup_event ||
4446 #if CONFIG_SECLUDED_MEMORY
4447 			    secluded_wakeup_event ||
4448 #endif /* CONFIG_SECLUDED_MEMORY */
4449 			    normal_wakeup_event) {
4450 				if (vps_dynamic_priority_enabled == TRUE) {
4451 					thread_t thread_woken = NULL;
4452 
4453 					if (priv_wakeup_all == TRUE) {
4454 						wakeup_all_with_inheritor(priv_wakeup_event, THREAD_AWAKENED);
4455 					}
4456 
4457 #if CONFIG_SECLUDED_MEMORY
4458 					if (secluded_wakeup_all == TRUE) {
4459 						wakeup_all_with_inheritor(secluded_wakeup_event, THREAD_AWAKENED);
4460 					}
4461 
4462 					while (need_wakeup_secluded-- != 0) {
4463 						/*
4464 						 * Wake up one waiter per page we just released.
4465 						 */
4466 						wakeup_one_with_inheritor(secluded_wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
4467 						thread_deallocate(thread_woken);
4468 					}
4469 #endif /* CONFIG_SECLUDED_MEMORY */
4470 
4471 					if (normal_wakeup_all == TRUE) {
4472 						wakeup_all_with_inheritor(normal_wakeup_event, THREAD_AWAKENED);
4473 					}
4474 
4475 					while (need_wakeup-- != 0) {
4476 						/*
4477 						 * Wake up one waiter per page we just released.
4478 						 */
4479 						wakeup_one_with_inheritor(normal_wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
4480 						thread_deallocate(thread_woken);
4481 					}
4482 				} else {
4483 					/*
4484 					 * Non-priority-aware wakeups.
4485 					 */
4486 
4487 					if (priv_wakeup_all == TRUE) {
4488 						thread_wakeup(priv_wakeup_event);
4489 					}
4490 
4491 #if CONFIG_SECLUDED_MEMORY
4492 					if (secluded_wakeup_all == TRUE) {
4493 						thread_wakeup(secluded_wakeup_event);
4494 					}
4495 
4496 					while (need_wakeup_secluded-- != 0) {
4497 						/*
4498 						 * Wake up one waiter per page we just released.
4499 						 */
4500 						thread_wakeup_one(secluded_wakeup_event);
4501 					}
4502 
4503 #endif /* CONFIG_SECLUDED_MEMORY */
4504 					if (normal_wakeup_all == TRUE) {
4505 						thread_wakeup(normal_wakeup_event);
4506 					}
4507 
4508 					while (need_wakeup-- != 0) {
4509 						/*
4510 						 * Wake up one waiter per page we just released.
4511 						 */
4512 						thread_wakeup_one(normal_wakeup_event);
4513 					}
4514 				}
4515 			}
4516 
4517 			VM_CHECK_MEMORYSTATUS;
4518 		}
4519 	}
4520 }
4521 
4522 
4523 /*
4524  *	vm_page_wire:
4525  *
4526  *	Mark this page as wired down by yet
4527  *	another map, removing it from paging queues
4528  *	as necessary.
4529  *
4530  *	The page's object and the page queues must be locked.
4531  */
4532 
4533 
4534 void
vm_page_wire(vm_page_t mem,vm_tag_t tag,boolean_t check_memorystatus)4535 vm_page_wire(
4536 	vm_page_t mem,
4537 	vm_tag_t           tag,
4538 	boolean_t          check_memorystatus)
4539 {
4540 	vm_object_t     m_object;
4541 
4542 	m_object = VM_PAGE_OBJECT(mem);
4543 
4544 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 1);	/* (TEST/DEBUG) */
4545 
4546 	VM_PAGE_CHECK(mem);
4547 	if (m_object) {
4548 		vm_object_lock_assert_exclusive(m_object);
4549 	} else {
4550 		/*
4551 		 * In theory, the page should be in an object before it
4552 		 * gets wired, since we need to hold the object lock
4553 		 * to update some fields in the page structure.
4554 		 * However, some code (i386 pmap, for example) might want
4555 		 * to wire a page before it gets inserted into an object.
4556 		 * That's somewhat OK, as long as nobody else can get to
4557 		 * that page and update it at the same time.
4558 		 */
4559 	}
4560 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4561 	if (!VM_PAGE_WIRED(mem)) {
4562 		if (mem->vmp_laundry) {
4563 			vm_pageout_steal_laundry(mem, TRUE);
4564 		}
4565 
4566 		vm_page_queues_remove(mem, TRUE);
4567 
4568 		assert(mem->vmp_wire_count == 0);
4569 		mem->vmp_q_state = VM_PAGE_IS_WIRED;
4570 
4571 		if (m_object) {
4572 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4573 			VM_OBJECT_WIRED_PAGE_ADD(m_object, mem);
4574 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, tag);
4575 
4576 			assert(m_object->resident_page_count >=
4577 			    m_object->wired_page_count);
4578 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4579 				assert(vm_page_purgeable_count > 0);
4580 				OSAddAtomic(-1, &vm_page_purgeable_count);
4581 				OSAddAtomic(1, &vm_page_purgeable_wired_count);
4582 			}
4583 			if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4584 			    m_object->purgable == VM_PURGABLE_EMPTY) &&
4585 			    m_object->vo_owner != TASK_NULL) {
4586 				task_t          owner;
4587 				int             ledger_idx_volatile;
4588 				int             ledger_idx_nonvolatile;
4589 				int             ledger_idx_volatile_compressed;
4590 				int             ledger_idx_nonvolatile_compressed;
4591 				boolean_t       do_footprint;
4592 
4593 				owner = VM_OBJECT_OWNER(m_object);
4594 				vm_object_ledger_tag_ledgers(
4595 					m_object,
4596 					&ledger_idx_volatile,
4597 					&ledger_idx_nonvolatile,
4598 					&ledger_idx_volatile_compressed,
4599 					&ledger_idx_nonvolatile_compressed,
4600 					&do_footprint);
4601 				/* less volatile bytes */
4602 				ledger_debit(owner->ledger,
4603 				    ledger_idx_volatile,
4604 				    PAGE_SIZE);
4605 				/* more not-quite-volatile bytes */
4606 				ledger_credit(owner->ledger,
4607 				    ledger_idx_nonvolatile,
4608 				    PAGE_SIZE);
4609 				if (do_footprint) {
4610 					/* more footprint */
4611 					ledger_credit(owner->ledger,
4612 					    task_ledgers.phys_footprint,
4613 					    PAGE_SIZE);
4614 				}
4615 			}
4616 			if (m_object->all_reusable) {
4617 				/*
4618 				 * Wired pages are not counted as "re-usable"
4619 				 * in "all_reusable" VM objects, so nothing
4620 				 * to do here.
4621 				 */
4622 			} else if (mem->vmp_reusable) {
4623 				/*
4624 				 * This page is not "re-usable" when it's
4625 				 * wired, so adjust its state and the
4626 				 * accounting.
4627 				 */
4628 				vm_object_reuse_pages(m_object,
4629 				    mem->vmp_offset,
4630 				    mem->vmp_offset + PAGE_SIZE_64,
4631 				    FALSE);
4632 			}
4633 		}
4634 		assert(!mem->vmp_reusable);
4635 
4636 		if (!mem->vmp_private && !mem->vmp_fictitious && !mem->vmp_gobbled) {
4637 			vm_page_wire_count++;
4638 		}
4639 		if (mem->vmp_gobbled) {
4640 			vm_page_gobble_count--;
4641 		}
4642 		mem->vmp_gobbled = FALSE;
4643 
4644 		if (check_memorystatus == TRUE) {
4645 			VM_CHECK_MEMORYSTATUS;
4646 		}
4647 	}
4648 	assert(!mem->vmp_gobbled);
4649 	assert(mem->vmp_q_state == VM_PAGE_IS_WIRED);
4650 	mem->vmp_wire_count++;
4651 	if (__improbable(mem->vmp_wire_count == 0)) {
4652 		panic("vm_page_wire(%p): wire_count overflow", mem);
4653 	}
4654 	VM_PAGE_CHECK(mem);
4655 }
4656 
4657 /*
4658  *	vm_page_unwire:
4659  *
4660  *	Release one wiring of this page, potentially
4661  *	enabling it to be paged again.
4662  *
4663  *	The page's object and the page queues must be locked.
4664  */
4665 void
vm_page_unwire(vm_page_t mem,boolean_t queueit)4666 vm_page_unwire(
4667 	vm_page_t       mem,
4668 	boolean_t       queueit)
4669 {
4670 	vm_object_t     m_object;
4671 
4672 	m_object = VM_PAGE_OBJECT(mem);
4673 
4674 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 0);	/* (TEST/DEBUG) */
4675 
4676 	VM_PAGE_CHECK(mem);
4677 	assert(VM_PAGE_WIRED(mem));
4678 	assert(mem->vmp_wire_count > 0);
4679 	assert(!mem->vmp_gobbled);
4680 	assert(m_object != VM_OBJECT_NULL);
4681 	vm_object_lock_assert_exclusive(m_object);
4682 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4683 	if (--mem->vmp_wire_count == 0) {
4684 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4685 
4686 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4687 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4688 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4689 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4690 			vm_page_wire_count--;
4691 		}
4692 
4693 		assert(m_object->resident_page_count >=
4694 		    m_object->wired_page_count);
4695 		if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4696 			OSAddAtomic(+1, &vm_page_purgeable_count);
4697 			assert(vm_page_purgeable_wired_count > 0);
4698 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4699 		}
4700 		if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4701 		    m_object->purgable == VM_PURGABLE_EMPTY) &&
4702 		    m_object->vo_owner != TASK_NULL) {
4703 			task_t          owner;
4704 			int             ledger_idx_volatile;
4705 			int             ledger_idx_nonvolatile;
4706 			int             ledger_idx_volatile_compressed;
4707 			int             ledger_idx_nonvolatile_compressed;
4708 			boolean_t       do_footprint;
4709 
4710 			owner = VM_OBJECT_OWNER(m_object);
4711 			vm_object_ledger_tag_ledgers(
4712 				m_object,
4713 				&ledger_idx_volatile,
4714 				&ledger_idx_nonvolatile,
4715 				&ledger_idx_volatile_compressed,
4716 				&ledger_idx_nonvolatile_compressed,
4717 				&do_footprint);
4718 			/* more volatile bytes */
4719 			ledger_credit(owner->ledger,
4720 			    ledger_idx_volatile,
4721 			    PAGE_SIZE);
4722 			/* less not-quite-volatile bytes */
4723 			ledger_debit(owner->ledger,
4724 			    ledger_idx_nonvolatile,
4725 			    PAGE_SIZE);
4726 			if (do_footprint) {
4727 				/* less footprint */
4728 				ledger_debit(owner->ledger,
4729 				    task_ledgers.phys_footprint,
4730 				    PAGE_SIZE);
4731 			}
4732 		}
4733 		assert(m_object != kernel_object);
4734 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
4735 
4736 		if (queueit == TRUE) {
4737 			if (m_object->purgable == VM_PURGABLE_EMPTY) {
4738 				vm_page_deactivate(mem);
4739 			} else {
4740 				vm_page_activate(mem);
4741 			}
4742 		}
4743 
4744 		VM_CHECK_MEMORYSTATUS;
4745 	}
4746 	VM_PAGE_CHECK(mem);
4747 }
4748 
4749 /*
4750  *	vm_page_deactivate:
4751  *
4752  *	Returns the given page to the inactive list,
4753  *	indicating that no physical maps have access
4754  *	to this page.  [Used by the physical mapping system.]
4755  *
4756  *	The page queues must be locked.
4757  */
4758 void
vm_page_deactivate(vm_page_t m)4759 vm_page_deactivate(
4760 	vm_page_t       m)
4761 {
4762 	vm_page_deactivate_internal(m, TRUE);
4763 }
4764 
4765 
4766 void
vm_page_deactivate_internal(vm_page_t m,boolean_t clear_hw_reference)4767 vm_page_deactivate_internal(
4768 	vm_page_t       m,
4769 	boolean_t       clear_hw_reference)
4770 {
4771 	vm_object_t     m_object;
4772 
4773 	m_object = VM_PAGE_OBJECT(m);
4774 
4775 	VM_PAGE_CHECK(m);
4776 	assert(m_object != kernel_object);
4777 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4778 
4779 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(m), vm_page_free_count, vm_page_wire_count, 6);	/* (TEST/DEBUG) */
4780 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4781 	/*
4782 	 *	This page is no longer very interesting.  If it was
4783 	 *	interesting (active or inactive/referenced), then we
4784 	 *	clear the reference bit and (re)enter it in the
4785 	 *	inactive queue.  Note wired pages should not have
4786 	 *	their reference bit cleared.
4787 	 */
4788 	assert( !(m->vmp_absent && !m->vmp_unusual));
4789 
4790 	if (m->vmp_gobbled) {           /* can this happen? */
4791 		assert( !VM_PAGE_WIRED(m));
4792 
4793 		if (!m->vmp_private && !m->vmp_fictitious) {
4794 			vm_page_wire_count--;
4795 		}
4796 		vm_page_gobble_count--;
4797 		m->vmp_gobbled = FALSE;
4798 	}
4799 	/*
4800 	 * if this page is currently on the pageout queue, we can't do the
4801 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4802 	 * and we can't remove it manually since we would need the object lock
4803 	 * (which is not required here) to decrement the activity_in_progress
4804 	 * reference which is held on the object while the page is in the pageout queue...
4805 	 * just let the normal laundry processing proceed
4806 	 */
4807 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4808 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
4809 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
4810 	    VM_PAGE_WIRED(m)) {
4811 		return;
4812 	}
4813 	if (!m->vmp_absent && clear_hw_reference == TRUE) {
4814 		pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(m));
4815 	}
4816 
4817 	m->vmp_reference = FALSE;
4818 	m->vmp_no_cache = FALSE;
4819 
4820 	if (!VM_PAGE_INACTIVE(m)) {
4821 		vm_page_queues_remove(m, FALSE);
4822 
4823 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
4824 		    m->vmp_dirty && m_object->internal &&
4825 		    (m_object->purgable == VM_PURGABLE_DENY ||
4826 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
4827 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
4828 			vm_page_check_pageable_safe(m);
4829 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
4830 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
4831 			vm_page_throttled_count++;
4832 		} else {
4833 			if (m_object->named && m_object->ref_count == 1) {
4834 				vm_page_speculate(m, FALSE);
4835 #if DEVELOPMENT || DEBUG
4836 				vm_page_speculative_recreated++;
4837 #endif
4838 			} else {
4839 				vm_page_enqueue_inactive(m, FALSE);
4840 			}
4841 		}
4842 	}
4843 }
4844 
4845 /*
4846  * vm_page_enqueue_cleaned
4847  *
4848  * Put the page on the cleaned queue, mark it cleaned, etc.
4849  * Being on the cleaned queue (and having m->clean_queue set)
4850  * does ** NOT ** guarantee that the page is clean!
4851  *
4852  * Call with the queues lock held.
4853  */
4854 
4855 void
vm_page_enqueue_cleaned(vm_page_t m)4856 vm_page_enqueue_cleaned(vm_page_t m)
4857 {
4858 	vm_object_t     m_object;
4859 
4860 	m_object = VM_PAGE_OBJECT(m);
4861 
4862 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4863 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4864 	assert( !(m->vmp_absent && !m->vmp_unusual));
4865 
4866 	if (VM_PAGE_WIRED(m)) {
4867 		return;
4868 	}
4869 
4870 	if (m->vmp_gobbled) {
4871 		if (!m->vmp_private && !m->vmp_fictitious) {
4872 			vm_page_wire_count--;
4873 		}
4874 		vm_page_gobble_count--;
4875 		m->vmp_gobbled = FALSE;
4876 	}
4877 	/*
4878 	 * if this page is currently on the pageout queue, we can't do the
4879 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4880 	 * and we can't remove it manually since we would need the object lock
4881 	 * (which is not required here) to decrement the activity_in_progress
4882 	 * reference which is held on the object while the page is in the pageout queue...
4883 	 * just let the normal laundry processing proceed
4884 	 */
4885 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4886 	    (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) ||
4887 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
4888 		return;
4889 	}
4890 	vm_page_queues_remove(m, FALSE);
4891 
4892 	vm_page_check_pageable_safe(m);
4893 	vm_page_queue_enter(&vm_page_queue_cleaned, m, vmp_pageq);
4894 	m->vmp_q_state = VM_PAGE_ON_INACTIVE_CLEANED_Q;
4895 	vm_page_cleaned_count++;
4896 
4897 	vm_page_inactive_count++;
4898 	if (m_object->internal) {
4899 		vm_page_pageable_internal_count++;
4900 	} else {
4901 		vm_page_pageable_external_count++;
4902 	}
4903 	vm_page_add_to_specialq(m, TRUE);
4904 	VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
4905 }
4906 
4907 /*
4908  *	vm_page_activate:
4909  *
4910  *	Put the specified page on the active list (if appropriate).
4911  *
4912  *	The page queues must be locked.
4913  */
4914 
4915 void
vm_page_activate(vm_page_t m)4916 vm_page_activate(
4917 	vm_page_t       m)
4918 {
4919 	vm_object_t     m_object;
4920 
4921 	m_object = VM_PAGE_OBJECT(m);
4922 
4923 	VM_PAGE_CHECK(m);
4924 #ifdef  FIXME_4778297
4925 	assert(m_object != kernel_object);
4926 #endif
4927 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4928 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4929 	assert( !(m->vmp_absent && !m->vmp_unusual));
4930 
4931 	if (m->vmp_gobbled) {
4932 		assert( !VM_PAGE_WIRED(m));
4933 		if (!m->vmp_private && !m->vmp_fictitious) {
4934 			vm_page_wire_count--;
4935 		}
4936 		vm_page_gobble_count--;
4937 		m->vmp_gobbled = FALSE;
4938 	}
4939 	/*
4940 	 * if this page is currently on the pageout queue, we can't do the
4941 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4942 	 * and we can't remove it manually since we would need the object lock
4943 	 * (which is not required here) to decrement the activity_in_progress
4944 	 * reference which is held on the object while the page is in the pageout queue...
4945 	 * just let the normal laundry processing proceed
4946 	 */
4947 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4948 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
4949 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
4950 		return;
4951 	}
4952 
4953 #if DEBUG
4954 	if (m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q) {
4955 		panic("vm_page_activate: already active");
4956 	}
4957 #endif
4958 
4959 	if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) {
4960 		DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL);
4961 		DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL);
4962 	}
4963 
4964 	/*
4965 	 * A freshly activated page should be promoted in the donation queue.
4966 	 * So we remove it here while preserving its hint and we will enqueue
4967 	 * it again in vm_page_enqueue_active.
4968 	 */
4969 	vm_page_queues_remove(m, ((m->vmp_on_specialq == VM_PAGE_SPECIAL_Q_DONATE) ? TRUE : FALSE));
4970 
4971 	if (!VM_PAGE_WIRED(m)) {
4972 		vm_page_check_pageable_safe(m);
4973 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
4974 		    m->vmp_dirty && m_object->internal &&
4975 		    (m_object->purgable == VM_PURGABLE_DENY ||
4976 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
4977 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
4978 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
4979 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
4980 			vm_page_throttled_count++;
4981 		} else {
4982 #if CONFIG_SECLUDED_MEMORY
4983 			if (secluded_for_filecache &&
4984 			    vm_page_secluded_target != 0 &&
4985 			    num_tasks_can_use_secluded_mem == 0 &&
4986 			    m_object->eligible_for_secluded) {
4987 				vm_page_queue_enter(&vm_page_queue_secluded, m, vmp_pageq);
4988 				m->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
4989 				vm_page_secluded_count++;
4990 				VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
4991 				vm_page_secluded_count_inuse++;
4992 				assert(!m_object->internal);
4993 //				vm_page_pageable_external_count++;
4994 			} else
4995 #endif /* CONFIG_SECLUDED_MEMORY */
4996 			vm_page_enqueue_active(m, FALSE);
4997 		}
4998 		m->vmp_reference = TRUE;
4999 		m->vmp_no_cache = FALSE;
5000 	}
5001 	VM_PAGE_CHECK(m);
5002 }
5003 
5004 
5005 /*
5006  *      vm_page_speculate:
5007  *
5008  *      Put the specified page on the speculative list (if appropriate).
5009  *
5010  *      The page queues must be locked.
5011  */
5012 void
vm_page_speculate(vm_page_t m,boolean_t new)5013 vm_page_speculate(
5014 	vm_page_t       m,
5015 	boolean_t       new)
5016 {
5017 	struct vm_speculative_age_q     *aq;
5018 	vm_object_t     m_object;
5019 
5020 	m_object = VM_PAGE_OBJECT(m);
5021 
5022 	VM_PAGE_CHECK(m);
5023 	vm_page_check_pageable_safe(m);
5024 
5025 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5026 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5027 	assert( !(m->vmp_absent && !m->vmp_unusual));
5028 	assert(m_object->internal == FALSE);
5029 
5030 	/*
5031 	 * if this page is currently on the pageout queue, we can't do the
5032 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5033 	 * and we can't remove it manually since we would need the object lock
5034 	 * (which is not required here) to decrement the activity_in_progress
5035 	 * reference which is held on the object while the page is in the pageout queue...
5036 	 * just let the normal laundry processing proceed
5037 	 */
5038 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
5039 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5040 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5041 		return;
5042 	}
5043 
5044 	vm_page_queues_remove(m, FALSE);
5045 
5046 	if (!VM_PAGE_WIRED(m)) {
5047 		mach_timespec_t         ts;
5048 		clock_sec_t sec;
5049 		clock_nsec_t nsec;
5050 
5051 		clock_get_system_nanotime(&sec, &nsec);
5052 		ts.tv_sec = (unsigned int) sec;
5053 		ts.tv_nsec = nsec;
5054 
5055 		if (vm_page_speculative_count == 0) {
5056 			speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5057 			speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5058 
5059 			aq = &vm_page_queue_speculative[speculative_age_index];
5060 
5061 			/*
5062 			 * set the timer to begin a new group
5063 			 */
5064 			aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5065 			aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5066 
5067 			ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5068 		} else {
5069 			aq = &vm_page_queue_speculative[speculative_age_index];
5070 
5071 			if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) {
5072 				speculative_age_index++;
5073 
5074 				if (speculative_age_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) {
5075 					speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5076 				}
5077 				if (speculative_age_index == speculative_steal_index) {
5078 					speculative_steal_index = speculative_age_index + 1;
5079 
5080 					if (speculative_steal_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) {
5081 						speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5082 					}
5083 				}
5084 				aq = &vm_page_queue_speculative[speculative_age_index];
5085 
5086 				if (!vm_page_queue_empty(&aq->age_q)) {
5087 					vm_page_speculate_ageit(aq);
5088 				}
5089 
5090 				aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5091 				aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5092 
5093 				ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5094 			}
5095 		}
5096 		vm_page_enqueue_tail(&aq->age_q, &m->vmp_pageq);
5097 		m->vmp_q_state = VM_PAGE_ON_SPECULATIVE_Q;
5098 		vm_page_speculative_count++;
5099 		vm_page_pageable_external_count++;
5100 
5101 		if (new == TRUE) {
5102 			vm_object_lock_assert_exclusive(m_object);
5103 
5104 			m_object->pages_created++;
5105 #if DEVELOPMENT || DEBUG
5106 			vm_page_speculative_created++;
5107 #endif
5108 		}
5109 	}
5110 	VM_PAGE_CHECK(m);
5111 }
5112 
5113 
5114 /*
5115  * move pages from the specified aging bin to
5116  * the speculative bin that pageout_scan claims from
5117  *
5118  *      The page queues must be locked.
5119  */
5120 void
vm_page_speculate_ageit(struct vm_speculative_age_q * aq)5121 vm_page_speculate_ageit(struct vm_speculative_age_q *aq)
5122 {
5123 	struct vm_speculative_age_q     *sq;
5124 	vm_page_t       t;
5125 
5126 	sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q];
5127 
5128 	if (vm_page_queue_empty(&sq->age_q)) {
5129 		sq->age_q.next = aq->age_q.next;
5130 		sq->age_q.prev = aq->age_q.prev;
5131 
5132 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.next);
5133 		t->vmp_pageq.prev = VM_PAGE_PACK_PTR(&sq->age_q);
5134 
5135 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5136 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5137 	} else {
5138 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5139 		t->vmp_pageq.next = aq->age_q.next;
5140 
5141 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.next);
5142 		t->vmp_pageq.prev = sq->age_q.prev;
5143 
5144 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.prev);
5145 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5146 
5147 		sq->age_q.prev = aq->age_q.prev;
5148 	}
5149 	vm_page_queue_init(&aq->age_q);
5150 }
5151 
5152 
5153 void
vm_page_lru(vm_page_t m)5154 vm_page_lru(
5155 	vm_page_t       m)
5156 {
5157 	VM_PAGE_CHECK(m);
5158 	assert(VM_PAGE_OBJECT(m) != kernel_object);
5159 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5160 
5161 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5162 
5163 	if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q) {
5164 		/*
5165 		 * we don't need to do all the other work that
5166 		 * vm_page_queues_remove and vm_page_enqueue_inactive
5167 		 * bring along for the ride
5168 		 */
5169 		assert(!m->vmp_laundry);
5170 		assert(!m->vmp_private);
5171 
5172 		m->vmp_no_cache = FALSE;
5173 
5174 		vm_page_queue_remove(&vm_page_queue_inactive, m, vmp_pageq);
5175 		vm_page_queue_enter(&vm_page_queue_inactive, m, vmp_pageq);
5176 
5177 		return;
5178 	}
5179 	/*
5180 	 * if this page is currently on the pageout queue, we can't do the
5181 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5182 	 * and we can't remove it manually since we would need the object lock
5183 	 * (which is not required here) to decrement the activity_in_progress
5184 	 * reference which is held on the object while the page is in the pageout queue...
