xref: /xnu-8792.81.2/osfmk/vm/vm_resident.c (revision 19c3b8c28c31cb8130e034cfb5df6bf9ba342d90)
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 unsigned int    vm_page_realtime_count;
339 
340 /*
341  *	Occasionally, the virtual memory system uses
342  *	resident page structures that do not refer to
343  *	real pages, for example to leave a page with
344  *	important state information in the VP table.
345  *
346  *	These page structures are allocated the way
347  *	most other kernel structures are.
348  */
349 SECURITY_READ_ONLY_LATE(zone_t) vm_page_zone;
350 vm_locks_array_t vm_page_locks;
351 
352 LCK_ATTR_DECLARE(vm_page_lck_attr, 0, 0);
353 LCK_GRP_DECLARE(vm_page_lck_grp_free, "vm_page_free");
354 LCK_GRP_DECLARE(vm_page_lck_grp_queue, "vm_page_queue");
355 LCK_GRP_DECLARE(vm_page_lck_grp_local, "vm_page_queue_local");
356 LCK_GRP_DECLARE(vm_page_lck_grp_purge, "vm_page_purge");
357 LCK_GRP_DECLARE(vm_page_lck_grp_alloc, "vm_page_alloc");
358 LCK_GRP_DECLARE(vm_page_lck_grp_bucket, "vm_page_bucket");
359 LCK_SPIN_DECLARE_ATTR(vm_objects_wired_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr);
360 LCK_TICKET_DECLARE(vm_allocation_sites_lock, &vm_page_lck_grp_bucket);
361 
362 unsigned int    vm_page_local_q_soft_limit = 250;
363 unsigned int    vm_page_local_q_hard_limit = 500;
364 struct vpl     *__zpercpu vm_page_local_q;
365 
366 /* N.B. Guard and fictitious pages must not
367  * be assigned a zero phys_page value.
368  */
369 /*
370  *	Fictitious pages don't have a physical address,
371  *	but we must initialize phys_page to something.
372  *	For debugging, this should be a strange value
373  *	that the pmap module can recognize in assertions.
374  */
375 const ppnum_t vm_page_fictitious_addr = (ppnum_t) -1;
376 
377 /*
378  *	Guard pages are not accessible so they don't
379  *      need a physical address, but we need to enter
380  *	one in the pmap.
381  *	Let's make it recognizable and make sure that
382  *	we don't use a real physical page with that
383  *	physical address.
384  */
385 const ppnum_t vm_page_guard_addr = (ppnum_t) -2;
386 
387 /*
388  *	Resident page structures are also chained on
389  *	queues that are used by the page replacement
390  *	system (pageout daemon).  These queues are
391  *	defined here, but are shared by the pageout
392  *	module.  The inactive queue is broken into
393  *	file backed and anonymous for convenience as the
394  *	pageout daemon often assignes a higher
395  *	importance to anonymous pages (less likely to pick)
396  */
397 vm_page_queue_head_t    vm_page_queue_active VM_PAGE_PACKED_ALIGNED;
398 vm_page_queue_head_t    vm_page_queue_inactive VM_PAGE_PACKED_ALIGNED;
399 #if CONFIG_SECLUDED_MEMORY
400 vm_page_queue_head_t    vm_page_queue_secluded VM_PAGE_PACKED_ALIGNED;
401 #endif /* CONFIG_SECLUDED_MEMORY */
402 vm_page_queue_head_t    vm_page_queue_anonymous VM_PAGE_PACKED_ALIGNED;  /* inactive memory queue for anonymous pages */
403 vm_page_queue_head_t    vm_page_queue_throttled VM_PAGE_PACKED_ALIGNED;
404 
405 queue_head_t    vm_objects_wired;
406 
407 void vm_update_darkwake_mode(boolean_t);
408 
409 vm_page_queue_head_t    vm_page_queue_donate VM_PAGE_PACKED_ALIGNED;
410 uint32_t        vm_page_donate_mode;
411 uint32_t        vm_page_donate_target, vm_page_donate_target_high, vm_page_donate_target_low;
412 uint32_t        vm_page_donate_count;
413 bool            vm_page_donate_queue_ripe;
414 
415 
416 vm_page_queue_head_t    vm_page_queue_background VM_PAGE_PACKED_ALIGNED;
417 uint32_t        vm_page_background_target;
418 uint32_t        vm_page_background_target_snapshot;
419 uint32_t        vm_page_background_count;
420 uint64_t        vm_page_background_promoted_count;
421 
422 uint32_t        vm_page_background_internal_count;
423 uint32_t        vm_page_background_external_count;
424 
425 uint32_t        vm_page_background_mode;
426 uint32_t        vm_page_background_exclude_external;
427 
428 unsigned int    vm_page_active_count;
429 unsigned int    vm_page_inactive_count;
430 unsigned int    vm_page_kernelcache_count;
431 #if CONFIG_SECLUDED_MEMORY
432 unsigned int    vm_page_secluded_count;
433 unsigned int    vm_page_secluded_count_free;
434 unsigned int    vm_page_secluded_count_inuse;
435 unsigned int    vm_page_secluded_count_over_target;
436 #endif /* CONFIG_SECLUDED_MEMORY */
437 unsigned int    vm_page_anonymous_count;
438 unsigned int    vm_page_throttled_count;
439 unsigned int    vm_page_speculative_count;
440 
441 unsigned int    vm_page_wire_count;
442 unsigned int    vm_page_wire_count_on_boot = 0;
443 unsigned int    vm_page_stolen_count = 0;
444 unsigned int    vm_page_wire_count_initial;
445 unsigned int    vm_page_gobble_count = 0;
446 unsigned int    vm_page_kern_lpage_count = 0;
447 
448 uint64_t        booter_size;  /* external so it can be found in core dumps */
449 
450 #define VM_PAGE_WIRE_COUNT_WARNING      0
451 #define VM_PAGE_GOBBLE_COUNT_WARNING    0
452 
453 unsigned int    vm_page_purgeable_count = 0; /* # of pages purgeable now */
454 unsigned int    vm_page_purgeable_wired_count = 0; /* # of purgeable pages that are wired now */
455 uint64_t        vm_page_purged_count = 0;    /* total count of purged pages */
456 
457 unsigned int    vm_page_xpmapped_external_count = 0;
458 unsigned int    vm_page_external_count = 0;
459 unsigned int    vm_page_internal_count = 0;
460 unsigned int    vm_page_pageable_external_count = 0;
461 unsigned int    vm_page_pageable_internal_count = 0;
462 
463 #if DEVELOPMENT || DEBUG
464 unsigned int    vm_page_speculative_recreated = 0;
465 unsigned int    vm_page_speculative_created = 0;
466 unsigned int    vm_page_speculative_used = 0;
467 #endif
468 
469 vm_page_queue_head_t    vm_page_queue_cleaned VM_PAGE_PACKED_ALIGNED;
470 
471 unsigned int    vm_page_cleaned_count = 0;
472 
473 uint64_t        max_valid_dma_address = 0xffffffffffffffffULL;
474 ppnum_t         max_valid_low_ppnum = PPNUM_MAX;
475 
476 
477 /*
478  *	Several page replacement parameters are also
479  *	shared with this module, so that page allocation
480  *	(done here in vm_page_alloc) can trigger the
481  *	pageout daemon.
482  */
483 unsigned int    vm_page_free_target = 0;
484 unsigned int    vm_page_free_min = 0;
485 unsigned int    vm_page_throttle_limit = 0;
486 unsigned int    vm_page_inactive_target = 0;
487 #if CONFIG_SECLUDED_MEMORY
488 unsigned int    vm_page_secluded_target = 0;
489 #endif /* CONFIG_SECLUDED_MEMORY */
490 unsigned int    vm_page_anonymous_min = 0;
491 unsigned int    vm_page_free_reserved = 0;
492 
493 
494 /*
495  *	The VM system has a couple of heuristics for deciding
496  *	that pages are "uninteresting" and should be placed
497  *	on the inactive queue as likely candidates for replacement.
498  *	These variables let the heuristics be controlled at run-time
499  *	to make experimentation easier.
500  */
501 
502 boolean_t vm_page_deactivate_hint = TRUE;
503 
504 struct vm_page_stats_reusable vm_page_stats_reusable;
505 
506 /*
507  *	vm_set_page_size:
508  *
509  *	Sets the page size, perhaps based upon the memory
510  *	size.  Must be called before any use of page-size
511  *	dependent functions.
512  *
513  *	Sets page_shift and page_mask from page_size.
514  */
515 void
vm_set_page_size(void)516 vm_set_page_size(void)
517 {
518 	page_size  = PAGE_SIZE;
519 	page_mask  = PAGE_MASK;
520 	page_shift = PAGE_SHIFT;
521 
522 	if ((page_mask & page_size) != 0) {
523 		panic("vm_set_page_size: page size not a power of two");
524 	}
525 
526 	for (page_shift = 0;; page_shift++) {
527 		if ((1U << page_shift) == page_size) {
528 			break;
529 		}
530 	}
531 }
532 
533 #if defined (__x86_64__)
534 
535 #define MAX_CLUMP_SIZE      16
536 #define DEFAULT_CLUMP_SIZE  4
537 
538 unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold;
539 
540 #if DEVELOPMENT || DEBUG
541 unsigned long vm_clump_stats[MAX_CLUMP_SIZE + 1];
542 unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes;
543 
544 static inline void
vm_clump_update_stats(unsigned int c)545 vm_clump_update_stats(unsigned int c)
546 {
547 	assert(c <= vm_clump_size);
548 	if (c > 0 && c <= vm_clump_size) {
549 		vm_clump_stats[c] += c;
550 	}
551 	vm_clump_allocs += c;
552 }
553 #endif  /*  if DEVELOPMENT || DEBUG */
554 
555 /* Called once to setup the VM clump knobs */
556 static void
vm_page_setup_clump(void)557 vm_page_setup_clump( void )
558 {
559 	unsigned int override, n;
560 
561 	vm_clump_size = DEFAULT_CLUMP_SIZE;
562 	if (PE_parse_boot_argn("clump_size", &override, sizeof(override))) {
563 		vm_clump_size = override;
564 	}
565 
566 	if (vm_clump_size > MAX_CLUMP_SIZE) {
567 		panic("vm_page_setup_clump:: clump_size is too large!");
568 	}
569 	if (vm_clump_size < 1) {
570 		panic("vm_page_setup_clump:: clump_size must be >= 1");
571 	}
572 	if ((vm_clump_size & (vm_clump_size - 1)) != 0) {
573 		panic("vm_page_setup_clump:: clump_size must be a power of 2");
574 	}
575 
576 	vm_clump_promote_threshold = vm_clump_size;
577 	vm_clump_mask = vm_clump_size - 1;
578 	for (vm_clump_shift = 0, n = vm_clump_size; n > 1; n >>= 1, vm_clump_shift++) {
579 		;
580 	}
581 
582 #if DEVELOPMENT || DEBUG
583 	bzero(vm_clump_stats, sizeof(vm_clump_stats));
584 	vm_clump_allocs = vm_clump_inserts = vm_clump_inrange = vm_clump_promotes = 0;
585 #endif  /*  if DEVELOPMENT || DEBUG */
586 }
587 
588 #endif  /* #if defined (__x86_64__) */
589 
590 #define COLOR_GROUPS_TO_STEAL   4
591 
592 /* Called once during statup, once the cache geometry is known.
593  */
594 static void
vm_page_set_colors(void)595 vm_page_set_colors( void )
596 {
597 	unsigned int    n, override;
598 
599 #if defined (__x86_64__)
600 	/* adjust #colors because we need to color outside the clump boundary */
601 	vm_cache_geometry_colors >>= vm_clump_shift;
602 #endif
603 	if (PE_parse_boot_argn("colors", &override, sizeof(override))) {                /* colors specified as a boot-arg? */
604 		n = override;
605 	} else if (vm_cache_geometry_colors) {                  /* do we know what the cache geometry is? */
606 		n = vm_cache_geometry_colors;
607 	} else {
608 		n = DEFAULT_COLORS;                             /* use default if all else fails */
609 	}
610 	if (n == 0) {
611 		n = 1;
612 	}
613 	if (n > MAX_COLORS) {
614 		n = MAX_COLORS;
615 	}
616 
617 	/* the count must be a power of 2  */
618 	if ((n & (n - 1)) != 0) {
619 		n = DEFAULT_COLORS;                             /* use default if all else fails */
620 	}
621 	vm_colors = n;
622 	vm_color_mask = n - 1;
623 
624 	vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL;
625 
626 #if defined (__x86_64__)
627 	/* adjust for reduction in colors due to clumping and multiple cores */
628 	if (real_ncpus) {
629 		vm_free_magazine_refill_limit *= (vm_clump_size * real_ncpus);
630 	}
631 #endif
632 }
633 
634 /*
635  * During single threaded early boot we don't initialize all pages.
636  * This avoids some delay during boot. They'll be initialized and
637  * added to the free list as needed or after we are multithreaded by
638  * what becomes the pageout thread.
639  */
640 static boolean_t fill = FALSE;
641 static unsigned int fillval;
642 uint_t vm_delayed_count = 0;    /* when non-zero, indicates we may have more pages to init */
643 ppnum_t delay_above_pnum = PPNUM_MAX;
644 
645 /*
646  * For x86 first 8 Gig initializes quickly and gives us lots of lowmem + mem above to start off with.
647  * If ARM ever uses delayed page initialization, this value may need to be quite different.
648  */
649 #define DEFAULT_DELAY_ABOVE_PHYS_GB (8)
650 
651 /*
652  * When we have to dip into more delayed pages due to low memory, free up
653  * a large chunk to get things back to normal. This avoids contention on the
654  * delayed code allocating page by page.
655  */
656 #define VM_DELAY_PAGE_CHUNK ((1024 * 1024 * 1024) / PAGE_SIZE)
657 
658 /*
659  * Get and initialize the next delayed page.
660  */
661 static vm_page_t
vm_get_delayed_page(int grab_options)662 vm_get_delayed_page(int grab_options)
663 {
664 	vm_page_t p;
665 	ppnum_t   pnum;
666 
667 	/*
668 	 * Get a new page if we have one.
669 	 */
670 	vm_free_page_lock();
671 	if (vm_delayed_count == 0) {
672 		vm_free_page_unlock();
673 		return NULL;
674 	}
675 
676 	if (!pmap_next_page(&pnum)) {
677 		vm_delayed_count = 0;
678 		vm_free_page_unlock();
679 		return NULL;
680 	}
681 
682 
683 	assert(vm_delayed_count > 0);
684 	--vm_delayed_count;
685 
686 #if defined(__x86_64__)
687 	/* x86 cluster code requires increasing phys_page in vm_pages[] */
688 	if (vm_pages_count > 0) {
689 		assert(pnum > vm_pages[vm_pages_count - 1].vmp_phys_page);
690 	}
691 #endif
692 	p = &vm_pages[vm_pages_count];
693 	assert(p < vm_page_array_ending_addr);
694 	vm_page_init(p, pnum, FALSE);
695 	++vm_pages_count;
696 	++vm_page_pages;
697 	vm_free_page_unlock();
698 
699 	/*
700 	 * These pages were initially counted as wired, undo that now.
701 	 */
702 	if (grab_options & VM_PAGE_GRAB_Q_LOCK_HELD) {
703 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
704 	} else {
705 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
706 		vm_page_lockspin_queues();
707 	}
708 	--vm_page_wire_count;
709 	--vm_page_wire_count_initial;
710 	if (vm_page_wire_count_on_boot != 0) {
711 		--vm_page_wire_count_on_boot;
712 	}
713 	if (!(grab_options & VM_PAGE_GRAB_Q_LOCK_HELD)) {
714 		vm_page_unlock_queues();
715 	}
716 
717 
718 	if (fill) {
719 		fillPage(pnum, fillval);
720 	}
721 	return p;
722 }
723 
724 static void vm_page_module_init_delayed(void);
725 
726 /*
727  * Free all remaining delayed pages to the free lists.
728  */
729 void
vm_free_delayed_pages(void)730 vm_free_delayed_pages(void)
731 {
732 	vm_page_t   p;
733 	vm_page_t   list = NULL;
734 	uint_t      cnt = 0;
735 	vm_offset_t start_free_va;
736 	int64_t     free_size;
737 
738 	while ((p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE)) != NULL) {
739 		if (vm_himemory_mode) {
740 			vm_page_release(p, FALSE);
741 		} else {
742 			p->vmp_snext = list;
743 			list = p;
744 		}
745 		++cnt;
746 	}
747 
748 	/*
749 	 * Free the pages in reverse order if not himemory mode.
750 	 * Hence the low memory pages will be first on free lists. (LIFO)
751 	 */
752 	while (list != NULL) {
753 		p = list;
754 		list = p->vmp_snext;
755 		p->vmp_snext = NULL;
756 		vm_page_release(p, FALSE);
757 	}
758 #if DEVELOPMENT || DEBUG
759 	kprintf("vm_free_delayed_pages: initialized %d free pages\n", cnt);
760 #endif
761 
762 	/*
763 	 * Free up any unused full pages at the end of the vm_pages[] array
764 	 */
765 	start_free_va = round_page((vm_offset_t)&vm_pages[vm_pages_count]);
766 
767 #if defined(__x86_64__)
768 	/*
769 	 * Since x86 might have used large pages for vm_pages[], we can't
770 	 * free starting in the middle of a partially used large page.