5185 	 * just let the normal laundry processing proceed
5186 	 */
5187 	if (m->vmp_laundry || m->vmp_private ||
5188 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5189 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
5190 	    VM_PAGE_WIRED(m)) {
5191 		return;
5192 	}
5193 
5194 	m->vmp_no_cache = FALSE;
5195 
5196 	vm_page_queues_remove(m, FALSE);
5197 
5198 	vm_page_enqueue_inactive(m, FALSE);
5199 }
5200 
5201 
5202 void
vm_page_reactivate_all_throttled(void)5203 vm_page_reactivate_all_throttled(void)
5204 {
5205 	vm_page_t       first_throttled, last_throttled;
5206 	vm_page_t       first_active;
5207 	vm_page_t       m;
5208 	int             extra_active_count;
5209 	int             extra_internal_count, extra_external_count;
5210 	vm_object_t     m_object;
5211 
5212 	if (!VM_DYNAMIC_PAGING_ENABLED()) {
5213 		return;
5214 	}
5215 
5216 	extra_active_count = 0;
5217 	extra_internal_count = 0;
5218 	extra_external_count = 0;
5219 	vm_page_lock_queues();
5220 	if (!vm_page_queue_empty(&vm_page_queue_throttled)) {
5221 		/*
5222 		 * Switch "throttled" pages to "active".
5223 		 */
5224 		vm_page_queue_iterate(&vm_page_queue_throttled, m, vmp_pageq) {
5225 			VM_PAGE_CHECK(m);
5226 			assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
5227 
5228 			m_object = VM_PAGE_OBJECT(m);
5229 
5230 			extra_active_count++;
5231 			if (m_object->internal) {
5232 				extra_internal_count++;
5233 			} else {
5234 				extra_external_count++;
5235 			}
5236 
5237 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5238 			VM_PAGE_CHECK(m);
5239 			vm_page_add_to_specialq(m, FALSE);
5240 		}
5241 
5242 		/*
5243 		 * Transfer the entire throttled queue to a regular LRU page queues.
5244 		 * We insert it at the head of the active queue, so that these pages
5245 		 * get re-evaluated by the LRU algorithm first, since they've been
5246 		 * completely out of it until now.
5247 		 */
5248 		first_throttled = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
5249 		last_throttled = (vm_page_t) vm_page_queue_last(&vm_page_queue_throttled);
5250 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5251 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5252 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5253 		} else {
5254 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5255 		}
5256 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_throttled);
5257 		first_throttled->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5258 		last_throttled->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5259 
5260 #if DEBUG
5261 		printf("reactivated %d throttled pages\n", vm_page_throttled_count);
5262 #endif
5263 		vm_page_queue_init(&vm_page_queue_throttled);
5264 		/*
5265 		 * Adjust the global page counts.
5266 		 */
5267 		vm_page_active_count += extra_active_count;
5268 		vm_page_pageable_internal_count += extra_internal_count;
5269 		vm_page_pageable_external_count += extra_external_count;
5270 		vm_page_throttled_count = 0;
5271 	}
5272 	assert(vm_page_throttled_count == 0);
5273 	assert(vm_page_queue_empty(&vm_page_queue_throttled));
5274 	vm_page_unlock_queues();
5275 }
5276 
5277 
5278 /*
5279  * move pages from the indicated local queue to the global active queue
5280  * its ok to fail if we're below the hard limit and force == FALSE
5281  * the nolocks == TRUE case is to allow this function to be run on
5282  * the hibernate path
5283  */
5284 
5285 void
vm_page_reactivate_local(uint32_t lid,boolean_t force,boolean_t nolocks)5286 vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks)
5287 {
5288 	struct vpl      *lq;
5289 	vm_page_t       first_local, last_local;
5290 	vm_page_t       first_active;
5291 	vm_page_t       m;
5292 	uint32_t        count = 0;
5293 
5294 	if (vm_page_local_q == NULL) {
5295 		return;
5296 	}
5297 
5298 	lq = zpercpu_get_cpu(vm_page_local_q, lid);
5299 
5300 	if (nolocks == FALSE) {
5301 		if (lq->vpl_count < vm_page_local_q_hard_limit && force == FALSE) {
5302 			if (!vm_page_trylockspin_queues()) {
5303 				return;
5304 			}
5305 		} else {
5306 			vm_page_lockspin_queues();
5307 		}
5308 
5309 		VPL_LOCK(&lq->vpl_lock);
5310 	}
5311 	if (lq->vpl_count) {
5312 		/*
5313 		 * Switch "local" pages to "active".
5314 		 */
5315 		assert(!vm_page_queue_empty(&lq->vpl_queue));
5316 
5317 		vm_page_queue_iterate(&lq->vpl_queue, m, vmp_pageq) {
5318 			VM_PAGE_CHECK(m);
5319 			vm_page_check_pageable_safe(m);
5320 			assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q);
5321 			assert(!m->vmp_fictitious);
5322 
5323 			if (m->vmp_local_id != lid) {
5324 				panic("vm_page_reactivate_local: found vm_page_t(%p) with wrong cpuid", m);
5325 			}
5326 
5327 			m->vmp_local_id = 0;
5328 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5329 			VM_PAGE_CHECK(m);
5330 			vm_page_add_to_specialq(m, FALSE);
5331 			count++;
5332 		}
5333 		if (count != lq->vpl_count) {
5334 			panic("vm_page_reactivate_local: count = %d, vm_page_local_count = %d", count, lq->vpl_count);
5335 		}
5336 
5337 		/*
5338 		 * Transfer the entire local queue to a regular LRU page queues.
5339 		 */
5340 		first_local = (vm_page_t) vm_page_queue_first(&lq->vpl_queue);
5341 		last_local = (vm_page_t) vm_page_queue_last(&lq->vpl_queue);
5342 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5343 
5344 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5345 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5346 		} else {
5347 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5348 		}
5349 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_local);
5350 		first_local->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5351 		last_local->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5352 
5353 		vm_page_queue_init(&lq->vpl_queue);
5354 		/*
5355 		 * Adjust the global page counts.
5356 		 */
5357 		vm_page_active_count += lq->vpl_count;
5358 		vm_page_pageable_internal_count += lq->vpl_internal_count;
5359 		vm_page_pageable_external_count += lq->vpl_external_count;
5360 		lq->vpl_count = 0;
5361 		lq->vpl_internal_count = 0;
5362 		lq->vpl_external_count = 0;
5363 	}
5364 	assert(vm_page_queue_empty(&lq->vpl_queue));
5365 
5366 	if (nolocks == FALSE) {
5367 		VPL_UNLOCK(&lq->vpl_lock);
5368 
5369 		vm_page_balance_inactive(count / 4);
5370 		vm_page_unlock_queues();
5371 	}
5372 }
5373 
5374 /*
5375  *	vm_page_part_zero_fill:
5376  *
5377  *	Zero-fill a part of the page.
5378  */
5379 #define PMAP_ZERO_PART_PAGE_IMPLEMENTED
5380 void
vm_page_part_zero_fill(vm_page_t m,vm_offset_t m_pa,vm_size_t len)5381 vm_page_part_zero_fill(
5382 	vm_page_t       m,
5383 	vm_offset_t     m_pa,
5384 	vm_size_t       len)
5385 {
5386 #if 0
5387 	/*
5388 	 * we don't hold the page queue lock
5389 	 * so this check isn't safe to make
5390 	 */
5391 	VM_PAGE_CHECK(m);
5392 #endif
5393 
5394 #ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED
5395 	pmap_zero_part_page(VM_PAGE_GET_PHYS_PAGE(m), m_pa, len);
5396 #else
5397 	vm_page_t       tmp;
5398 	while (1) {
5399 		tmp = vm_page_grab();
5400 		if (tmp == VM_PAGE_NULL) {
5401 			vm_page_wait(THREAD_UNINT);
5402 			continue;
5403 		}
5404 		break;
5405 	}
5406 	vm_page_zero_fill(tmp);
5407 	if (m_pa != 0) {
5408 		vm_page_part_copy(m, 0, tmp, 0, m_pa);
5409 	}
5410 	if ((m_pa + len) < PAGE_SIZE) {
5411 		vm_page_part_copy(m, m_pa + len, tmp,
5412 		    m_pa + len, PAGE_SIZE - (m_pa + len));
5413 	}
5414 	vm_page_copy(tmp, m);
5415 	VM_PAGE_FREE(tmp);
5416 #endif
5417 }
5418 
5419 /*
5420  *	vm_page_zero_fill:
5421  *
5422  *	Zero-fill the specified page.
5423  */
5424 void
vm_page_zero_fill(vm_page_t m)5425 vm_page_zero_fill(
5426 	vm_page_t       m)
5427 {
5428 #if 0
5429 	/*
5430 	 * we don't hold the page queue lock
5431 	 * so this check isn't safe to make
5432 	 */
5433 	VM_PAGE_CHECK(m);
5434 #endif
5435 
5436 //	dbgTrace(0xAEAEAEAE, VM_PAGE_GET_PHYS_PAGE(m), 0);		/* (BRINGUP) */
5437 	pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m));
5438 }
5439 
5440 /*
5441  *	vm_page_part_copy:
5442  *
5443  *	copy part of one page to another
5444  */
5445 
5446 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)5447 vm_page_part_copy(
5448 	vm_page_t       src_m,
5449 	vm_offset_t     src_pa,
5450 	vm_page_t       dst_m,
5451 	vm_offset_t     dst_pa,
5452 	vm_size_t       len)
5453 {
5454 #if 0
5455 	/*
5456 	 * we don't hold the page queue lock
5457 	 * so this check isn't safe to make
5458 	 */
5459 	VM_PAGE_CHECK(src_m);
5460 	VM_PAGE_CHECK(dst_m);
5461 #endif
5462 	pmap_copy_part_page(VM_PAGE_GET_PHYS_PAGE(src_m), src_pa,
5463 	    VM_PAGE_GET_PHYS_PAGE(dst_m), dst_pa, len);
5464 }
5465 
5466 /*
5467  *	vm_page_copy:
5468  *
5469  *	Copy one page to another
5470  */
5471 
5472 int vm_page_copy_cs_validations = 0;
5473 int vm_page_copy_cs_tainted = 0;
5474 
5475 void
vm_page_copy(vm_page_t src_m,vm_page_t dest_m)5476 vm_page_copy(
5477 	vm_page_t       src_m,
5478 	vm_page_t       dest_m)
5479 {
5480 	vm_object_t     src_m_object;
5481 
5482 	src_m_object = VM_PAGE_OBJECT(src_m);
5483 
5484 #if 0
5485 	/*
5486 	 * we don't hold the page queue lock
5487 	 * so this check isn't safe to make
5488 	 */
5489 	VM_PAGE_CHECK(src_m);
5490 	VM_PAGE_CHECK(dest_m);
5491 #endif
5492 	vm_object_lock_assert_held(src_m_object);
5493 
5494 	if (src_m_object != VM_OBJECT_NULL &&
5495 	    src_m_object->code_signed) {
5496 		/*
5497 		 * We're copying a page from a code-signed object.
5498 		 * Whoever ends up mapping the copy page might care about
5499 		 * the original page's integrity, so let's validate the
5500 		 * source page now.
5501 		 */
5502 		vm_page_copy_cs_validations++;
5503 		vm_page_validate_cs(src_m, PAGE_SIZE, 0);
5504 #if DEVELOPMENT || DEBUG
5505 		DTRACE_VM4(codesigned_copy,
5506 		    vm_object_t, src_m_object,
5507 		    vm_object_offset_t, src_m->vmp_offset,
5508 		    int, src_m->vmp_cs_validated,
5509 		    int, src_m->vmp_cs_tainted);
5510 #endif /* DEVELOPMENT || DEBUG */
5511 	}
5512 
5513 	/*
5514 	 * Propagate the cs_tainted bit to the copy page. Do not propagate
5515 	 * the cs_validated bit.
5516 	 */
5517 	dest_m->vmp_cs_tainted = src_m->vmp_cs_tainted;
5518 	dest_m->vmp_cs_nx = src_m->vmp_cs_nx;
5519 	if (dest_m->vmp_cs_tainted) {
5520 		vm_page_copy_cs_tainted++;
5521 	}
5522 	dest_m->vmp_error = VMP_ERROR_GET(src_m); /* sliding src_m might have failed... */
5523 	pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(src_m), VM_PAGE_GET_PHYS_PAGE(dest_m));
5524 }
5525 
5526 #if MACH_ASSERT
5527 static void
_vm_page_print(vm_page_t p)5528 _vm_page_print(
5529 	vm_page_t       p)
5530 {
5531 	printf("vm_page %p: \n", p);
5532 	printf("  pageq: next=%p prev=%p\n",
5533 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.next),
5534 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev));
5535 	printf("  listq: next=%p prev=%p\n",
5536 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.next)),
5537 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.prev)));
5538 	printf("  next=%p\n", (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m)));
5539 	printf("  object=%p offset=0x%llx\n", VM_PAGE_OBJECT(p), p->vmp_offset);
5540 	printf("  wire_count=%u\n", p->vmp_wire_count);
5541 	printf("  q_state=%u\n", p->vmp_q_state);
5542 
5543 	printf("  %slaundry, %sref, %sgobbled, %sprivate\n",
5544 	    (p->vmp_laundry ? "" : "!"),
5545 	    (p->vmp_reference ? "" : "!"),
5546 	    (p->vmp_gobbled ? "" : "!"),
5547 	    (p->vmp_private ? "" : "!"));
5548 	printf("  %sbusy, %swanted, %stabled, %sfictitious, %spmapped, %swpmapped\n",
5549 	    (p->vmp_busy ? "" : "!"),
5550 	    (p->vmp_wanted ? "" : "!"),
5551 	    (p->vmp_tabled ? "" : "!"),
5552 	    (p->vmp_fictitious ? "" : "!"),
5553 	    (p->vmp_pmapped ? "" : "!"),
5554 	    (p->vmp_wpmapped ? "" : "!"));
5555 	printf("  %sfree_when_done, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n",
5556 	    (p->vmp_free_when_done ? "" : "!"),
5557 	    (p->vmp_absent ? "" : "!"),
5558 	    (VMP_ERROR_GET(p) ? "" : "!"),
5559 	    (p->vmp_dirty ? "" : "!"),
5560 	    (p->vmp_cleaning ? "" : "!"),
5561 	    (p->vmp_precious ? "" : "!"),
5562 	    (p->vmp_clustered ? "" : "!"));
5563 	printf("  %soverwriting, %srestart, %sunusual\n",
5564 	    (p->vmp_overwriting ? "" : "!"),
5565 	    (p->vmp_restart ? "" : "!"),
5566 	    (p->vmp_unusual ? "" : "!"));
5567 	printf("  cs_validated=%d, cs_tainted=%d, cs_nx=%d, %sno_cache\n",
5568 	    p->vmp_cs_validated,
5569 	    p->vmp_cs_tainted,
5570 	    p->vmp_cs_nx,
5571 	    (p->vmp_no_cache ? "" : "!"));
5572 
5573 	printf("phys_page=0x%x\n", VM_PAGE_GET_PHYS_PAGE(p));
5574 }
5575 
5576 /*
5577  *	Check that the list of pages is ordered by
5578  *	ascending physical address and has no holes.
5579  */
5580 static int
vm_page_verify_contiguous(vm_page_t pages,unsigned int npages)5581 vm_page_verify_contiguous(
5582 	vm_page_t       pages,
5583 	unsigned int    npages)
5584 {
5585 	vm_page_t               m;
5586 	unsigned int            page_count;
5587 	vm_offset_t             prev_addr;
5588 
5589 	prev_addr = VM_PAGE_GET_PHYS_PAGE(pages);
5590 	page_count = 1;
5591 	for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
5592 		if (VM_PAGE_GET_PHYS_PAGE(m) != prev_addr + 1) {
5593 			printf("m %p prev_addr 0x%lx, current addr 0x%x\n",
5594 			    m, (long)prev_addr, VM_PAGE_GET_PHYS_PAGE(m));
5595 			printf("pages %p page_count %d npages %d\n", pages, page_count, npages);
5596 			panic("vm_page_verify_contiguous:  not contiguous!");
5597 		}
5598 		prev_addr = VM_PAGE_GET_PHYS_PAGE(m);
5599 		++page_count;
5600 	}
5601 	if (page_count != npages) {
5602 		printf("pages %p actual count 0x%x but requested 0x%x\n",
5603 		    pages, page_count, npages);
5604 		panic("vm_page_verify_contiguous:  count error");
5605 	}
5606 	return 1;
5607 }
5608 
5609 
5610 /*
5611  *	Check the free lists for proper length etc.
5612  */
5613 static boolean_t vm_page_verify_this_free_list_enabled = FALSE;
5614 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)5615 vm_page_verify_free_list(
5616 	vm_page_queue_head_t    *vm_page_queue,
5617 	unsigned int    color,
5618 	vm_page_t       look_for_page,
5619 	boolean_t       expect_page)
5620 {
5621 	unsigned int    npages;
5622 	vm_page_t       m;
5623 	vm_page_t       prev_m;
5624 	boolean_t       found_page;
5625 
5626 	if (!vm_page_verify_this_free_list_enabled) {
5627 		return 0;
5628 	}
5629 
5630 	found_page = FALSE;
5631 	npages = 0;
5632 	prev_m = (vm_page_t)((uintptr_t)vm_page_queue);
5633 
5634 	vm_page_queue_iterate(vm_page_queue, m, vmp_pageq) {
5635 		if (m == look_for_page) {
5636 			found_page = TRUE;
5637 		}
5638 		if ((vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev) != prev_m) {
5639 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p corrupted prev ptr %p instead of %p",
5640 			    color, npages, m, (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev), prev_m);
5641 		}
5642 		if (!m->vmp_busy) {
5643 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy",
5644 			    color, npages, m);
5645 		}
5646 		if (color != (unsigned int) -1) {
5647 			if (VM_PAGE_GET_COLOR(m) != color) {
5648 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u",
5649 				    color, npages, m, VM_PAGE_GET_COLOR(m), color);
5650 			}
5651 			if (m->vmp_q_state != VM_PAGE_ON_FREE_Q) {
5652 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p - expecting q_state == VM_PAGE_ON_FREE_Q, found %d",
5653 				    color, npages, m, m->vmp_q_state);
5654 			}
5655 		} else {
5656 			if (m->vmp_q_state != VM_PAGE_ON_FREE_LOCAL_Q) {
5657 				panic("vm_page_verify_free_list(npages=%u): local page %p - expecting q_state == VM_PAGE_ON_FREE_LOCAL_Q, found %d",
5658 				    npages, m, m->vmp_q_state);
5659 			}
5660 		}
5661 		++npages;
5662 		prev_m = m;
5663 	}
5664 	if (look_for_page != VM_PAGE_NULL) {
5665 		unsigned int other_color;
5666 
5667 		if (expect_page && !found_page) {
5668 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p not found phys=%u\n",
5669 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5670 			_vm_page_print(look_for_page);
5671 			for (other_color = 0;
5672 			    other_color < vm_colors;
5673 			    other_color++) {
5674 				if (other_color == color) {
5675 					continue;
5676 				}
5677 				vm_page_verify_free_list(&vm_page_queue_free[other_color].qhead,
5678 				    other_color, look_for_page, FALSE);
5679 			}
5680 			if (color == (unsigned int) -1) {
5681 				vm_page_verify_free_list(&vm_lopage_queue_free,
5682 				    (unsigned int) -1, look_for_page, FALSE);
5683 			}
5684 			panic("vm_page_verify_free_list(color=%u)", color);
5685 		}
5686 		if (!expect_page && found_page) {
5687 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p found phys=%u\n",
5688 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5689 		}
5690 	}
5691 	return npages;
5692 }
5693 
5694 static boolean_t vm_page_verify_all_free_lists_enabled = FALSE;
5695 static void
vm_page_verify_free_lists(void)5696 vm_page_verify_free_lists( void )
5697 {
5698 	unsigned int    color, npages, nlopages;
5699 	boolean_t       toggle = TRUE;
5700 
5701 	if (!vm_page_verify_all_free_lists_enabled) {
5702 		return;
5703 	}
5704 
5705 	npages = 0;
5706 
5707 	vm_free_page_lock();
5708 
5709 	if (vm_page_verify_this_free_list_enabled == TRUE) {
5710 		/*
5711 		 * This variable has been set globally for extra checking of
5712 		 * each free list Q. Since we didn't set it, we don't own it
5713 		 * and we shouldn't toggle it.
5714 		 */
5715 		toggle = FALSE;
5716 	}
5717 
5718 	if (toggle == TRUE) {
5719 		vm_page_verify_this_free_list_enabled = TRUE;
5720 	}
5721 
5722 	for (color = 0; color < vm_colors; color++) {
5723 		npages += vm_page_verify_free_list(&vm_page_queue_free[color].qhead,
5724 		    color, VM_PAGE_NULL, FALSE);
5725 	}
5726 	nlopages = vm_page_verify_free_list(&vm_lopage_queue_free,
5727 	    (unsigned int) -1,
5728 	    VM_PAGE_NULL, FALSE);
5729 	if (npages != vm_page_free_count || nlopages != vm_lopage_free_count) {
5730 		panic("vm_page_verify_free_lists:  "
5731 		    "npages %u free_count %d nlopages %u lo_free_count %u",
5732 		    npages, vm_page_free_count, nlopages, vm_lopage_free_count);
5733 	}
5734 
5735 	if (toggle == TRUE) {
5736 		vm_page_verify_this_free_list_enabled = FALSE;
5737 	}
5738 
5739 	vm_free_page_unlock();
5740 }
5741 
5742 #endif  /* MACH_ASSERT */
5743 
5744 
5745 extern boolean_t(*volatile consider_buffer_cache_collect)(int);
5746 
5747 /*
5748  *	CONTIGUOUS PAGE ALLOCATION
5749  *
5750  *	Find a region large enough to contain at least n pages
5751  *	of contiguous physical memory.
5752  *
5753  *	This is done by traversing the vm_page_t array in a linear fashion
5754  *	we assume that the vm_page_t array has the avaiable physical pages in an
5755  *	ordered, ascending list... this is currently true of all our implementations
5756  *      and must remain so... there can be 'holes' in the array...  we also can
5757  *	no longer tolerate the vm_page_t's in the list being 'freed' and reclaimed
5758  *      which use to happen via 'vm_page_convert'... that function was no longer
5759  *      being called and was removed...
5760  *
5761  *	The basic flow consists of stabilizing some of the interesting state of
5762  *	a vm_page_t behind the vm_page_queue and vm_page_free locks... we start our
5763  *	sweep at the beginning of the array looking for pages that meet our criterea
5764  *	for a 'stealable' page... currently we are pretty conservative... if the page
5765  *	meets this criterea and is physically contiguous to the previous page in the 'run'
5766  *      we keep developing it.  If we hit a page that doesn't fit, we reset our state
5767  *	and start to develop a new run... if at this point we've already considered
5768  *      at least MAX_CONSIDERED_BEFORE_YIELD pages, we'll drop the 2 locks we hold,
5769  *	and mutex_pause (which will yield the processor), to keep the latency low w/r
5770  *	to other threads trying to acquire free pages (or move pages from q to q),
5771  *	and then continue from the spot we left off... we only make 1 pass through the
5772  *	array.  Once we have a 'run' that is long enough, we'll go into the loop which
5773  *      which steals the pages from the queues they're currently on... pages on the free
5774  *	queue can be stolen directly... pages that are on any of the other queues
5775  *	must be removed from the object they are tabled on... this requires taking the
5776  *      object lock... we do this as a 'try' to prevent deadlocks... if the 'try' fails
5777  *	or if the state of the page behind the vm_object lock is no longer viable, we'll
5778  *	dump the pages we've currently stolen back to the free list, and pick up our
5779  *	scan from the point where we aborted the 'current' run.
5780  *
5781  *
5782  *	Requirements:
5783  *		- neither vm_page_queue nor vm_free_list lock can be held on entry
5784  *
5785  *	Returns a pointer to a list of gobbled/wired pages or VM_PAGE_NULL.
5786  *
5787  * Algorithm:
5788  */
5789 
5790 #define MAX_CONSIDERED_BEFORE_YIELD     1000
5791 
5792 
5793 #define RESET_STATE_OF_RUN()    \
5794 	MACRO_BEGIN             \
5795 	prevcontaddr = -2;      \
5796 	start_pnum = -1;        \
5797 	free_considered = 0;    \
5798 	substitute_needed = 0;  \
5799 	npages = 0;             \
5800 	MACRO_END
5801 
5802 /*
5803  * Can we steal in-use (i.e. not free) pages when searching for
5804  * physically-contiguous pages ?
5805  */
5806 #define VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL 1
5807 
5808 static unsigned int vm_page_find_contiguous_last_idx = 0, vm_page_lomem_find_contiguous_last_idx = 0;
5809 #if DEBUG
5810 int vm_page_find_contig_debug = 0;
5811 #endif
5812 
5813 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)5814 vm_page_find_contiguous(
5815 	unsigned int    contig_pages,
5816 	ppnum_t         max_pnum,
5817 	ppnum_t     pnum_mask,
5818 	boolean_t       wire,
5819 	int             flags)
5820 {
5821 	vm_page_t       m = NULL;
5822 	ppnum_t         prevcontaddr = 0;
5823 	ppnum_t         start_pnum = 0;
5824 	unsigned int    npages = 0, considered = 0, scanned = 0;
5825 	unsigned int    page_idx = 0, start_idx = 0, last_idx = 0, orig_last_idx = 0;
5826 	unsigned int    idx_last_contig_page_found = 0;
5827 	int             free_considered = 0, free_available = 0;
5828 	int             substitute_needed = 0;
5829 	int             zone_gc_called = 0;
5830 	boolean_t       wrapped;
5831 	kern_return_t   kr;
5832 #if DEBUG
5833 	clock_sec_t     tv_start_sec = 0, tv_end_sec = 0;
5834 	clock_usec_t    tv_start_usec = 0, tv_end_usec = 0;
5835 #endif
5836 
5837 	int             yielded = 0;
5838 	int             dumped_run = 0;
5839 	int             stolen_pages = 0;
5840 	int             compressed_pages = 0;
5841 
5842 
5843 	if (contig_pages == 0) {
5844 		return VM_PAGE_NULL;
5845 	}
5846 
5847 full_scan_again:
5848 
5849 #if MACH_ASSERT
5850 	vm_page_verify_free_lists();
5851 #endif
5852 #if DEBUG
5853 	clock_get_system_microtime(&tv_start_sec, &tv_start_usec);
5854 #endif
5855 	PAGE_REPLACEMENT_ALLOWED(TRUE);
5856 
5857 	/*
5858 	 * If there are still delayed pages, try to free up some that match.
5859 	 */
5860 	if (__improbable(vm_delayed_count != 0 && contig_pages != 0)) {
5861 		vm_free_delayed_pages_contig(contig_pages, max_pnum, pnum_mask);
5862 	}
5863 
5864 	vm_page_lock_queues();
5865 	vm_free_page_lock();
5866 
5867 	RESET_STATE_OF_RUN();
5868 
5869 	scanned = 0;
5870 	considered = 0;
5871 	free_available = vm_page_free_count - vm_page_free_reserved;
5872 
5873 	wrapped = FALSE;
5874 
5875 	if (flags & KMA_LOMEM) {
5876 		idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx;
5877 	} else {
5878 		idx_last_contig_page_found =  vm_page_find_contiguous_last_idx;
5879 	}
5880 
5881 	orig_last_idx = idx_last_contig_page_found;
5882 	last_idx = orig_last_idx;
5883 
5884 	for (page_idx = last_idx, start_idx = last_idx;
5885 	    npages < contig_pages && page_idx < vm_pages_count;
5886 	    page_idx++) {
5887 retry:
5888 		if (wrapped &&
5889 		    npages == 0 &&
5890 		    page_idx >= orig_last_idx) {
5891 			/*
5892 			 * We're back where we started and we haven't
5893 			 * found any suitable contiguous range.  Let's
5894 			 * give up.
5895 			 */
5896 			break;
5897 		}
5898 		scanned++;
5899 		m = &vm_pages[page_idx];
5900 
5901 		assert(!m->vmp_fictitious);
5902 		assert(!m->vmp_private);
5903 
5904 		if (max_pnum && VM_PAGE_GET_PHYS_PAGE(m) > max_pnum) {
5905 			/* no more low pages... */
5906 			break;
5907 		}
5908 		if (!npages & ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0)) {
5909 			/*
5910 			 * not aligned
5911 			 */
5912 			RESET_STATE_OF_RUN();
5913 		} else if (VM_PAGE_WIRED(m) || m->vmp_gobbled ||
5914 		    m->vmp_laundry || m->vmp_wanted ||
5915 		    m->vmp_cleaning || m->vmp_overwriting || m->vmp_free_when_done) {
5916 			/*
5917 			 * page is in a transient state
5918 			 * or a state we don't want to deal
5919 			 * with, so don't consider it which
5920 			 * means starting a new run
5921 			 */
5922 			RESET_STATE_OF_RUN();
5923 		} else if ((m->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
5924 		    (m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q) ||
5925 		    (m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q) ||
5926 		    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5927 			/*
5928 			 * page needs to be on one of our queues (other then the pageout or special free queues)
5929 			 * or it needs to belong to the compressor pool (which is now indicated
5930 			 * by vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR and falls out
5931 			 * from the check for VM_PAGE_NOT_ON_Q)
5932 			 * in order for it to be stable behind the
5933 			 * locks we hold at this point...
5934 			 * if not, don't consider it which
5935 			 * means starting a new run
5936 			 */
5937 			RESET_STATE_OF_RUN();
5938 		} else if ((m->vmp_q_state != VM_PAGE_ON_FREE_Q) && (!m->vmp_tabled || m->vmp_busy)) {
5939 			/*
5940 			 * pages on the free list are always 'busy'
5941 			 * so we couldn't test for 'busy' in the check
5942 			 * for the transient states... pages that are
5943 			 * 'free' are never 'tabled', so we also couldn't
5944 			 * test for 'tabled'.  So we check here to make
5945 			 * sure that a non-free page is not busy and is
5946 			 * tabled on an object...
5947 			 * if not, don't consider it which
5948 			 * means starting a new run
5949 			 */
5950 			RESET_STATE_OF_RUN();
5951 		} else {
5952 			if (VM_PAGE_GET_PHYS_PAGE(m) != prevcontaddr + 1) {
5953 				if ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0) {
5954 					RESET_STATE_OF_RUN();
5955 					goto did_consider;
5956 				} else {
5957 					npages = 1;
5958 					start_idx = page_idx;
5959 					start_pnum = VM_PAGE_GET_PHYS_PAGE(m);
5960 				}
5961 			} else {
5962 				npages++;
5963 			}
5964 			prevcontaddr = VM_PAGE_GET_PHYS_PAGE(m);
5965 
5966 			VM_PAGE_CHECK(m);
5967 			if (m->vmp_q_state == VM_PAGE_ON_FREE_Q) {
5968 				free_considered++;
5969 			} else {
5970 				/*
5971 				 * This page is not free.
5972 				 * If we can't steal used pages,
5973 				 * we have to give up this run
5974 				 * and keep looking.
5975 				 * Otherwise, we might need to
5976 				 * move the contents of this page
5977 				 * into a substitute page.
5978 				 */
5979 #if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
5980 				if (m->vmp_pmapped || m->vmp_dirty || m->vmp_precious) {
5981 					substitute_needed++;
5982 				}
5983 #else
5984 				RESET_STATE_OF_RUN();
5985 #endif
5986 			}
5987 
5988 			if ((free_considered + substitute_needed) > free_available) {
5989 				/*
5990 				 * if we let this run continue
5991 				 * we will end up dropping the vm_page_free_count
5992 				 * below the reserve limit... we need to abort
5993 				 * this run, but we can at least re-consider this
5994 				 * page... thus the jump back to 'retry'
5995 				 */
5996 				RESET_STATE_OF_RUN();
5997 
5998 				if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) {
5999 					considered++;
6000 					goto retry;
6001 				}
6002 				/*
6003 				 * free_available == 0
6004 				 * so can't consider any free pages... if
6005 				 * we went to retry in this case, we'd
6006 				 * get stuck looking at the same page
6007 				 * w/o making any forward progress
6008 				 * we also want to take this path if we've already
6009 				 * reached our limit that controls the lock latency
6010 				 */
6011 			}
6012 		}
6013 did_consider:
6014 		if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) {
6015 			PAGE_REPLACEMENT_ALLOWED(FALSE);
6016 
6017 			vm_free_page_unlock();
6018 			vm_page_unlock_queues();
6019 
6020 			mutex_pause(0);
6021 
6022 			PAGE_REPLACEMENT_ALLOWED(TRUE);
6023 
6024 			vm_page_lock_queues();
6025 			vm_free_page_lock();
6026 
6027 			RESET_STATE_OF_RUN();
6028 			/*
6029 			 * reset our free page limit since we
6030 			 * dropped the lock protecting the vm_page_free_queue
6031 			 */
6032 			free_available = vm_page_free_count - vm_page_free_reserved;
6033 			considered = 0;
6034 
6035 			yielded++;
6036 
6037 			goto retry;
6038 		}
6039 		considered++;
6040 	}
6041 	m = VM_PAGE_NULL;
6042 
6043 	if (npages != contig_pages) {
6044 		if (!wrapped) {
6045 			/*
6046 			 * We didn't find a contiguous range but we didn't
6047 			 * start from the very first page.
6048 			 * Start again from the very first page.
6049 			 */
6050 			RESET_STATE_OF_RUN();
6051 			if (flags & KMA_LOMEM) {
6052 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = 0;
6053 			} else {
6054 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0;
6055 			}
6056 			last_idx = 0;
6057 			page_idx = last_idx;
6058 			wrapped = TRUE;
6059 			goto retry;
6060 		}
6061 		vm_free_page_unlock();
6062 	} else {
6063 		vm_page_t       m1;
6064 		vm_page_t       m2;
6065 		unsigned int    cur_idx;
6066 		unsigned int    tmp_start_idx;
6067 		vm_object_t     locked_object = VM_OBJECT_NULL;
6068 		boolean_t       abort_run = FALSE;
6069 
6070 		assert(page_idx - start_idx == contig_pages);
6071 
6072 		tmp_start_idx = start_idx;
6073 
6074 		/*
6075 		 * first pass through to pull the free pages
6076 		 * off of the free queue so that in case we
6077 		 * need substitute pages, we won't grab any
6078 		 * of the free pages in the run... we'll clear
6079 		 * the 'free' bit in the 2nd pass, and even in
6080 		 * an abort_run case, we'll collect all of the
6081 		 * free pages in this run and return them to the free list
6082 		 */
6083 		while (start_idx < page_idx) {
6084 			m1 = &vm_pages[start_idx++];
6085 
6086 #if !VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
6087 			assert(m1->vmp_q_state == VM_PAGE_ON_FREE_Q);
6088 #endif
6089 
6090 			if (m1->vmp_q_state == VM_PAGE_ON_FREE_Q) {
6091 				unsigned int color;
6092 
6093 				color = VM_PAGE_GET_COLOR(m1);
6094 #if MACH_ASSERT
6095 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, m1, TRUE);
6096 #endif
6097 				vm_page_queue_remove(&vm_page_queue_free[color].qhead, m1, vmp_pageq);
6098 
6099 				VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6100 #if MACH_ASSERT
6101 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, VM_PAGE_NULL, FALSE);
6102 #endif
6103 				/*
6104 				 * Clear the "free" bit so that this page
6105 				 * does not get considered for another
6106 				 * concurrent physically-contiguous allocation.
6107 				 */
6108 				m1->vmp_q_state = VM_PAGE_NOT_ON_Q;
6109 				assert(m1->vmp_busy);
6110 
6111 				vm_page_free_count--;
6112 			}
6113 		}
6114 		if (flags & KMA_LOMEM) {
6115 			vm_page_lomem_find_contiguous_last_idx = page_idx;
6116 		} else {
6117 			vm_page_find_contiguous_last_idx = page_idx;
6118 		}
6119 
6120 		/*
6121 		 * we can drop the free queue lock at this point since
6122 		 * we've pulled any 'free' candidates off of the list
6123 		 * we need it dropped so that we can do a vm_page_grab
6124 		 * when substituing for pmapped/dirty pages
6125 		 */
6126 		vm_free_page_unlock();
6127 
6128 		start_idx = tmp_start_idx;
6129 		cur_idx = page_idx - 1;
6130 
6131 		while (start_idx++ < page_idx) {
6132 			/*
6133 			 * must go through the list from back to front
6134 			 * so that the page list is created in the
6135 			 * correct order - low -> high phys addresses
6136 			 */
6137 			m1 = &vm_pages[cur_idx--];
6138 
6139 			if (m1->vmp_object == 0) {
6140 				/*
6141 				 * page has already been removed from
6142 				 * the free list in the 1st pass
6143 				 */
6144 				assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6145 				assert(m1->vmp_offset == (vm_object_offset_t) -1);
6146 				assert(m1->vmp_busy);
6147 				assert(!m1->vmp_wanted);
6148 				assert(!m1->vmp_laundry);
6149 			} else {
6150 				vm_object_t object;
6151 				int refmod;
6152 				boolean_t disconnected, reusable;
6153 
6154 				if (abort_run == TRUE) {
6155 					continue;
6156 				}
6157 
6158 				assert(m1->vmp_q_state != VM_PAGE_NOT_ON_Q);
6159 
6160 				object = VM_PAGE_OBJECT(m1);
6161 
6162 				if (object != locked_object) {
6163 					if (locked_object) {
6164 						vm_object_unlock(locked_object);
6165 						locked_object = VM_OBJECT_NULL;
6166 					}
6167 					if (vm_object_lock_try(object)) {
6168 						locked_object = object;
6169 					}
6170 				}
6171 				if (locked_object == VM_OBJECT_NULL ||
6172 				    (VM_PAGE_WIRED(m1) || m1->vmp_gobbled ||
6173 				    m1->vmp_laundry || m1->vmp_wanted ||
6174 				    m1->vmp_cleaning || m1->vmp_overwriting || m1->vmp_free_when_done || m1->vmp_busy) ||
6175 				    (m1->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
6176 					if (locked_object) {
6177 						vm_object_unlock(locked_object);
6178 						locked_object = VM_OBJECT_NULL;
6179 					}
6180 					tmp_start_idx = cur_idx;
6181 					abort_run = TRUE;
6182 					continue;
6183 				}
6184 
6185 				disconnected = FALSE;
6186 				reusable = FALSE;
6187 
6188 				if ((m1->vmp_reusable ||
6189 				    object->all_reusable) &&
6190 				    (m1->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q) &&
6191 				    !m1->vmp_dirty &&
6192 				    !m1->vmp_reference) {
6193 					/* reusable page... */
6194 					refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6195 					disconnected = TRUE;
6196 					if (refmod == 0) {
6197 						/*
6198 						 * ... not reused: can steal
6199 						 * without relocating contents.
6200 						 */
6201 						reusable = TRUE;
6202 					}
6203 				}
6204 
6205 				if ((m1->vmp_pmapped &&
6206 				    !reusable) ||
6207 				    m1->vmp_dirty ||
6208 				    m1->vmp_precious) {
6209 					vm_object_offset_t offset;
6210 
6211 					m2 = vm_page_grab_options(VM_PAGE_GRAB_Q_LOCK_HELD);
6212 
6213 					if (m2 == VM_PAGE_NULL) {
6214 						if (locked_object) {
6215 							vm_object_unlock(locked_object);
6216 							locked_object = VM_OBJECT_NULL;
6217 						}
6218 						tmp_start_idx = cur_idx;
6219 						abort_run = TRUE;
6220 						continue;
6221 					}
6222 					if (!disconnected) {
6223 						if (m1->vmp_pmapped) {
6224 							refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6225 						} else {
6226 							refmod = 0;
6227 						}
6228 					}
6229 
6230 					/* copy the page's contents */
6231 					pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(m1), VM_PAGE_GET_PHYS_PAGE(m2));
6232 					/* copy the page's state */
6233 					assert(!VM_PAGE_WIRED(m1));
6234 					assert(m1->vmp_q_state != VM_PAGE_ON_FREE_Q);
6235 					assert(m1->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q);
6236 					assert(!m1->vmp_laundry);
6237 					m2->vmp_reference       = m1->vmp_reference;
6238 					assert(!m1->vmp_gobbled);
6239 					assert(!m1->vmp_private);
6240 					m2->vmp_no_cache        = m1->vmp_no_cache;
6241 					m2->vmp_xpmapped        = 0;
6242 					assert(!m1->vmp_busy);
6243 					assert(!m1->vmp_wanted);
6244 					assert(!m1->vmp_fictitious);
6245 					m2->vmp_pmapped = m1->vmp_pmapped; /* should flush cache ? */
6246 					m2->vmp_wpmapped        = m1->vmp_wpmapped;
6247 					assert(!m1->vmp_free_when_done);
6248 					m2->vmp_absent  = m1->vmp_absent;
6249 					m2->vmp_error   = VMP_ERROR_GET(m1);
6250 					m2->vmp_dirty   = m1->vmp_dirty;
6251 					assert(!m1->vmp_cleaning);
6252 					m2->vmp_precious        = m1->vmp_precious;
6253 					m2->vmp_clustered       = m1->vmp_clustered;
6254 					assert(!m1->vmp_overwriting);
6255 					m2->vmp_restart = m1->vmp_restart;
6256 					m2->vmp_unusual = m1->vmp_unusual;
6257 					m2->vmp_cs_validated = m1->vmp_cs_validated;
6258 					m2->vmp_cs_tainted      = m1->vmp_cs_tainted;
6259 					m2->vmp_cs_nx   = m1->vmp_cs_nx;
6260 
6261 					/*
6262 					 * If m1 had really been reusable,
6263 					 * we would have just stolen it, so
6264 					 * let's not propagate it's "reusable"
6265 					 * bit and assert that m2 is not
6266 					 * marked as "reusable".
6267 					 */
6268 					// m2->vmp_reusable	= m1->vmp_reusable;
6269 					assert(!m2->vmp_reusable);
6270 
6271 					// assert(!m1->vmp_lopage);
6272 
6273 					if (m1->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6274 						m2->vmp_q_state = VM_PAGE_USED_BY_COMPRESSOR;
6275 						/*
6276 						 * We just grabbed m2 up above and so it isn't
6277 						 * going to be on any special Q as yet and so
6278 						 * we don't need to 'remove' it from the special
6279 						 * queues. Just resetting the state should be enough.
6280 						 */
6281 						m2->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
6282 					}
6283 
6284 					/*
6285 					 * page may need to be flushed if
6286 					 * it is marshalled into a UPL
6287 					 * that is going to be used by a device
6288 					 * that doesn't support coherency
6289 					 */
6290 					m2->vmp_written_by_kernel = TRUE;
6291 
6292 					/*
6293 					 * make sure we clear the ref/mod state
6294 					 * from the pmap layer... else we risk
6295 					 * inheriting state from the last time
6296 					 * this page was used...
6297 					 */
6298 					pmap_clear_refmod(VM_PAGE_GET_PHYS_PAGE(m2), VM_MEM_MODIFIED | VM_MEM_REFERENCED);
6299 
6300 					if (refmod & VM_MEM_REFERENCED) {
6301 						m2->vmp_reference = TRUE;
6302 					}
6303 					if (refmod & VM_MEM_MODIFIED) {
6304 						SET_PAGE_DIRTY(m2, TRUE);
6305 					}
6306 					offset = m1->vmp_offset;
6307 
6308 					/*
6309 					 * completely cleans up the state
6310 					 * of the page so that it is ready
6311 					 * to be put onto the free list, or
6312 					 * for this purpose it looks like it
6313 					 * just came off of the free list
6314 					 */
6315 					vm_page_free_prepare(m1);
6316 
6317 					/*
6318 					 * now put the substitute page
6319 					 * on the object
6320 					 */
6321 					vm_page_insert_internal(m2, locked_object, offset, VM_KERN_MEMORY_NONE, TRUE, TRUE, FALSE, FALSE, NULL);
6322 
6323 					if (m2->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6324 						m2->vmp_pmapped = TRUE;
6325 						m2->vmp_wpmapped = TRUE;
6326 
6327 						PMAP_ENTER(kernel_pmap, (vm_map_offset_t)m2->vmp_offset, m2,
6328 						    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE, kr);
6329 
6330 						assert(kr == KERN_SUCCESS);
6331 
6332 						compressed_pages++;
6333 					} else {
6334 						if (m2->vmp_reference) {
6335 							vm_page_activate(m2);
6336 						} else {
6337 							vm_page_deactivate(m2);
6338 						}
6339 					}
6340 					PAGE_WAKEUP_DONE(m2);
6341 				} else {
6342 					assert(m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR);
6343 
6344 					/*
6345 					 * completely cleans up the state
6346 					 * of the page so that it is ready
6347 					 * to be put onto the free list, or
6348 					 * for this purpose it looks like it
6349 					 * just came off of the free list
6350 					 */
6351 					vm_page_free_prepare(m1);
6352 				}
6353 
6354 				stolen_pages++;
6355 			}
6356 			if (m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR) {
6357 				/*
6358 				 * The Q state is preserved on m1 because vm_page_queues_remove doesn't
6359 				 * change it for pages marked as used-by-compressor.
6360 				 */
6361 				vm_page_assign_special_state(m1, VM_PAGE_SPECIAL_Q_BG);
6362 			}
6363 			VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6364 			m1->vmp_snext = m;
6365 			m = m1;
6366 		}
6367 		if (locked_object) {
6368 			vm_object_unlock(locked_object);
6369 			locked_object = VM_OBJECT_NULL;
6370 		}
6371 
6372 		if (abort_run == TRUE) {
6373 			/*
6374 			 * want the index of the last
6375 			 * page in this run that was
6376 			 * successfully 'stolen', so back
6377 			 * it up 1 for the auto-decrement on use
6378 			 * and 1 more to bump back over this page
6379 			 */
6380 			page_idx = tmp_start_idx + 2;
6381 			if (page_idx >= vm_pages_count) {
6382 				if (wrapped) {
6383 					if (m != VM_PAGE_NULL) {
6384 						vm_page_unlock_queues();
6385 						vm_page_free_list(m, FALSE);
6386 						vm_page_lock_queues();
6387 						m = VM_PAGE_NULL;
6388 					}
6389 					dumped_run++;
6390 					goto done_scanning;
6391 				}
6392 				page_idx = last_idx = 0;
6393 				wrapped = TRUE;
6394 			}
6395 			abort_run = FALSE;
6396 
6397 			/*
6398 			 * We didn't find a contiguous range but we didn't
6399 			 * start from the very first page.
6400 			 * Start again from the very first page.
6401 			 */
6402 			RESET_STATE_OF_RUN();
6403 
6404 			if (flags & KMA_LOMEM) {
6405 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = page_idx;
6406 			} else {
6407 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx;
6408 			}
6409 
6410 			last_idx = page_idx;
6411 
6412 			if (m != VM_PAGE_NULL) {
6413 				vm_page_unlock_queues();
6414 				vm_page_free_list(m, FALSE);
6415 				vm_page_lock_queues();
6416 				m = VM_PAGE_NULL;
6417 			}
6418 			dumped_run++;
6419 
6420 			vm_free_page_lock();
6421 			/*
6422 			 * reset our free page limit since we
6423 			 * dropped the lock protecting the vm_page_free_queue
6424 			 */
6425 			free_available = vm_page_free_count - vm_page_free_reserved;
6426 			goto retry;
6427 		}
6428 
6429 		for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
6430 			assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6431 			assert(m1->vmp_wire_count == 0);
6432 
6433 			if (wire == TRUE) {
6434 				m1->vmp_wire_count++;
6435 				m1->vmp_q_state = VM_PAGE_IS_WIRED;
6436 			} else {
6437 				m1->vmp_gobbled = TRUE;
6438 			}
6439 		}
6440 		if (wire == FALSE) {
6441 			vm_page_gobble_count += npages;
6442 		}
6443 
6444 		/*
6445 		 * gobbled pages are also counted as wired pages
6446 		 */
6447 		vm_page_wire_count += npages;
6448 
6449 		assert(vm_page_verify_contiguous(m, npages));
6450 	}
6451 done_scanning:
6452 	PAGE_REPLACEMENT_ALLOWED(FALSE);
6453 
6454 	vm_page_unlock_queues();
6455 
6456 #if DEBUG
6457 	clock_get_system_microtime(&tv_end_sec, &tv_end_usec);
6458 
6459 	tv_end_sec -= tv_start_sec;
6460 	if (tv_end_usec < tv_start_usec) {
6461 		tv_end_sec--;
6462 		tv_end_usec += 1000000;
6463 	}
6464 	tv_end_usec -= tv_start_usec;
6465 	if (tv_end_usec >= 1000000) {
6466 		tv_end_sec++;
6467 		tv_end_sec -= 1000000;
6468 	}
6469 	if (vm_page_find_contig_debug) {
6470 		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",
6471 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6472 		    (long)tv_end_sec, tv_end_usec, orig_last_idx,
6473 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages);
6474 	}
6475 
6476 #endif
6477 #if MACH_ASSERT
6478 	vm_page_verify_free_lists();
6479 #endif
6480 	if (m == NULL && zone_gc_called < 2) {
6481 		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",
6482 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6483 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages, vm_page_wire_count);
6484 
6485 		if (consider_buffer_cache_collect != NULL) {
6486 			(void)(*consider_buffer_cache_collect)(1);
6487 		}
6488 
6489 		zone_gc(zone_gc_called ? ZONE_GC_DRAIN : ZONE_GC_TRIM);
6490 
6491 		zone_gc_called++;
6492 
6493 		printf("vm_page_find_contiguous: zone_gc called... wired count is %d\n", vm_page_wire_count);
6494 		goto full_scan_again;
6495 	}
6496 
6497 	return m;
6498 }
6499 
6500 /*
6501  *	Allocate a list of contiguous, wired pages.
6502  */
6503 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)6504 cpm_allocate(
6505 	vm_size_t       size,
6506 	vm_page_t       *list,
6507 	ppnum_t         max_pnum,
6508 	ppnum_t         pnum_mask,
6509 	boolean_t       wire,
6510 	int             flags)
6511 {
6512 	vm_page_t               pages;
6513 	unsigned int            npages;
6514 
6515 	if (size % PAGE_SIZE != 0) {
6516 		return KERN_INVALID_ARGUMENT;
6517 	}
6518 
6519 	npages = (unsigned int) (size / PAGE_SIZE);
6520 	if (npages != size / PAGE_SIZE) {
6521 		/* 32-bit overflow */
6522 		return KERN_INVALID_ARGUMENT;
6523 	}
6524 
6525 	/*
6526 	 *	Obtain a pointer to a subset of the free
6527 	 *	list large enough to satisfy the request;
6528 	 *	the region will be physically contiguous.
6529 	 */
6530 	pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags);
6531 
6532 	if (pages == VM_PAGE_NULL) {
6533 		return KERN_NO_SPACE;
6534 	}
6535 	/*
6536 	 * determine need for wakeups
6537 	 */
6538 	if (vm_page_free_count < vm_page_free_min) {
6539 		vm_free_page_lock();
6540 		if (vm_pageout_running == FALSE) {
6541 			vm_free_page_unlock();
6542 			thread_wakeup((event_t) &vm_page_free_wanted);
6543 		} else {
6544 			vm_free_page_unlock();
6545 		}
6546 	}
6547 
6548 	VM_CHECK_MEMORYSTATUS;
6549 
6550 	/*
6551 	 *	The CPM pages should now be available and
6552 	 *	ordered by ascending physical address.