771 	 */
772 	if (pmap_query_pagesize(kernel_pmap, start_free_va) == I386_LPGBYTES) {
773 		start_free_va = ((start_free_va + I386_LPGMASK) & ~I386_LPGMASK);
774 	}
775 #endif
776 	if (start_free_va < (vm_offset_t)vm_page_array_ending_addr) {
777 		free_size = trunc_page((vm_offset_t)vm_page_array_ending_addr - start_free_va);
778 		if (free_size > 0) {
779 			ml_static_mfree(start_free_va, (vm_offset_t)free_size);
780 			vm_page_array_ending_addr = (void *)start_free_va;
781 
782 			/*
783 			 * Note there's no locking here, as only this thread will ever change this value.
784 			 * The reader, vm_page_diagnose, doesn't grab any locks for the counts it looks at.
785 			 */
786 			vm_page_stolen_count -= (free_size >> PAGE_SHIFT);
787 
788 #if DEVELOPMENT || DEBUG
789 			kprintf("Freeing final unused %ld bytes from vm_pages[] at 0x%lx\n",
790 			    (long)free_size, (long)start_free_va);
791 #endif
792 		}
793 	}
794 
795 
796 	/*
797 	 * now we can create the VM page array zone
798 	 */
799 	vm_page_module_init_delayed();
800 }
801 
802 /*
803  * Try and free up enough delayed pages to match a contig memory allocation.
804  */
805 static void
vm_free_delayed_pages_contig(uint_t npages,ppnum_t max_pnum,ppnum_t pnum_mask)806 vm_free_delayed_pages_contig(
807 	uint_t    npages,
808 	ppnum_t   max_pnum,
809 	ppnum_t   pnum_mask)
810 {
811 	vm_page_t p;
812 	ppnum_t   pnum;
813 	uint_t    cnt = 0;
814 
815 	/*
816 	 * Treat 0 as the absolute max page number.
817 	 */
818 	if (max_pnum == 0) {
819 		max_pnum = PPNUM_MAX;
820 	}
821 
822 	/*
823 	 * Free till we get a properly aligned start page
824 	 */
825 	for (;;) {
826 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
827 		if (p == NULL) {
828 			return;
829 		}
830 		pnum = VM_PAGE_GET_PHYS_PAGE(p);
831 		vm_page_release(p, FALSE);
832 		if (pnum >= max_pnum) {
833 			return;
834 		}
835 		if ((pnum & pnum_mask) == 0) {
836 			break;
837 		}
838 	}
839 
840 	/*
841 	 * Having a healthy pool of free pages will help performance. We don't
842 	 * want to fall back to the delayed code for every page allocation.
843 	 */
844 	if (vm_page_free_count < VM_DELAY_PAGE_CHUNK) {
845 		npages += VM_DELAY_PAGE_CHUNK;
846 	}
847 
848 	/*
849 	 * Now free up the pages
850 	 */
851 	for (cnt = 1; cnt < npages; ++cnt) {
852 		p = vm_get_delayed_page(VM_PAGE_GRAB_OPTIONS_NONE);
853 		if (p == NULL) {
854 			return;
855 		}
856 		vm_page_release(p, FALSE);
857 	}
858 }
859 
860 #define ROUNDUP_NEXTP2(X) (1U << (32 - __builtin_clz((X) - 1)))
861 
862 void
vm_page_init_local_q(unsigned int num_cpus)863 vm_page_init_local_q(unsigned int num_cpus)
864 {
865 	struct vpl *t_local_q;
866 
867 	/*
868 	 * no point in this for a uni-processor system
869 	 */
870 	if (num_cpus >= 2) {
871 		ml_cpu_info_t cpu_info;
872 
873 		/*
874 		 * Force the allocation alignment to a cacheline,
875 		 * because the `vpl` struct has a lock and will be taken
876 		 * cross CPU so we want to isolate the rest of the per-CPU
877 		 * data to avoid false sharing due to this lock being taken.
878 		 */
879 
880 		ml_cpu_get_info(&cpu_info);
881 
882 		t_local_q = zalloc_percpu_permanent(sizeof(struct vpl),
883 		    cpu_info.cache_line_size - 1);
884 
885 		zpercpu_foreach(lq, t_local_q) {
886 			VPL_LOCK_INIT(lq, &vm_page_lck_grp_local, &vm_page_lck_attr);
887 			vm_page_queue_init(&lq->vpl_queue);
888 		}
889 
890 		/* make the initialization visible to all cores */
891 		os_atomic_store(&vm_page_local_q, t_local_q, release);
892 	}
893 }
894 
895 /*
896  * vm_init_before_launchd
897  *
898  * This should be called right before launchd is loaded.
899  */
900 void
vm_init_before_launchd()901 vm_init_before_launchd()
902 {
903 	vm_page_lockspin_queues();
904 	vm_page_wire_count_on_boot = vm_page_wire_count;
905 	vm_page_unlock_queues();
906 }
907 
908 
909 /*
910  *	vm_page_bootstrap:
911  *
912  *	Initializes the resident memory module.
913  *
914  *	Allocates memory for the page cells, and
915  *	for the object/offset-to-page hash table headers.
916  *	Each page cell is initialized and placed on the free list.
917  *	Returns the range of available kernel virtual memory.
918  */
919 __startup_func
920 void
vm_page_bootstrap(vm_offset_t * startp,vm_offset_t * endp)921 vm_page_bootstrap(
922 	vm_offset_t             *startp,
923 	vm_offset_t             *endp)
924 {
925 	unsigned int            i;
926 	unsigned int            log1;
927 	unsigned int            log2;
928 	unsigned int            size;
929 
930 	/*
931 	 *	Initialize the page queues.
932 	 */
933 
934 	lck_mtx_init(&vm_page_queue_free_lock, &vm_page_lck_grp_free, &vm_page_lck_attr);
935 	lck_mtx_init(&vm_page_queue_lock, &vm_page_lck_grp_queue, &vm_page_lck_attr);
936 	lck_mtx_init(&vm_purgeable_queue_lock, &vm_page_lck_grp_purge, &vm_page_lck_attr);
937 
938 	for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) {
939 		int group;
940 
941 		purgeable_queues[i].token_q_head = 0;
942 		purgeable_queues[i].token_q_tail = 0;
943 		for (group = 0; group < NUM_VOLATILE_GROUPS; group++) {
944 			queue_init(&purgeable_queues[i].objq[group]);
945 		}
946 
947 		purgeable_queues[i].type = i;
948 		purgeable_queues[i].new_pages = 0;
949 #if MACH_ASSERT
950 		purgeable_queues[i].debug_count_tokens = 0;
951 		purgeable_queues[i].debug_count_objects = 0;
952 #endif
953 	}
954 	;
955 	purgeable_nonvolatile_count = 0;
956 	queue_init(&purgeable_nonvolatile_queue);
957 
958 	for (i = 0; i < MAX_COLORS; i++) {
959 		vm_page_queue_init(&vm_page_queue_free[i].qhead);
960 	}
961 
962 	vm_page_queue_init(&vm_lopage_queue_free);
963 	vm_page_queue_init(&vm_page_queue_active);
964 	vm_page_queue_init(&vm_page_queue_inactive);
965 #if CONFIG_SECLUDED_MEMORY
966 	vm_page_queue_init(&vm_page_queue_secluded);
967 #endif /* CONFIG_SECLUDED_MEMORY */
968 	vm_page_queue_init(&vm_page_queue_cleaned);
969 	vm_page_queue_init(&vm_page_queue_throttled);
970 	vm_page_queue_init(&vm_page_queue_anonymous);
971 	queue_init(&vm_objects_wired);
972 
973 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
974 		vm_page_queue_init(&vm_page_queue_speculative[i].age_q);
975 
976 		vm_page_queue_speculative[i].age_ts.tv_sec = 0;
977 		vm_page_queue_speculative[i].age_ts.tv_nsec = 0;
978 	}
979 
980 	vm_page_queue_init(&vm_page_queue_donate);
981 	vm_page_queue_init(&vm_page_queue_background);
982 
983 	vm_page_background_count = 0;
984 	vm_page_background_internal_count = 0;
985 	vm_page_background_external_count = 0;
986 	vm_page_background_promoted_count = 0;
987 
988 	vm_page_background_target = (unsigned int)(atop_64(max_mem) / 25);
989 
990 	if (vm_page_background_target > VM_PAGE_BACKGROUND_TARGET_MAX) {
991 		vm_page_background_target = VM_PAGE_BACKGROUND_TARGET_MAX;
992 	}
993 
994 #if    defined(__LP64__)
995 	vm_page_background_mode = VM_PAGE_BG_ENABLED;
996 	vm_page_donate_mode = VM_PAGE_DONATE_ENABLED;
997 #else
998 	vm_page_background_mode = VM_PAGE_BG_DISABLED;
999 	vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
1000 #endif
1001 	vm_page_background_exclude_external = 0;
1002 
1003 	PE_parse_boot_argn("vm_page_bg_mode", &vm_page_background_mode, sizeof(vm_page_background_mode));
1004 	PE_parse_boot_argn("vm_page_bg_exclude_external", &vm_page_background_exclude_external, sizeof(vm_page_background_exclude_external));
1005 	PE_parse_boot_argn("vm_page_bg_target", &vm_page_background_target, sizeof(vm_page_background_target));
1006 
1007 	if (vm_page_background_mode != VM_PAGE_BG_DISABLED && vm_page_background_mode != VM_PAGE_BG_ENABLED) {
1008 		vm_page_background_mode = VM_PAGE_BG_DISABLED;
1009 	}
1010 
1011 	PE_parse_boot_argn("vm_page_donate_mode", &vm_page_donate_mode, sizeof(vm_page_donate_mode));
1012 	if (vm_page_donate_mode != VM_PAGE_DONATE_DISABLED && vm_page_donate_mode != VM_PAGE_DONATE_ENABLED) {
1013 		vm_page_donate_mode = VM_PAGE_DONATE_DISABLED;
1014 	}
1015 
1016 	vm_page_donate_target_high = VM_PAGE_DONATE_TARGET_HIGHWATER;
1017 	vm_page_donate_target_low = VM_PAGE_DONATE_TARGET_LOWWATER;
1018 	vm_page_donate_target = vm_page_donate_target_high;
1019 	vm_page_donate_count = 0;
1020 
1021 	vm_page_free_wanted = 0;
1022 	vm_page_free_wanted_privileged = 0;
1023 #if CONFIG_SECLUDED_MEMORY
1024 	vm_page_free_wanted_secluded = 0;
1025 #endif /* CONFIG_SECLUDED_MEMORY */
1026 
1027 #if defined (__x86_64__)
1028 	/* this must be called before vm_page_set_colors() */
1029 	vm_page_setup_clump();
1030 #endif
1031 
1032 	vm_page_set_colors();
1033 
1034 	bzero(vm_page_inactive_states, sizeof(vm_page_inactive_states));
1035 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1036 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1037 	vm_page_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1038 
1039 	bzero(vm_page_pageable_states, sizeof(vm_page_pageable_states));
1040 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1041 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1042 	vm_page_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1043 	vm_page_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1044 	vm_page_pageable_states[VM_PAGE_ON_SPECULATIVE_Q] = 1;
1045 	vm_page_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1046 #if CONFIG_SECLUDED_MEMORY
1047 	vm_page_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1048 #endif /* CONFIG_SECLUDED_MEMORY */
1049 
1050 	bzero(vm_page_non_speculative_pageable_states, sizeof(vm_page_non_speculative_pageable_states));
1051 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1052 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1053 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1054 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1055 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1;
1056 #if CONFIG_SECLUDED_MEMORY
1057 	vm_page_non_speculative_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1058 #endif /* CONFIG_SECLUDED_MEMORY */
1059 
1060 	bzero(vm_page_active_or_inactive_states, sizeof(vm_page_active_or_inactive_states));
1061 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1;
1062 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1;
1063 	vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1;
1064 	vm_page_active_or_inactive_states[VM_PAGE_ON_ACTIVE_Q] = 1;
1065 #if CONFIG_SECLUDED_MEMORY
1066 	vm_page_active_or_inactive_states[VM_PAGE_ON_SECLUDED_Q] = 1;
1067 #endif /* CONFIG_SECLUDED_MEMORY */
1068 
1069 	for (vm_tag_t t = 0; t < VM_KERN_MEMORY_FIRST_DYNAMIC; t++) {
1070 		vm_allocation_sites_static[t].refcount = 2;
1071 		vm_allocation_sites_static[t].tag = t;
1072 		vm_allocation_sites[t] = &vm_allocation_sites_static[t];
1073 	}
1074 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].refcount = 2;
1075 	vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].tag = VM_KERN_MEMORY_ANY;
1076 	vm_allocation_sites[VM_KERN_MEMORY_ANY] = &vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC];
1077 
1078 	/*
1079 	 *	Steal memory for the map and zone subsystems.
1080 	 */
1081 	kernel_startup_initialize_upto(STARTUP_SUB_PMAP_STEAL);
1082 
1083 	/*
1084 	 *	Allocate (and initialize) the virtual-to-physical
1085 	 *	table hash buckets.
1086 	 *
1087 	 *	The number of buckets should be a power of two to
1088 	 *	get a good hash function.  The following computation
1089 	 *	chooses the first power of two that is greater
1090 	 *	than the number of physical pages in the system.
1091 	 */
1092 
1093 	if (vm_page_bucket_count == 0) {
1094 		unsigned int npages = pmap_free_pages();
1095 
1096 		vm_page_bucket_count = 1;
1097 		while (vm_page_bucket_count < npages) {
1098 			vm_page_bucket_count <<= 1;
1099 		}
1100 	}
1101 	vm_page_bucket_lock_count = (vm_page_bucket_count + BUCKETS_PER_LOCK - 1) / BUCKETS_PER_LOCK;
1102 
1103 	vm_page_hash_mask = vm_page_bucket_count - 1;
1104 
1105 	/*
1106 	 *	Calculate object shift value for hashing algorithm:
1107 	 *		O = log2(sizeof(struct vm_object))
1108 	 *		B = log2(vm_page_bucket_count)
1109 	 *	        hash shifts the object left by
1110 	 *		B/2 - O
1111 	 */
1112 	size = vm_page_bucket_count;
1113 	for (log1 = 0; size > 1; log1++) {
1114 		size /= 2;
1115 	}
1116 	size = sizeof(struct vm_object);
1117 	for (log2 = 0; size > 1; log2++) {
1118 		size /= 2;
1119 	}
1120 	vm_page_hash_shift = log1 / 2 - log2 + 1;
1121 
1122 	vm_page_bucket_hash = 1 << ((log1 + 1) >> 1);           /* Get (ceiling of sqrt of table size) */
1123 	vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2);          /* Get (ceiling of quadroot of table size) */
1124 	vm_page_bucket_hash |= 1;                                                       /* Set bit and add 1 - always must be 1 to insure unique series */
1125 
1126 	if (vm_page_hash_mask & vm_page_bucket_count) {
1127 		printf("vm_page_bootstrap: WARNING -- strange page hash\n");
1128 	}
1129 
1130 #if VM_PAGE_BUCKETS_CHECK
1131 #if VM_PAGE_FAKE_BUCKETS
1132 	/*
1133 	 * Allocate a decoy set of page buckets, to detect
1134 	 * any stomping there.
1135 	 */
1136 	vm_page_fake_buckets = (vm_page_bucket_t *)
1137 	    pmap_steal_memory(vm_page_bucket_count *
1138 	    sizeof(vm_page_bucket_t));
1139 	vm_page_fake_buckets_start = (vm_map_offset_t) vm_page_fake_buckets;
1140 	vm_page_fake_buckets_end =
1141 	    vm_map_round_page((vm_page_fake_buckets_start +
1142 	    (vm_page_bucket_count *
1143 	    sizeof(vm_page_bucket_t))),
1144 	    PAGE_MASK);
1145 	char *cp;
1146 	for (cp = (char *)vm_page_fake_buckets_start;
1147 	    cp < (char *)vm_page_fake_buckets_end;
1148 	    cp++) {
1149 		*cp = 0x5a;
1150 	}
1151 #endif /* VM_PAGE_FAKE_BUCKETS */
1152 #endif /* VM_PAGE_BUCKETS_CHECK */
1153 
1154 	kernel_debug_string_early("vm_page_buckets");
1155 	vm_page_buckets = (vm_page_bucket_t *)
1156 	    pmap_steal_memory(vm_page_bucket_count *
1157 	    sizeof(vm_page_bucket_t));
1158 
1159 	kernel_debug_string_early("vm_page_bucket_locks");
1160 	vm_page_bucket_locks = (lck_spin_t *)
1161 	    pmap_steal_memory(vm_page_bucket_lock_count *
1162 	    sizeof(lck_spin_t));
1163 
1164 	for (i = 0; i < vm_page_bucket_count; i++) {
1165 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
1166 
1167 		bucket->page_list = VM_PAGE_PACK_PTR(VM_PAGE_NULL);
1168 #if     MACH_PAGE_HASH_STATS
1169 		bucket->cur_count = 0;
1170 		bucket->hi_count = 0;
1171 #endif /* MACH_PAGE_HASH_STATS */
1172 	}
1173 
1174 	for (i = 0; i < vm_page_bucket_lock_count; i++) {
1175 		lck_spin_init(&vm_page_bucket_locks[i], &vm_page_lck_grp_bucket, &vm_page_lck_attr);
1176 	}
1177 
1178 	vm_tag_init();
1179 
1180 #if VM_PAGE_BUCKETS_CHECK
1181 	vm_page_buckets_check_ready = TRUE;
1182 #endif /* VM_PAGE_BUCKETS_CHECK */
1183 
1184 	/*
1185 	 *	Machine-dependent code allocates the resident page table.