6553 	 */
6554 	assert(vm_page_verify_contiguous(pages, npages));
6555 
6556 	*list = pages;
6557 	return KERN_SUCCESS;
6558 }
6559 
6560 
6561 unsigned int vm_max_delayed_work_limit = DEFAULT_DELAYED_WORK_LIMIT;
6562 
6563 /*
6564  * when working on a 'run' of pages, it is necessary to hold
6565  * the vm_page_queue_lock (a hot global lock) for certain operations
6566  * on the page... however, the majority of the work can be done
6567  * while merely holding the object lock... in fact there are certain
6568  * collections of pages that don't require any work brokered by the
6569  * vm_page_queue_lock... to mitigate the time spent behind the global
6570  * lock, go to a 2 pass algorithm... collect pages up to DELAYED_WORK_LIMIT
6571  * while doing all of the work that doesn't require the vm_page_queue_lock...
6572  * then call vm_page_do_delayed_work to acquire the vm_page_queue_lock and do the
6573  * necessary work for each page... we will grab the busy bit on the page
6574  * if it's not already held so that vm_page_do_delayed_work can drop the object lock
6575  * if it can't immediately take the vm_page_queue_lock in order to compete
6576  * for the locks in the same order that vm_pageout_scan takes them.
6577  * the operation names are modeled after the names of the routines that
6578  * need to be called in order to make the changes very obvious in the
6579  * original loop
6580  */
6581 
6582 void
vm_page_do_delayed_work(vm_object_t object,vm_tag_t tag,struct vm_page_delayed_work * dwp,int dw_count)6583 vm_page_do_delayed_work(
6584 	vm_object_t     object,
6585 	vm_tag_t        tag,
6586 	struct vm_page_delayed_work *dwp,
6587 	int             dw_count)
6588 {
6589 	int             j;
6590 	vm_page_t       m;
6591 	vm_page_t       local_free_q = VM_PAGE_NULL;
6592 
6593 	/*
6594 	 * pageout_scan takes the vm_page_lock_queues first
6595 	 * then tries for the object lock... to avoid what
6596 	 * is effectively a lock inversion, we'll go to the
6597 	 * trouble of taking them in that same order... otherwise
6598 	 * if this object contains the majority of the pages resident
6599 	 * in the UBC (or a small set of large objects actively being
6600 	 * worked on contain the majority of the pages), we could
6601 	 * cause the pageout_scan thread to 'starve' in its attempt
6602 	 * to find pages to move to the free queue, since it has to
6603 	 * successfully acquire the object lock of any candidate page
6604 	 * before it can steal/clean it.
6605 	 */
6606 	if (!vm_page_trylockspin_queues()) {
6607 		vm_object_unlock(object);
6608 
6609 		/*
6610 		 * "Turnstile enabled vm_pageout_scan" can be runnable
6611 		 * for a very long time without getting on a core.
6612 		 * If this is a higher priority thread it could be
6613 		 * waiting here for a very long time respecting the fact
6614 		 * that pageout_scan would like its object after VPS does
6615 		 * a mutex_pause(0).
6616 		 * So we cap the number of yields in the vm_object_lock_avoid()
6617 		 * case to a single mutex_pause(0) which will give vm_pageout_scan
6618 		 * 10us to run and grab the object if needed.
6619 		 */
6620 		vm_page_lockspin_queues();
6621 
6622 		for (j = 0;; j++) {
6623 			if ((!vm_object_lock_avoid(object) ||
6624 			    (vps_dynamic_priority_enabled && (j > 0))) &&
6625 			    _vm_object_lock_try(object)) {
6626 				break;
6627 			}
6628 			vm_page_unlock_queues();
6629 			mutex_pause(j);
6630 			vm_page_lockspin_queues();
6631 		}
6632 	}
6633 	for (j = 0; j < dw_count; j++, dwp++) {
6634 		m = dwp->dw_m;
6635 
6636 		if (dwp->dw_mask & DW_vm_pageout_throttle_up) {
6637 			vm_pageout_throttle_up(m);
6638 		}
6639 #if CONFIG_PHANTOM_CACHE
6640 		if (dwp->dw_mask & DW_vm_phantom_cache_update) {
6641 			vm_phantom_cache_update(m);
6642 		}
6643 #endif
6644 		if (dwp->dw_mask & DW_vm_page_wire) {
6645 			vm_page_wire(m, tag, FALSE);
6646 		} else if (dwp->dw_mask & DW_vm_page_unwire) {
6647 			boolean_t       queueit;
6648 
6649 			queueit = (dwp->dw_mask & (DW_vm_page_free | DW_vm_page_deactivate_internal)) ? FALSE : TRUE;
6650 
6651 			vm_page_unwire(m, queueit);
6652 		}
6653 		if (dwp->dw_mask & DW_vm_page_free) {
6654 			vm_page_free_prepare_queues(m);
6655 
6656 			assert(m->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0);
6657 			/*
6658 			 * Add this page to our list of reclaimed pages,
6659 			 * to be freed later.
6660 			 */
6661 			m->vmp_snext = local_free_q;
6662 			local_free_q = m;
6663 		} else {
6664 			if (dwp->dw_mask & DW_vm_page_deactivate_internal) {
6665 				vm_page_deactivate_internal(m, FALSE);
6666 			} else if (dwp->dw_mask & DW_vm_page_activate) {
6667 				if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6668 					vm_page_activate(m);
6669 				}
6670 			} else if (dwp->dw_mask & DW_vm_page_speculate) {
6671 				vm_page_speculate(m, TRUE);
6672 			} else if (dwp->dw_mask & DW_enqueue_cleaned) {
6673 				/*
6674 				 * if we didn't hold the object lock and did this,
6675 				 * we might disconnect the page, then someone might
6676 				 * soft fault it back in, then we would put it on the
6677 				 * cleaned queue, and so we would have a referenced (maybe even dirty)
6678 				 * page on that queue, which we don't want
6679 				 */
6680 				int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
6681 
6682 				if ((refmod_state & VM_MEM_REFERENCED)) {
6683 					/*
6684 					 * this page has been touched since it got cleaned; let's activate it
6685 					 * if it hasn't already been
6686 					 */
6687 					VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
6688 					VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1);
6689 
6690 					if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6691 						vm_page_activate(m);
6692 					}
6693 				} else {
6694 					m->vmp_reference = FALSE;
6695 					vm_page_enqueue_cleaned(m);
6696 				}
6697 			} else if (dwp->dw_mask & DW_vm_page_lru) {
6698 				vm_page_lru(m);
6699 			} else if (dwp->dw_mask & DW_VM_PAGE_QUEUES_REMOVE) {
6700 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
6701 					vm_page_queues_remove(m, TRUE);
6702 				}
6703 			}
6704 			if (dwp->dw_mask & DW_set_reference) {
6705 				m->vmp_reference = TRUE;
6706 			} else if (dwp->dw_mask & DW_clear_reference) {
6707 				m->vmp_reference = FALSE;
6708 			}
6709 
6710 			if (dwp->dw_mask & DW_move_page) {
6711 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
6712 					vm_page_queues_remove(m, FALSE);
6713 
6714 					assert(VM_PAGE_OBJECT(m) != kernel_object);
6715 
6716 					vm_page_enqueue_inactive(m, FALSE);
6717 				}
6718 			}
6719 			if (dwp->dw_mask & DW_clear_busy) {
6720 				m->vmp_busy = FALSE;
6721 			}
6722 
6723 			if (dwp->dw_mask & DW_PAGE_WAKEUP) {
6724 				PAGE_WAKEUP(m);
6725 			}
6726 		}
6727 	}
6728 	vm_page_unlock_queues();
6729 
6730 	if (local_free_q) {
6731 		vm_page_free_list(local_free_q, TRUE);
6732 	}
6733 
6734 	VM_CHECK_MEMORYSTATUS;
6735 }
6736 
6737 __abortlike
6738 static void
__vm_page_alloc_list_failed_panic(vm_size_t page_count,kma_flags_t flags,kern_return_t kr)6739 __vm_page_alloc_list_failed_panic(
6740 	vm_size_t       page_count,
6741 	kma_flags_t     flags,
6742 	kern_return_t   kr)
6743 {
6744 	panic("vm_page_alloc_list(%zd, 0x%x) failed unexpectedly with %d",
6745 	    (size_t)page_count, flags, kr);
6746 }
6747 
6748 kern_return_t
vm_page_alloc_list(vm_size_t page_count,kma_flags_t flags,vm_page_t * list)6749 vm_page_alloc_list(
6750 	vm_size_t   page_count,
6751 	kma_flags_t flags,
6752 	vm_page_t  *list)
6753 {
6754 	vm_page_t       page_list = VM_PAGE_NULL;
6755 	vm_page_t       mem;
6756 	kern_return_t   kr = KERN_SUCCESS;
6757 	int             page_grab_count = 0;
6758 #if DEVELOPMENT || DEBUG
6759 	task_t          task;
6760 #endif /* DEVELOPMENT || DEBUG */
6761 
6762 	for (vm_size_t i = 0; i < page_count; i++) {
6763 		for (;;) {
6764 			if (flags & KMA_LOMEM) {
6765 				mem = vm_page_grablo();
6766 			} else {
6767 				mem = vm_page_grab();
6768 			}
6769 
6770 			if (mem != VM_PAGE_NULL) {
6771 				break;
6772 			}
6773 
6774 			if (flags & KMA_NOPAGEWAIT) {
6775 				kr = KERN_RESOURCE_SHORTAGE;
6776 				goto out;
6777 			}
6778 			if ((flags & KMA_LOMEM) && (vm_lopage_needed == TRUE)) {
6779 				kr = KERN_RESOURCE_SHORTAGE;
6780 				goto out;
6781 			}
6782 
6783 			/* VM privileged threads should have waited in vm_page_grab() and not get here. */
6784 			assert(!(current_thread()->options & TH_OPT_VMPRIV));
6785 
6786 			if ((flags & KMA_NOFAIL) == 0) {
6787 				uint64_t unavailable = ptoa_64(vm_page_wire_count + vm_page_free_target);
6788 				if (unavailable > max_mem || ptoa_64(page_count) > (max_mem - unavailable)) {
6789 					kr = KERN_RESOURCE_SHORTAGE;
6790 					goto out;
6791 				}
6792 			}
6793 			VM_PAGE_WAIT();
6794 		}
6795 
6796 		page_grab_count++;
6797 		mem->vmp_snext = page_list;
6798 		page_list = mem;
6799 	}
6800 
6801 	if ((KMA_ZERO | KMA_NOENCRYPT) & flags) {
6802 		for (mem = page_list; mem; mem = mem->vmp_snext) {
6803 			vm_page_zero_fill(mem);
6804 		}
6805 	}
6806 
6807 out:
6808 #if DEBUG || DEVELOPMENT
6809 	task = current_task_early();
6810 	if (task != NULL) {
6811 		ledger_credit(task->ledger, task_ledgers.pages_grabbed_kern, page_grab_count);
6812 	}
6813 #endif
6814 
6815 	if (kr == KERN_SUCCESS) {
6816 		*list = page_list;
6817 	} else if (flags & KMA_NOFAIL) {
6818 		__vm_page_alloc_list_failed_panic(page_count, flags, kr);
6819 	} else {
6820 		vm_page_free_list(page_list, FALSE);
6821 	}
6822 
6823 	return kr;
6824 }
6825 
6826 void
vm_page_set_offset(vm_page_t page,vm_object_offset_t offset)6827 vm_page_set_offset(vm_page_t page, vm_object_offset_t offset)
6828 {
6829 	page->vmp_offset = offset;
6830 }
6831 
6832 vm_page_t
vm_page_get_next(vm_page_t page)6833 vm_page_get_next(vm_page_t page)
6834 {
6835 	return page->vmp_snext;
6836 }
6837 
6838 vm_object_offset_t
vm_page_get_offset(vm_page_t page)6839 vm_page_get_offset(vm_page_t page)
6840 {
6841 	return page->vmp_offset;
6842 }
6843 
6844 ppnum_t
vm_page_get_phys_page(vm_page_t page)6845 vm_page_get_phys_page(vm_page_t page)
6846 {
6847 	return VM_PAGE_GET_PHYS_PAGE(page);
6848 }
6849 
6850 
6851 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
6852 
6853 #if HIBERNATION
6854 
6855 static vm_page_t hibernate_gobble_queue;
6856 
6857 static int  hibernate_drain_pageout_queue(struct vm_pageout_queue *);
6858 static int  hibernate_flush_dirty_pages(int);
6859 static int  hibernate_flush_queue(vm_page_queue_head_t *, int);
6860 
6861 void hibernate_flush_wait(void);
6862 void hibernate_mark_in_progress(void);
6863 void hibernate_clear_in_progress(void);
6864 
6865 void            hibernate_free_range(int, int);
6866 void            hibernate_hash_insert_page(vm_page_t);
6867 uint32_t        hibernate_mark_as_unneeded(addr64_t, addr64_t, hibernate_page_list_t *, hibernate_page_list_t *);
6868 uint32_t        hibernate_teardown_vm_structs(hibernate_page_list_t *, hibernate_page_list_t *);
6869 ppnum_t         hibernate_lookup_paddr(unsigned int);
6870 
6871 struct hibernate_statistics {
6872 	int hibernate_considered;
6873 	int hibernate_reentered_on_q;
6874 	int hibernate_found_dirty;
6875 	int hibernate_skipped_cleaning;
6876 	int hibernate_skipped_transient;
6877 	int hibernate_skipped_precious;
6878 	int hibernate_skipped_external;
6879 	int hibernate_queue_nolock;
6880 	int hibernate_queue_paused;
6881 	int hibernate_throttled;
6882 	int hibernate_throttle_timeout;
6883 	int hibernate_drained;
6884 	int hibernate_drain_timeout;
6885 	int cd_lock_failed;
6886 	int cd_found_precious;
6887 	int cd_found_wired;
6888 	int cd_found_busy;
6889 	int cd_found_unusual;
6890 	int cd_found_cleaning;
6891 	int cd_found_laundry;
6892 	int cd_found_dirty;
6893 	int cd_found_xpmapped;
6894 	int cd_skipped_xpmapped;
6895 	int cd_local_free;
6896 	int cd_total_free;
6897 	int cd_vm_page_wire_count;
6898 	int cd_vm_struct_pages_unneeded;
6899 	int cd_pages;
6900 	int cd_discarded;
6901 	int cd_count_wire;
6902 } hibernate_stats;
6903 
6904 
6905 /*
6906  * clamp the number of 'xpmapped' pages we'll sweep into the hibernation image
6907  * so that we don't overrun the estimated image size, which would
6908  * result in a hibernation failure.
6909  *
6910  * We use a size value instead of pages because we don't want to take up more space
6911  * on disk if the system has a 16K page size vs 4K. Also, we are not guaranteed
6912  * to have that additional space available.
6913  *
6914  * Since this was set at 40000 pages on X86 we are going to use 160MB as our
6915  * xpmapped size.
6916  */
6917 #define HIBERNATE_XPMAPPED_LIMIT        ((160 * 1024 * 1024ULL) / PAGE_SIZE)
6918 
6919 
6920 static int
hibernate_drain_pageout_queue(struct vm_pageout_queue * q)6921 hibernate_drain_pageout_queue(struct vm_pageout_queue *q)
6922 {
6923 	wait_result_t   wait_result;
6924 
6925 	vm_page_lock_queues();
6926 
6927 	while (!vm_page_queue_empty(&q->pgo_pending)) {
6928 		q->pgo_draining = TRUE;
6929 
6930 		assert_wait_timeout((event_t) (&q->pgo_laundry + 1), THREAD_INTERRUPTIBLE, 5000, 1000 * NSEC_PER_USEC);
6931 
6932 		vm_page_unlock_queues();
6933 
6934 		wait_result = thread_block(THREAD_CONTINUE_NULL);
6935 
6936 		if (wait_result == THREAD_TIMED_OUT && !vm_page_queue_empty(&q->pgo_pending)) {
6937 			hibernate_stats.hibernate_drain_timeout++;
6938 
6939 			if (q == &vm_pageout_queue_external) {
6940 				return 0;
6941 			}
6942 
6943 			return 1;
6944 		}
6945 		vm_page_lock_queues();
6946 
6947 		hibernate_stats.hibernate_drained++;
6948 	}
6949 	vm_page_unlock_queues();
6950 
6951 	return 0;
6952 }
6953 
6954 
6955 boolean_t hibernate_skip_external = FALSE;
6956 
6957 static int
hibernate_flush_queue(vm_page_queue_head_t * q,int qcount)6958 hibernate_flush_queue(vm_page_queue_head_t *q, int qcount)
6959 {
6960 	vm_page_t       m;
6961 	vm_object_t     l_object = NULL;
6962 	vm_object_t     m_object = NULL;
6963 	int             refmod_state = 0;
6964 	int             try_failed_count = 0;
6965 	int             retval = 0;
6966 	int             current_run = 0;
6967 	struct  vm_pageout_queue *iq;
6968 	struct  vm_pageout_queue *eq;
6969 	struct  vm_pageout_queue *tq;
6970 
6971 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START,
6972 	    VM_KERNEL_UNSLIDE_OR_PERM(q), qcount);
6973 
6974 	iq = &vm_pageout_queue_internal;
6975 	eq = &vm_pageout_queue_external;
6976 
6977 	vm_page_lock_queues();
6978 
6979 	while (qcount && !vm_page_queue_empty(q)) {
6980 		if (current_run++ == 1000) {
6981 			if (hibernate_should_abort()) {
6982 				retval = 1;
6983 				break;
6984 			}
6985 			current_run = 0;
6986 		}
6987 
6988 		m = (vm_page_t) vm_page_queue_first(q);
6989 		m_object = VM_PAGE_OBJECT(m);
6990 
6991 		/*
6992 		 * check to see if we currently are working
6993 		 * with the same object... if so, we've
6994 		 * already got the lock
6995 		 */
6996 		if (m_object != l_object) {
6997 			/*
6998 			 * the object associated with candidate page is
6999 			 * different from the one we were just working
7000 			 * with... dump the lock if we still own it
7001 			 */
7002 			if (l_object != NULL) {
7003 				vm_object_unlock(l_object);
7004 				l_object = NULL;
7005 			}
7006 			/*
7007 			 * Try to lock object; since we've alread got the
7008 			 * page queues lock, we can only 'try' for this one.
7009 			 * if the 'try' fails, we need to do a mutex_pause
7010 			 * to allow the owner of the object lock a chance to
7011 			 * run...
7012 			 */
7013 			if (!vm_object_lock_try_scan(m_object)) {
7014 				if (try_failed_count > 20) {
7015 					hibernate_stats.hibernate_queue_nolock++;
7016 
7017 					goto reenter_pg_on_q;
7018 				}
7019 
7020 				vm_page_unlock_queues();
7021 				mutex_pause(try_failed_count++);
7022 				vm_page_lock_queues();
7023 
7024 				hibernate_stats.hibernate_queue_paused++;
7025 				continue;
7026 			} else {
7027 				l_object = m_object;
7028 			}
7029 		}
7030 		if (!m_object->alive || m->vmp_cleaning || m->vmp_laundry || m->vmp_busy || m->vmp_absent || VMP_ERROR_GET(m)) {
7031 			/*
7032 			 * page is not to be cleaned
7033 			 * put it back on the head of its queue
7034 			 */
7035 			if (m->vmp_cleaning) {
7036 				hibernate_stats.hibernate_skipped_cleaning++;
7037 			} else {
7038 				hibernate_stats.hibernate_skipped_transient++;
7039 			}
7040 
7041 			goto reenter_pg_on_q;
7042 		}
7043 		if (m_object->copy == VM_OBJECT_NULL) {
7044 			if (m_object->purgable == VM_PURGABLE_VOLATILE || m_object->purgable == VM_PURGABLE_EMPTY) {
7045 				/*
7046 				 * let the normal hibernate image path
7047 				 * deal with these
7048 				 */
7049 				goto reenter_pg_on_q;
7050 			}
7051 		}
7052 		if (!m->vmp_dirty && m->vmp_pmapped) {
7053 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7054 
7055 			if ((refmod_state & VM_MEM_MODIFIED)) {
7056 				SET_PAGE_DIRTY(m, FALSE);
7057 			}
7058 		} else {
7059 			refmod_state = 0;
7060 		}
7061 
7062 		if (!m->vmp_dirty) {
7063 			/*
7064 			 * page is not to be cleaned
7065 			 * put it back on the head of its queue
7066 			 */
7067 			if (m->vmp_precious) {
7068 				hibernate_stats.hibernate_skipped_precious++;
7069 			}
7070 
7071 			goto reenter_pg_on_q;
7072 		}
7073 
7074 		if (hibernate_skip_external == TRUE && !m_object->internal) {
7075 			hibernate_stats.hibernate_skipped_external++;
7076 
7077 			goto reenter_pg_on_q;
7078 		}
7079 		tq = NULL;
7080 
7081 		if (m_object->internal) {
7082 			if (VM_PAGE_Q_THROTTLED(iq)) {
7083 				tq = iq;
7084 			}
7085 		} else if (VM_PAGE_Q_THROTTLED(eq)) {
7086 			tq = eq;
7087 		}
7088 
7089 		if (tq != NULL) {
7090 			wait_result_t   wait_result;
7091 			int             wait_count = 5;
7092 
7093 			if (l_object != NULL) {
7094 				vm_object_unlock(l_object);
7095 				l_object = NULL;
7096 			}
7097 
7098 			while (retval == 0) {
7099 				tq->pgo_throttled = TRUE;
7100 
7101 				assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000 * NSEC_PER_USEC);
7102 
7103 				vm_page_unlock_queues();
7104 
7105 				wait_result = thread_block(THREAD_CONTINUE_NULL);
7106 
7107 				vm_page_lock_queues();
7108 
7109 				if (wait_result != THREAD_TIMED_OUT) {
7110 					break;
7111 				}
7112 				if (!VM_PAGE_Q_THROTTLED(tq)) {
7113 					break;
7114 				}
7115 
7116 				if (hibernate_should_abort()) {
7117 					retval = 1;
7118 				}
7119 
7120 				if (--wait_count == 0) {
7121 					hibernate_stats.hibernate_throttle_timeout++;
7122 
7123 					if (tq == eq) {
7124 						hibernate_skip_external = TRUE;
7125 						break;
7126 					}
7127 					retval = 1;
7128 				}
7129 			}
7130 			if (retval) {
7131 				break;
7132 			}
7133 
7134 			hibernate_stats.hibernate_throttled++;
7135 
7136 			continue;
7137 		}
7138 		/*
7139 		 * we've already factored out pages in the laundry which
7140 		 * means this page can't be on the pageout queue so it's
7141 		 * safe to do the vm_page_queues_remove
7142 		 */
7143 		vm_page_queues_remove(m, TRUE);
7144 
7145 		if (m_object->internal == TRUE) {
7146 			pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(m), PMAP_OPTIONS_COMPRESSOR, NULL);
7147 		}
7148 
7149 		vm_pageout_cluster(m);
7150 
7151 		hibernate_stats.hibernate_found_dirty++;
7152 
7153 		goto next_pg;
7154 
7155 reenter_pg_on_q:
7156 		vm_page_queue_remove(q, m, vmp_pageq);
7157 		vm_page_queue_enter(q, m, vmp_pageq);
7158 
7159 		hibernate_stats.hibernate_reentered_on_q++;
7160 next_pg:
7161 		hibernate_stats.hibernate_considered++;
7162 
7163 		qcount--;
7164 		try_failed_count = 0;
7165 	}
7166 	if (l_object != NULL) {
7167 		vm_object_unlock(l_object);
7168 		l_object = NULL;
7169 	}
7170 
7171 	vm_page_unlock_queues();
7172 
7173 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_END, hibernate_stats.hibernate_found_dirty, retval, 0, 0, 0);
7174 
7175 	return retval;
7176 }
7177 
7178 
7179 static int
hibernate_flush_dirty_pages(int pass)7180 hibernate_flush_dirty_pages(int pass)
7181 {
7182 	struct vm_speculative_age_q     *aq;
7183 	uint32_t        i;
7184 
7185 	if (vm_page_local_q) {
7186 		zpercpu_foreach_cpu(lid) {
7187 			vm_page_reactivate_local(lid, TRUE, FALSE);
7188 		}
7189 	}
7190 
7191 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
7192 		int             qcount;
7193 		vm_page_t       m;
7194 
7195 		aq = &vm_page_queue_speculative[i];
7196 
7197 		if (vm_page_queue_empty(&aq->age_q)) {
7198 			continue;
7199 		}
7200 		qcount = 0;
7201 
7202 		vm_page_lockspin_queues();
7203 
7204 		vm_page_queue_iterate(&aq->age_q, m, vmp_pageq) {
7205 			qcount++;
7206 		}
7207 		vm_page_unlock_queues();
7208 
7209 		if (qcount) {
7210 			if (hibernate_flush_queue(&aq->age_q, qcount)) {
7211 				return 1;
7212 			}
7213 		}
7214 	}
7215 	if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_page_anonymous_count - vm_page_cleaned_count)) {
7216 		return 1;
7217 	}
7218 	/* XXX FBDP TODO: flush secluded queue */
7219 	if (hibernate_flush_queue(&vm_page_queue_anonymous, vm_page_anonymous_count)) {
7220 		return 1;
7221 	}
7222 	if (hibernate_flush_queue(&vm_page_queue_cleaned, vm_page_cleaned_count)) {
7223 		return 1;
7224 	}
7225 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7226 		return 1;
7227 	}
7228 
7229 	if (pass == 1) {
7230 		vm_compressor_record_warmup_start();
7231 	}
7232 
7233 	if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) {
7234 		if (pass == 1) {
7235 			vm_compressor_record_warmup_end();
7236 		}
7237 		return 1;
7238 	}
7239 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7240 		if (pass == 1) {
7241 			vm_compressor_record_warmup_end();
7242 		}
7243 		return 1;
7244 	}
7245 	if (pass == 1) {
7246 		vm_compressor_record_warmup_end();
7247 	}
7248 
7249 	if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external)) {
7250 		return 1;
7251 	}
7252 
7253 	return 0;
7254 }
7255 
7256 
7257 void
hibernate_reset_stats()7258 hibernate_reset_stats()
7259 {
7260 	bzero(&hibernate_stats, sizeof(struct hibernate_statistics));
7261 }
7262 
7263 
7264 int
hibernate_flush_memory()7265 hibernate_flush_memory()
7266 {
7267 	int     retval;
7268 
7269 	assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
7270 
7271 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0);
7272 
7273 	hibernate_cleaning_in_progress = TRUE;
7274 	hibernate_skip_external = FALSE;
7275 
7276 	if ((retval = hibernate_flush_dirty_pages(1)) == 0) {
7277 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_START, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7278 
7279 		vm_compressor_flush();
7280 
7281 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7282 
7283 		if (consider_buffer_cache_collect != NULL) {
7284 			unsigned int orig_wire_count;
7285 
7286 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, 0, 0, 0, 0, 0);
7287 			orig_wire_count = vm_page_wire_count;
7288 
7289 			(void)(*consider_buffer_cache_collect)(1);
7290 			zone_gc(ZONE_GC_DRAIN);
7291 
7292 			HIBLOG("hibernate_flush_memory: buffer_cache_gc freed up %d wired pages\n", orig_wire_count - vm_page_wire_count);
7293 
7294 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0);
7295 		}
7296 	}
7297 	hibernate_cleaning_in_progress = FALSE;
7298 
7299 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0);
7300 
7301 	if (retval) {
7302 		HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n", VM_PAGE_COMPRESSOR_COUNT);
7303 	}
7304 
7305 
7306 	HIBPRINT("hibernate_flush_memory() considered(%d) reentered_on_q(%d) found_dirty(%d)\n",
7307 	    hibernate_stats.hibernate_considered,
7308 	    hibernate_stats.hibernate_reentered_on_q,
7309 	    hibernate_stats.hibernate_found_dirty);
7310 	HIBPRINT("   skipped_cleaning(%d) skipped_transient(%d) skipped_precious(%d) skipped_external(%d) queue_nolock(%d)\n",
7311 	    hibernate_stats.hibernate_skipped_cleaning,
7312 	    hibernate_stats.hibernate_skipped_transient,
7313 	    hibernate_stats.hibernate_skipped_precious,
7314 	    hibernate_stats.hibernate_skipped_external,
7315 	    hibernate_stats.hibernate_queue_nolock);
7316 	HIBPRINT("   queue_paused(%d) throttled(%d) throttle_timeout(%d) drained(%d) drain_timeout(%d)\n",
7317 	    hibernate_stats.hibernate_queue_paused,
7318 	    hibernate_stats.hibernate_throttled,
7319 	    hibernate_stats.hibernate_throttle_timeout,
7320 	    hibernate_stats.hibernate_drained,
7321 	    hibernate_stats.hibernate_drain_timeout);
7322 
7323 	return retval;
7324 }
7325 
7326 
7327 static void
hibernate_page_list_zero(hibernate_page_list_t * list)7328 hibernate_page_list_zero(hibernate_page_list_t *list)
7329 {
7330 	uint32_t             bank;
7331 	hibernate_bitmap_t * bitmap;
7332 
7333 	bitmap = &list->bank_bitmap[0];
7334 	for (bank = 0; bank < list->bank_count; bank++) {
7335 		uint32_t last_bit;
7336 
7337 		bzero((void *) &bitmap->bitmap[0], bitmap->bitmapwords << 2);
7338 		// set out-of-bound bits at end of bitmap.
7339 		last_bit = ((bitmap->last_page - bitmap->first_page + 1) & 31);
7340 		if (last_bit) {
7341 			bitmap->bitmap[bitmap->bitmapwords - 1] = (0xFFFFFFFF >> last_bit);
7342 		}
7343 
7344 		bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords];
7345 	}
7346 }
7347 
7348 void
hibernate_free_gobble_pages(void)7349 hibernate_free_gobble_pages(void)
7350 {
7351 	vm_page_t m, next;
7352 	uint32_t  count = 0;
7353 
7354 	m = (vm_page_t) hibernate_gobble_queue;
7355 	while (m) {
7356 		next = m->vmp_snext;
7357 		vm_page_free(m);
7358 		count++;
7359 		m = next;
7360 	}
7361 	hibernate_gobble_queue = VM_PAGE_NULL;
7362 
7363 	if (count) {
7364 		HIBLOG("Freed %d pages\n", count);
7365 	}
7366 }
7367 
7368 static boolean_t
hibernate_consider_discard(vm_page_t m,boolean_t preflight)7369 hibernate_consider_discard(vm_page_t m, boolean_t preflight)
7370 {
7371 	vm_object_t object = NULL;
7372 	int                  refmod_state;
7373 	boolean_t            discard = FALSE;
7374 
7375 	do{
7376 		if (m->vmp_private) {
7377 			panic("hibernate_consider_discard: private");
7378 		}
7379 
7380 		object = VM_PAGE_OBJECT(m);
7381 
7382 		if (!vm_object_lock_try(object)) {
7383 			object = NULL;
7384 			if (!preflight) {
7385 				hibernate_stats.cd_lock_failed++;
7386 			}
7387 			break;
7388 		}
7389 		if (VM_PAGE_WIRED(m)) {
7390 			if (!preflight) {
7391 				hibernate_stats.cd_found_wired++;
7392 			}
7393 			break;
7394 		}
7395 		if (m->vmp_precious) {
7396 			if (!preflight) {
7397 				hibernate_stats.cd_found_precious++;
7398 			}
7399 			break;
7400 		}
7401 		if (m->vmp_busy || !object->alive) {
7402 			/*
7403 			 *	Somebody is playing with this page.
7404 			 */
7405 			if (!preflight) {
7406 				hibernate_stats.cd_found_busy++;
7407 			}
7408 			break;
7409 		}
7410 		if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7411 			/*
7412 			 * If it's unusual in anyway, ignore it
7413 			 */
7414 			if (!preflight) {
7415 				hibernate_stats.cd_found_unusual++;
7416 			}
7417 			break;
7418 		}
7419 		if (m->vmp_cleaning) {
7420 			if (!preflight) {
7421 				hibernate_stats.cd_found_cleaning++;
7422 			}
7423 			break;
7424 		}
7425 		if (m->vmp_laundry) {
7426 			if (!preflight) {
7427 				hibernate_stats.cd_found_laundry++;
7428 			}
7429 			break;
7430 		}
7431 		if (!m->vmp_dirty) {
7432 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7433 
7434 			if (refmod_state & VM_MEM_REFERENCED) {
7435 				m->vmp_reference = TRUE;
7436 			}
7437 			if (refmod_state & VM_MEM_MODIFIED) {
7438 				SET_PAGE_DIRTY(m, FALSE);
7439 			}
7440 		}
7441 
7442 		/*
7443 		 * If it's clean or purgeable we can discard the page on wakeup.
7444 		 */
7445 		discard = (!m->vmp_dirty)
7446 		    || (VM_PURGABLE_VOLATILE == object->purgable)
7447 		    || (VM_PURGABLE_EMPTY == object->purgable);
7448 
7449 
7450 		if (discard == FALSE) {
7451 			if (!preflight) {
7452 				hibernate_stats.cd_found_dirty++;
7453 			}
7454 		} else if (m->vmp_xpmapped && m->vmp_reference && !object->internal) {
7455 			if (hibernate_stats.cd_found_xpmapped < HIBERNATE_XPMAPPED_LIMIT) {
7456 				if (!preflight) {
7457 					hibernate_stats.cd_found_xpmapped++;
7458 				}
7459 				discard = FALSE;
7460 			} else {
7461 				if (!preflight) {
7462 					hibernate_stats.cd_skipped_xpmapped++;
7463 				}
7464 			}
7465 		}
7466 	}while (FALSE);
7467 
7468 	if (object) {
7469 		vm_object_unlock(object);
7470 	}
7471 
7472 	return discard;
7473 }
7474 
7475 
7476 static void
hibernate_discard_page(vm_page_t m)7477 hibernate_discard_page(vm_page_t m)
7478 {
7479 	vm_object_t m_object;
7480 
7481 	if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7482 		/*
7483 		 * If it's unusual in anyway, ignore
7484 		 */
7485 		return;
7486 	}
7487 
7488 	m_object = VM_PAGE_OBJECT(m);
7489 
7490 #if MACH_ASSERT || DEBUG
7491 	if (!vm_object_lock_try(m_object)) {
7492 		panic("hibernate_discard_page(%p) !vm_object_lock_try", m);
7493 	}
7494 #else
7495 	/* No need to lock page queue for token delete, hibernate_vm_unlock()
7496 	 *  makes sure these locks are uncontended before sleep */
7497 #endif /* MACH_ASSERT || DEBUG */
7498 
7499 	if (m->vmp_pmapped == TRUE) {
7500 		__unused int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
7501 	}
7502 
7503 	if (m->vmp_laundry) {
7504 		panic("hibernate_discard_page(%p) laundry", m);
7505 	}
7506 	if (m->vmp_private) {
7507 		panic("hibernate_discard_page(%p) private", m);
7508 	}
7509 	if (m->vmp_fictitious) {
7510 		panic("hibernate_discard_page(%p) fictitious", m);
7511 	}
7512 
7513 	if (VM_PURGABLE_VOLATILE == m_object->purgable) {
7514 		/* object should be on a queue */
7515 		assert((m_object->objq.next != NULL) && (m_object->objq.prev != NULL));
7516 		purgeable_q_t old_queue = vm_purgeable_object_remove(m_object);
7517 		assert(old_queue);
7518 		if (m_object->purgeable_when_ripe) {
7519 			vm_purgeable_token_delete_first(old_queue);
7520 		}
7521 		vm_object_lock_assert_exclusive(m_object);
7522 		m_object->purgable = VM_PURGABLE_EMPTY;
7523 
7524 		/*
7525 		 * Purgeable ledgers:  pages of VOLATILE and EMPTY objects are
7526 		 * accounted in the "volatile" ledger, so no change here.
7527 		 * We have to update vm_page_purgeable_count, though, since we're
7528 		 * effectively purging this object.
7529 		 */
7530 		unsigned int delta;
7531 		assert(m_object->resident_page_count >= m_object->wired_page_count);
7532 		delta = (m_object->resident_page_count - m_object->wired_page_count);
7533 		assert(vm_page_purgeable_count >= delta);
7534 		assert(delta > 0);
7535 		OSAddAtomic(-delta, (SInt32 *)&vm_page_purgeable_count);
7536 	}
7537 
7538 	vm_page_free(m);
7539 
7540 #if MACH_ASSERT || DEBUG
7541 	vm_object_unlock(m_object);
7542 #endif  /* MACH_ASSERT || DEBUG */
7543 }
7544 
7545 /*
7546  *  Grab locks for hibernate_page_list_setall()
7547  */
7548 void
hibernate_vm_lock_queues(void)7549 hibernate_vm_lock_queues(void)
7550 {
7551 	vm_object_lock(compressor_object);
7552 	vm_page_lock_queues();
7553 	vm_free_page_lock();
7554 	lck_mtx_lock(&vm_purgeable_queue_lock);
7555 
7556 	if (vm_page_local_q) {
7557 		zpercpu_foreach(lq, vm_page_local_q) {
7558 			VPL_LOCK(&lq->vpl_lock);
7559 		}
7560 	}
7561 }
7562 
7563 void
hibernate_vm_unlock_queues(void)7564 hibernate_vm_unlock_queues(void)
7565 {
7566 	if (vm_page_local_q) {
7567 		zpercpu_foreach(lq, vm_page_local_q) {
7568 			VPL_UNLOCK(&lq->vpl_lock);
7569 		}
7570 	}
7571 	lck_mtx_unlock(&vm_purgeable_queue_lock);
7572 	vm_free_page_unlock();
7573 	vm_page_unlock_queues();
7574 	vm_object_unlock(compressor_object);
7575 }
7576 
7577 /*
7578  *  Bits zero in the bitmaps => page needs to be saved. All pages default to be saved,
7579  *  pages known to VM to not need saving are subtracted.
7580  *  Wired pages to be saved are present in page_list_wired, pageable in page_list.
7581  */
7582 
7583 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)7584 hibernate_page_list_setall(hibernate_page_list_t * page_list,
7585     hibernate_page_list_t * page_list_wired,
7586     hibernate_page_list_t * page_list_pal,
7587     boolean_t preflight,
7588     boolean_t will_discard,
7589     uint32_t * pagesOut)
7590 {
7591 	uint64_t start, end, nsec;
7592 	vm_page_t m;
7593 	vm_page_t next;
7594 	uint32_t pages = page_list->page_count;
7595 	uint32_t count_anonymous = 0, count_throttled = 0, count_compressor = 0;
7596 	uint32_t count_inactive = 0, count_active = 0, count_speculative = 0, count_cleaned = 0;
7597 	uint32_t count_wire = pages;
7598 	uint32_t count_discard_active    = 0;
7599 	uint32_t count_discard_inactive  = 0;
7600 	uint32_t count_retired = 0;
7601 	uint32_t count_discard_cleaned   = 0;
7602 	uint32_t count_discard_purgeable = 0;
7603 	uint32_t count_discard_speculative = 0;
7604 	uint32_t count_discard_vm_struct_pages = 0;
7605 	uint32_t i;
7606 	uint32_t             bank;
7607 	hibernate_bitmap_t * bitmap;
7608 	hibernate_bitmap_t * bitmap_wired;
7609 	boolean_t                    discard_all;
7610 	boolean_t            discard = FALSE;
7611 
7612 	HIBLOG("hibernate_page_list_setall(preflight %d) start\n", preflight);
7613 
7614 	if (preflight) {
7615 		page_list       = NULL;
7616 		page_list_wired = NULL;
7617 		page_list_pal   = NULL;
7618 		discard_all     = FALSE;
7619 	} else {
7620 		discard_all     = will_discard;
7621 	}
7622 
7623 #if MACH_ASSERT || DEBUG
7624 	if (!preflight) {
7625 		assert(hibernate_vm_locks_are_safe());
7626 		vm_page_lock_queues();
7627 		if (vm_page_local_q) {
7628 			zpercpu_foreach(lq, vm_page_local_q) {
7629 				VPL_LOCK(&lq->vpl_lock);
7630 			}
7631 		}
7632 	}
7633 #endif  /* MACH_ASSERT || DEBUG */
7634 
7635 
7636 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0);
7637 
7638 	clock_get_uptime(&start);
7639 
7640 	if (!preflight) {
7641 		hibernate_page_list_zero(page_list);
7642 		hibernate_page_list_zero(page_list_wired);
7643 		hibernate_page_list_zero(page_list_pal);
7644 
7645 		hibernate_stats.cd_vm_page_wire_count = vm_page_wire_count;
7646 		hibernate_stats.cd_pages = pages;
7647 	}
7648 
7649 	if (vm_page_local_q) {
7650 		zpercpu_foreach_cpu(lid) {
7651 			vm_page_reactivate_local(lid, TRUE, !preflight);
7652 		}
7653 	}
7654 
7655 	if (preflight) {
7656 		vm_object_lock(compressor_object);
7657 		vm_page_lock_queues();
7658 		vm_free_page_lock();
7659 	}
7660 
7661 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
7662 
7663 	hibernation_vmqueues_inspection = TRUE;
7664 
7665 	m = (vm_page_t) hibernate_gobble_queue;
7666 	while (m) {
7667 		pages--;
7668 		count_wire--;
7669 		if (!preflight) {
7670 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7671 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7672 		}
7673 		m = m->vmp_snext;
7674 	}
7675 
7676 	if (!preflight) {
7677 		percpu_foreach(free_pages_head, free_pages) {
7678 			for (m = *free_pages_head; m; m = m->vmp_snext) {
7679 				assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
7680 
7681 				pages--;
7682 				count_wire--;
7683 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7684 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7685 
7686 				hibernate_stats.cd_local_free++;
7687 				hibernate_stats.cd_total_free++;
7688 			}
7689 		}
7690 	}
7691 
7692 	for (i = 0; i < vm_colors; i++) {
7693 		vm_page_queue_iterate(&vm_page_queue_free[i].qhead, m, vmp_pageq) {
7694 			assert(m->vmp_q_state == VM_PAGE_ON_FREE_Q);
7695 
7696 			pages--;
7697 			count_wire--;
7698 			if (!preflight) {
7699 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7700 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7701 
7702 				hibernate_stats.cd_total_free++;
7703 			}
7704 		}
7705 	}
7706 
7707 	vm_page_queue_iterate(&vm_lopage_queue_free, m, vmp_pageq) {
7708 		assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
7709 
7710 		pages--;
7711 		count_wire--;
7712 		if (!preflight) {
7713 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7714 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7715 
7716 			hibernate_stats.cd_total_free++;
7717 		}
7718 	}
7719 
7720 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
7721 	while (m && !vm_page_queue_end(&vm_page_queue_throttled, (vm_page_queue_entry_t)m)) {
7722 		assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
7723 
7724 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7725 		discard = FALSE;
7726 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
7727 		    && hibernate_consider_discard(m, preflight)) {
7728 			if (!preflight) {
7729 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7730 			}
7731 			count_discard_inactive++;
7732 			discard = discard_all;
7733 		} else {
7734 			count_throttled++;
7735 		}
7736 		count_wire--;
7737 		if (!preflight) {
7738 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7739 		}
7740 
7741 		if (discard) {
7742 			hibernate_discard_page(m);
7743 		}
7744 		m = next;
7745 	}
7746 
7747 	m = (vm_page_t)vm_page_queue_first(&vm_page_queue_anonymous);
7748 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
7749 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
7750 
7751 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7752 		discard = FALSE;
7753 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7754 		    hibernate_consider_discard(m, preflight)) {
7755 			if (!preflight) {
7756 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7757 			}
7758 			if (m->vmp_dirty) {
7759 				count_discard_purgeable++;
7760 			} else {
7761 				count_discard_inactive++;
7762 			}
7763 			discard = discard_all;
7764 		} else {
7765 			count_anonymous++;
7766 		}
7767 		count_wire--;
7768 		if (!preflight) {
7769 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7770 		}
7771 		if (discard) {
7772 			hibernate_discard_page(m);
7773 		}
7774 		m = next;
7775 	}
7776 
7777 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
7778 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
7779 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
7780 
7781 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7782 		discard = FALSE;
7783 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7784 		    hibernate_consider_discard(m, preflight)) {
7785 			if (!preflight) {
7786 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7787 			}
7788 			if (m->vmp_dirty) {
7789 				count_discard_purgeable++;
7790 			} else {
7791 				count_discard_cleaned++;
7792 			}
7793 			discard = discard_all;
7794 		} else {
7795 			count_cleaned++;
7796 		}
7797 		count_wire--;
7798 		if (!preflight) {
7799 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7800 		}
7801 		if (discard) {
7802 			hibernate_discard_page(m);
7803 		}
7804 		m = next;
7805 	}
7806 
7807 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
7808 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
7809 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
7810 
7811 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7812 		discard = FALSE;
7813 		if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) &&
7814 		    hibernate_consider_discard(m, preflight)) {
7815 			if (!preflight) {
7816 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7817 			}
7818 			if (m->vmp_dirty) {
7819 				count_discard_purgeable++;
7820 			} else {
7821 				count_discard_active++;
7822 			}
7823 			discard = discard_all;
7824 		} else {
7825 			count_active++;
7826 		}
7827 		count_wire--;
7828 		if (!preflight) {
7829 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7830 		}
7831 		if (discard) {
7832 			hibernate_discard_page(m);
7833 		}
7834 		m = next;
7835 	}
7836 
7837 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
7838 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
7839 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
7840 
7841 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7842 		discard = FALSE;
7843 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7844 		    hibernate_consider_discard(m, preflight)) {
7845 			if (!preflight) {
7846 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7847 			}
7848 			if (m->vmp_dirty) {
7849 				count_discard_purgeable++;
7850 			} else {
7851 				count_discard_inactive++;
7852 			}
7853 			discard = discard_all;
7854 		} else {
7855 			count_inactive++;
7856 		}
7857 		count_wire--;
7858 		if (!preflight) {
7859 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7860 		}
7861 		if (discard) {
7862 			hibernate_discard_page(m);
7863 		}
7864 		m = next;
7865 	}
7866 	/* XXX FBDP TODO: secluded queue */
7867 
7868 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
7869 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
7870 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
7871 			assertf(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q,
7872 			    "Bad page: %p (0x%x:0x%x) on queue %d has state: %d (Discard: %d, Preflight: %d)",
7873 			    m, m->vmp_pageq.next, m->vmp_pageq.prev, i, m->vmp_q_state, discard, preflight);
7874 
7875 			next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7876 			discard = FALSE;
7877 			if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7878 			    hibernate_consider_discard(m, preflight)) {
7879 				if (!preflight) {
7880 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7881 				}
7882 				count_discard_speculative++;
7883 				discard = discard_all;
7884 			} else {
7885 				count_speculative++;
7886 			}
7887 			count_wire--;
7888 			if (!preflight) {
7889 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7890 			}
7891 			if (discard) {
7892 				hibernate_discard_page(m);
7893 			}
7894 			m = next;
7895 		}
7896 	}
7897 
7898 	vm_page_queue_iterate(&compressor_object->memq, m, vmp_listq) {
7899 		assert(m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR);
7900 
7901 		count_compressor++;
7902 		count_wire--;
7903 		if (!preflight) {
7904 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7905 		}
7906 	}
7907 
7908 
7909 	if (preflight == FALSE && discard_all == TRUE) {
7910 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START);
7911 
7912 		HIBLOG("hibernate_teardown started\n");
7913 		count_discard_vm_struct_pages = hibernate_teardown_vm_structs(page_list, page_list_wired);
7914 		HIBLOG("hibernate_teardown completed - discarded %d\n", count_discard_vm_struct_pages);
7915 
7916 		pages -= count_discard_vm_struct_pages;
7917 		count_wire -= count_discard_vm_struct_pages;
7918 
7919 		hibernate_stats.cd_vm_struct_pages_unneeded = count_discard_vm_struct_pages;
7920 
7921 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_END);
7922 	}
7923 
7924 	if (!preflight) {
7925 		// pull wired from hibernate_bitmap
7926 		bitmap = &page_list->bank_bitmap[0];
7927 		bitmap_wired = &page_list_wired->bank_bitmap[0];
7928 		for (bank = 0; bank < page_list->bank_count; bank++) {
7929 			for (i = 0; i < bitmap->bitmapwords; i++) {
7930 				bitmap->bitmap[i] = bitmap->bitmap[i] | ~bitmap_wired->bitmap[i];
7931 			}
7932 			bitmap = (hibernate_bitmap_t *)&bitmap->bitmap[bitmap->bitmapwords];
7933 			bitmap_wired = (hibernate_bitmap_t *) &bitmap_wired->bitmap[bitmap_wired->bitmapwords];
7934 		}
7935 	}
7936 
7937 	// machine dependent adjustments
7938 	hibernate_page_list_setall_machine(page_list, page_list_wired, preflight, &pages);
7939 
7940 	if (!preflight) {
7941 		hibernate_stats.cd_count_wire = count_wire;
7942 		hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable +
7943 		    count_discard_speculative + count_discard_cleaned + count_discard_vm_struct_pages;
7944 	}
7945 
7946 	clock_get_uptime(&end);
7947 	absolutetime_to_nanoseconds(end - start, &nsec);
7948 	HIBLOG("hibernate_page_list_setall time: %qd ms\n", nsec / 1000000ULL);
7949 
7950 	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",
7951 	    pages, count_wire, count_active, count_inactive, count_cleaned, count_speculative, count_anonymous, count_throttled, count_compressor, hibernate_stats.cd_found_xpmapped,
7952 	    discard_all ? "did" : "could",
7953 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned, count_retired);
7954 
7955 	if (hibernate_stats.cd_skipped_xpmapped) {
7956 		HIBLOG("WARNING: hibernate_page_list_setall skipped %d xpmapped pages\n", hibernate_stats.cd_skipped_xpmapped);
7957 	}
7958 
7959 	*pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative - count_discard_cleaned - count_retired;
7960 
7961 	if (preflight && will_discard) {
7962 		*pagesOut -= count_compressor + count_throttled + count_anonymous + count_inactive + count_cleaned + count_speculative + count_active;
7963 		/*
7964 		 * We try to keep max HIBERNATE_XPMAPPED_LIMIT pages around in the hibernation image
7965 		 * even if these are clean and so we need to size the hibernation image accordingly.
7966 		 *
7967 		 * NB: We have to assume all HIBERNATE_XPMAPPED_LIMIT pages might show up because 'dirty'
7968 		 * xpmapped pages aren't distinguishable from other 'dirty' pages in preflight. So we might
7969 		 * only see part of the xpmapped pages if we look at 'cd_found_xpmapped' which solely tracks
7970 		 * clean xpmapped pages.