1186 	 *	It uses vm_page_init to initialize the page frames.
1187 	 *	The code also returns to us the virtual space available
1188 	 *	to the kernel.  We don't trust the pmap module
1189 	 *	to get the alignment right.
1190 	 */
1191 
1192 	kernel_debug_string_early("pmap_startup");
1193 	pmap_startup(&virtual_space_start, &virtual_space_end);
1194 	virtual_space_start = round_page(virtual_space_start);
1195 	virtual_space_end = trunc_page(virtual_space_end);
1196 
1197 	*startp = virtual_space_start;
1198 	*endp = virtual_space_end;
1199 
1200 	/*
1201 	 *	Compute the initial "wire" count.
1202 	 *	Up until now, the pages which have been set aside are not under
1203 	 *	the VM system's control, so although they aren't explicitly
1204 	 *	wired, they nonetheless can't be moved. At this moment,
1205 	 *	all VM managed pages are "free", courtesy of pmap_startup.
1206 	 */
1207 	assert((unsigned int) atop_64(max_mem) == atop_64(max_mem));
1208 	vm_page_wire_count = ((unsigned int) atop_64(max_mem)) -
1209 	    vm_page_free_count - vm_lopage_free_count;
1210 #if CONFIG_SECLUDED_MEMORY
1211 	vm_page_wire_count -= vm_page_secluded_count;
1212 #endif
1213 	vm_page_wire_count_initial = vm_page_wire_count;
1214 
1215 	/* capture this for later use */
1216 	booter_size = ml_get_booter_memory_size();
1217 
1218 	printf("vm_page_bootstrap: %d free pages, %d wired pages, (up to %d of which are delayed free)\n",
1219 	    vm_page_free_count, vm_page_wire_count, vm_delayed_count);
1220 
1221 	kernel_debug_string_early("vm_page_bootstrap complete");
1222 }
1223 
1224 #ifndef MACHINE_PAGES
1225 /*
1226  * This is the early boot time allocator for data structures needed to bootstrap the VM system.
1227  * On x86 it will allocate large pages if size is sufficiently large. We don't need to do this
1228  * on ARM yet, due to the combination of a large base page size and smaller RAM devices.
1229  */
1230 static void *
pmap_steal_memory_internal(vm_size_t size,boolean_t might_free)1231 pmap_steal_memory_internal(
1232 	vm_size_t size,
1233 	boolean_t might_free)
1234 {
1235 	kern_return_t kr;
1236 	vm_offset_t addr;
1237 	vm_offset_t map_addr;
1238 	ppnum_t phys_page;
1239 
1240 	/*
1241 	 * Size needs to be aligned to word size.
1242 	 */
1243 	size = (size + sizeof(void *) - 1) & ~(sizeof(void *) - 1);
1244 
1245 	/*
1246 	 * On the first call, get the initial values for virtual address space
1247 	 * and page align them.
1248 	 */
1249 	if (virtual_space_start == virtual_space_end) {
1250 		pmap_virtual_space(&virtual_space_start, &virtual_space_end);
1251 		virtual_space_start = round_page(virtual_space_start);
1252 		virtual_space_end = trunc_page(virtual_space_end);
1253 
1254 #if defined(__x86_64__)
1255 		/*
1256 		 * Release remaining unused section of preallocated KVA and the 4K page tables
1257 		 * that map it. This makes the VA available for large page mappings.
1258 		 */
1259 		Idle_PTs_release(virtual_space_start, virtual_space_end);
1260 #endif
1261 	}
1262 
1263 	/*
1264 	 * Allocate the virtual space for this request. On x86, we'll align to a large page
1265 	 * address if the size is big enough to back with at least 1 large page.
1266 	 */
1267 #if defined(__x86_64__)
1268 	if (size >= I386_LPGBYTES) {
1269 		virtual_space_start = ((virtual_space_start + I386_LPGMASK) & ~I386_LPGMASK);
1270 	}
1271 #endif
1272 	addr = virtual_space_start;
1273 	virtual_space_start += size;
1274 
1275 	//kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)size);	/* (TEST/DEBUG) */
1276 
1277 	/*
1278 	 * Allocate and map physical pages to back the new virtual space.
1279 	 */
1280 	map_addr = round_page(addr);
1281 	while (map_addr < addr + size) {
1282 #if defined(__x86_64__)
1283 		/*
1284 		 * Back with a large page if properly aligned on x86
1285 		 */
1286 		if ((map_addr & I386_LPGMASK) == 0 &&
1287 		    map_addr + I386_LPGBYTES <= addr + size &&
1288 		    pmap_pre_expand_large(kernel_pmap, map_addr) == KERN_SUCCESS &&
1289 		    pmap_next_page_large(&phys_page) == KERN_SUCCESS) {
1290 			kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1291 			    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1292 			    VM_WIMG_USE_DEFAULT | VM_MEM_SUPERPAGE, FALSE);
1293 
1294 			if (kr != KERN_SUCCESS) {
1295 				panic("pmap_steal_memory: pmap_enter() large failed, new_addr=%#lx, phys_page=%u",
1296 				    (unsigned long)map_addr, phys_page);
1297 			}
1298 			map_addr += I386_LPGBYTES;
1299 			vm_page_wire_count += I386_LPGBYTES >> PAGE_SHIFT;
1300 			vm_page_stolen_count += I386_LPGBYTES >> PAGE_SHIFT;
1301 			vm_page_kern_lpage_count++;
1302 			continue;
1303 		}
1304 #endif
1305 
1306 		if (!pmap_next_page_hi(&phys_page, might_free)) {
1307 			panic("pmap_steal_memory() size: 0x%llx", (uint64_t)size);
1308 		}
1309 
1310 #if defined(__x86_64__)
1311 		pmap_pre_expand(kernel_pmap, map_addr);
1312 #endif
1313 
1314 		kr = pmap_enter(kernel_pmap, map_addr, phys_page,
1315 		    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE,
1316 		    VM_WIMG_USE_DEFAULT, FALSE);
1317 
1318 		if (kr != KERN_SUCCESS) {
1319 			panic("pmap_steal_memory() pmap_enter failed, map_addr=%#lx, phys_page=%u",
1320 			    (unsigned long)map_addr, phys_page);
1321 		}
1322 		map_addr += PAGE_SIZE;
1323 
1324 		/*
1325 		 * Account for newly stolen memory
1326 		 */
1327 		vm_page_wire_count++;
1328 		vm_page_stolen_count++;
1329 	}
1330 
1331 #if defined(__x86_64__)
1332 	/*
1333 	 * The call with might_free is currently the last use of pmap_steal_memory*().
1334 	 * Notify the pmap layer to record which high pages were allocated so far.
1335 	 */
1336 	if (might_free) {
1337 		pmap_hi_pages_done();
1338 	}
1339 #endif
1340 #if KASAN
1341 	kasan_notify_address(round_page(addr), size);
1342 #endif
1343 	return (void *) addr;
1344 }
1345 
1346 void *
pmap_steal_memory(vm_size_t size)1347 pmap_steal_memory(
1348 	vm_size_t size)
1349 {
1350 	return pmap_steal_memory_internal(size, FALSE);
1351 }
1352 
1353 void *
pmap_steal_freeable_memory(vm_size_t size)1354 pmap_steal_freeable_memory(
1355 	vm_size_t size)
1356 {
1357 	return pmap_steal_memory_internal(size, TRUE);
1358 }
1359 
1360 
1361 #if CONFIG_SECLUDED_MEMORY
1362 /* boot-args to control secluded memory */
1363 unsigned int secluded_mem_mb = 0;       /* # of MBs of RAM to seclude */
1364 int secluded_for_iokit = 1;             /* IOKit can use secluded memory */
1365 int secluded_for_apps = 1;              /* apps can use secluded memory */
1366 int secluded_for_filecache = 2;         /* filecache can use seclude memory */
1367 #if 11
1368 int secluded_for_fbdp = 0;
1369 #endif
1370 uint64_t secluded_shutoff_trigger = 0;
1371 uint64_t secluded_shutoff_headroom = 150 * 1024 * 1024; /* original value from N56 */
1372 #endif /* CONFIG_SECLUDED_MEMORY */
1373 
1374 
1375 #if defined(__arm64__)
1376 extern void patch_low_glo_vm_page_info(void *, void *, uint32_t);
1377 unsigned int vm_first_phys_ppnum = 0;
1378 #endif
1379 
1380 void vm_page_release_startup(vm_page_t mem);
1381 void
pmap_startup(vm_offset_t * startp,vm_offset_t * endp)1382 pmap_startup(
1383 	vm_offset_t     *startp,
1384 	vm_offset_t     *endp)
1385 {
1386 	unsigned int    i, npages;
1387 	ppnum_t         phys_page;
1388 	uint64_t        mem_sz;
1389 	uint64_t        start_ns;
1390 	uint64_t        now_ns;
1391 	uint_t          low_page_count = 0;
1392 
1393 #if    defined(__LP64__)
1394 	/*
1395 	 * make sure we are aligned on a 64 byte boundary
1396 	 * for VM_PAGE_PACK_PTR (it clips off the low-order
1397 	 * 6 bits of the pointer)
1398 	 */
1399 	if (virtual_space_start != virtual_space_end) {
1400 		virtual_space_start = round_page(virtual_space_start);
1401 	}
1402 #endif
1403 
1404 	/*
1405 	 * We calculate how many page frames we will have
1406 	 * and then allocate the page structures in one chunk.
1407 	 *
1408 	 * Note that the calculation here doesn't take into account
1409 	 * the memory needed to map what's being allocated, i.e. the page
1410 	 * table entries. So the actual number of pages we get will be
1411 	 * less than this. To do someday: include that in the computation.
1412 	 *
1413 	 * Also for ARM, we don't use the count of free_pages, but rather the
1414 	 * range from last page to first page (ignore holes due to retired pages).
1415 	 */
1416 #if defined(__arm64__)
1417 	mem_sz = pmap_free_pages_span() * (uint64_t)PAGE_SIZE;
1418 #else /* defined(__arm64__) */
1419 	mem_sz = pmap_free_pages() * (uint64_t)PAGE_SIZE;
1420 #endif /* defined(__arm64__) */
1421 	mem_sz += round_page(virtual_space_start) - virtual_space_start;        /* Account for any slop */
1422 	npages = (uint_t)(mem_sz / (PAGE_SIZE + sizeof(*vm_pages)));    /* scaled to include the vm_page_ts */
1423 
1424 
1425 	vm_pages = (vm_page_t) pmap_steal_freeable_memory(npages * sizeof *vm_pages);
1426 
1427 	/*
1428 	 * Check if we want to initialize pages to a known value
1429 	 */
1430 	if (PE_parse_boot_argn("fill", &fillval, sizeof(fillval))) {
1431 		fill = TRUE;
1432 	}
1433 #if     DEBUG
1434 	/* This slows down booting the DEBUG kernel, particularly on
1435 	 * large memory systems, but is worthwhile in deterministically
1436 	 * trapping uninitialized memory usage.
1437 	 */
1438 	if (!fill) {
1439 		fill = TRUE;
1440 		fillval = 0xDEB8F177;
1441 	}
1442 #endif
1443 	if (fill) {
1444 		kprintf("Filling vm_pages with pattern: 0x%x\n", fillval);
1445 	}
1446 
1447 #if CONFIG_SECLUDED_MEMORY
1448 	/*
1449 	 * Figure out how much secluded memory to have before we start
1450 	 * release pages to free lists.
1451 	 * The default, if specified nowhere else, is no secluded mem.
1452 	 */
1453 	secluded_mem_mb = 0;
1454 	if (max_mem > 1 * 1024 * 1024 * 1024) {
1455 		/* default to 90MB for devices with > 1GB of RAM */
1456 		secluded_mem_mb = 90;
1457 	}
1458 	/* override with value from device tree, if provided */
1459 	PE_get_default("kern.secluded_mem_mb",
1460 	    &secluded_mem_mb, sizeof(secluded_mem_mb));
1461 	/* override with value from boot-args, if provided */
1462 	PE_parse_boot_argn("secluded_mem_mb",
1463 	    &secluded_mem_mb,
1464 	    sizeof(secluded_mem_mb));
1465 
1466 	vm_page_secluded_target = (unsigned int)
1467 	    ((secluded_mem_mb * 1024ULL * 1024ULL) / PAGE_SIZE);
1468 	PE_parse_boot_argn("secluded_for_iokit",
1469 	    &secluded_for_iokit,
1470 	    sizeof(secluded_for_iokit));
1471 	PE_parse_boot_argn("secluded_for_apps",
1472 	    &secluded_for_apps,
1473 	    sizeof(secluded_for_apps));
1474 	PE_parse_boot_argn("secluded_for_filecache",
1475 	    &secluded_for_filecache,
1476 	    sizeof(secluded_for_filecache));
1477 #if 11
1478 	PE_parse_boot_argn("secluded_for_fbdp",
1479 	    &secluded_for_fbdp,
1480 	    sizeof(secluded_for_fbdp));
1481 #endif
1482 
1483 	/*
1484 	 * Allow a really large app to effectively use secluded memory until it exits.
1485 	 */
1486 	if (vm_page_secluded_target != 0) {
1487 		/*
1488 		 * Get an amount from boot-args, else use 1/2 of max_mem.
1489 		 * 1/2 max_mem was chosen from a Peace daemon tentpole test which
1490 		 * used munch to induce jetsam thrashing of false idle daemons on N56.
1491 		 */
1492 		int secluded_shutoff_mb;
1493 		if (PE_parse_boot_argn("secluded_shutoff_mb", &secluded_shutoff_mb,
1494 		    sizeof(secluded_shutoff_mb))) {
1495 			secluded_shutoff_trigger = (uint64_t)secluded_shutoff_mb * 1024 * 1024;
1496 		} else {
1497 			secluded_shutoff_trigger = max_mem / 2;
1498 		}
1499 
1500 		/* ensure the headroom value is sensible and avoid underflows */
1501 		assert(secluded_shutoff_trigger == 0 || secluded_shutoff_trigger > secluded_shutoff_headroom);
1502 	}
1503 
1504 #endif /* CONFIG_SECLUDED_MEMORY */
1505 
1506 #if defined(__x86_64__)
1507 
1508 	/*
1509 	 * Decide how much memory we delay freeing at boot time.
1510 	 */
1511 	uint32_t delay_above_gb;
1512 	if (!PE_parse_boot_argn("delay_above_gb", &delay_above_gb, sizeof(delay_above_gb))) {
1513 		delay_above_gb = DEFAULT_DELAY_ABOVE_PHYS_GB;
1514 	}
1515 
1516 	if (delay_above_gb == 0) {
1517 		delay_above_pnum = PPNUM_MAX;
1518 	} else {
1519 		delay_above_pnum = delay_above_gb * (1024 * 1024 * 1024 / PAGE_SIZE);
1520 	}
1521 
1522 	/* make sure we have sane breathing room: 1G above low memory */
1523 	if (delay_above_pnum <= max_valid_low_ppnum) {
1524 		delay_above_pnum = max_valid_low_ppnum + ((1024 * 1024 * 1024) >> PAGE_SHIFT);
1525 	}
1526 
1527 	if (delay_above_pnum < PPNUM_MAX) {
1528 		printf("pmap_startup() delaying init/free of page nums > 0x%x\n", delay_above_pnum);
1529 	}
1530 
1531 #endif /* defined(__x86_64__) */
1532 
1533 	/*
1534 	 * Initialize and release the page frames.