7971 		 *
7972 		 * Since these pages are all cleaned by the time we are in the post-preflight phase, we might
7973 		 * see a much larger number in 'cd_found_xpmapped' now than we did in the preflight phase
7974 		 */
7975 		*pagesOut +=  HIBERNATE_XPMAPPED_LIMIT;
7976 	}
7977 
7978 	hibernation_vmqueues_inspection = FALSE;
7979 
7980 #if MACH_ASSERT || DEBUG
7981 	if (!preflight) {
7982 		if (vm_page_local_q) {
7983 			zpercpu_foreach(lq, vm_page_local_q) {
7984 				VPL_UNLOCK(&lq->vpl_lock);
7985 			}
7986 		}
7987 		vm_page_unlock_queues();
7988 	}
7989 #endif  /* MACH_ASSERT || DEBUG */
7990 
7991 	if (preflight) {
7992 		vm_free_page_unlock();
7993 		vm_page_unlock_queues();
7994 		vm_object_unlock(compressor_object);
7995 	}
7996 
7997 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_END, count_wire, *pagesOut, 0, 0, 0);
7998 }
7999 
8000 void
hibernate_page_list_discard(hibernate_page_list_t * page_list)8001 hibernate_page_list_discard(hibernate_page_list_t * page_list)
8002 {
8003 	uint64_t  start, end, nsec;
8004 	vm_page_t m;
8005 	vm_page_t next;
8006 	uint32_t  i;
8007 	uint32_t  count_discard_active    = 0;
8008 	uint32_t  count_discard_inactive  = 0;
8009 	uint32_t  count_discard_purgeable = 0;
8010 	uint32_t  count_discard_cleaned   = 0;
8011 	uint32_t  count_discard_speculative = 0;
8012 
8013 
8014 #if MACH_ASSERT || DEBUG
8015 	vm_page_lock_queues();
8016 	if (vm_page_local_q) {
8017 		zpercpu_foreach(lq, vm_page_local_q) {
8018 			VPL_LOCK(&lq->vpl_lock);
8019 		}
8020 	}
8021 #endif  /* MACH_ASSERT || DEBUG */
8022 
8023 	clock_get_uptime(&start);
8024 
8025 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_anonymous);
8026 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
8027 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
8028 
8029 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8030 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8031 			if (m->vmp_dirty) {
8032 				count_discard_purgeable++;
8033 			} else {
8034 				count_discard_inactive++;
8035 			}
8036 			hibernate_discard_page(m);
8037 		}
8038 		m = next;
8039 	}
8040 
8041 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
8042 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
8043 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
8044 			assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q);
8045 
8046 			next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8047 			if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8048 				count_discard_speculative++;
8049 				hibernate_discard_page(m);
8050 			}
8051 			m = next;
8052 		}
8053 	}
8054 
8055 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
8056 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
8057 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
8058 
8059 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8060 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8061 			if (m->vmp_dirty) {
8062 				count_discard_purgeable++;
8063 			} else {
8064 				count_discard_inactive++;
8065 			}
8066 			hibernate_discard_page(m);
8067 		}
8068 		m = next;
8069 	}
8070 	/* XXX FBDP TODO: secluded queue */
8071 
8072 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
8073 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
8074 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
8075 
8076 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8077 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8078 			if (m->vmp_dirty) {
8079 				count_discard_purgeable++;
8080 			} else {
8081 				count_discard_active++;
8082 			}
8083 			hibernate_discard_page(m);
8084 		}
8085 		m = next;
8086 	}
8087 
8088 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
8089 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
8090 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
8091 
8092 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8093 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8094 			if (m->vmp_dirty) {
8095 				count_discard_purgeable++;
8096 			} else {
8097 				count_discard_cleaned++;
8098 			}
8099 			hibernate_discard_page(m);
8100 		}
8101 		m = next;
8102 	}
8103 
8104 #if MACH_ASSERT || DEBUG
8105 	if (vm_page_local_q) {
8106 		zpercpu_foreach(lq, vm_page_local_q) {
8107 			VPL_UNLOCK(&lq->vpl_lock);
8108 		}
8109 	}
8110 	vm_page_unlock_queues();
8111 #endif  /* MACH_ASSERT || DEBUG */
8112 
8113 	clock_get_uptime(&end);
8114 	absolutetime_to_nanoseconds(end - start, &nsec);
8115 	HIBLOG("hibernate_page_list_discard time: %qd ms, discarded act %d inact %d purgeable %d spec %d cleaned %d\n",
8116 	    nsec / 1000000ULL,
8117 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned);
8118 }
8119 
8120 boolean_t       hibernate_paddr_map_inited = FALSE;
8121 unsigned int    hibernate_teardown_last_valid_compact_indx = -1;
8122 vm_page_t       hibernate_rebuild_hash_list = NULL;
8123 
8124 unsigned int    hibernate_teardown_found_tabled_pages = 0;
8125 unsigned int    hibernate_teardown_found_created_pages = 0;
8126 unsigned int    hibernate_teardown_found_free_pages = 0;
8127 unsigned int    hibernate_teardown_vm_page_free_count;
8128 
8129 
8130 struct ppnum_mapping {
8131 	struct ppnum_mapping    *ppnm_next;
8132 	ppnum_t                 ppnm_base_paddr;
8133 	unsigned int            ppnm_sindx;
8134 	unsigned int            ppnm_eindx;
8135 };
8136 
8137 struct ppnum_mapping    *ppnm_head;
8138 struct ppnum_mapping    *ppnm_last_found = NULL;
8139 
8140 
8141 void
hibernate_create_paddr_map(void)8142 hibernate_create_paddr_map(void)
8143 {
8144 	unsigned int    i;
8145 	ppnum_t         next_ppnum_in_run = 0;
8146 	struct ppnum_mapping *ppnm = NULL;
8147 
8148 	if (hibernate_paddr_map_inited == FALSE) {
8149 		for (i = 0; i < vm_pages_count; i++) {
8150 			if (ppnm) {
8151 				ppnm->ppnm_eindx = i;
8152 			}
8153 
8154 			if (ppnm == NULL || VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) != next_ppnum_in_run) {
8155 				ppnm = zalloc_permanent_type(struct ppnum_mapping);
8156 
8157 				ppnm->ppnm_next = ppnm_head;
8158 				ppnm_head = ppnm;
8159 
8160 				ppnm->ppnm_sindx = i;
8161 				ppnm->ppnm_base_paddr = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]);
8162 			}
8163 			next_ppnum_in_run = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) + 1;
8164 		}
8165 		ppnm->ppnm_eindx = vm_pages_count;
8166 
8167 		hibernate_paddr_map_inited = TRUE;
8168 	}
8169 }
8170 
8171 ppnum_t
hibernate_lookup_paddr(unsigned int indx)8172 hibernate_lookup_paddr(unsigned int indx)
8173 {
8174 	struct ppnum_mapping *ppnm = NULL;
8175 
8176 	ppnm = ppnm_last_found;
8177 
8178 	if (ppnm) {
8179 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8180 			goto done;
8181 		}
8182 	}
8183 	for (ppnm = ppnm_head; ppnm; ppnm = ppnm->ppnm_next) {
8184 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8185 			ppnm_last_found = ppnm;
8186 			break;
8187 		}
8188 	}
8189 	if (ppnm == NULL) {
8190 		panic("hibernate_lookup_paddr of %d failed", indx);
8191 	}
8192 done:
8193 	return ppnm->ppnm_base_paddr + (indx - ppnm->ppnm_sindx);
8194 }
8195 
8196 
8197 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)8198 hibernate_mark_as_unneeded(addr64_t saddr, addr64_t eaddr, hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8199 {
8200 	addr64_t        saddr_aligned;
8201 	addr64_t        eaddr_aligned;
8202 	addr64_t        addr;
8203 	ppnum_t         paddr;
8204 	unsigned int    mark_as_unneeded_pages = 0;
8205 
8206 	saddr_aligned = (saddr + PAGE_MASK_64) & ~PAGE_MASK_64;
8207 	eaddr_aligned = eaddr & ~PAGE_MASK_64;
8208 
8209 	for (addr = saddr_aligned; addr < eaddr_aligned; addr += PAGE_SIZE_64) {
8210 		paddr = pmap_find_phys(kernel_pmap, addr);
8211 
8212 		assert(paddr);
8213 
8214 		hibernate_page_bitset(page_list, TRUE, paddr);
8215 		hibernate_page_bitset(page_list_wired, TRUE, paddr);
8216 
8217 		mark_as_unneeded_pages++;
8218 	}
8219 	return mark_as_unneeded_pages;
8220 }
8221 
8222 
8223 void
hibernate_hash_insert_page(vm_page_t mem)8224 hibernate_hash_insert_page(vm_page_t mem)
8225 {
8226 	vm_page_bucket_t *bucket;
8227 	int             hash_id;
8228 	vm_object_t     m_object;
8229 
8230 	m_object = VM_PAGE_OBJECT(mem);
8231 
8232 	assert(mem->vmp_hashed);
8233 	assert(m_object);
8234 	assert(mem->vmp_offset != (vm_object_offset_t) -1);
8235 
8236 	/*
8237 	 *	Insert it into the object_object/offset hash table
8238 	 */
8239 	hash_id = vm_page_hash(m_object, mem->vmp_offset);
8240 	bucket = &vm_page_buckets[hash_id];
8241 
8242 	mem->vmp_next_m = bucket->page_list;
8243 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
8244 }
8245 
8246 
8247 void
hibernate_free_range(int sindx,int eindx)8248 hibernate_free_range(int sindx, int eindx)
8249 {
8250 	vm_page_t       mem;
8251 	unsigned int    color;
8252 
8253 	while (sindx < eindx) {
8254 		mem = &vm_pages[sindx];
8255 
8256 		vm_page_init(mem, hibernate_lookup_paddr(sindx), FALSE);
8257 
8258 		mem->vmp_lopage = FALSE;
8259 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8260 
8261 		color = VM_PAGE_GET_COLOR(mem);
8262 #if defined(__x86_64__)
8263 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
8264 #else
8265 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8266 #endif
8267 		vm_page_free_count++;
8268 
8269 		sindx++;
8270 	}
8271 }
8272 
8273 void
hibernate_rebuild_vm_structs(void)8274 hibernate_rebuild_vm_structs(void)
8275 {
8276 	int             i, cindx, sindx, eindx;
8277 	vm_page_t       mem, tmem, mem_next;
8278 	AbsoluteTime    startTime, endTime;
8279 	uint64_t        nsec;
8280 
8281 	if (hibernate_rebuild_needed == FALSE) {
8282 		return;
8283 	}
8284 
8285 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START);
8286 	HIBLOG("hibernate_rebuild started\n");
8287 
8288 	clock_get_uptime(&startTime);
8289 
8290 	pal_hib_rebuild_pmap_structs();
8291 
8292 	bzero(&vm_page_buckets[0], vm_page_bucket_count * sizeof(vm_page_bucket_t));
8293 	eindx = vm_pages_count;
8294 
8295 	/*
8296 	 * Mark all the vm_pages[] that have not been initialized yet as being
8297 	 * transient. This is needed to ensure that buddy page search is corrrect.
8298 	 * Without this random data in these vm_pages[] can trip the buddy search
8299 	 */
8300 	for (i = hibernate_teardown_last_valid_compact_indx + 1; i < eindx; ++i) {
8301 		vm_pages[i].vmp_q_state = VM_PAGE_NOT_ON_Q;
8302 	}
8303 
8304 	for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) {
8305 		mem = &vm_pages[cindx];
8306 		assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
8307 		/*
8308 		 * hibernate_teardown_vm_structs leaves the location where
8309 		 * this vm_page_t must be located in "next".
8310 		 */
8311 		tmem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8312 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
8313 
8314 		sindx = (int)(tmem - &vm_pages[0]);
8315 
8316 		if (mem != tmem) {
8317 			/*
8318 			 * this vm_page_t was moved by hibernate_teardown_vm_structs,
8319 			 * so move it back to its real location
8320 			 */
8321 			*tmem = *mem;
8322 			mem = tmem;
8323 		}
8324 		if (mem->vmp_hashed) {
8325 			hibernate_hash_insert_page(mem);
8326 		}
8327 		/*
8328 		 * the 'hole' between this vm_page_t and the previous
8329 		 * vm_page_t we moved needs to be initialized as
8330 		 * a range of free vm_page_t's
8331 		 */
8332 		hibernate_free_range(sindx + 1, eindx);
8333 
8334 		eindx = sindx;
8335 	}
8336 	if (sindx) {
8337 		hibernate_free_range(0, sindx);
8338 	}
8339 
8340 	assert(vm_page_free_count == hibernate_teardown_vm_page_free_count);
8341 
8342 	/*
8343 	 * process the list of vm_page_t's that were entered in the hash,
8344 	 * but were not located in the vm_pages arrary... these are
8345 	 * vm_page_t's that were created on the fly (i.e. fictitious)
8346 	 */
8347 	for (mem = hibernate_rebuild_hash_list; mem; mem = mem_next) {
8348 		mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8349 
8350 		mem->vmp_next_m = 0;
8351 		hibernate_hash_insert_page(mem);
8352 	}
8353 	hibernate_rebuild_hash_list = NULL;
8354 
8355 	clock_get_uptime(&endTime);
8356 	SUB_ABSOLUTETIME(&endTime, &startTime);
8357 	absolutetime_to_nanoseconds(endTime, &nsec);
8358 
8359 	HIBLOG("hibernate_rebuild completed - took %qd msecs\n", nsec / 1000000ULL);
8360 
8361 	hibernate_rebuild_needed = FALSE;
8362 
8363 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END);
8364 }
8365 
8366 uint32_t
hibernate_teardown_vm_structs(hibernate_page_list_t * page_list,hibernate_page_list_t * page_list_wired)8367 hibernate_teardown_vm_structs(hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8368 {
8369 	unsigned int    i;
8370 	unsigned int    compact_target_indx;
8371 	vm_page_t       mem, mem_next;
8372 	vm_page_bucket_t *bucket;
8373 	unsigned int    mark_as_unneeded_pages = 0;
8374 	unsigned int    unneeded_vm_page_bucket_pages = 0;
8375 	unsigned int    unneeded_vm_pages_pages = 0;
8376 	unsigned int    unneeded_pmap_pages = 0;
8377 	addr64_t        start_of_unneeded = 0;
8378 	addr64_t        end_of_unneeded = 0;
8379 
8380 
8381 	if (hibernate_should_abort()) {
8382 		return 0;
8383 	}
8384 
8385 	hibernate_rebuild_needed = TRUE;
8386 
8387 	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",
8388 	    vm_page_wire_count, vm_page_free_count, vm_page_active_count, vm_page_inactive_count, vm_page_speculative_count,
8389 	    vm_page_cleaned_count, compressor_object->resident_page_count);
8390 
8391 	for (i = 0; i < vm_page_bucket_count; i++) {
8392 		bucket = &vm_page_buckets[i];
8393 
8394 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); mem != VM_PAGE_NULL; mem = mem_next) {
8395 			assert(mem->vmp_hashed);
8396 
8397 			mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8398 
8399 			if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) {
8400 				mem->vmp_next_m = VM_PAGE_PACK_PTR(hibernate_rebuild_hash_list);
8401 				hibernate_rebuild_hash_list = mem;
8402 			}
8403 		}
8404 	}
8405 	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);
8406 	mark_as_unneeded_pages += unneeded_vm_page_bucket_pages;
8407 
8408 	hibernate_teardown_vm_page_free_count = vm_page_free_count;
8409 
8410 	compact_target_indx = 0;
8411 
8412 	for (i = 0; i < vm_pages_count; i++) {
8413 		mem = &vm_pages[i];
8414 
8415 		if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
8416 			unsigned int color;
8417 
8418 			assert(mem->vmp_busy);
8419 			assert(!mem->vmp_lopage);
8420 
8421 			color = VM_PAGE_GET_COLOR(mem);
8422 
8423 			vm_page_queue_remove(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8424 
8425 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
8426 
8427 			vm_page_free_count--;
8428 
8429 			hibernate_teardown_found_free_pages++;
8430 
8431 			if (vm_pages[compact_target_indx].vmp_q_state != VM_PAGE_ON_FREE_Q) {
8432 				compact_target_indx = i;
8433 			}
8434 		} else {
8435 			/*
8436 			 * record this vm_page_t's original location
8437 			 * we need this even if it doesn't get moved
8438 			 * as an indicator to the rebuild function that
8439 			 * we don't have to move it
8440 			 */
8441 			mem->vmp_next_m = VM_PAGE_PACK_PTR(mem);
8442 
8443 			if (vm_pages[compact_target_indx].vmp_q_state == VM_PAGE_ON_FREE_Q) {
8444 				/*
8445 				 * we've got a hole to fill, so
8446 				 * move this vm_page_t to it's new home
8447 				 */
8448 				vm_pages[compact_target_indx] = *mem;
8449 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8450 
8451 				hibernate_teardown_last_valid_compact_indx = compact_target_indx;
8452 				compact_target_indx++;
8453 			} else {
8454 				hibernate_teardown_last_valid_compact_indx = i;
8455 			}
8456 		}
8457 	}
8458 	unneeded_vm_pages_pages = hibernate_mark_as_unneeded((addr64_t)&vm_pages[hibernate_teardown_last_valid_compact_indx + 1],
8459 	    (addr64_t)&vm_pages[vm_pages_count - 1], page_list, page_list_wired);
8460 	mark_as_unneeded_pages += unneeded_vm_pages_pages;
8461 
8462 	pal_hib_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded);
8463 
8464 	if (start_of_unneeded) {
8465 		unneeded_pmap_pages = hibernate_mark_as_unneeded(start_of_unneeded, end_of_unneeded, page_list, page_list_wired);
8466 		mark_as_unneeded_pages += unneeded_pmap_pages;
8467 	}
8468 	HIBLOG("hibernate_teardown: mark_as_unneeded_pages %d, %d, %d\n", unneeded_vm_page_bucket_pages, unneeded_vm_pages_pages, unneeded_pmap_pages);
8469 
8470 	return mark_as_unneeded_pages;
8471 }
8472 
8473 
8474 #endif /* HIBERNATION */
8475 
8476 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
8477 
8478 #include <mach_vm_debug.h>
8479 #if     MACH_VM_DEBUG
8480 
8481 #include <mach_debug/hash_info.h>
8482 #include <vm/vm_debug.h>
8483 
8484 /*
8485  *	Routine:	vm_page_info
8486  *	Purpose:
8487  *		Return information about the global VP table.
8488  *		Fills the buffer with as much information as possible
8489  *		and returns the desired size of the buffer.
8490  *	Conditions:
8491  *		Nothing locked.  The caller should provide
8492  *		possibly-pageable memory.
8493  */
8494 
8495 unsigned int
vm_page_info(hash_info_bucket_t * info,unsigned int count)8496 vm_page_info(
8497 	hash_info_bucket_t *info,
8498 	unsigned int count)
8499 {
8500 	unsigned int i;
8501 	lck_spin_t      *bucket_lock;
8502 
8503 	if (vm_page_bucket_count < count) {
8504 		count = vm_page_bucket_count;
8505 	}
8506 
8507 	for (i = 0; i < count; i++) {
8508 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
8509 		unsigned int bucket_count = 0;
8510 		vm_page_t m;
8511 
8512 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8513 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8514 
8515 		for (m = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8516 		    m != VM_PAGE_NULL;
8517 		    m = (vm_page_t)(VM_PAGE_UNPACK_PTR(m->vmp_next_m))) {
8518 			bucket_count++;
8519 		}
8520 
8521 		lck_spin_unlock(bucket_lock);
8522 
8523 		/* don't touch pageable memory while holding locks */
8524 		info[i].hib_count = bucket_count;
8525 	}
8526 
8527 	return vm_page_bucket_count;
8528 }
8529 #endif  /* MACH_VM_DEBUG */
8530 
8531 #if VM_PAGE_BUCKETS_CHECK
8532 void
vm_page_buckets_check(void)8533 vm_page_buckets_check(void)
8534 {
8535 	unsigned int i;
8536 	vm_page_t p;
8537 	unsigned int p_hash;
8538 	vm_page_bucket_t *bucket;
8539 	lck_spin_t      *bucket_lock;
8540 
8541 	if (!vm_page_buckets_check_ready) {
8542 		return;
8543 	}
8544 
8545 #if HIBERNATION
8546 	if (hibernate_rebuild_needed ||
8547 	    hibernate_rebuild_hash_list) {
8548 		panic("BUCKET_CHECK: hibernation in progress: "
8549 		    "rebuild_needed=%d rebuild_hash_list=%p\n",
8550 		    hibernate_rebuild_needed,
8551 		    hibernate_rebuild_hash_list);
8552 	}
8553 #endif /* HIBERNATION */
8554 
8555 #if VM_PAGE_FAKE_BUCKETS
8556 	char *cp;
8557 	for (cp = (char *) vm_page_fake_buckets_start;
8558 	    cp < (char *) vm_page_fake_buckets_end;
8559 	    cp++) {
8560 		if (*cp != 0x5a) {
8561 			panic("BUCKET_CHECK: corruption at %p in fake buckets "
8562 			    "[0x%llx:0x%llx]\n",
8563 			    cp,
8564 			    (uint64_t) vm_page_fake_buckets_start,
8565 			    (uint64_t) vm_page_fake_buckets_end);
8566 		}
8567 	}
8568 #endif /* VM_PAGE_FAKE_BUCKETS */
8569 
8570 	for (i = 0; i < vm_page_bucket_count; i++) {
8571 		vm_object_t     p_object;
8572 
8573 		bucket = &vm_page_buckets[i];
8574 		if (!bucket->page_list) {
8575 			continue;
8576 		}
8577 
8578 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8579 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8580 		p = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8581 
8582 		while (p != VM_PAGE_NULL) {
8583 			p_object = VM_PAGE_OBJECT(p);
8584 
8585 			if (!p->vmp_hashed) {
8586 				panic("BUCKET_CHECK: page %p (%p,0x%llx) "
8587 				    "hash %d in bucket %d at %p "
8588 				    "is not hashed\n",
8589 				    p, p_object, p->vmp_offset,
8590 				    p_hash, i, bucket);
8591 			}
8592 			p_hash = vm_page_hash(p_object, p->vmp_offset);
8593 			if (p_hash != i) {
8594 				panic("BUCKET_CHECK: corruption in bucket %d "
8595 				    "at %p: page %p object %p offset 0x%llx "
8596 				    "hash %d\n",
8597 				    i, bucket, p, p_object, p->vmp_offset,
8598 				    p_hash);
8599 			}
8600 			p = (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m));
8601 		}
8602 		lck_spin_unlock(bucket_lock);
8603 	}
8604 
8605 //	printf("BUCKET_CHECK: checked buckets\n");
8606 }
8607 #endif /* VM_PAGE_BUCKETS_CHECK */
8608 
8609 /*
8610  * 'vm_fault_enter' will place newly created pages (zero-fill and COW) onto the
8611  * local queues if they exist... its the only spot in the system where we add pages
8612  * to those queues...  once on those queues, those pages can only move to one of the
8613  * global page queues or the free queues... they NEVER move from local q to local q.
8614  * the 'local' state is stable when vm_page_queues_remove is called since we're behind
8615  * the global vm_page_queue_lock at this point...  we still need to take the local lock
8616  * in case this operation is being run on a different CPU then the local queue's identity,
8617  * but we don't have to worry about the page moving to a global queue or becoming wired
8618  * while we're grabbing the local lock since those operations would require the global
8619  * vm_page_queue_lock to be held, and we already own it.
8620  *
8621  * this is why its safe to utilze the wire_count field in the vm_page_t as the local_id...
8622  * 'wired' and local are ALWAYS mutually exclusive conditions.
8623  */
8624 
8625 void
vm_page_queues_remove(vm_page_t mem,boolean_t remove_from_specialq)8626 vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_specialq)
8627 {
8628 	boolean_t       was_pageable = TRUE;
8629 	vm_object_t     m_object;
8630 
8631 	m_object = VM_PAGE_OBJECT(mem);
8632 
8633 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8634 
8635 	if (mem->vmp_q_state == VM_PAGE_NOT_ON_Q) {
8636 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8637 		if (remove_from_specialq == TRUE) {
8638 			vm_page_remove_from_specialq(mem);
8639 		}
8640 		/*if (mem->vmp_on_specialq != VM_PAGE_SPECIAL_Q_EMPTY) {
8641 		 *       assert(mem->vmp_specialq.next != 0);
8642 		 *       assert(mem->vmp_specialq.prev != 0);
8643 		 *  } else {*/
8644 		if (mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
8645 			assert(mem->vmp_specialq.next == 0);
8646 			assert(mem->vmp_specialq.prev == 0);
8647 		}
8648 		return;
8649 	}
8650 
8651 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
8652 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8653 		assert(mem->vmp_specialq.next == 0 &&
8654 		    mem->vmp_specialq.prev == 0 &&
8655 		    mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
8656 		return;
8657 	}
8658 	if (mem->vmp_q_state == VM_PAGE_IS_WIRED) {
8659 		/*
8660 		 * might put these guys on a list for debugging purposes
8661 		 * if we do, we'll need to remove this assert
8662 		 */
8663 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8664 		assert(mem->vmp_specialq.next == 0 &&
8665 		    mem->vmp_specialq.prev == 0);
8666 		/*
8667 		 * Recall that vmp_on_specialq also means a request to put
8668 		 * it on the special Q. So we don't want to reset that bit
8669 		 * just because a wiring request came in. We might want to
8670 		 * put it on the special queue post-unwiring.