1535 	 */
1536 	kernel_debug_string_early("page_frame_init");
1537 
1538 	vm_page_array_beginning_addr = &vm_pages[0];
1539 	vm_page_array_ending_addr = &vm_pages[npages];  /* used by ptr packing/unpacking code */
1540 #if VM_PAGE_PACKED_FROM_ARRAY
1541 	if (npages >= VM_PAGE_PACKED_FROM_ARRAY) {
1542 		panic("pmap_startup(): too many pages to support vm_page packing");
1543 	}
1544 #endif
1545 
1546 	vm_delayed_count = 0;
1547 
1548 	absolutetime_to_nanoseconds(mach_absolute_time(), &start_ns);
1549 	vm_pages_count = 0;
1550 	for (i = 0; i < npages; i++) {
1551 		/* Did we run out of pages? */
1552 		if (!pmap_next_page(&phys_page)) {
1553 			break;
1554 		}
1555 
1556 		if (phys_page < max_valid_low_ppnum) {
1557 			++low_page_count;
1558 		}
1559 
1560 		/* Are we at high enough pages to delay the rest? */
1561 		if (low_page_count > vm_lopage_free_limit && phys_page > delay_above_pnum) {
1562 			vm_delayed_count = pmap_free_pages();
1563 			break;
1564 		}
1565 
1566 #if defined(__arm64__)
1567 		if (i == 0) {
1568 			vm_first_phys_ppnum = phys_page;
1569 			patch_low_glo_vm_page_info((void *)vm_page_array_beginning_addr,
1570 			    (void *)vm_page_array_ending_addr, vm_first_phys_ppnum);
1571 		}
1572 #endif /* defined(__arm64__) */
1573 
1574 #if defined(__x86_64__)
1575 		/* The x86 clump freeing code requires increasing ppn's to work correctly */
1576 		if (i > 0) {
1577 			assert(phys_page > vm_pages[i - 1].vmp_phys_page);
1578 		}
1579 #endif
1580 		++vm_pages_count;
1581 		vm_page_init(&vm_pages[i], phys_page, FALSE);
1582 		if (fill) {
1583 			fillPage(phys_page, fillval);
1584 		}
1585 		if (vm_himemory_mode) {
1586 			vm_page_release_startup(&vm_pages[i]);
1587 		}
1588 	}
1589 	vm_page_pages = vm_pages_count; /* used to report to user space */
1590 
1591 	if (!vm_himemory_mode) {
1592 		do {
1593 			if (!VMP_ERROR_GET(&vm_pages[--i])) {               /* skip retired pages */
1594 				vm_page_release_startup(&vm_pages[i]);
1595 			}
1596 		} while (i != 0);
1597 	}
1598 
1599 	absolutetime_to_nanoseconds(mach_absolute_time(), &now_ns);
1600 	printf("pmap_startup() init/release time: %lld microsec\n", (now_ns - start_ns) / NSEC_PER_USEC);
1601 	printf("pmap_startup() delayed init/release of %d pages\n", vm_delayed_count);
1602 
1603 #if defined(__LP64__)
1604 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[0]))) != &vm_pages[0]) {
1605 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[0] - %p", (void *)&vm_pages[0]);
1606 	}
1607 
1608 	if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[vm_pages_count - 1]))) != &vm_pages[vm_pages_count - 1]) {
1609 		panic("VM_PAGE_PACK_PTR failed on &vm_pages[vm_pages_count-1] - %p", (void *)&vm_pages[vm_pages_count - 1]);
1610 	}
1611 #endif
1612 
1613 	VM_CHECK_MEMORYSTATUS;
1614 
1615 	/*
1616 	 * We have to re-align virtual_space_start,
1617 	 * because pmap_steal_memory has been using it.
1618 	 */
1619 	virtual_space_start = round_page(virtual_space_start);
1620 	*startp = virtual_space_start;
1621 	*endp = virtual_space_end;
1622 }
1623 #endif  /* MACHINE_PAGES */
1624 
1625 /*
1626  * Create the zone that represents the vm_pages[] array. Nothing ever allocates
1627  * or frees to this zone. It's just here for reporting purposes via zprint command.
1628  * This needs to be done after all initially delayed pages are put on the free lists.
1629  */
1630 static void
vm_page_module_init_delayed(void)1631 vm_page_module_init_delayed(void)
1632 {
1633 	(void)zone_create_ext("vm pages array", sizeof(struct vm_page),
1634 	    ZC_NONE, ZONE_ID_VM_PAGES, ^(zone_t z) {
1635 		uint64_t vm_page_zone_pages, vm_page_array_zone_data_size;
1636 
1637 		zone_set_exhaustible(z, 0);
1638 		/*
1639 		 * Reflect size and usage information for vm_pages[].
1640 		 */
1641 
1642 		z->z_elems_avail = (uint32_t)(vm_page_array_ending_addr - vm_pages);
1643 		z->z_elems_free = z->z_elems_avail - vm_pages_count;
1644 		zpercpu_get_cpu(z->z_stats, 0)->zs_mem_allocated =
1645 		vm_pages_count * sizeof(struct vm_page);
1646 		vm_page_array_zone_data_size = (uint64_t)vm_page_array_ending_addr - (uint64_t)vm_pages;
1647 		vm_page_zone_pages = atop(round_page((vm_offset_t)vm_page_array_zone_data_size));
1648 		z->z_wired_cur += vm_page_zone_pages;
1649 		z->z_wired_hwm = z->z_wired_cur;
1650 		z->z_va_cur = z->z_wired_cur;
1651 		/* since zone accounts for these, take them out of stolen */
1652 		VM_PAGE_MOVE_STOLEN(vm_page_zone_pages);
1653 	});
1654 }
1655 
1656 /*
1657  * Create the vm_pages zone. This is used for the vm_page structures for the pages
1658  * that are scavanged from other boot time usages by ml_static_mfree(). As such,
1659  * this needs to happen in early VM bootstrap.
1660  */
1661 
1662 __startup_func
1663 static void
vm_page_module_init(void)1664 vm_page_module_init(void)
1665 {
1666 	vm_size_t vm_page_with_ppnum_size;
1667 
1668 	/*
1669 	 * Since the pointers to elements in this zone will be packed, they
1670 	 * must have appropriate size. Not strictly what sizeof() reports.
1671 	 */
1672 	vm_page_with_ppnum_size =
1673 	    (sizeof(struct vm_page_with_ppnum) + (VM_PAGE_PACKED_PTR_ALIGNMENT - 1)) &
1674 	    ~(VM_PAGE_PACKED_PTR_ALIGNMENT - 1);
1675 
1676 	vm_page_zone = zone_create_ext("vm pages", vm_page_with_ppnum_size,
1677 	    ZC_ALIGNMENT_REQUIRED | ZC_VM_LP64 | ZC_NOTBITAG,
1678 	    ZONE_ID_ANY, ^(zone_t z) {
1679 		/*
1680 		 * The number "10" is a small number that is larger than the number
1681 		 * of fictitious pages that any single caller will attempt to allocate
1682 		 * without blocking.
1683 		 *
1684 		 * The largest such number at the moment is kmem_alloc()
1685 		 * when 2 guard pages are asked. 10 is simply a somewhat larger number,
1686 		 * taking into account the 50% hysteresis the zone allocator uses.
1687 		 *
1688 		 * Note: this works at all because the zone allocator
1689 		 *       doesn't ever allocate fictitious pages.
1690 		 */
1691 		zone_raise_reserve(z, 10);
1692 	});
1693 }
1694 STARTUP(ZALLOC, STARTUP_RANK_SECOND, vm_page_module_init);
1695 
1696 /*
1697  *	Routine:	vm_page_create
1698  *	Purpose:
1699  *		After the VM system is up, machine-dependent code
1700  *		may stumble across more physical memory.  For example,
1701  *		memory that it was reserving for a frame buffer.
1702  *		vm_page_create turns this memory into available pages.
1703  */
1704 
1705 void
vm_page_create(ppnum_t start,ppnum_t end)1706 vm_page_create(
1707 	ppnum_t start,
1708 	ppnum_t end)
1709 {
1710 	ppnum_t         phys_page;
1711 	vm_page_t       m;
1712 
1713 	for (phys_page = start;
1714 	    phys_page < end;
1715 	    phys_page++) {
1716 		m = vm_page_grab_fictitious_common(phys_page, TRUE);
1717 		m->vmp_fictitious = FALSE;
1718 		pmap_clear_noencrypt(phys_page);
1719 
1720 
1721 		vm_free_page_lock();
1722 		vm_page_pages++;
1723 		vm_free_page_unlock();
1724 		vm_page_release(m, FALSE);
1725 	}
1726 }
1727 
1728 
1729 /*
1730  *	vm_page_hash:
1731  *
1732  *	Distributes the object/offset key pair among hash buckets.
1733  *
1734  *	NOTE:	The bucket count must be a power of 2
1735  */
1736 #define vm_page_hash(object, offset) (\
1737 	( (natural_t)((uintptr_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
1738 	 & vm_page_hash_mask)
1739 
1740 
1741 /*
1742  *	vm_page_insert:		[ internal use only ]
1743  *
1744  *	Inserts the given mem entry into the object/object-page
1745  *	table and object list.
1746  *
1747  *	The object must be locked.
1748  */
1749 void
vm_page_insert(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)1750 vm_page_insert(
1751 	vm_page_t               mem,
1752 	vm_object_t             object,
1753 	vm_object_offset_t      offset)
1754 {
1755 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, FALSE, FALSE, NULL);
1756 }
1757 
1758 void
vm_page_insert_wired(vm_page_t mem,vm_object_t object,vm_object_offset_t offset,vm_tag_t tag)1759 vm_page_insert_wired(
1760 	vm_page_t               mem,
1761 	vm_object_t             object,
1762 	vm_object_offset_t      offset,
1763 	vm_tag_t                tag)
1764 {
1765 	vm_page_insert_internal(mem, object, offset, tag, FALSE, TRUE, FALSE, FALSE, NULL);
1766 }
1767 
1768 void
vm_page_insert_internal(vm_page_t mem,vm_object_t object,vm_object_offset_t offset,vm_tag_t tag,boolean_t queues_lock_held,boolean_t insert_in_hash,boolean_t batch_pmap_op,boolean_t batch_accounting,uint64_t * delayed_ledger_update)1769 vm_page_insert_internal(
1770 	vm_page_t               mem,
1771 	vm_object_t             object,
1772 	vm_object_offset_t      offset,
1773 	vm_tag_t                tag,
1774 	boolean_t               queues_lock_held,
1775 	boolean_t               insert_in_hash,
1776 	boolean_t               batch_pmap_op,
1777 	boolean_t               batch_accounting,
1778 	uint64_t                *delayed_ledger_update)
1779 {
1780 	vm_page_bucket_t        *bucket;
1781 	lck_spin_t              *bucket_lock;
1782 	int                     hash_id;
1783 	task_t                  owner;
1784 	int                     ledger_idx_volatile;
1785 	int                     ledger_idx_nonvolatile;
1786 	int                     ledger_idx_volatile_compressed;
1787 	int                     ledger_idx_nonvolatile_compressed;
1788 	boolean_t               do_footprint;
1789 
1790 #if 0
1791 	/*
1792 	 * we may not hold the page queue lock
1793 	 * so this check isn't safe to make
1794 	 */
1795 	VM_PAGE_CHECK(mem);
1796 #endif
1797 
1798 	assertf(page_aligned(offset), "0x%llx\n", offset);
1799 
1800 	assert(!VM_PAGE_WIRED(mem) || mem->vmp_private || mem->vmp_fictitious || (tag != VM_KERN_MEMORY_NONE));
1801 
1802 	vm_object_lock_assert_exclusive(object);
1803 	LCK_MTX_ASSERT(&vm_page_queue_lock,
1804 	    queues_lock_held ? LCK_MTX_ASSERT_OWNED
1805 	    : LCK_MTX_ASSERT_NOTOWNED);
1806 
1807 	if (queues_lock_held == FALSE) {
1808 		assert(!VM_PAGE_PAGEABLE(mem));
1809 	}
1810 
1811 	if (insert_in_hash == TRUE) {
1812 #if DEBUG || VM_PAGE_BUCKETS_CHECK
1813 		if (mem->vmp_tabled || mem->vmp_object) {
1814 			panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
1815 			    "already in (obj=%p,off=0x%llx)",
1816 			    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
1817 		}
1818 #endif
1819 		if (object->internal && (offset >= object->vo_size)) {
1820 			panic("vm_page_insert_internal: (page=%p,obj=%p,off=0x%llx,size=0x%llx) inserted at offset past object bounds",
1821 			    mem, object, offset, object->vo_size);
1822 		}
1823 
1824 		assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
1825 
1826 		/*
1827 		 *	Record the object/offset pair in this page
1828 		 */
1829 
1830 		mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
1831 		mem->vmp_offset = offset;
1832 
1833 #if CONFIG_SECLUDED_MEMORY
1834 		if (object->eligible_for_secluded) {
1835 			vm_page_secluded.eligible_for_secluded++;
1836 		}
1837 #endif /* CONFIG_SECLUDED_MEMORY */
1838 
1839 		/*
1840 		 *	Insert it into the object_object/offset hash table
1841 		 */
1842 		hash_id = vm_page_hash(object, offset);
1843 		bucket = &vm_page_buckets[hash_id];
1844 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
1845 
1846 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
1847 
1848 		mem->vmp_next_m = bucket->page_list;
1849 		bucket->page_list = VM_PAGE_PACK_PTR(mem);
1850 		assert(mem == (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)));
1851 
1852 #if     MACH_PAGE_HASH_STATS
1853 		if (++bucket->cur_count > bucket->hi_count) {
1854 			bucket->hi_count = bucket->cur_count;
1855 		}
1856 #endif /* MACH_PAGE_HASH_STATS */
1857 		mem->vmp_hashed = TRUE;
1858 		lck_spin_unlock(bucket_lock);
1859 	}
1860 
1861 	{
1862 		unsigned int    cache_attr;
1863 
1864 		cache_attr = object->wimg_bits & VM_WIMG_MASK;
1865 
1866 		if (cache_attr != VM_WIMG_USE_DEFAULT) {
1867 			PMAP_SET_CACHE_ATTR(mem, object, cache_attr, batch_pmap_op);
1868 		}
1869 	}
1870 	/*
1871 	 *	Now link into the object's list of backed pages.
1872 	 */
1873 	vm_page_queue_enter(&object->memq, mem, vmp_listq);
1874 	object->memq_hint = mem;
1875 	mem->vmp_tabled = TRUE;
1876 
1877 	/*
1878 	 *	Show that the object has one more resident page.
1879 	 */
1880 
1881 	object->resident_page_count++;
1882 	if (VM_PAGE_WIRED(mem)) {
1883 		assert(mem->vmp_wire_count > 0);
1884 		VM_OBJECT_WIRED_PAGE_UPDATE_START(object);
1885 		VM_OBJECT_WIRED_PAGE_ADD(object, mem);
1886 		VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag);
1887 	}
1888 	assert(object->resident_page_count >= object->wired_page_count);
1889 
1890 #if DEVELOPMENT || DEBUG
1891 	if (object->object_is_shared_cache &&
1892 	    object->pager != NULL &&
1893 	    object->pager->mo_pager_ops == &shared_region_pager_ops) {
1894 		int new, old;
1895 		assert(!object->internal);
1896 		new = OSAddAtomic(+1, &shared_region_pagers_resident_count);
1897 		do {
1898 			old = shared_region_pagers_resident_peak;
1899 		} while (old < new &&
1900 		    !OSCompareAndSwap(old, new, &shared_region_pagers_resident_peak));
1901 	}
1902 #endif /* DEVELOPMENT || DEBUG */
1903 
1904 	if (batch_accounting == FALSE) {
1905 		if (object->internal) {
1906 			OSAddAtomic(1, &vm_page_internal_count);
1907 		} else {
1908 			OSAddAtomic(1, &vm_page_external_count);
1909 		}
1910 	}
1911 
1912 	/*
1913 	 * It wouldn't make sense to insert a "reusable" page in
1914 	 * an object (the page would have been marked "reusable" only
1915 	 * at the time of a madvise(MADV_FREE_REUSABLE) if it was already
1916 	 * in the object at that time).
1917 	 * But a page could be inserted in a "all_reusable" object, if
1918 	 * something faults it in (a vm_read() from another task or a
1919 	 * "use-after-free" issue in user space, for example).  It can
1920 	 * also happen if we're relocating a page from that object to
1921 	 * a different physical page during a physically-contiguous
1922 	 * allocation.
1923 	 */
1924 	assert(!mem->vmp_reusable);
1925 	if (object->all_reusable) {
1926 		OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count);
1927 	}
1928 
1929 	if (object->purgable == VM_PURGABLE_DENY &&
1930 	    !object->vo_ledger_tag) {
1931 		owner = TASK_NULL;
1932 	} else {
1933 		owner = VM_OBJECT_OWNER(object);
1934 		vm_object_ledger_tag_ledgers(object,
1935 		    &ledger_idx_volatile,
1936 		    &ledger_idx_nonvolatile,
1937 		    &ledger_idx_volatile_compressed,
1938 		    &ledger_idx_nonvolatile_compressed,
1939 		    &do_footprint);
1940 	}
1941 	if (owner &&
1942 	    (object->purgable == VM_PURGABLE_NONVOLATILE ||
1943 	    object->purgable == VM_PURGABLE_DENY ||
1944 	    VM_PAGE_WIRED(mem))) {
1945 		if (delayed_ledger_update) {
1946 			*delayed_ledger_update += PAGE_SIZE;
1947 		} else {
1948 			/* more non-volatile bytes */
1949 			ledger_credit(owner->ledger,
1950 			    ledger_idx_nonvolatile,
1951 			    PAGE_SIZE);
1952 			if (do_footprint) {
1953 				/* more footprint */
1954 				ledger_credit(owner->ledger,
1955 				    task_ledgers.phys_footprint,
1956 				    PAGE_SIZE);
1957 			}
1958 		}
1959 	} else if (owner &&
1960 	    (object->purgable == VM_PURGABLE_VOLATILE ||
1961 	    object->purgable == VM_PURGABLE_EMPTY)) {
1962 		assert(!VM_PAGE_WIRED(mem));
1963 		/* more volatile bytes */
1964 		ledger_credit(owner->ledger,
1965 		    ledger_idx_volatile,
1966 		    PAGE_SIZE);
1967 	}
1968 
1969 	if (object->purgable == VM_PURGABLE_VOLATILE) {
1970 		if (VM_PAGE_WIRED(mem)) {
1971 			OSAddAtomic(+1, &vm_page_purgeable_wired_count);
1972 		} else {
1973 			OSAddAtomic(+1, &vm_page_purgeable_count);
1974 		}
1975 	} else if (object->purgable == VM_PURGABLE_EMPTY &&
1976 	    mem->vmp_q_state == VM_PAGE_ON_THROTTLED_Q) {
1977 		/*
1978 		 * This page belongs to a purged VM object but hasn't
1979 		 * been purged (because it was "busy").