8671 		 *
8672 		 * &&
8673 		 * mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
8674 		 */
8675 		return;
8676 	}
8677 
8678 	assert(m_object != compressor_object);
8679 	assert(m_object != kernel_object);
8680 	assert(!mem->vmp_fictitious);
8681 
8682 	switch (mem->vmp_q_state) {
8683 	case VM_PAGE_ON_ACTIVE_LOCAL_Q:
8684 	{
8685 		struct vpl      *lq;
8686 
8687 		lq = zpercpu_get_cpu(vm_page_local_q, mem->vmp_local_id);
8688 		VPL_LOCK(&lq->vpl_lock);
8689 		vm_page_queue_remove(&lq->vpl_queue, mem, vmp_pageq);
8690 		mem->vmp_local_id = 0;
8691 		lq->vpl_count--;
8692 		if (m_object->internal) {
8693 			lq->vpl_internal_count--;
8694 		} else {
8695 			lq->vpl_external_count--;
8696 		}
8697 		VPL_UNLOCK(&lq->vpl_lock);
8698 		was_pageable = FALSE;
8699 		break;
8700 	}
8701 	case VM_PAGE_ON_ACTIVE_Q:
8702 	{
8703 		vm_page_queue_remove(&vm_page_queue_active, mem, vmp_pageq);
8704 		vm_page_active_count--;
8705 		break;
8706 	}
8707 
8708 	case VM_PAGE_ON_INACTIVE_INTERNAL_Q:
8709 	{
8710 		assert(m_object->internal == TRUE);
8711 
8712 		vm_page_inactive_count--;
8713 		vm_page_queue_remove(&vm_page_queue_anonymous, mem, vmp_pageq);
8714 		vm_page_anonymous_count--;
8715 
8716 		vm_purgeable_q_advance_all();
8717 		vm_page_balance_inactive(3);
8718 		break;
8719 	}
8720 
8721 	case VM_PAGE_ON_INACTIVE_EXTERNAL_Q:
8722 	{
8723 		assert(m_object->internal == FALSE);
8724 
8725 		vm_page_inactive_count--;
8726 		vm_page_queue_remove(&vm_page_queue_inactive, mem, vmp_pageq);
8727 		vm_purgeable_q_advance_all();
8728 		vm_page_balance_inactive(3);
8729 		break;
8730 	}
8731 
8732 	case VM_PAGE_ON_INACTIVE_CLEANED_Q:
8733 	{
8734 		assert(m_object->internal == FALSE);
8735 
8736 		vm_page_inactive_count--;
8737 		vm_page_queue_remove(&vm_page_queue_cleaned, mem, vmp_pageq);
8738 		vm_page_cleaned_count--;
8739 		vm_page_balance_inactive(3);
8740 		break;
8741 	}
8742 
8743 	case VM_PAGE_ON_THROTTLED_Q:
8744 	{
8745 		assert(m_object->internal == TRUE);
8746 
8747 		vm_page_queue_remove(&vm_page_queue_throttled, mem, vmp_pageq);
8748 		vm_page_throttled_count--;
8749 		was_pageable = FALSE;
8750 		break;
8751 	}
8752 
8753 	case VM_PAGE_ON_SPECULATIVE_Q:
8754 	{
8755 		assert(m_object->internal == FALSE);
8756 
8757 		vm_page_remque(&mem->vmp_pageq);
8758 		vm_page_speculative_count--;
8759 		vm_page_balance_inactive(3);
8760 		break;
8761 	}
8762 
8763 #if CONFIG_SECLUDED_MEMORY
8764 	case VM_PAGE_ON_SECLUDED_Q:
8765 	{
8766 		vm_page_queue_remove(&vm_page_queue_secluded, mem, vmp_pageq);
8767 		vm_page_secluded_count--;
8768 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
8769 		if (m_object == VM_OBJECT_NULL) {
8770 			vm_page_secluded_count_free--;
8771 			was_pageable = FALSE;
8772 		} else {
8773 			assert(!m_object->internal);
8774 			vm_page_secluded_count_inuse--;
8775 			was_pageable = FALSE;
8776 //			was_pageable = TRUE;
8777 		}
8778 		break;
8779 	}
8780 #endif /* CONFIG_SECLUDED_MEMORY */
8781 
8782 	default:
8783 	{
8784 		/*
8785 		 *	if (mem->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)
8786 		 *              NOTE: vm_page_queues_remove does not deal with removing pages from the pageout queue...
8787 		 *              the caller is responsible for determing if the page is on that queue, and if so, must
8788 		 *              either first remove it (it needs both the page queues lock and the object lock to do
8789 		 *              this via vm_pageout_steal_laundry), or avoid the call to vm_page_queues_remove
8790 		 *
8791 		 *	we also don't expect to encounter VM_PAGE_ON_FREE_Q, VM_PAGE_ON_FREE_LOCAL_Q, VM_PAGE_ON_FREE_LOPAGE_Q
8792 		 *	or any of the undefined states
8793 		 */
8794 		panic("vm_page_queues_remove - bad page q_state (%p, %d)", mem, mem->vmp_q_state);
8795 		break;
8796 	}
8797 	}
8798 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
8799 	mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
8800 
8801 	if (remove_from_specialq == TRUE) {
8802 		vm_page_remove_from_specialq(mem);
8803 	}
8804 	if (was_pageable) {
8805 		if (m_object->internal) {
8806 			vm_page_pageable_internal_count--;
8807 		} else {
8808 			vm_page_pageable_external_count--;
8809 		}
8810 	}
8811 }
8812 
8813 void
vm_page_remove_internal(vm_page_t page)8814 vm_page_remove_internal(vm_page_t page)
8815 {
8816 	vm_object_t __object = VM_PAGE_OBJECT(page);
8817 	if (page == __object->memq_hint) {
8818 		vm_page_t       __new_hint;
8819 		vm_page_queue_entry_t   __qe;
8820 		__qe = (vm_page_queue_entry_t)vm_page_queue_next(&page->vmp_listq);
8821 		if (vm_page_queue_end(&__object->memq, __qe)) {
8822 			__qe = (vm_page_queue_entry_t)vm_page_queue_prev(&page->vmp_listq);
8823 			if (vm_page_queue_end(&__object->memq, __qe)) {
8824 				__qe = NULL;
8825 			}
8826 		}
8827 		__new_hint = (vm_page_t)((uintptr_t) __qe);
8828 		__object->memq_hint = __new_hint;
8829 	}
8830 	vm_page_queue_remove(&__object->memq, page, vmp_listq);
8831 #if CONFIG_SECLUDED_MEMORY
8832 	if (__object->eligible_for_secluded) {
8833 		vm_page_secluded.eligible_for_secluded--;
8834 	}
8835 #endif /* CONFIG_SECLUDED_MEMORY */
8836 }
8837 
8838 void
vm_page_enqueue_inactive(vm_page_t mem,boolean_t first)8839 vm_page_enqueue_inactive(vm_page_t mem, boolean_t first)
8840 {
8841 	vm_object_t     m_object;
8842 
8843 	m_object = VM_PAGE_OBJECT(mem);
8844 
8845 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8846 	assert(!mem->vmp_fictitious);
8847 	assert(!mem->vmp_laundry);
8848 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
8849 	vm_page_check_pageable_safe(mem);
8850 
8851 	if (m_object->internal) {
8852 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_INTERNAL_Q;
8853 
8854 		if (first == TRUE) {
8855 			vm_page_queue_enter_first(&vm_page_queue_anonymous, mem, vmp_pageq);
8856 		} else {
8857 			vm_page_queue_enter(&vm_page_queue_anonymous, mem, vmp_pageq);
8858 		}
8859 
8860 		vm_page_anonymous_count++;
8861 		vm_page_pageable_internal_count++;
8862 	} else {
8863 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_EXTERNAL_Q;
8864 
8865 		if (first == TRUE) {
8866 			vm_page_queue_enter_first(&vm_page_queue_inactive, mem, vmp_pageq);
8867 		} else {
8868 			vm_page_queue_enter(&vm_page_queue_inactive, mem, vmp_pageq);
8869 		}
8870 
8871 		vm_page_pageable_external_count++;
8872 	}
8873 	vm_page_inactive_count++;
8874 	token_new_pagecount++;
8875 
8876 	vm_page_add_to_specialq(mem, FALSE);
8877 }
8878 
8879 void
vm_page_enqueue_active(vm_page_t mem,boolean_t first)8880 vm_page_enqueue_active(vm_page_t mem, boolean_t first)
8881 {
8882 	vm_object_t     m_object;
8883 
8884 	m_object = VM_PAGE_OBJECT(mem);
8885 
8886 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8887 	assert(!mem->vmp_fictitious);
8888 	assert(!mem->vmp_laundry);
8889 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
8890 	vm_page_check_pageable_safe(mem);
8891 
8892 	mem->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
8893 	if (first == TRUE) {
8894 		vm_page_queue_enter_first(&vm_page_queue_active, mem, vmp_pageq);
8895 	} else {
8896 		vm_page_queue_enter(&vm_page_queue_active, mem, vmp_pageq);
8897 	}
8898 	vm_page_active_count++;
8899 
8900 	if (m_object->internal) {
8901 		vm_page_pageable_internal_count++;
8902 	} else {
8903 		vm_page_pageable_external_count++;
8904 	}
8905 
8906 	vm_page_add_to_specialq(mem, FALSE);
8907 	vm_page_balance_inactive(3);
8908 }
8909 
8910 /*
8911  * Pages from special kernel objects shouldn't
8912  * be placed on pageable queues.
8913  */
8914 void
vm_page_check_pageable_safe(vm_page_t page)8915 vm_page_check_pageable_safe(vm_page_t page)
8916 {
8917 	vm_object_t     page_object;
8918 
8919 	page_object = VM_PAGE_OBJECT(page);
8920 
8921 	if (page_object == kernel_object) {
8922 		panic("vm_page_check_pageable_safe: trying to add page"
8923 		    "from kernel object (%p) to pageable queue", kernel_object);
8924 	}
8925 
8926 	if (page_object == compressor_object) {
8927 		panic("vm_page_check_pageable_safe: trying to add page"
8928 		    "from compressor object (%p) to pageable queue", compressor_object);
8929 	}
8930 }
8931 
8932 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
8933 * wired page diagnose
8934 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
8935 
8936 #include <libkern/OSKextLibPrivate.h>
8937 
8938 #define KA_SIZE(namelen, subtotalscount)        \
8939 	(sizeof(struct vm_allocation_site) + (namelen) + 1 + ((subtotalscount) * sizeof(struct vm_allocation_total)))
8940 
8941 #define KA_NAME(alloc)  \
8942 	((char *)(&(alloc)->subtotals[(alloc->subtotalscount)]))
8943 
8944 #define KA_NAME_LEN(alloc)      \
8945     (VM_TAG_NAME_LEN_MAX & (alloc->flags >> VM_TAG_NAME_LEN_SHIFT))
8946 
8947 vm_tag_t
vm_tag_bt(void)8948 vm_tag_bt(void)
8949 {
8950 	uintptr_t* frameptr;
8951 	uintptr_t* frameptr_next;
8952 	uintptr_t retaddr;
8953 	uintptr_t kstackb, kstackt;
8954 	const vm_allocation_site_t * site;
8955 	thread_t cthread;
8956 	kern_allocation_name_t name;
8957 
8958 	cthread = current_thread();
8959 	if (__improbable(cthread == NULL)) {
8960 		return VM_KERN_MEMORY_OSFMK;
8961 	}
8962 
8963 	if ((name = thread_get_kernel_state(cthread)->allocation_name)) {
8964 		if (!name->tag) {
8965 			vm_tag_alloc(name);
8966 		}
8967 		return name->tag;
8968 	}
8969 
8970 	kstackb = cthread->kernel_stack;
8971 	kstackt = kstackb + kernel_stack_size;
8972 
8973 	/* Load stack frame pointer (EBP on x86) into frameptr */
8974 	frameptr = __builtin_frame_address(0);
8975 	site = NULL;
8976 	while (frameptr != NULL) {
8977 		/* Verify thread stack bounds */
8978 		if (((uintptr_t)(frameptr + 2) > kstackt) || ((uintptr_t)frameptr < kstackb)) {
8979 			break;
8980 		}
8981 
8982 		/* Next frame pointer is pointed to by the previous one */
8983 		frameptr_next = (uintptr_t*) *frameptr;
8984 
8985 		/* Pull return address from one spot above the frame pointer */
8986 		retaddr = *(frameptr + 1);
8987 
8988 #if defined(HAS_APPLE_PAC)
8989 		retaddr = (uintptr_t) ptrauth_strip((void *)retaddr, ptrauth_key_return_address);
8990 #endif
8991 
8992 		if (((retaddr < vm_kernel_builtinkmod_text_end) && (retaddr >= vm_kernel_builtinkmod_text))
8993 		    || (retaddr < vm_kernel_stext) || (retaddr > vm_kernel_top)) {
8994 			site = OSKextGetAllocationSiteForCaller(retaddr);
8995 			break;
8996 		}
8997 		frameptr = frameptr_next;
8998 	}
8999 
9000 	return site ? site->tag : VM_KERN_MEMORY_NONE;
9001 }
9002 
9003 static uint64_t free_tag_bits[VM_MAX_TAG_VALUE / 64];
9004 
9005 void
vm_tag_alloc_locked(vm_allocation_site_t * site,vm_allocation_site_t ** releasesiteP)9006 vm_tag_alloc_locked(vm_allocation_site_t * site, vm_allocation_site_t ** releasesiteP)
9007 {
9008 	vm_tag_t tag;
9009 	uint64_t avail;
9010 	uint32_t idx;
9011 	vm_allocation_site_t * prev;
9012 
9013 	if (site->tag) {
9014 		return;
9015 	}
9016 
9017 	idx = 0;
9018 	while (TRUE) {
9019 		avail = free_tag_bits[idx];
9020 		if (avail) {
9021 			tag = (vm_tag_t)__builtin_clzll(avail);
9022 			avail &= ~(1ULL << (63 - tag));
9023 			free_tag_bits[idx] = avail;
9024 			tag += (idx << 6);
9025 			break;
9026 		}
9027 		idx++;
9028 		if (idx >= ARRAY_COUNT(free_tag_bits)) {
9029 			for (idx = 0; idx < ARRAY_COUNT(vm_allocation_sites); idx++) {
9030 				prev = vm_allocation_sites[idx];
9031 				if (!prev) {
9032 					continue;
9033 				}
9034 				if (!KA_NAME_LEN(prev)) {
9035 					continue;
9036 				}
9037 				if (!prev->tag) {
9038 					continue;
9039 				}
9040 				if (prev->total) {
9041 					continue;
9042 				}
9043 				if (1 != prev->refcount) {
9044 					continue;
9045 				}
9046 
9047 				assert(idx == prev->tag);
9048 				tag = (vm_tag_t)idx;
9049 				prev->tag = VM_KERN_MEMORY_NONE;
9050 				*releasesiteP = prev;
9051 				break;
9052 			}
9053 			if (idx >= ARRAY_COUNT(vm_allocation_sites)) {
9054 				tag = VM_KERN_MEMORY_ANY;
9055 			}
9056 			break;
9057 		}
9058 	}
9059 	site->tag = tag;
9060 
9061 	OSAddAtomic16(1, &site->refcount);
9062 
9063 	if (VM_KERN_MEMORY_ANY != tag) {
9064 		vm_allocation_sites[tag] = site;
9065 	}
9066 
9067 	if (tag > vm_allocation_tag_highest) {
9068 		vm_allocation_tag_highest = tag;
9069 	}
9070 }
9071 
9072 static void
vm_tag_free_locked(vm_tag_t tag)9073 vm_tag_free_locked(vm_tag_t tag)
9074 {
9075 	uint64_t avail;
9076 	uint32_t idx;
9077 	uint64_t bit;
9078 
9079 	if (VM_KERN_MEMORY_ANY == tag) {
9080 		return;
9081 	}
9082 
9083 	idx = (tag >> 6);
9084 	avail = free_tag_bits[idx];
9085 	tag &= 63;
9086 	bit = (1ULL << (63 - tag));
9087 	assert(!(avail & bit));
9088 	free_tag_bits[idx] = (avail | bit);
9089 }
9090 
9091 static void
vm_tag_init(void)9092 vm_tag_init(void)
9093 {
9094 	vm_tag_t tag;
9095 	for (tag = VM_KERN_MEMORY_FIRST_DYNAMIC; tag < VM_KERN_MEMORY_ANY; tag++) {
9096 		vm_tag_free_locked(tag);
9097 	}
9098 
9099 	for (tag = VM_KERN_MEMORY_ANY + 1; tag < VM_MAX_TAG_VALUE; tag++) {
9100 		vm_tag_free_locked(tag);
9101 	}
9102 }
9103 
9104 vm_tag_t
vm_tag_alloc(vm_allocation_site_t * site)9105 vm_tag_alloc(vm_allocation_site_t * site)
9106 {
9107 	vm_allocation_site_t * releasesite;
9108 
9109 	if (!site->tag) {
9110 		releasesite = NULL;
9111 		lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9112 		vm_tag_alloc_locked(site, &releasesite);
9113 		lck_ticket_unlock(&vm_allocation_sites_lock);
9114 		if (releasesite) {
9115 			kern_allocation_name_release(releasesite);
9116 		}
9117 	}
9118 
9119 	return site->tag;
9120 }
9121 
9122 void
vm_tag_update_size(vm_tag_t tag,int64_t delta)9123 vm_tag_update_size(vm_tag_t tag, int64_t delta)
9124 {
9125 	vm_allocation_site_t * allocation;
9126 	uint64_t value;
9127 
9128 	assert(VM_KERN_MEMORY_NONE != tag);
9129 	assert(tag < VM_MAX_TAG_VALUE);
9130 
9131 	allocation = vm_allocation_sites[tag];
9132 	assert(allocation);
9133 
9134 	value = os_atomic_add(&allocation->total, delta, relaxed);
9135 	if (delta < 0) {
9136 		assertf(value + (uint64_t)-delta > value,
9137 		    "tag %d, site %p", tag, allocation);
9138 	}
9139 
9140 #if DEBUG || DEVELOPMENT
9141 	if (value > allocation->peak) {
9142 		os_atomic_max(&allocation->peak, value, relaxed);
9143 	}
9144 #endif /* DEBUG || DEVELOPMENT */
9145 
9146 	if (tag < VM_KERN_MEMORY_FIRST_DYNAMIC) {
9147 		return;
9148 	}
9149 
9150 	if (value == (uint64_t)delta && !allocation->tag) {
9151 		vm_tag_alloc(allocation);
9152 	}
9153 }
9154 
9155 uint64_t
vm_tag_get_size(vm_tag_t tag)9156 vm_tag_get_size(vm_tag_t tag)
9157 {
9158 	vm_allocation_site_t * allocation;
9159 
9160 	assert(VM_KERN_MEMORY_NONE != tag);
9161 	assert(tag < VM_MAX_TAG_VALUE);
9162 
9163 	allocation = vm_allocation_sites[tag];
9164 	return allocation ? os_atomic_load(&allocation->total, relaxed) : 0;
9165 }
9166 
9167 void
kern_allocation_update_size(kern_allocation_name_t allocation,int64_t delta)9168 kern_allocation_update_size(kern_allocation_name_t allocation, int64_t delta)
9169 {
9170 	uint64_t prior;
9171 
9172 	if (delta < 0) {
9173 		assertf(allocation->total >= ((uint64_t)-delta), "name %p", allocation);
9174 	}
9175 	prior = OSAddAtomic64(delta, &allocation->total);
9176 
9177 #if DEBUG || DEVELOPMENT
9178 
9179 	uint64_t new, peak;
9180 	new = prior + delta;
9181 	do{
9182 		peak = allocation->peak;
9183 		if (new <= peak) {
9184 			break;
9185 		}
9186 	}while (!OSCompareAndSwap64(peak, new, &allocation->peak));
9187 
9188 #endif /* DEBUG || DEVELOPMENT */
9189 
9190 	if (!prior && !allocation->tag) {
9191 		vm_tag_alloc(allocation);
9192 	}
9193 }
9194 
9195 #if VM_TAG_SIZECLASSES
9196 
9197 void
vm_allocation_zones_init(void)9198 vm_allocation_zones_init(void)
9199 {
9200 	vm_offset_t   addr;
9201 	vm_size_t     size;
9202 
9203 	const vm_tag_t early_tags[] = {
9204 		VM_KERN_MEMORY_DIAG,
9205 		VM_KERN_MEMORY_KALLOC,
9206 		VM_KERN_MEMORY_KALLOC_DATA,
9207 		VM_KERN_MEMORY_KALLOC_TYPE,
9208 		VM_KERN_MEMORY_LIBKERN,
9209 		VM_KERN_MEMORY_OSFMK,
9210 		VM_KERN_MEMORY_RECOUNT,
9211 	};
9212 
9213 	size = VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *)
9214 	    + ARRAY_COUNT(early_tags) * VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9215 
9216 	kmem_alloc(kernel_map, &addr, round_page(size),
9217 	    KMA_NOFAIL | KMA_KOBJECT | KMA_ZERO | KMA_PERMANENT,
9218 	    VM_KERN_MEMORY_DIAG);
9219 
9220 	vm_allocation_zone_totals = (vm_allocation_zone_total_t **) addr;
9221 	addr += VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *);
9222 
9223 	// prepopulate early tag ranges so allocations
9224 	// in vm_tag_update_zone_size() and early boot won't recurse
9225 	for (size_t i = 0; i < ARRAY_COUNT(early_tags); i++) {
9226 		vm_allocation_zone_totals[early_tags[i]] = (vm_allocation_zone_total_t *)addr;
9227 		addr += VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9228 	}
9229 }
9230 
9231 __attribute__((noinline))
9232 static vm_tag_t
vm_tag_zone_stats_alloc(vm_tag_t tag,zalloc_flags_t flags)9233 vm_tag_zone_stats_alloc(vm_tag_t tag, zalloc_flags_t flags)
9234 {
9235 	vm_allocation_zone_total_t *stats;
9236 	vm_size_t size = sizeof(*stats) * VM_TAG_SIZECLASSES;
9237 
9238 	flags = Z_VM_TAG(Z_ZERO | flags, VM_KERN_MEMORY_DIAG);
9239 	stats = kalloc_data(size, flags);
9240 	if (!stats) {
9241 		return VM_KERN_MEMORY_NONE;
9242 	}
9243 	if (!os_atomic_cmpxchg(&vm_allocation_zone_totals[tag], NULL, stats, release)) {
9244 		kfree_data(stats, size);
9245 	}
9246 	return tag;
9247 }
9248 
9249 vm_tag_t
vm_tag_will_update_zone(vm_tag_t tag,uint32_t zidx,uint32_t zflags)9250 vm_tag_will_update_zone(vm_tag_t tag, uint32_t zidx, uint32_t zflags)
9251 {
9252 	assert(VM_KERN_MEMORY_NONE != tag);
9253 	assert(tag < VM_MAX_TAG_VALUE);
9254 
9255 	if (zidx >= VM_TAG_SIZECLASSES) {
9256 		return VM_KERN_MEMORY_NONE;
9257 	}
9258 
9259 	if (__probable(vm_allocation_zone_totals[tag])) {
9260 		return tag;
9261 	}
9262 	return vm_tag_zone_stats_alloc(tag, zflags);
9263 }
9264 
9265 void
vm_tag_update_zone_size(vm_tag_t tag,uint32_t zidx,long delta)9266 vm_tag_update_zone_size(vm_tag_t tag, uint32_t zidx, long delta)
9267 {
9268 	vm_allocation_zone_total_t *stats;
9269 	vm_size_t value;
9270 
9271 	assert(VM_KERN_MEMORY_NONE != tag);
9272 	assert(tag < VM_MAX_TAG_VALUE);
9273 
9274 	if (zidx >= VM_TAG_SIZECLASSES) {
9275 		return;
9276 	}
9277 
9278 	stats = vm_allocation_zone_totals[tag];
9279 	assert(stats);
9280 	stats += zidx;
9281 
9282 	value = os_atomic_add(&stats->vazt_total, delta, relaxed);
9283 	if (delta < 0) {
9284 		assertf((long)value >= 0, "zidx %d, tag %d, %p", zidx, tag, stats);
9285 		return;
9286 	} else if (os_atomic_load(&stats->vazt_peak, relaxed) < value) {
9287 		os_atomic_max(&stats->vazt_peak, value, relaxed);
9288 	}
9289 }
9290 
9291 #endif /* VM_TAG_SIZECLASSES */
9292 
9293 void
kern_allocation_update_subtotal(kern_allocation_name_t allocation,uint32_t subtag,int64_t delta)9294 kern_allocation_update_subtotal(kern_allocation_name_t allocation, uint32_t subtag, int64_t delta)
9295 {
9296 	kern_allocation_name_t other;
9297 	struct vm_allocation_total * total;
9298 	uint32_t subidx;
9299 
9300 	subidx = 0;
9301 	assert(VM_KERN_MEMORY_NONE != subtag);
9302 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9303 	for (; subidx < allocation->subtotalscount; subidx++) {
9304 		if (VM_KERN_MEMORY_NONE == allocation->subtotals[subidx].tag) {
9305 			allocation->subtotals[subidx].tag = (vm_tag_t)subtag;
9306 			break;
9307 		}
9308 		if (subtag == allocation->subtotals[subidx].tag) {
9309 			break;
9310 		}
9311 	}
9312 	lck_ticket_unlock(&vm_allocation_sites_lock);
9313 	assert(subidx < allocation->subtotalscount);
9314 	if (subidx >= allocation->subtotalscount) {
9315 		return;
9316 	}
9317 
9318 	total = &allocation->subtotals[subidx];
9319 	other = vm_allocation_sites[subtag];
9320 	assert(other);
9321 
9322 	if (delta < 0) {
9323 		assertf(total->total >= ((uint64_t)-delta), "name %p", allocation);
9324 		assertf(other->mapped >= ((uint64_t)-delta), "other %p", other);
9325 	}
9326 	OSAddAtomic64(delta, &other->mapped);
9327 	OSAddAtomic64(delta, &total->total);
9328 }
9329 
9330 const char *
kern_allocation_get_name(kern_allocation_name_t allocation)9331 kern_allocation_get_name(kern_allocation_name_t allocation)
9332 {
9333 	return KA_NAME(allocation);
9334 }
9335 
9336 kern_allocation_name_t
kern_allocation_name_allocate(const char * name,uint16_t subtotalscount)9337 kern_allocation_name_allocate(const char * name, uint16_t subtotalscount)
9338 {
9339 	kern_allocation_name_t allocation;
9340 	uint16_t namelen;
9341 
9342 	namelen = (uint16_t)strnlen(name, MACH_MEMORY_INFO_NAME_MAX_LEN - 1);
9343 
9344 	allocation = kalloc_data(KA_SIZE(namelen, subtotalscount), Z_WAITOK | Z_ZERO);
9345 	allocation->refcount       = 1;
9346 	allocation->subtotalscount = subtotalscount;
9347 	allocation->flags          = (uint16_t)(namelen << VM_TAG_NAME_LEN_SHIFT);
9348 	strlcpy(KA_NAME(allocation), name, namelen + 1);
9349 
9350 	vm_tag_alloc(allocation);
9351 	return allocation;
9352 }
9353 
9354 void
kern_allocation_name_release(kern_allocation_name_t allocation)9355 kern_allocation_name_release(kern_allocation_name_t allocation)