1980 		 * It's in the "throttled" queue and hence not
1981 		 * visible to vm_pageout_scan().  Move it to a pageable
1982 		 * queue, so that it can eventually be reclaimed, instead
1983 		 * of lingering in the "empty" object.
1984 		 */
1985 		if (queues_lock_held == FALSE) {
1986 			vm_page_lockspin_queues();
1987 		}
1988 		vm_page_deactivate(mem);
1989 		if (queues_lock_held == FALSE) {
1990 			vm_page_unlock_queues();
1991 		}
1992 	}
1993 
1994 #if VM_OBJECT_TRACKING_OP_MODIFIED
1995 	if (vm_object_tracking_btlog &&
1996 	    object->internal &&
1997 	    object->resident_page_count == 0 &&
1998 	    object->pager == NULL &&
1999 	    object->shadow != NULL &&
2000 	    object->shadow->copy == object) {
2001 		btlog_record(vm_object_tracking_btlog, object,
2002 		    VM_OBJECT_TRACKING_OP_MODIFIED,
2003 		    btref_get(__builtin_frame_address(0), 0));
2004 	}
2005 #endif /* VM_OBJECT_TRACKING_OP_MODIFIED */
2006 }
2007 
2008 /*
2009  *	vm_page_replace:
2010  *
2011  *	Exactly like vm_page_insert, except that we first
2012  *	remove any existing page at the given offset in object.
2013  *
2014  *	The object must be locked.
2015  */
2016 void
vm_page_replace(vm_page_t mem,vm_object_t object,vm_object_offset_t offset)2017 vm_page_replace(
2018 	vm_page_t               mem,
2019 	vm_object_t             object,
2020 	vm_object_offset_t      offset)
2021 {
2022 	vm_page_bucket_t *bucket;
2023 	vm_page_t        found_m = VM_PAGE_NULL;
2024 	lck_spin_t      *bucket_lock;
2025 	int             hash_id;
2026 
2027 #if 0
2028 	/*
2029 	 * we don't hold the page queue lock
2030 	 * so this check isn't safe to make
2031 	 */
2032 	VM_PAGE_CHECK(mem);
2033 #endif
2034 	vm_object_lock_assert_exclusive(object);
2035 #if DEBUG || VM_PAGE_BUCKETS_CHECK
2036 	if (mem->vmp_tabled || mem->vmp_object) {
2037 		panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) "
2038 		    "already in (obj=%p,off=0x%llx)",
2039 		    mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset);
2040 	}
2041 #endif
2042 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2043 
2044 	assert(!VM_PAGE_PAGEABLE(mem));
2045 
2046 	/*
2047 	 *	Record the object/offset pair in this page
2048 	 */
2049 	mem->vmp_object = VM_PAGE_PACK_OBJECT(object);
2050 	mem->vmp_offset = offset;
2051 
2052 	/*
2053 	 *	Insert it into the object_object/offset hash table,
2054 	 *	replacing any page that might have been there.
2055 	 */
2056 
2057 	hash_id = vm_page_hash(object, offset);
2058 	bucket = &vm_page_buckets[hash_id];
2059 	bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2060 
2061 	lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2062 
2063 	if (bucket->page_list) {
2064 		vm_page_packed_t *mp = &bucket->page_list;
2065 		vm_page_t m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp));
2066 
2067 		do {
2068 			/*
2069 			 * compare packed object pointers
2070 			 */
2071 			if (m->vmp_object == mem->vmp_object && m->vmp_offset == offset) {
2072 				/*
2073 				 * Remove old page from hash list
2074 				 */
2075 				*mp = m->vmp_next_m;
2076 				m->vmp_hashed = FALSE;
2077 				m->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2078 
2079 				found_m = m;
2080 				break;
2081 			}
2082 			mp = &m->vmp_next_m;
2083 		} while ((m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp))));
2084 
2085 		mem->vmp_next_m = bucket->page_list;
2086 	} else {
2087 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2088 	}
2089 	/*
2090 	 * insert new page at head of hash list
2091 	 */
2092 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
2093 	mem->vmp_hashed = TRUE;
2094 
2095 	lck_spin_unlock(bucket_lock);
2096 
2097 	if (found_m) {
2098 		/*
2099 		 * there was already a page at the specified
2100 		 * offset for this object... remove it from
2101 		 * the object and free it back to the free list
2102 		 */
2103 		vm_page_free_unlocked(found_m, FALSE);
2104 	}
2105 	vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL);
2106 }
2107 
2108 /*
2109  *	vm_page_remove:		[ internal use only ]
2110  *
2111  *	Removes the given mem entry from the object/offset-page
2112  *	table and the object page list.
2113  *
2114  *	The object must be locked.
2115  */
2116 
2117 void
vm_page_remove(vm_page_t mem,boolean_t remove_from_hash)2118 vm_page_remove(
2119 	vm_page_t       mem,
2120 	boolean_t       remove_from_hash)
2121 {
2122 	vm_page_bucket_t *bucket;
2123 	vm_page_t       this;
2124 	lck_spin_t      *bucket_lock;
2125 	int             hash_id;
2126 	task_t          owner;
2127 	vm_object_t     m_object;
2128 	int             ledger_idx_volatile;
2129 	int             ledger_idx_nonvolatile;
2130 	int             ledger_idx_volatile_compressed;
2131 	int             ledger_idx_nonvolatile_compressed;
2132 	int             do_footprint;
2133 
2134 	m_object = VM_PAGE_OBJECT(mem);
2135 
2136 	vm_object_lock_assert_exclusive(m_object);
2137 	assert(mem->vmp_tabled);
2138 	assert(!mem->vmp_cleaning);
2139 	assert(!mem->vmp_laundry);
2140 
2141 	if (VM_PAGE_PAGEABLE(mem)) {
2142 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2143 	}
2144 #if 0
2145 	/*
2146 	 * we don't hold the page queue lock
2147 	 * so this check isn't safe to make
2148 	 */
2149 	VM_PAGE_CHECK(mem);
2150 #endif
2151 	if (remove_from_hash == TRUE) {
2152 		/*
2153 		 *	Remove from the object_object/offset hash table
2154 		 */
2155 		hash_id = vm_page_hash(m_object, mem->vmp_offset);
2156 		bucket = &vm_page_buckets[hash_id];
2157 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2158 
2159 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2160 
2161 		if ((this = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))) == mem) {
2162 			/* optimize for common case */
2163 
2164 			bucket->page_list = mem->vmp_next_m;
2165 		} else {
2166 			vm_page_packed_t        *prev;
2167 
2168 			for (prev = &this->vmp_next_m;
2169 			    (this = (vm_page_t)(VM_PAGE_UNPACK_PTR(*prev))) != mem;
2170 			    prev = &this->vmp_next_m) {
2171 				continue;
2172 			}
2173 			*prev = this->vmp_next_m;
2174 		}
2175 #if     MACH_PAGE_HASH_STATS
2176 		bucket->cur_count--;
2177 #endif /* MACH_PAGE_HASH_STATS */
2178 		mem->vmp_hashed = FALSE;
2179 		this->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
2180 		lck_spin_unlock(bucket_lock);
2181 	}
2182 	/*
2183 	 *	Now remove from the object's list of backed pages.
2184 	 */
2185 
2186 	vm_page_remove_internal(mem);
2187 
2188 	/*
2189 	 *	And show that the object has one fewer resident
2190 	 *	page.
2191 	 */
2192 
2193 	assert(m_object->resident_page_count > 0);
2194 	m_object->resident_page_count--;
2195 
2196 #if DEVELOPMENT || DEBUG
2197 	if (m_object->object_is_shared_cache &&
2198 	    m_object->pager != NULL &&
2199 	    m_object->pager->mo_pager_ops == &shared_region_pager_ops) {
2200 		assert(!m_object->internal);
2201 		OSAddAtomic(-1, &shared_region_pagers_resident_count);
2202 	}
2203 #endif /* DEVELOPMENT || DEBUG */
2204 
2205 	if (m_object->internal) {
2206 #if DEBUG
2207 		assert(vm_page_internal_count);
2208 #endif /* DEBUG */
2209 
2210 		OSAddAtomic(-1, &vm_page_internal_count);
2211 	} else {
2212 		assert(vm_page_external_count);
2213 		OSAddAtomic(-1, &vm_page_external_count);
2214 
2215 		if (mem->vmp_xpmapped) {
2216 			assert(vm_page_xpmapped_external_count);
2217 			OSAddAtomic(-1, &vm_page_xpmapped_external_count);
2218 		}
2219 	}
2220 	if (!m_object->internal &&
2221 	    m_object->cached_list.next &&
2222 	    m_object->cached_list.prev) {
2223 		if (m_object->resident_page_count == 0) {
2224 			vm_object_cache_remove(m_object);
2225 		}
2226 	}
2227 
2228 	if (VM_PAGE_WIRED(mem)) {
2229 		assert(mem->vmp_wire_count > 0);
2230 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
2231 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
2232 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
2233 	}
2234 	assert(m_object->resident_page_count >=
2235 	    m_object->wired_page_count);
2236 	if (mem->vmp_reusable) {
2237 		assert(m_object->reusable_page_count > 0);
2238 		m_object->reusable_page_count--;
2239 		assert(m_object->reusable_page_count <=
2240 		    m_object->resident_page_count);
2241 		mem->vmp_reusable = FALSE;
2242 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2243 		vm_page_stats_reusable.reused_remove++;
2244 	} else if (m_object->all_reusable) {
2245 		OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count);
2246 		vm_page_stats_reusable.reused_remove++;
2247 	}
2248 
2249 	if (m_object->purgable == VM_PURGABLE_DENY &&
2250 	    !m_object->vo_ledger_tag) {
2251 		owner = TASK_NULL;
2252 	} else {
2253 		owner = VM_OBJECT_OWNER(m_object);
2254 		vm_object_ledger_tag_ledgers(m_object,
2255 		    &ledger_idx_volatile,
2256 		    &ledger_idx_nonvolatile,
2257 		    &ledger_idx_volatile_compressed,
2258 		    &ledger_idx_nonvolatile_compressed,
2259 		    &do_footprint);
2260 	}
2261 	if (owner &&
2262 	    (m_object->purgable == VM_PURGABLE_NONVOLATILE ||
2263 	    m_object->purgable == VM_PURGABLE_DENY ||
2264 	    VM_PAGE_WIRED(mem))) {
2265 		/* less non-volatile bytes */
2266 		ledger_debit(owner->ledger,
2267 		    ledger_idx_nonvolatile,
2268 		    PAGE_SIZE);
2269 		if (do_footprint) {
2270 			/* less footprint */
2271 			ledger_debit(owner->ledger,
2272 			    task_ledgers.phys_footprint,
2273 			    PAGE_SIZE);
2274 		}
2275 	} else if (owner &&
2276 	    (m_object->purgable == VM_PURGABLE_VOLATILE ||
2277 	    m_object->purgable == VM_PURGABLE_EMPTY)) {
2278 		assert(!VM_PAGE_WIRED(mem));
2279 		/* less volatile bytes */
2280 		ledger_debit(owner->ledger,
2281 		    ledger_idx_volatile,
2282 		    PAGE_SIZE);
2283 	}
2284 	if (m_object->purgable == VM_PURGABLE_VOLATILE) {
2285 		if (VM_PAGE_WIRED(mem)) {
2286 			assert(vm_page_purgeable_wired_count > 0);
2287 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
2288 		} else {
2289 			assert(vm_page_purgeable_count > 0);
2290 			OSAddAtomic(-1, &vm_page_purgeable_count);
2291 		}
2292 	}
2293 
2294 	if (m_object->set_cache_attr == TRUE) {
2295 		pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(mem), 0);
2296 	}
2297 
2298 	mem->vmp_tabled = FALSE;
2299 	mem->vmp_object = 0;
2300 	mem->vmp_offset = (vm_object_offset_t) -1;
2301 }
2302 
2303 
2304 /*
2305  *	vm_page_lookup:
2306  *
2307  *	Returns the page associated with the object/offset
2308  *	pair specified; if none is found, VM_PAGE_NULL is returned.
2309  *
2310  *	The object must be locked.  No side effects.
2311  */
2312 
2313 #define VM_PAGE_HASH_LOOKUP_THRESHOLD   10
2314 
2315 #if DEBUG_VM_PAGE_LOOKUP
2316 
2317 struct {
2318 	uint64_t        vpl_total;
2319 	uint64_t        vpl_empty_obj;
2320 	uint64_t        vpl_bucket_NULL;
2321 	uint64_t        vpl_hit_hint;
2322 	uint64_t        vpl_hit_hint_next;
2323 	uint64_t        vpl_hit_hint_prev;
2324 	uint64_t        vpl_fast;
2325 	uint64_t        vpl_slow;
2326 	uint64_t        vpl_hit;
2327 	uint64_t        vpl_miss;
2328 
2329 	uint64_t        vpl_fast_elapsed;
2330 	uint64_t        vpl_slow_elapsed;
2331 } vm_page_lookup_stats __attribute__((aligned(8)));
2332 
2333 #endif
2334 
2335 #define KDP_VM_PAGE_WALK_MAX    1000
2336 
2337 vm_page_t
kdp_vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2338 kdp_vm_page_lookup(
2339 	vm_object_t             object,
2340 	vm_object_offset_t      offset)
2341 {
2342 	vm_page_t cur_page;
2343 	int num_traversed = 0;
2344 
2345 	if (not_in_kdp) {
2346 		panic("panic: kdp_vm_page_lookup done outside of kernel debugger");
2347 	}
2348 
2349 	vm_page_queue_iterate(&object->memq, cur_page, vmp_listq) {
2350 		if (cur_page->vmp_offset == offset) {
2351 			return cur_page;
2352 		}
2353 		num_traversed++;
2354 
2355 		if (num_traversed >= KDP_VM_PAGE_WALK_MAX) {
2356 			return VM_PAGE_NULL;
2357 		}
2358 	}
2359 
2360 	return VM_PAGE_NULL;
2361 }
2362 
2363 vm_page_t
vm_page_lookup(vm_object_t object,vm_object_offset_t offset)2364 vm_page_lookup(
2365 	vm_object_t             object,
2366 	vm_object_offset_t      offset)
2367 {
2368 	vm_page_t       mem;
2369 	vm_page_bucket_t *bucket;
2370 	vm_page_queue_entry_t   qe;
2371 	lck_spin_t      *bucket_lock = NULL;
2372 	int             hash_id;
2373 #if DEBUG_VM_PAGE_LOOKUP
2374 	uint64_t        start, elapsed;
2375 
2376 	OSAddAtomic64(1, &vm_page_lookup_stats.vpl_total);
2377 #endif
2378 
2379 #if CONFIG_KERNEL_TBI
2380 	if (VM_KERNEL_ADDRESS(offset)) {
2381 		offset = VM_KERNEL_STRIP_UPTR(offset);
2382 	}
2383 #endif /* CONFIG_KERNEL_TBI */
2384 
2385 	vm_object_lock_assert_held(object);
2386 	assertf(page_aligned(offset), "offset 0x%llx\n", offset);
2387 
2388 	if (object->resident_page_count == 0) {
2389 #if DEBUG_VM_PAGE_LOOKUP
2390 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_empty_obj);
2391 #endif
2392 		return VM_PAGE_NULL;
2393 	}
2394 
2395 	mem = object->memq_hint;
2396 
2397 	if (mem != VM_PAGE_NULL) {
2398 		assert(VM_PAGE_OBJECT(mem) == object);
2399 
2400 		if (mem->vmp_offset == offset) {
2401 #if DEBUG_VM_PAGE_LOOKUP
2402 			OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint);
2403 #endif
2404 			return mem;
2405 		}
2406 		qe = (vm_page_queue_entry_t)vm_page_queue_next(&mem->vmp_listq);
2407 
2408 		if (!vm_page_queue_end(&object->memq, qe)) {
2409 			vm_page_t       next_page;
2410 
2411 			next_page = (vm_page_t)((uintptr_t)qe);
2412 			assert(VM_PAGE_OBJECT(next_page) == object);
2413 
2414 			if (next_page->vmp_offset == offset) {
2415 				object->memq_hint = next_page; /* new hint */
2416 #if DEBUG_VM_PAGE_LOOKUP
2417 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next);
2418 #endif
2419 				return next_page;
2420 			}
2421 		}
2422 		qe = (vm_page_queue_entry_t)vm_page_queue_prev(&mem->vmp_listq);
2423 
2424 		if (!vm_page_queue_end(&object->memq, qe)) {
2425 			vm_page_t prev_page;
2426 
2427 			prev_page = (vm_page_t)((uintptr_t)qe);
2428 			assert(VM_PAGE_OBJECT(prev_page) == object);
2429 
2430 			if (prev_page->vmp_offset == offset) {
2431 				object->memq_hint = prev_page; /* new hint */
2432 #if DEBUG_VM_PAGE_LOOKUP
2433 				OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev);
2434 #endif
2435 				return prev_page;
2436 			}
2437 		}
2438 	}
2439 	/*
2440 	 * Search the hash table for this object/offset pair
2441 	 */
2442 	hash_id = vm_page_hash(object, offset);
2443 	bucket = &vm_page_buckets[hash_id];
2444 
2445 	/*
2446 	 * since we hold the object lock, we are guaranteed that no
2447 	 * new pages can be inserted into this object... this in turn
2448 	 * guarantess that the page we're looking for can't exist
2449 	 * if the bucket it hashes to is currently NULL even when looked
2450 	 * at outside the scope of the hash bucket lock... this is a
2451 	 * really cheap optimiztion to avoid taking the lock
2452 	 */
2453 	if (!bucket->page_list) {
2454 #if DEBUG_VM_PAGE_LOOKUP
2455 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL);
2456 #endif
2457 		return VM_PAGE_NULL;
2458 	}
2459 
2460 #if DEBUG_VM_PAGE_LOOKUP
2461 	start = mach_absolute_time();
2462 #endif
2463 	if (object->resident_page_count <= VM_PAGE_HASH_LOOKUP_THRESHOLD) {
2464 		/*
2465 		 * on average, it's roughly 3 times faster to run a short memq list
2466 		 * than to take the spin lock and go through the hash list
2467 		 */
2468 		mem = (vm_page_t)vm_page_queue_first(&object->memq);
2469 
2470 		while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2471 			if (mem->vmp_offset == offset) {
2472 				break;
2473 			}
2474 
2475 			mem = (vm_page_t)vm_page_queue_next(&mem->vmp_listq);
2476 		}
2477 		if (vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) {
2478 			mem = NULL;
2479 		}
2480 	} else {
2481 		vm_page_object_t        packed_object;
2482 
2483 		packed_object = VM_PAGE_PACK_OBJECT(object);
2484 
2485 		bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK];
2486 
2487 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
2488 
2489 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
2490 		    mem != VM_PAGE_NULL;
2491 		    mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m))) {
2492 #if 0
2493 			/*
2494 			 * we don't hold the page queue lock
2495 			 * so this check isn't safe to make
2496 			 */
2497 			VM_PAGE_CHECK(mem);
2498 #endif
2499 			if ((mem->vmp_object == packed_object) && (mem->vmp_offset == offset)) {
2500 				break;
2501 			}
2502 		}
2503 		lck_spin_unlock(bucket_lock);
2504 	}
2505 
2506 #if DEBUG_VM_PAGE_LOOKUP
2507 	elapsed = mach_absolute_time() - start;
2508 
2509 	if (bucket_lock) {
2510 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_slow);
2511 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_slow_elapsed);
2512 	} else {
2513 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_fast);
2514 		OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_fast_elapsed);
2515 	}
2516 	if (mem != VM_PAGE_NULL) {
2517 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit);
2518 	} else {
2519 		OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss);
2520 	}
2521 #endif
2522 	if (mem != VM_PAGE_NULL) {
2523 		assert(VM_PAGE_OBJECT(mem) == object);
2524 
2525 		object->memq_hint = mem;
2526 	}
2527 	return mem;
2528 }
2529 
2530 
2531 /*
2532  *	vm_page_rename:
2533  *
2534  *	Move the given memory entry from its
2535  *	current object to the specified target object/offset.
2536  *
2537  *	The object must be locked.
2538  */
2539 void
vm_page_rename(vm_page_t mem,vm_object_t new_object,vm_object_offset_t new_offset)2540 vm_page_rename(
2541 	vm_page_t               mem,
2542 	vm_object_t             new_object,
2543 	vm_object_offset_t      new_offset)
2544 {
2545 	boolean_t       internal_to_external, external_to_internal;
2546 	vm_tag_t        tag;
2547 	vm_object_t     m_object;
2548 
2549 	m_object = VM_PAGE_OBJECT(mem);
2550 
2551 	assert(m_object != new_object);
2552 	assert(m_object);
2553 
2554 	/*
2555 	 *	Changes to mem->vmp_object require the page lock because
2556 	 *	the pageout daemon uses that lock to get the object.
2557 	 */
2558 	vm_page_lockspin_queues();
2559 
2560 	internal_to_external = FALSE;
2561 	external_to_internal = FALSE;
2562 
2563 	if (mem->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q) {
2564 		/*
2565 		 * it's much easier to get the vm_page_pageable_xxx accounting correct
2566 		 * if we first move the page to the active queue... it's going to end
2567 		 * up there anyway, and we don't do vm_page_rename's frequently enough
2568 		 * for this to matter.
2569 		 */
2570 		vm_page_queues_remove(mem, FALSE);
2571 		vm_page_activate(mem);
2572 	}
2573 	if (VM_PAGE_PAGEABLE(mem)) {
2574 		if (m_object->internal && !new_object->internal) {
2575 			internal_to_external = TRUE;
2576 		}
2577 		if (!m_object->internal && new_object->internal) {
2578 			external_to_internal = TRUE;
2579 		}
2580 	}
2581 
2582 	tag = m_object->wire_tag;
2583 	vm_page_remove(mem, TRUE);
2584 	vm_page_insert_internal(mem, new_object, new_offset, tag, TRUE, TRUE, FALSE, FALSE, NULL);
2585 
2586 	if (internal_to_external) {
2587 		vm_page_pageable_internal_count--;
2588 		vm_page_pageable_external_count++;
2589 	} else if (external_to_internal) {
2590 		vm_page_pageable_external_count--;
2591 		vm_page_pageable_internal_count++;
2592 	}
2593 
2594 	vm_page_unlock_queues();
2595 }
2596 
2597 /*
2598  *	vm_page_init:
2599  *
2600  *	Initialize the fields in a new page.
2601  *	This takes a structure with random values and initializes it
2602  *	so that it can be given to vm_page_release or vm_page_insert.
2603  */
2604 void
vm_page_init(vm_page_t mem,ppnum_t phys_page,boolean_t lopage)2605 vm_page_init(
2606 	vm_page_t mem,
2607 	ppnum_t   phys_page,
2608 	boolean_t lopage)
2609 {
2610 	uint_t    i;
2611 	uintptr_t *p;
2612 
2613 	assert(phys_page);
2614 
2615 #if DEBUG
2616 	if ((phys_page != vm_page_fictitious_addr) && (phys_page != vm_page_guard_addr)) {
2617 		if (!(pmap_valid_page(phys_page))) {
2618 			panic("vm_page_init: non-DRAM phys_page 0x%x", phys_page);
2619 		}
2620 	}
2621 #endif /* DEBUG */
2622 
2623 	/*
2624 	 * Initialize the fields of the vm_page. If adding any new fields to vm_page,
2625 	 * try to use initial values which match 0. This minimizes the number of writes
2626 	 * needed for boot-time initialization.
2627 	 *
2628 	 * Kernel bzero() isn't an inline yet, so do it by hand for performance.
2629 	 */
2630 	assert(VM_PAGE_NOT_ON_Q == 0);
2631 	assert(sizeof(*mem) % sizeof(uintptr_t) == 0);
2632 	for (p = (uintptr_t *)(void *)mem, i = sizeof(*mem) / sizeof(uintptr_t); i != 0; --i) {
2633 		*p++ = 0;
2634 	}
2635 	mem->vmp_offset = (vm_object_offset_t)-1;
2636 	mem->vmp_busy = TRUE;
2637 	mem->vmp_lopage = lopage;
2638 
2639 	VM_PAGE_SET_PHYS_PAGE(mem, phys_page);
2640 #if 0
2641 	/*
2642 	 * we're leaving this turned off for now... currently pages
2643 	 * come off the free list and are either immediately dirtied/referenced
2644 	 * due to zero-fill or COW faults, or are used to read or write files...
2645 	 * in the file I/O case, the UPL mechanism takes care of clearing
2646 	 * the state of the HW ref/mod bits in a somewhat fragile way.
2647 	 * Since we may change the way this works in the future (to toughen it up),
2648 	 * I'm leaving this as a reminder of where these bits could get cleared
2649 	 */
2650 
2651 	/*
2652 	 * make sure both the h/w referenced and modified bits are
2653 	 * clear at this point... we are especially dependent on
2654 	 * not finding a 'stale' h/w modified in a number of spots
2655 	 * once this page goes back into use
2656 	 */
2657 	pmap_clear_refmod(phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
2658 #endif
2659 }
2660 
2661 /*
2662  *	vm_page_grab_fictitious:
2663  *
2664  *	Remove a fictitious page from the free list.
2665  *	Returns VM_PAGE_NULL if there are no free pages.
2666  */
2667 
2668 static vm_page_t
vm_page_grab_fictitious_common(ppnum_t phys_addr,boolean_t canwait)2669 vm_page_grab_fictitious_common(ppnum_t phys_addr, boolean_t canwait)
2670 {
2671 	vm_page_t m;
2672 
2673 	m = zalloc_flags(vm_page_zone, canwait ? Z_WAITOK : Z_NOWAIT);
2674 	if (m) {
2675 		vm_page_init(m, phys_addr, FALSE);
2676 		m->vmp_fictitious = TRUE;
2677 	}
2678 	return m;
2679 }
2680 
2681 vm_page_t
vm_page_grab_fictitious(boolean_t canwait)2682 vm_page_grab_fictitious(boolean_t canwait)
2683 {
2684 	return vm_page_grab_fictitious_common(vm_page_fictitious_addr, canwait);
2685 }
2686 
2687 int vm_guard_count;
2688 
2689 
2690 vm_page_t
vm_page_grab_guard(boolean_t canwait)2691 vm_page_grab_guard(boolean_t canwait)
2692 {
2693 	vm_page_t page;
2694 	page = vm_page_grab_fictitious_common(vm_page_guard_addr, canwait);
2695 	if (page) {
2696 		OSAddAtomic(1, &vm_guard_count);
2697 	}
2698 	return page;
2699 }
2700 
2701 
2702 /*
2703  *	vm_page_release_fictitious:
2704  *
2705  *	Release a fictitious page to the zone pool
2706  */
2707 void
vm_page_release_fictitious(vm_page_t m)2708 vm_page_release_fictitious(
2709 	vm_page_t m)
2710 {
2711 	assert((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || (m->vmp_q_state == VM_PAGE_IS_WIRED));
2712 	assert(m->vmp_fictitious);
2713 	assert(VM_PAGE_GET_PHYS_PAGE(m) == vm_page_fictitious_addr ||
2714 	    VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr);
2715 	assert(!m->vmp_realtime);
2716 
2717 	if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) {
2718 		OSAddAtomic(-1, &vm_guard_count);
2719 	}
2720 
2721 	zfree(vm_page_zone, m);
2722 }
2723 
2724 /*
2725  *	vm_pool_low():
2726  *
2727  *	Return true if it is not likely that a non-vm_privileged thread
2728  *	can get memory without blocking.  Advisory only, since the
2729  *	situation may change under us.
2730  */
2731 bool
vm_pool_low(void)2732 vm_pool_low(void)
2733 {
2734 	/* No locking, at worst we will fib. */
2735 	return vm_page_free_count <= vm_page_free_reserved;
2736 }
2737 
2738 boolean_t vm_darkwake_mode = FALSE;
2739 
2740 /*
2741  * vm_update_darkwake_mode():
2742  *
2743  * Tells the VM that the system is in / out of darkwake.
2744  *
2745  * Today, the VM only lowers/raises the background queue target
2746  * so as to favor consuming more/less background pages when
2747  * darwake is ON/OFF.
2748  *
2749  * We might need to do more things in the future.
2750  */
2751 
2752 void
vm_update_darkwake_mode(boolean_t darkwake_mode)2753 vm_update_darkwake_mode(boolean_t darkwake_mode)
2754 {
2755 #if XNU_TARGET_OS_OSX && defined(__arm64__)
2756 #pragma unused(darkwake_mode)
2757 	assert(vm_darkwake_mode == FALSE);
2758 	/*
2759 	 * Darkwake mode isn't supported for AS macOS.
2760 	 */
2761 	return;
2762 #else /* XNU_TARGET_OS_OSX && __arm64__ */
2763 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
2764 
2765 	vm_page_lockspin_queues();
2766 
2767 	if (vm_darkwake_mode == darkwake_mode) {
2768 		/*
2769 		 * No change.
2770 		 */
2771 		vm_page_unlock_queues();
2772 		return;
2773 	}
2774 
2775 	vm_darkwake_mode = darkwake_mode;
2776 
2777 	if (vm_darkwake_mode == TRUE) {
2778 		/* save background target to restore later */
2779 		vm_page_background_target_snapshot = vm_page_background_target;
2780 
2781 		/* target is set to 0...no protection for background pages */
2782 		vm_page_background_target = 0;
2783 	} else if (vm_darkwake_mode == FALSE) {
2784 		if (vm_page_background_target_snapshot) {
2785 			vm_page_background_target = vm_page_background_target_snapshot;
2786 		}
2787 	}
2788 	vm_page_unlock_queues();
2789 #endif
2790 }
2791 
2792 void
vm_page_update_special_state(vm_page_t mem)2793 vm_page_update_special_state(vm_page_t mem)
2794 {
2795 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR || mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
2796 		return;
2797 	}
2798 
2799 	int mode = mem->vmp_on_specialq;
2800 
2801 	switch (mode) {
2802 	case VM_PAGE_SPECIAL_Q_BG:
2803 	{
2804 		task_t  my_task = current_task_early();
2805 
2806 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2807 			return;
2808 		}
2809 
2810 		if (my_task) {
2811 			if (task_get_darkwake_mode(my_task)) {
2812 				return;
2813 			}
2814 		}
2815 
2816 		if (my_task) {
2817 			if (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG)) {
2818 				return;
2819 			}
2820 		}
2821 		vm_page_lockspin_queues();
2822 
2823 		vm_page_background_promoted_count++;
2824 
2825 		vm_page_remove_from_specialq(mem);
2826 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2827 
2828 		vm_page_unlock_queues();
2829 		break;
2830 	}
2831 
2832 	case VM_PAGE_SPECIAL_Q_DONATE:
2833 	{
2834 		task_t  my_task = current_task_early();
2835 
2836 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2837 			return;
2838 		}
2839 
2840 		if (my_task->donates_own_pages == false) {
2841 			vm_page_lockspin_queues();
2842 
2843 			vm_page_remove_from_specialq(mem);
2844 			mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
2845 
2846 			vm_page_unlock_queues();
2847 		}
2848 		break;
2849 	}
2850 
2851 	default:
2852 	{
2853 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2854 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2855 		break;
2856 	}
2857 	}
2858 }
2859 
2860 
2861 void
vm_page_assign_special_state(vm_page_t mem,int mode)2862 vm_page_assign_special_state(vm_page_t mem, int mode)
2863 {
2864 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
2865 		return;
2866 	}
2867 
2868 	switch (mode) {
2869 	case VM_PAGE_SPECIAL_Q_BG:
2870 	{
2871 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2872 			return;
2873 		}
2874 
2875 		task_t  my_task = current_task_early();
2876 
2877 		if (my_task) {
2878 			if (task_get_darkwake_mode(my_task)) {
2879 				mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
2880 				return;
2881 			}
2882 		}
2883 
2884 		if (my_task) {
2885 			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);
2886 		}
2887 		break;
2888 	}
2889 
2890 	case VM_PAGE_SPECIAL_Q_DONATE:
2891 	{
2892 		if (vm_page_donate_mode == VM_PAGE_DONATE_DISABLED) {
2893 			return;
2894 		}
2895 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
2896 		break;
2897 	}
2898 
2899 	default:
2900 		break;
2901 	}
2902 }
2903 
2904 
2905 void
vm_page_remove_from_specialq(vm_page_t mem)2906 vm_page_remove_from_specialq(
2907 	vm_page_t       mem)
2908 {
2909 	vm_object_t     m_object;
2910 	unsigned short  mode;
2911 
2912 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2913 
2914 	mode = mem->vmp_on_specialq;
2915 
2916 	switch (mode) {
2917 	case VM_PAGE_SPECIAL_Q_BG:
2918 	{
2919 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2920 			vm_page_queue_remove(&vm_page_queue_background, mem, vmp_specialq);
2921 
2922 			mem->vmp_specialq.next = 0;
2923 			mem->vmp_specialq.prev = 0;
2924 
2925 			vm_page_background_count--;
2926 
2927 			m_object = VM_PAGE_OBJECT(mem);
2928 
2929 			if (m_object->internal) {
2930 				vm_page_background_internal_count--;
2931 			} else {
2932 				vm_page_background_external_count--;
2933 			}
2934 		}
2935 		break;
2936 	}
2937 
2938 	case VM_PAGE_SPECIAL_Q_DONATE:
2939 	{
2940 		if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2941 			vm_page_queue_remove((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
2942 			mem->vmp_specialq.next = 0;
2943 			mem->vmp_specialq.prev = 0;
2944 			vm_page_donate_count--;
2945 			if (vm_page_donate_queue_ripe && (vm_page_donate_count < vm_page_donate_target)) {
2946 				assert(vm_page_donate_target == vm_page_donate_target_low);
2947 				vm_page_donate_target = vm_page_donate_target_high;
2948 				vm_page_donate_queue_ripe = false;
2949 			}
2950 		}
2951 
2952 		break;
2953 	}
2954 
2955 	default:
2956 	{
2957 		assert(VM_PAGE_UNPACK_PTR(mem->vmp_specialq.next) == (uintptr_t)NULL &&
2958 		    VM_PAGE_UNPACK_PTR(mem->vmp_specialq.prev) == (uintptr_t)NULL);
2959 		break;
2960 	}
2961 	}
2962 }
2963 
2964 
2965 void
vm_page_add_to_specialq(vm_page_t mem,boolean_t first)2966 vm_page_add_to_specialq(
2967 	vm_page_t       mem,
2968 	boolean_t       first)
2969 {
2970 	vm_object_t     m_object;
2971 
2972 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
2973 
2974 	if (mem->vmp_specialq.next && mem->vmp_specialq.prev) {
2975 		return;
2976 	}
2977 
2978 	int mode = mem->vmp_on_specialq;
2979 
2980 	switch (mode) {
2981 	case VM_PAGE_SPECIAL_Q_BG:
2982 	{
2983 		if (vm_page_background_mode == VM_PAGE_BG_DISABLED) {
2984 			return;
2985 		}
2986 
2987 		m_object = VM_PAGE_OBJECT(mem);
2988 
2989 		if (vm_page_background_exclude_external && !m_object->internal) {
2990 			return;
2991 		}
2992 
2993 		if (first == TRUE) {
2994 			vm_page_queue_enter_first(&vm_page_queue_background, mem, vmp_specialq);
2995 		} else {
2996 			vm_page_queue_enter(&vm_page_queue_background, mem, vmp_specialq);
2997 		}
2998 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_BG;
2999 
3000 		vm_page_background_count++;
3001 
3002 		if (m_object->internal) {
3003 			vm_page_background_internal_count++;
3004 		} else {
3005 			vm_page_background_external_count++;
3006 		}
3007 		break;
3008 	}
3009 
3010 	case VM_PAGE_SPECIAL_Q_DONATE:
3011 	{
3012 		if (first == TRUE) {
3013 			vm_page_queue_enter_first((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3014 		} else {
3015 			vm_page_queue_enter((vm_page_queue_head_t*)&vm_page_queue_donate, mem, vmp_specialq);
3016 		}
3017 		vm_page_donate_count++;
3018 		if (!vm_page_donate_queue_ripe && (vm_page_donate_count > vm_page_donate_target)) {
3019 			assert(vm_page_donate_target == vm_page_donate_target_high);
3020 			vm_page_donate_target = vm_page_donate_target_low;
3021 			vm_page_donate_queue_ripe = true;
3022 		}
3023 		mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_DONATE;
3024 		break;
3025 	}
3026 
3027 	default:
3028 		break;
3029 	}
3030 }
3031 
3032 /*
3033  * This can be switched to FALSE to help debug drivers
3034  * that are having problems with memory > 4G.