9356 {
9357 	assert(allocation->refcount > 0);
9358 	if (1 == OSAddAtomic16(-1, &allocation->refcount)) {
9359 		kfree_data(allocation,
9360 		    KA_SIZE(KA_NAME_LEN(allocation), allocation->subtotalscount));
9361 	}
9362 }
9363 
9364 vm_tag_t
kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation)9365 kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation)
9366 {
9367 	return vm_tag_alloc(allocation);
9368 }
9369 
9370 #if !VM_TAG_ACTIVE_UPDATE
9371 static void
vm_page_count_object(mach_memory_info_t * info,unsigned int __unused num_info,vm_object_t object)9372 vm_page_count_object(mach_memory_info_t * info, unsigned int __unused num_info, vm_object_t object)
9373 {
9374 	if (!object->wired_page_count) {
9375 		return;
9376 	}
9377 	if (object != kernel_object) {
9378 		assert(object->wire_tag < num_info);
9379 		info[object->wire_tag].size += ptoa_64(object->wired_page_count);
9380 	}
9381 }
9382 
9383 typedef void (*vm_page_iterate_proc)(mach_memory_info_t * info,
9384     unsigned int num_info, vm_object_t object);
9385 
9386 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)9387 vm_page_iterate_purgeable_objects(mach_memory_info_t * info, unsigned int num_info,
9388     vm_page_iterate_proc proc, purgeable_q_t queue,
9389     int group)
9390 {
9391 	vm_object_t object;
9392 
9393 	for (object = (vm_object_t) queue_first(&queue->objq[group]);
9394 	    !queue_end(&queue->objq[group], (queue_entry_t) object);
9395 	    object = (vm_object_t) queue_next(&object->objq)) {
9396 		proc(info, num_info, object);
9397 	}
9398 }
9399 
9400 static void
vm_page_iterate_objects(mach_memory_info_t * info,unsigned int num_info,vm_page_iterate_proc proc)9401 vm_page_iterate_objects(mach_memory_info_t * info, unsigned int num_info,
9402     vm_page_iterate_proc proc)
9403 {
9404 	vm_object_t     object;
9405 
9406 	lck_spin_lock_grp(&vm_objects_wired_lock, &vm_page_lck_grp_bucket);
9407 	queue_iterate(&vm_objects_wired,
9408 	    object,
9409 	    vm_object_t,
9410 	    wired_objq)
9411 	{
9412 		proc(info, num_info, object);
9413 	}
9414 	lck_spin_unlock(&vm_objects_wired_lock);
9415 }
9416 #endif /* ! VM_TAG_ACTIVE_UPDATE */
9417 
9418 static uint64_t
process_account(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes,boolean_t iterated)9419 process_account(mach_memory_info_t * info, unsigned int num_info,
9420     uint64_t zones_collectable_bytes, boolean_t iterated)
9421 {
9422 	size_t                 namelen;
9423 	unsigned int           idx, count, nextinfo;
9424 	vm_allocation_site_t * site;
9425 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9426 
9427 	for (idx = 0; idx <= vm_allocation_tag_highest; idx++) {
9428 		site = vm_allocation_sites[idx];
9429 		if (!site) {
9430 			continue;
9431 		}
9432 		info[idx].mapped = site->mapped;
9433 		info[idx].tag    = site->tag;
9434 		if (!iterated) {
9435 			info[idx].size = site->total;
9436 #if DEBUG || DEVELOPMENT
9437 			info[idx].peak = site->peak;
9438 #endif /* DEBUG || DEVELOPMENT */
9439 		} else {
9440 			if (!site->subtotalscount && (site->total != info[idx].size)) {
9441 				printf("tag mismatch[%d] 0x%qx, iter 0x%qx\n", idx, site->total, info[idx].size);
9442 				info[idx].size = site->total;
9443 			}
9444 		}
9445 		info[idx].flags |= VM_KERN_SITE_WIRED;
9446 		if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC) {
9447 			info[idx].site   = idx;
9448 			info[idx].flags |= VM_KERN_SITE_TAG;
9449 			if (VM_KERN_MEMORY_ZONE == idx) {
9450 				info[idx].flags |= VM_KERN_SITE_HIDE;
9451 				info[idx].flags &= ~VM_KERN_SITE_WIRED;
9452 				info[idx].collectable_bytes = zones_collectable_bytes;
9453 			}
9454 		} else if ((namelen = (VM_TAG_NAME_LEN_MAX & (site->flags >> VM_TAG_NAME_LEN_SHIFT)))) {
9455 			info[idx].site   = 0;
9456 			info[idx].flags |= VM_KERN_SITE_NAMED;
9457 			if (namelen > sizeof(info[idx].name)) {
9458 				namelen = sizeof(info[idx].name);
9459 			}
9460 			strncpy(&info[idx].name[0], KA_NAME(site), namelen);
9461 		} else if (VM_TAG_KMOD & site->flags) {
9462 			info[idx].site   = OSKextGetKmodIDForSite(site, NULL, 0);
9463 			info[idx].flags |= VM_KERN_SITE_KMOD;
9464 		} else {
9465 			info[idx].site   = VM_KERNEL_UNSLIDE(site);
9466 			info[idx].flags |= VM_KERN_SITE_KERNEL;
9467 		}
9468 	}
9469 
9470 	nextinfo = (vm_allocation_tag_highest + 1);
9471 	count    = nextinfo;
9472 	if (count >= num_info) {
9473 		count = num_info;
9474 	}
9475 
9476 	for (idx = 0; idx < count; idx++) {
9477 		site = vm_allocation_sites[idx];
9478 		if (!site) {
9479 			continue;
9480 		}
9481 #if VM_TAG_SIZECLASSES
9482 		vm_allocation_zone_total_t * zone;
9483 		unsigned int                 zidx;
9484 
9485 		if (vm_allocation_zone_totals
9486 		    && (zone = vm_allocation_zone_totals[idx])
9487 		    && (nextinfo < num_info)) {
9488 			for (zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
9489 				if (!zone[zidx].vazt_peak) {
9490 					continue;
9491 				}
9492 				info[nextinfo]        = info[idx];
9493 				info[nextinfo].zone   = (uint16_t)zone_index_from_tag_index(zidx);
9494 				info[nextinfo].flags  &= ~VM_KERN_SITE_WIRED;
9495 				info[nextinfo].flags  |= VM_KERN_SITE_ZONE;
9496 				info[nextinfo].flags  |= VM_KERN_SITE_KALLOC;
9497 				info[nextinfo].size   = zone[zidx].vazt_total;
9498 				info[nextinfo].peak   = zone[zidx].vazt_peak;
9499 				info[nextinfo].mapped = 0;
9500 				nextinfo++;
9501 			}
9502 		}
9503 #endif /* VM_TAG_SIZECLASSES */
9504 		if (site->subtotalscount) {
9505 			uint64_t mapped, mapcost, take;
9506 			uint32_t sub;
9507 			vm_tag_t alloctag;
9508 
9509 			info[idx].size = site->total;
9510 			mapped = info[idx].size;
9511 			info[idx].mapped = mapped;
9512 			mapcost = 0;
9513 			for (sub = 0; sub < site->subtotalscount; sub++) {
9514 				alloctag = site->subtotals[sub].tag;
9515 				assert(alloctag < num_info);
9516 				if (info[alloctag].name[0]) {
9517 					continue;
9518 				}
9519 				take = site->subtotals[sub].total;
9520 				if (take > info[alloctag].size) {
9521 					take = info[alloctag].size;
9522 				}
9523 				if (take > mapped) {
9524 					take = mapped;
9525 				}
9526 				info[alloctag].mapped  -= take;
9527 				info[alloctag].size    -= take;
9528 				mapped                 -= take;
9529 				mapcost                += take;
9530 			}
9531 			info[idx].size = mapcost;
9532 		}
9533 	}
9534 	lck_ticket_unlock(&vm_allocation_sites_lock);
9535 
9536 	return 0;
9537 }
9538 
9539 uint32_t
vm_page_diagnose_estimate(void)9540 vm_page_diagnose_estimate(void)
9541 {
9542 	vm_allocation_site_t * site;
9543 	uint32_t               count = zone_view_count;
9544 	uint32_t               idx;
9545 
9546 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9547 	for (idx = 0; idx < VM_MAX_TAG_VALUE; idx++) {
9548 		site = vm_allocation_sites[idx];
9549 		if (!site) {
9550 			continue;
9551 		}
9552 		count++;
9553 #if VM_TAG_SIZECLASSES
9554 		if (vm_allocation_zone_totals) {
9555 			vm_allocation_zone_total_t * zone;
9556 			zone = vm_allocation_zone_totals[idx];
9557 			if (!zone) {
9558 				continue;
9559 			}
9560 			for (uint32_t zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
9561 				count += (zone[zidx].vazt_peak != 0);
9562 			}
9563 		}
9564 #endif
9565 	}
9566 	lck_ticket_unlock(&vm_allocation_sites_lock);
9567 
9568 	/* some slop for new tags created */
9569 	count += 8;
9570 	count += VM_KERN_COUNTER_COUNT;
9571 
9572 	return count;
9573 }
9574 
9575 static void
vm_page_diagnose_zone_stats(mach_memory_info_t * info,zone_stats_t zstats,bool percpu)9576 vm_page_diagnose_zone_stats(mach_memory_info_t *info, zone_stats_t zstats,
9577     bool percpu)
9578 {
9579 	zpercpu_foreach(zs, zstats) {
9580 		info->size += zs->zs_mem_allocated - zs->zs_mem_freed;
9581 	}
9582 	if (percpu) {
9583 		info->size *= zpercpu_count();
9584 	}
9585 	info->flags |= VM_KERN_SITE_NAMED | VM_KERN_SITE_ZONE_VIEW;
9586 }
9587 
9588 static void
vm_page_diagnose_zone(mach_memory_info_t * info,zone_t z)9589 vm_page_diagnose_zone(mach_memory_info_t *info, zone_t z)
9590 {
9591 	vm_page_diagnose_zone_stats(info, z->z_stats, z->z_percpu);
9592 	snprintf(info->name, sizeof(info->name),
9593 	    "%s%s[raw]", zone_heap_name(z), z->z_name);
9594 }
9595 
9596 static int
vm_page_diagnose_heap(mach_memory_info_t * info,kalloc_heap_t kheap)9597 vm_page_diagnose_heap(mach_memory_info_t *info, kalloc_heap_t kheap)
9598 {
9599 	struct kheap_zones *zones = kheap->kh_zones;
9600 	int i = 0;
9601 
9602 	for (; i < zones->max_k_zone; i++) {
9603 		vm_page_diagnose_zone(info + i, zones->k_zone[i]);
9604 	}
9605 
9606 	for (kalloc_heap_t kh = zones->views; kh; kh = kh->kh_next, i++) {
9607 		vm_page_diagnose_zone_stats(info + i, kh->kh_stats, false);
9608 		snprintf(info[i].name, sizeof(info[i].name),
9609 		    "%skalloc[%s]", kheap->kh_name, kh->kh_name);
9610 	}
9611 
9612 	return i;
9613 }
9614 
9615 static int
vm_page_diagnose_kt_heaps(mach_memory_info_t * info)9616 vm_page_diagnose_kt_heaps(mach_memory_info_t *info)
9617 {
9618 	uint32_t idx = 0;
9619 	vm_page_diagnose_zone_stats(info + idx, KHEAP_KT_VAR->kh_stats, false);
9620 	snprintf(info[idx].name, sizeof(info[idx].name),
9621 	    "%s[raw]", KHEAP_KT_VAR->kh_name);
9622 	idx++;
9623 
9624 	for (uint32_t i = 0; i < KT_VAR_MAX_HEAPS; i++) {
9625 		struct kt_heap_zones heap = kalloc_type_heap_array[i];
9626 
9627 		for (kalloc_type_var_view_t ktv = heap.views; ktv;
9628 		    ktv = (kalloc_type_var_view_t) ktv->kt_next) {
9629 			if (ktv->kt_stats && ktv->kt_stats != KHEAP_KT_VAR->kh_stats) {
9630 				vm_page_diagnose_zone_stats(info + idx, ktv->kt_stats, false);
9631 				snprintf(info[idx].name, sizeof(info[idx].name),
9632 				    "%s[%s]", KHEAP_KT_VAR->kh_name, ktv->kt_name);
9633 				idx++;
9634 			}
9635 		}
9636 	}
9637 
9638 	return idx;
9639 }
9640 
9641 kern_return_t
vm_page_diagnose(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes)9642 vm_page_diagnose(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes)
9643 {
9644 	uint64_t                 wired_size;
9645 	uint64_t                 wired_managed_size;
9646 	uint64_t                 wired_reserved_size;
9647 	boolean_t                iterate;
9648 	mach_memory_info_t     * counts;
9649 	uint32_t                 i;
9650 
9651 	bzero(info, num_info * sizeof(mach_memory_info_t));
9652 
9653 	if (!vm_page_wire_count_initial) {
9654 		return KERN_ABORTED;
9655 	}
9656 
9657 #if !XNU_TARGET_OS_OSX
9658 	wired_size          = ptoa_64(vm_page_wire_count);
9659 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count);
9660 #else /* !XNU_TARGET_OS_OSX */
9661 	wired_size          = ptoa_64(vm_page_wire_count + vm_lopage_free_count + vm_page_throttled_count);
9662 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count + vm_page_throttled_count);
9663 #endif /* !XNU_TARGET_OS_OSX */
9664 	wired_managed_size  = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial);
9665 
9666 	wired_size += booter_size;
9667 
9668 	assert(num_info >= VM_KERN_COUNTER_COUNT);
9669 	num_info -= VM_KERN_COUNTER_COUNT;
9670 	counts = &info[num_info];
9671 
9672 #define SET_COUNT(xcount, xsize, xflags)                        \
9673     counts[xcount].tag   = VM_MAX_TAG_VALUE + xcount;   \
9674     counts[xcount].site  = (xcount);                            \
9675     counts[xcount].size  = (xsize);                                 \
9676     counts[xcount].mapped  = (xsize);                           \
9677     counts[xcount].flags = VM_KERN_SITE_COUNTER | xflags;
9678 
9679 	SET_COUNT(VM_KERN_COUNT_MANAGED, ptoa_64(vm_page_pages), 0);
9680 	SET_COUNT(VM_KERN_COUNT_WIRED, wired_size, 0);
9681 	SET_COUNT(VM_KERN_COUNT_WIRED_MANAGED, wired_managed_size, 0);
9682 	SET_COUNT(VM_KERN_COUNT_RESERVED, wired_reserved_size, VM_KERN_SITE_WIRED);
9683 	SET_COUNT(VM_KERN_COUNT_STOLEN, ptoa_64(vm_page_stolen_count), VM_KERN_SITE_WIRED);
9684 	SET_COUNT(VM_KERN_COUNT_LOPAGE, ptoa_64(vm_lopage_free_count), VM_KERN_SITE_WIRED);
9685 	SET_COUNT(VM_KERN_COUNT_WIRED_BOOT, ptoa_64(vm_page_wire_count_on_boot), 0);
9686 	SET_COUNT(VM_KERN_COUNT_BOOT_STOLEN, booter_size, VM_KERN_SITE_WIRED);
9687 	SET_COUNT(VM_KERN_COUNT_WIRED_STATIC_KERNELCACHE, ptoa_64(vm_page_kernelcache_count), 0);
9688 
9689 #define SET_MAP(xcount, xsize, xfree, xlargest) \
9690     counts[xcount].site    = (xcount);                  \
9691     counts[xcount].size    = (xsize);                   \
9692     counts[xcount].mapped  = (xsize);                   \
9693     counts[xcount].free    = (xfree);                   \
9694     counts[xcount].largest = (xlargest);                \
9695     counts[xcount].flags   = VM_KERN_SITE_COUNTER;
9696 
9697 	vm_map_size_t map_size, map_free, map_largest;
9698 
9699 	vm_map_sizes(kernel_map, &map_size, &map_free, &map_largest);
9700 	SET_MAP(VM_KERN_COUNT_MAP_KERNEL, map_size, map_free, map_largest);
9701 
9702 	zone_map_sizes(&map_size, &map_free, &map_largest);
9703 	SET_MAP(VM_KERN_COUNT_MAP_ZONE, map_size, map_free, map_largest);
9704 
9705 	assert(num_info >= zone_view_count);
9706 	num_info -= zone_view_count;
9707 	counts = &info[num_info];
9708 	i = 0;
9709 
9710 	i += vm_page_diagnose_heap(counts + i, KHEAP_DEFAULT);
9711 	if (KHEAP_DATA_BUFFERS->kh_heap_id == KHEAP_ID_DATA_BUFFERS) {
9712 		i += vm_page_diagnose_heap(counts + i, KHEAP_DATA_BUFFERS);
9713 	}
9714 	if (KHEAP_KT_VAR->kh_heap_id == KHEAP_ID_KT_VAR) {
9715 		i += vm_page_diagnose_kt_heaps(counts + i);
9716 	}
9717 	assert(i <= zone_view_count);
9718 
9719 	zone_index_foreach(zidx) {
9720 		zone_t z = &zone_array[zidx];
9721 		zone_security_flags_t zsflags = zone_security_array[zidx];
9722 		zone_view_t zv = z->z_views;
9723 
9724 		if (zv == NULL) {
9725 			continue;
9726 		}
9727 
9728 		zone_stats_t zv_stats_head = z->z_stats;
9729 		bool has_raw_view = false;
9730 
9731 		for (; zv; zv = zv->zv_next) {
9732 			/*
9733 			 * kalloc_types that allocate from the same zone are linked
9734 			 * as views. Only print the ones that have their own stats.
9735 			 */
9736 			if (zv->zv_stats == zv_stats_head) {
9737 				continue;
9738 			}
9739 			has_raw_view = true;
9740 			vm_page_diagnose_zone_stats(counts + i, zv->zv_stats,
9741 			    z->z_percpu);
9742 			snprintf(counts[i].name, sizeof(counts[i].name), "%s%s[%s]",
9743 			    zone_heap_name(z), z->z_name, zv->zv_name);
9744 			i++;
9745 			assert(i <= zone_view_count);
9746 		}
9747 
9748 		/*
9749 		 * Print raw views for non kalloc or kalloc_type zones
9750 		 */
9751 		bool kalloc_type = zsflags.z_kalloc_type;
9752 		if ((zsflags.z_kheap_id == KHEAP_ID_NONE && !kalloc_type) ||
9753 		    (kalloc_type && has_raw_view)) {
9754 			vm_page_diagnose_zone(counts + i, z);
9755 			i++;
9756 			assert(i <= zone_view_count);
9757 		}
9758 	}
9759 
9760 	iterate = !VM_TAG_ACTIVE_UPDATE;
9761 	if (iterate) {
9762 		enum                       { kMaxKernelDepth = 1 };
9763 		vm_map_t                     maps[kMaxKernelDepth];
9764 		vm_map_entry_t               entries[kMaxKernelDepth];
9765 		vm_map_t                     map;
9766 		vm_map_entry_t               entry;
9767 		vm_object_offset_t           offset;
9768 		vm_page_t                    page;
9769 		int                          stackIdx, count;
9770 
9771 #if !VM_TAG_ACTIVE_UPDATE
9772 		vm_page_iterate_objects(info, num_info, &vm_page_count_object);
9773 #endif /* ! VM_TAG_ACTIVE_UPDATE */
9774 
9775 		map = kernel_map;
9776 		stackIdx = 0;
9777 		while (map) {
9778 			vm_map_lock(map);
9779 			for (entry = map->hdr.links.next; map; entry = entry->links.next) {
9780 				if (entry->is_sub_map) {
9781 					assert(stackIdx < kMaxKernelDepth);
9782 					maps[stackIdx] = map;
9783 					entries[stackIdx] = entry;
9784 					stackIdx++;
9785 					map = VME_SUBMAP(entry);
9786 					entry = NULL;
9787 					break;
9788 				}
9789 				if (VME_OBJECT(entry) == kernel_object) {
9790 					count = 0;
9791 					vm_object_lock(VME_OBJECT(entry));
9792 					for (offset = entry->links.start; offset < entry->links.end; offset += page_size) {
9793 						page = vm_page_lookup(VME_OBJECT(entry), offset);
9794 						if (page && VM_PAGE_WIRED(page)) {
9795 							count++;
9796 						}
9797 					}
9798 					vm_object_unlock(VME_OBJECT(entry));
9799 
9800 					if (count) {
9801 						assert(VME_ALIAS(entry) != VM_KERN_MEMORY_NONE);
9802 						assert(VME_ALIAS(entry) < num_info);
9803 						info[VME_ALIAS(entry)].size += ptoa_64(count);
9804 					}
9805 				}
9806 				while (map && (entry == vm_map_last_entry(map))) {
9807 					vm_map_unlock(map);
9808 					if (!stackIdx) {
9809 						map = NULL;
9810 					} else {
9811 						--stackIdx;
9812 						map = maps[stackIdx];
9813 						entry = entries[stackIdx];
9814 					}
9815 				}
9816 			}
9817 		}
9818 	}
9819 
9820 	process_account(info, num_info, zones_collectable_bytes, iterate);
9821 
9822 	return KERN_SUCCESS;
9823 }
9824 
9825 #if DEBUG || DEVELOPMENT
9826 
9827 kern_return_t
vm_kern_allocation_info(uintptr_t addr,vm_size_t * size,vm_tag_t * tag,vm_size_t * zone_size)9828 vm_kern_allocation_info(uintptr_t addr, vm_size_t * size, vm_tag_t * tag, vm_size_t * zone_size)
9829 {
9830 	kern_return_t  ret;
9831 	vm_size_t      zsize;
9832 	vm_map_t       map;
9833 	vm_map_entry_t entry;
9834 
9835 	zsize = zone_element_info((void *) addr, tag);
9836 	if (zsize) {
9837 		*zone_size = *size = zsize;
9838 		return KERN_SUCCESS;
9839 	}
9840 
9841 	*zone_size = 0;
9842 	ret = KERN_INVALID_ADDRESS;
9843 	for (map = kernel_map; map;) {
9844 		vm_map_lock(map);
9845 		if (!vm_map_lookup_entry_allow_pgz(map, addr, &entry)) {
9846 			break;
9847 		}
9848 		if (entry->is_sub_map) {
9849 			if (map != kernel_map) {
9850 				break;
9851 			}
9852 			map = VME_SUBMAP(entry);
9853 			continue;
9854 		}
9855 		if (entry->vme_start != addr) {
9856 			break;
9857 		}
9858 		*tag = (vm_tag_t)VME_ALIAS(entry);
9859 		*size = (entry->vme_end - addr);
9860 		ret = KERN_SUCCESS;
9861 		break;
9862 	}
9863 	if (map != kernel_map) {
9864 		vm_map_unlock(map);
9865 	}
9866 	vm_map_unlock(kernel_map);
9867 
9868 	return ret;
9869 }
9870 
9871 #endif /* DEBUG || DEVELOPMENT */
9872 
9873 uint32_t
vm_tag_get_kext(vm_tag_t tag,char * name,vm_size_t namelen)9874 vm_tag_get_kext(vm_tag_t tag, char * name, vm_size_t namelen)
9875 {
9876 	vm_allocation_site_t * site;
9877 	uint32_t               kmodId;
9878 
9879 	kmodId = 0;
9880 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9881 	if ((site = vm_allocation_sites[tag])) {
9882 		if (VM_TAG_KMOD & site->flags) {
9883 			kmodId = OSKextGetKmodIDForSite(site, name, namelen);
9884 		}
9885 	}
9886 	lck_ticket_unlock(&vm_allocation_sites_lock);
9887 
9888 	return kmodId;
9889 }
9890 
9891 
9892 #if CONFIG_SECLUDED_MEMORY
9893 /*
9894  * Note that there's no locking around other accesses to vm_page_secluded_target.
9895  * That should be OK, since these are the only place where it can be changed after
9896  * initialization. Other users (like vm_pageout) may see the wrong value briefly,
9897  * but will eventually get the correct value. This brief mismatch is OK as pageout
9898  * and page freeing will auto-adjust the vm_page_secluded_count to match the target
9899  * over time.
9900  */
9901 unsigned int vm_page_secluded_suppress_cnt = 0;
9902 unsigned int vm_page_secluded_save_target;
9903 
9904 LCK_GRP_DECLARE(secluded_suppress_slock_grp, "secluded_suppress_slock");
9905 LCK_SPIN_DECLARE(secluded_suppress_slock, &secluded_suppress_slock_grp);
9906 
9907 void
start_secluded_suppression(task_t task)9908 start_secluded_suppression(task_t task)
9909 {
9910 	if (task->task_suppressed_secluded) {
9911 		return;
9912 	}
9913 	lck_spin_lock(&secluded_suppress_slock);
9914 	if (!task->task_suppressed_secluded && vm_page_secluded_suppress_cnt++ == 0) {
9915 		task->task_suppressed_secluded = TRUE;
9916 		vm_page_secluded_save_target = vm_page_secluded_target;
9917 		vm_page_secluded_target = 0;
9918 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
9919 	}
9920 	lck_spin_unlock(&secluded_suppress_slock);
9921 }
9922 
9923 void
stop_secluded_suppression(task_t task)9924 stop_secluded_suppression(task_t task)
9925 {
9926 	lck_spin_lock(&secluded_suppress_slock);
9927 	if (task->task_suppressed_secluded && --vm_page_secluded_suppress_cnt == 0) {
9928 		task->task_suppressed_secluded = FALSE;
9929 		vm_page_secluded_target = vm_page_secluded_save_target;
9930 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
9931 	}
9932 	lck_spin_unlock(&secluded_suppress_slock);
9933 }
9934 
9935 #endif /* CONFIG_SECLUDED_MEMORY */
9936 
9937 /*
9938  * Move the list of retired pages on the vm_page_queue_retired to
9939  * their final resting place on retired_pages_object.
9940  */
9941 void
vm_retire_boot_pages(void)9942 vm_retire_boot_pages(void)
9943 {
9944 }
9945 
9946 /*
9947  * This holds the reported physical address if an ECC error leads to a panic.
9948  * SMC will store it in PMU SRAM under the 'sECC' key.
9949  */
9950 uint64_t ecc_panic_physical_address = 0;
9951 
9952