3035  */
3036 boolean_t       vm_himemory_mode = TRUE;
3037 
3038 /*
3039  * this interface exists to support hardware controllers
3040  * incapable of generating DMAs with more than 32 bits
3041  * of address on platforms with physical memory > 4G...
3042  */
3043 unsigned int    vm_lopages_allocated_q = 0;
3044 unsigned int    vm_lopages_allocated_cpm_success = 0;
3045 unsigned int    vm_lopages_allocated_cpm_failed = 0;
3046 vm_page_queue_head_t    vm_lopage_queue_free VM_PAGE_PACKED_ALIGNED;
3047 
3048 vm_page_t
vm_page_grablo(void)3049 vm_page_grablo(void)
3050 {
3051 	vm_page_t       mem;
3052 
3053 	if (vm_lopage_needed == FALSE) {
3054 		return vm_page_grab();
3055 	}
3056 
3057 	vm_free_page_lock_spin();
3058 
3059 	if (!vm_page_queue_empty(&vm_lopage_queue_free)) {
3060 		vm_page_queue_remove_first(&vm_lopage_queue_free, mem, vmp_pageq);
3061 		assert(vm_lopage_free_count);
3062 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
3063 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3064 
3065 		vm_lopage_free_count--;
3066 		vm_lopages_allocated_q++;
3067 
3068 		if (vm_lopage_free_count < vm_lopage_lowater) {
3069 			vm_lopage_refill = TRUE;
3070 		}
3071 
3072 		vm_free_page_unlock();
3073 
3074 		if (current_task()->donates_own_pages) {
3075 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3076 		} else {
3077 			vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3078 		}
3079 	} else {
3080 		vm_free_page_unlock();
3081 
3082 		if (cpm_allocate(PAGE_SIZE, &mem, atop(PPNUM_MAX), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) {
3083 			vm_free_page_lock_spin();
3084 			vm_lopages_allocated_cpm_failed++;
3085 			vm_free_page_unlock();
3086 
3087 			return VM_PAGE_NULL;
3088 		}
3089 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3090 
3091 		mem->vmp_busy = TRUE;
3092 
3093 		vm_page_lockspin_queues();
3094 
3095 		mem->vmp_gobbled = FALSE;
3096 		vm_page_gobble_count--;
3097 		vm_page_wire_count--;
3098 
3099 		vm_lopages_allocated_cpm_success++;
3100 		vm_page_unlock_queues();
3101 	}
3102 	assert(mem->vmp_busy);
3103 	assert(!mem->vmp_pmapped);
3104 	assert(!mem->vmp_wpmapped);
3105 	assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3106 
3107 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3108 
3109 	counter_inc(&vm_page_grab_count);
3110 	VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, 0, 1, 0, 0);
3111 
3112 	return mem;
3113 }
3114 
3115 /*
3116  *	vm_page_grab:
3117  *
3118  *	first try to grab a page from the per-cpu free list...
3119  *	this must be done while pre-emption is disabled... if
3120  *      a page is available, we're done...
3121  *	if no page is available, grab the vm_page_queue_free_lock
3122  *	and see if current number of free pages would allow us
3123  *      to grab at least 1... if not, return VM_PAGE_NULL as before...
3124  *	if there are pages available, disable preemption and
3125  *      recheck the state of the per-cpu free list... we could
3126  *	have been preempted and moved to a different cpu, or
3127  *      some other thread could have re-filled it... if still
3128  *	empty, figure out how many pages we can steal from the
3129  *	global free queue and move to the per-cpu queue...
3130  *	return 1 of these pages when done... only wakeup the
3131  *      pageout_scan thread if we moved pages from the global
3132  *	list... no need for the wakeup if we've satisfied the
3133  *	request from the per-cpu queue.
3134  */
3135 
3136 #if CONFIG_SECLUDED_MEMORY
3137 vm_page_t vm_page_grab_secluded(void);
3138 #endif /* CONFIG_SECLUDED_MEMORY */
3139 
3140 static inline void
3141 vm_page_grab_diags(void);
3142 
3143 vm_page_t
vm_page_grab(void)3144 vm_page_grab(void)
3145 {
3146 	return vm_page_grab_options(VM_PAGE_GRAB_OPTIONS_NONE);
3147 }
3148 
3149 #if HIBERNATION
3150 boolean_t       hibernate_rebuild_needed = FALSE;
3151 #endif /* HIBERNATION */
3152 
3153 vm_page_t
vm_page_grab_options(int grab_options)3154 vm_page_grab_options(
3155 	int grab_options)
3156 {
3157 	vm_page_t       mem;
3158 
3159 restart:
3160 	disable_preemption();
3161 
3162 	if ((mem = *PERCPU_GET(free_pages))) {
3163 		assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
3164 
3165 #if HIBERNATION
3166 		if (hibernate_rebuild_needed) {
3167 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3168 		}
3169 #endif /* HIBERNATION */
3170 
3171 		vm_page_grab_diags();
3172 
3173 		vm_offset_t pcpu_base = current_percpu_base();
3174 		counter_inc_preemption_disabled(&vm_page_grab_count);
3175 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = mem->vmp_snext;
3176 		VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3177 
3178 		VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3179 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
3180 		enable_preemption();
3181 
3182 		assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3183 		assert(mem->vmp_tabled == FALSE);
3184 		assert(mem->vmp_object == 0);
3185 		assert(!mem->vmp_laundry);
3186 		ASSERT_PMAP_FREE(mem);
3187 		assert(mem->vmp_busy);
3188 		assert(!mem->vmp_pmapped);
3189 		assert(!mem->vmp_wpmapped);
3190 		assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3191 		assert(!mem->vmp_realtime);
3192 
3193 		task_t  cur_task = current_task_early();
3194 		if (cur_task && cur_task != kernel_task) {
3195 			if (cur_task->donates_own_pages) {
3196 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3197 			} else {
3198 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3199 			}
3200 		}
3201 		return mem;
3202 	}
3203 	enable_preemption();
3204 
3205 
3206 	/*
3207 	 *	Optionally produce warnings if the wire or gobble
3208 	 *	counts exceed some threshold.
3209 	 */
3210 #if VM_PAGE_WIRE_COUNT_WARNING
3211 	if (vm_page_wire_count >= VM_PAGE_WIRE_COUNT_WARNING) {
3212 		printf("mk: vm_page_grab(): high wired page count of %d\n",
3213 		    vm_page_wire_count);
3214 	}
3215 #endif
3216 #if VM_PAGE_GOBBLE_COUNT_WARNING
3217 	if (vm_page_gobble_count >= VM_PAGE_GOBBLE_COUNT_WARNING) {
3218 		printf("mk: vm_page_grab(): high gobbled page count of %d\n",
3219 		    vm_page_gobble_count);
3220 	}
3221 #endif
3222 
3223 	/*
3224 	 * If free count is low and we have delayed pages from early boot,
3225 	 * get one of those instead.
3226 	 */
3227 	if (__improbable(vm_delayed_count > 0 &&
3228 	    vm_page_free_count <= vm_page_free_target &&
3229 	    (mem = vm_get_delayed_page(grab_options)) != NULL)) {
3230 		assert(!mem->vmp_realtime);
3231 		return mem;
3232 	}
3233 
3234 	vm_free_page_lock_spin();
3235 
3236 	/*
3237 	 *	Only let privileged threads (involved in pageout)
3238 	 *	dip into the reserved pool.
3239 	 */
3240 	if ((vm_page_free_count < vm_page_free_reserved) &&
3241 	    !(current_thread()->options & TH_OPT_VMPRIV)) {
3242 		/* no page for us in the free queue... */
3243 		vm_free_page_unlock();
3244 		mem = VM_PAGE_NULL;
3245 
3246 #if CONFIG_SECLUDED_MEMORY
3247 		/* ... but can we try and grab from the secluded queue? */
3248 		if (vm_page_secluded_count > 0 &&
3249 		    ((grab_options & VM_PAGE_GRAB_SECLUDED) ||
3250 		    task_can_use_secluded_mem(current_task(), TRUE))) {
3251 			mem = vm_page_grab_secluded();
3252 			if (grab_options & VM_PAGE_GRAB_SECLUDED) {
3253 				vm_page_secluded.grab_for_iokit++;
3254 				if (mem) {
3255 					vm_page_secluded.grab_for_iokit_success++;
3256 				}
3257 			}
3258 			if (mem) {
3259 				VM_CHECK_MEMORYSTATUS;
3260 
3261 				vm_page_grab_diags();
3262 				counter_inc(&vm_page_grab_count);
3263 				VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0);
3264 
3265 				assert(!mem->vmp_realtime);
3266 				return mem;
3267 			}
3268 		}
3269 #else /* CONFIG_SECLUDED_MEMORY */
3270 		(void) grab_options;
3271 #endif /* CONFIG_SECLUDED_MEMORY */
3272 	} else {
3273 		vm_page_t        head;
3274 		vm_page_t        tail;
3275 		unsigned int     pages_to_steal;
3276 		unsigned int     color;
3277 		unsigned int clump_end, sub_count;
3278 
3279 		while (vm_page_free_count == 0) {
3280 			vm_free_page_unlock();
3281 			/*
3282 			 * must be a privileged thread to be
3283 			 * in this state since a non-privileged
3284 			 * thread would have bailed if we were
3285 			 * under the vm_page_free_reserved mark
3286 			 */
3287 			VM_PAGE_WAIT();
3288 			vm_free_page_lock_spin();
3289 		}
3290 
3291 		/*
3292 		 * Need to repopulate the per-CPU free list from the global free list.
3293 		 * Note we don't do any processing of pending retirement pages here.
3294 		 * That'll happen in the code above when the page comes off the per-CPU list.
3295 		 */
3296 		disable_preemption();
3297 
3298 		/*
3299 		 * If we got preempted the cache might now have pages.
3300 		 */
3301 		if ((mem = *PERCPU_GET(free_pages))) {
3302 			vm_free_page_unlock();
3303 			enable_preemption();
3304 			goto restart;
3305 		}
3306 
3307 		if (vm_page_free_count <= vm_page_free_reserved) {
3308 			pages_to_steal = 1;
3309 		} else {
3310 			if (vm_free_magazine_refill_limit <= (vm_page_free_count - vm_page_free_reserved)) {
3311 				pages_to_steal = vm_free_magazine_refill_limit;
3312 			} else {
3313 				pages_to_steal = (vm_page_free_count - vm_page_free_reserved);
3314 			}
3315 		}
3316 		color = *PERCPU_GET(start_color);
3317 		head = tail = NULL;
3318 
3319 		vm_page_free_count -= pages_to_steal;
3320 		clump_end = sub_count = 0;
3321 
3322 		while (pages_to_steal--) {
3323 			while (vm_page_queue_empty(&vm_page_queue_free[color].qhead)) {
3324 				color = (color + 1) & vm_color_mask;
3325 			}
3326 #if defined(__x86_64__)
3327 			vm_page_queue_remove_first_with_clump(&vm_page_queue_free[color].qhead,
3328 			    mem, clump_end);
3329 #else
3330 			vm_page_queue_remove_first(&vm_page_queue_free[color].qhead,
3331 			    mem, vmp_pageq);
3332 #endif
3333 
3334 			assert(mem->vmp_q_state == VM_PAGE_ON_FREE_Q);
3335 
3336 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
3337 
3338 #if defined(__arm64__)
3339 			color = (color + 1) & vm_color_mask;
3340 #else
3341 
3342 #if DEVELOPMENT || DEBUG
3343 
3344 			sub_count++;
3345 			if (clump_end) {
3346 				vm_clump_update_stats(sub_count);
3347 				sub_count = 0;
3348 				color = (color + 1) & vm_color_mask;
3349 			}
3350 #else
3351 			if (clump_end) {
3352 				color = (color + 1) & vm_color_mask;
3353 			}
3354 
3355 #endif /* if DEVELOPMENT || DEBUG */
3356 
3357 #endif  /* if defined(__arm64__) */
3358 
3359 			if (head == NULL) {
3360 				head = mem;
3361 			} else {
3362 				tail->vmp_snext = mem;
3363 			}
3364 			tail = mem;
3365 
3366 			assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3367 			assert(mem->vmp_tabled == FALSE);
3368 			assert(mem->vmp_object == 0);
3369 			assert(!mem->vmp_laundry);
3370 
3371 			mem->vmp_q_state = VM_PAGE_ON_FREE_LOCAL_Q;
3372 
3373 			ASSERT_PMAP_FREE(mem);
3374 			assert(mem->vmp_busy);
3375 			assert(!mem->vmp_pmapped);
3376 			assert(!mem->vmp_wpmapped);
3377 			assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)));
3378 			assert(!mem->vmp_realtime);
3379 		}
3380 #if defined (__x86_64__) && (DEVELOPMENT || DEBUG)
3381 		vm_clump_update_stats(sub_count);
3382 #endif
3383 
3384 #if HIBERNATION
3385 		if (hibernate_rebuild_needed) {
3386 			panic("%s:%d should not modify cpu->free_pages while hibernating", __FUNCTION__, __LINE__);
3387 		}
3388 #endif /* HIBERNATION */
3389 		vm_offset_t pcpu_base = current_percpu_base();
3390 		*PERCPU_GET_WITH_BASE(pcpu_base, free_pages) = head;
3391 		*PERCPU_GET_WITH_BASE(pcpu_base, start_color) = color;
3392 
3393 		vm_free_page_unlock();
3394 		enable_preemption();
3395 		goto restart;
3396 	}
3397 
3398 	/*
3399 	 *	Decide if we should poke the pageout daemon.
3400 	 *	We do this if the free count is less than the low
3401 	 *	water mark. VM Pageout Scan will keep running till
3402 	 *	the free_count > free_target (& hence above free_min).
3403 	 *	This wakeup is to catch the possibility of the counts
3404 	 *	dropping between VM Pageout Scan parking and this check.
3405 	 *
3406 	 *	We don't have the counts locked ... if they change a little,
3407 	 *	it doesn't really matter.
3408 	 */
3409 	if (vm_page_free_count < vm_page_free_min) {
3410 		vm_free_page_lock();
3411 		if (vm_pageout_running == FALSE) {
3412 			vm_free_page_unlock();
3413 			thread_wakeup((event_t) &vm_page_free_wanted);
3414 		} else {
3415 			vm_free_page_unlock();
3416 		}
3417 	}
3418 
3419 	VM_CHECK_MEMORYSTATUS;
3420 
3421 	if (mem) {
3422 		assert(!mem->vmp_realtime);
3423 //		dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 4);	/* (TEST/DEBUG) */
3424 
3425 		task_t  cur_task = current_task_early();
3426 		if (cur_task && cur_task != kernel_task) {
3427 			if (cur_task->donates_own_pages) {
3428 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_DONATE);
3429 			} else {
3430 				vm_page_assign_special_state(mem, VM_PAGE_SPECIAL_Q_BG);
3431 			}
3432 		}
3433 	}
3434 	return mem;
3435 }
3436 
3437 #if CONFIG_SECLUDED_MEMORY
3438 vm_page_t
vm_page_grab_secluded(void)3439 vm_page_grab_secluded(void)
3440 {
3441 	vm_page_t       mem;
3442 	vm_object_t     object;
3443 	int             refmod_state;
3444 
3445 	if (vm_page_secluded_count == 0) {
3446 		/* no secluded pages to grab... */
3447 		return VM_PAGE_NULL;
3448 	}
3449 
3450 	/* secluded queue is protected by the VM page queue lock */
3451 	vm_page_lock_queues();
3452 
3453 	if (vm_page_secluded_count == 0) {
3454 		/* no secluded pages to grab... */
3455 		vm_page_unlock_queues();
3456 		return VM_PAGE_NULL;
3457 	}
3458 
3459 #if 00
3460 	/* can we grab from the secluded queue? */
3461 	if (vm_page_secluded_count > vm_page_secluded_target ||
3462 	    (vm_page_secluded_count > 0 &&
3463 	    task_can_use_secluded_mem(current_task(), TRUE))) {
3464 		/* OK */
3465 	} else {
3466 		/* can't grab from secluded queue... */
3467 		vm_page_unlock_queues();
3468 		return VM_PAGE_NULL;
3469 	}
3470 #endif
3471 
3472 	/* we can grab a page from secluded queue! */
3473 	assert((vm_page_secluded_count_free +
3474 	    vm_page_secluded_count_inuse) ==
3475 	    vm_page_secluded_count);
3476 	if (current_task()->task_can_use_secluded_mem) {
3477 		assert(num_tasks_can_use_secluded_mem > 0);
3478 	}
3479 	assert(!vm_page_queue_empty(&vm_page_queue_secluded));
3480 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3481 	mem = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3482 	assert(mem->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3483 	vm_page_queues_remove(mem, TRUE);
3484 
3485 	object = VM_PAGE_OBJECT(mem);
3486 
3487 	assert(!mem->vmp_fictitious);
3488 	assert(!VM_PAGE_WIRED(mem));
3489 	if (object == VM_OBJECT_NULL) {
3490 		/* free for grab! */
3491 		vm_page_unlock_queues();
3492 		vm_page_secluded.grab_success_free++;
3493 
3494 		assert(mem->vmp_busy);
3495 		assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3496 		assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3497 		assert(mem->vmp_pageq.next == 0);
3498 		assert(mem->vmp_pageq.prev == 0);
3499 		assert(mem->vmp_listq.next == 0);
3500 		assert(mem->vmp_listq.prev == 0);
3501 		assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3502 		assert(mem->vmp_specialq.next == 0);
3503 		assert(mem->vmp_specialq.prev == 0);
3504 		return mem;
3505 	}
3506 
3507 	assert(!object->internal);
3508 //	vm_page_pageable_external_count--;
3509 
3510 	if (!vm_object_lock_try(object)) {
3511 //		printf("SECLUDED: page %p: object %p locked\n", mem, object);
3512 		vm_page_secluded.grab_failure_locked++;
3513 reactivate_secluded_page:
3514 		vm_page_activate(mem);
3515 		vm_page_unlock_queues();
3516 		return VM_PAGE_NULL;
3517 	}
3518 	if (mem->vmp_busy ||
3519 	    mem->vmp_cleaning ||
3520 	    mem->vmp_laundry) {
3521 		/* can't steal page in this state... */
3522 		vm_object_unlock(object);
3523 		vm_page_secluded.grab_failure_state++;
3524 		goto reactivate_secluded_page;
3525 	}
3526 	if (mem->vmp_realtime) {
3527 		/* don't steal pages used by realtime threads... */
3528 		vm_object_unlock(object);
3529 		vm_page_secluded.grab_failure_realtime++;
3530 		goto reactivate_secluded_page;
3531 	}
3532 
3533 	mem->vmp_busy = TRUE;
3534 	refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem));
3535 	if (refmod_state & VM_MEM_REFERENCED) {
3536 		mem->vmp_reference = TRUE;
3537 	}
3538 	if (refmod_state & VM_MEM_MODIFIED) {
3539 		SET_PAGE_DIRTY(mem, FALSE);
3540 	}
3541 	if (mem->vmp_dirty || mem->vmp_precious) {
3542 		/* can't grab a dirty page; re-activate */
3543 //		printf("SECLUDED: dirty page %p\n", mem);
3544 		PAGE_WAKEUP_DONE(mem);
3545 		vm_page_secluded.grab_failure_dirty++;
3546 		vm_object_unlock(object);
3547 		goto reactivate_secluded_page;
3548 	}
3549 	if (mem->vmp_reference) {
3550 		/* it's been used but we do need to grab a page... */
3551 	}
3552 
3553 	vm_page_unlock_queues();
3554 
3555 
3556 	/* finish what vm_page_free() would have done... */
3557 	vm_page_free_prepare_object(mem, TRUE);
3558 	vm_object_unlock(object);
3559 	object = VM_OBJECT_NULL;
3560 	if (vm_page_free_verify) {
3561 		ASSERT_PMAP_FREE(mem);
3562 	}
3563 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3564 	vm_page_secluded.grab_success_other++;
3565 
3566 	assert(mem->vmp_busy);
3567 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3568 	assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL);
3569 	assert(mem->vmp_pageq.next == 0);
3570 	assert(mem->vmp_pageq.prev == 0);
3571 	assert(mem->vmp_listq.next == 0);
3572 	assert(mem->vmp_listq.prev == 0);
3573 	assert(mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
3574 	assert(mem->vmp_specialq.next == 0);
3575 	assert(mem->vmp_specialq.prev == 0);
3576 
3577 	return mem;
3578 }
3579 
3580 uint64_t
vm_page_secluded_drain(void)3581 vm_page_secluded_drain(void)
3582 {
3583 	vm_page_t local_freeq;
3584 	int local_freed;
3585 	uint64_t num_reclaimed;
3586 	unsigned int saved_secluded_count, saved_secluded_target;
3587 
3588 	num_reclaimed = 0;
3589 	local_freeq = NULL;
3590 	local_freed = 0;
3591 
3592 	vm_page_lock_queues();
3593 
3594 	saved_secluded_count = vm_page_secluded_count;
3595 	saved_secluded_target = vm_page_secluded_target;
3596 	vm_page_secluded_target = 0;
3597 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3598 	while (vm_page_secluded_count) {
3599 		vm_page_t secluded_page;
3600 
3601 		assert((vm_page_secluded_count_free +
3602 		    vm_page_secluded_count_inuse) ==
3603 		    vm_page_secluded_count);
3604 		secluded_page = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded);
3605 		assert(secluded_page->vmp_q_state == VM_PAGE_ON_SECLUDED_Q);
3606 
3607 		vm_page_queues_remove(secluded_page, FALSE);
3608 		assert(!secluded_page->vmp_fictitious);
3609 		assert(!VM_PAGE_WIRED(secluded_page));
3610 
3611 		if (secluded_page->vmp_object == 0) {
3612 			/* transfer to free queue */
3613 			assert(secluded_page->vmp_busy);
3614 			secluded_page->vmp_snext = local_freeq;
3615 			local_freeq = secluded_page;
3616 			local_freed += 1;
3617 		} else {
3618 			/* transfer to head of active queue */
3619 			vm_page_enqueue_active(secluded_page, FALSE);
3620 			secluded_page = VM_PAGE_NULL;
3621 		}
3622 		num_reclaimed++;
3623 	}
3624 	vm_page_secluded_target = saved_secluded_target;
3625 	VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3626 
3627 //	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);
3628 
3629 	vm_page_unlock_queues();
3630 
3631 	if (local_freed) {
3632 		vm_page_free_list(local_freeq, TRUE);
3633 		local_freeq = NULL;
3634 		local_freed = 0;
3635 	}
3636 
3637 	return num_reclaimed;
3638 }
3639 #endif /* CONFIG_SECLUDED_MEMORY */
3640 
3641 
3642 static inline void
vm_page_grab_diags()3643 vm_page_grab_diags()
3644 {
3645 #if DEVELOPMENT || DEBUG
3646 	task_t task = current_task_early();
3647 	if (task == NULL) {
3648 		return;
3649 	}
3650 
3651 	ledger_credit(task->ledger, task_ledgers.pages_grabbed, 1);
3652 #endif /* DEVELOPMENT || DEBUG */
3653 }
3654 
3655 /*
3656  *	vm_page_release:
3657  *
3658  *	Return a page to the free list.
3659  */
3660 
3661 void
vm_page_release(vm_page_t mem,boolean_t page_queues_locked)3662 vm_page_release(
3663 	vm_page_t       mem,
3664 	boolean_t       page_queues_locked)
3665 {
3666 	unsigned int    color;
3667 	int     need_wakeup = 0;
3668 	int     need_priv_wakeup = 0;
3669 #if CONFIG_SECLUDED_MEMORY
3670 	int     need_secluded_wakeup = 0;
3671 #endif /* CONFIG_SECLUDED_MEMORY */
3672 	event_t wakeup_event = NULL;
3673 
3674 	if (page_queues_locked) {
3675 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3676 	} else {
3677 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
3678 	}
3679 
3680 	assert(!mem->vmp_private && !mem->vmp_fictitious);
3681 	if (vm_page_free_verify) {
3682 		ASSERT_PMAP_FREE(mem);
3683 	}
3684 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 5);	/* (TEST/DEBUG) */
3685 
3686 	pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
3687 
3688 	if (__improbable(mem->vmp_realtime)) {
3689 		if (!page_queues_locked) {
3690 			vm_page_lock_queues();
3691 		}
3692 		if (mem->vmp_realtime) {
3693 			mem->vmp_realtime = false;
3694 			vm_page_realtime_count--;
3695 		}
3696 		if (!page_queues_locked) {
3697 			vm_page_unlock_queues();
3698 		}
3699 	}
3700 
3701 	vm_free_page_lock_spin();
3702 
3703 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
3704 	assert(mem->vmp_busy);
3705 	assert(!mem->vmp_laundry);
3706 	assert(mem->vmp_object == 0);
3707 	assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
3708 	assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0);
3709 	assert(mem->vmp_specialq.next == 0 && mem->vmp_specialq.prev == 0);
3710 
3711 	/* Clear any specialQ hints before releasing page to the free pool*/
3712 	mem->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
3713 
3714 	if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
3715 	    vm_lopage_free_count < vm_lopage_free_limit &&
3716 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3717 		/*
3718 		 * this exists to support hardware controllers
3719 		 * incapable of generating DMAs with more than 32 bits
3720 		 * of address on platforms with physical memory > 4G...
3721 		 */
3722 		vm_page_queue_enter_first(&vm_lopage_queue_free, mem, vmp_pageq);
3723 		vm_lopage_free_count++;
3724 
3725 		if (vm_lopage_free_count >= vm_lopage_free_limit) {
3726 			vm_lopage_refill = FALSE;
3727 		}
3728 
3729 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3730 		mem->vmp_lopage = TRUE;
3731 #if CONFIG_SECLUDED_MEMORY
3732 	} else if (vm_page_free_count > vm_page_free_reserved &&
3733 	    vm_page_secluded_count < vm_page_secluded_target &&
3734 	    num_tasks_can_use_secluded_mem == 0) {
3735 		/*
3736 		 * XXX FBDP TODO: also avoid refilling secluded queue
3737 		 * when some IOKit objects are already grabbing from it...
3738 		 */
3739 		if (!page_queues_locked) {
3740 			if (!vm_page_trylock_queues()) {
3741 				/* take locks in right order */
3742 				vm_free_page_unlock();
3743 				vm_page_lock_queues();
3744 				vm_free_page_lock_spin();
3745 			}
3746 		}
3747 		mem->vmp_lopage = FALSE;
3748 		LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
3749 		vm_page_queue_enter_first(&vm_page_queue_secluded, mem, vmp_pageq);
3750 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3751 		vm_page_secluded_count++;
3752 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3753 		vm_page_secluded_count_free++;
3754 		if (!page_queues_locked) {
3755 			vm_page_unlock_queues();
3756 		}
3757 		LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_OWNED);
3758 		if (vm_page_free_wanted_secluded > 0) {
3759 			vm_page_free_wanted_secluded--;
3760 			need_secluded_wakeup = 1;
3761 		}
3762 #endif /* CONFIG_SECLUDED_MEMORY */
3763 	} else {
3764 		mem->vmp_lopage = FALSE;
3765 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3766 
3767 		color = VM_PAGE_GET_COLOR(mem);
3768 #if defined(__x86_64__)
3769 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
3770 #else
3771 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
3772 #endif
3773 		vm_page_free_count++;
3774 		/*
3775 		 *	Check if we should wake up someone waiting for page.
3776 		 *	But don't bother waking them unless they can allocate.
3777 		 *
3778 		 *	We wakeup only one thread, to prevent starvation.
3779 		 *	Because the scheduling system handles wait queues FIFO,
3780 		 *	if we wakeup all waiting threads, one greedy thread
3781 		 *	can starve multiple niceguy threads.  When the threads
3782 		 *	all wakeup, the greedy threads runs first, grabs the page,
3783 		 *	and waits for another page.  It will be the first to run
3784 		 *	when the next page is freed.
3785 		 *
3786 		 *	However, there is a slight danger here.
3787 		 *	The thread we wake might not use the free page.
3788 		 *	Then the other threads could wait indefinitely
3789 		 *	while the page goes unused.  To forestall this,
3790 		 *	the pageout daemon will keep making free pages
3791 		 *	as long as vm_page_free_wanted is non-zero.
3792 		 */
3793 
3794 		assert(vm_page_free_count > 0);
3795 		if (vm_page_free_wanted_privileged > 0) {
3796 			vm_page_free_wanted_privileged--;
3797 			need_priv_wakeup = 1;
3798 #if CONFIG_SECLUDED_MEMORY
3799 		} else if (vm_page_free_wanted_secluded > 0 &&
3800 		    vm_page_free_count > vm_page_free_reserved) {
3801 			vm_page_free_wanted_secluded--;
3802 			need_secluded_wakeup = 1;
3803 #endif /* CONFIG_SECLUDED_MEMORY */
3804 		} else if (vm_page_free_wanted > 0 &&
3805 		    vm_page_free_count > vm_page_free_reserved) {
3806 			vm_page_free_wanted--;
3807 			need_wakeup = 1;
3808 		}
3809 	}
3810 	vm_pageout_vminfo.vm_page_pages_freed++;
3811 
3812 	vm_free_page_unlock();
3813 
3814 	VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, 1, 0, 0, 0);
3815 
3816 	if (need_priv_wakeup) {
3817 		wakeup_event = &vm_page_free_wanted_privileged;
3818 	}
3819 #if CONFIG_SECLUDED_MEMORY
3820 	else if (need_secluded_wakeup) {
3821 		wakeup_event = &vm_page_free_wanted_secluded;
3822 	}
3823 #endif /* CONFIG_SECLUDED_MEMORY */
3824 	else if (need_wakeup) {
3825 		wakeup_event = &vm_page_free_count;
3826 	}
3827 
3828 	if (wakeup_event) {
3829 		if (vps_dynamic_priority_enabled == TRUE) {
3830 			thread_t thread_woken = NULL;
3831 			wakeup_one_with_inheritor((event_t) wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
3832 			/*
3833 			 * (80947592) if this is the last reference on this
3834 			 * thread, calling thread_deallocate() here
3835 			 * might take the tasks_threads_lock,
3836 			 * sadly thread_create_internal is doing several
3837 			 * allocations under this lock, which can result in
3838 			 * deadlocks with the pageout scan daemon.
3839 			 *
3840 			 * FIXME: we should disallow allocations under the
3841 			 * task_thread_locks, but that is a larger fix to make.
3842 			 */
3843 			thread_deallocate_safe(thread_woken);
3844 		} else {
3845 			thread_wakeup_one((event_t) wakeup_event);
3846 		}
3847 	}
3848 
3849 	VM_CHECK_MEMORYSTATUS;
3850 }
3851 
3852 /*
3853  * This version of vm_page_release() is used only at startup
3854  * when we are single-threaded and pages are being released
3855  * for the first time. Hence, no locking or unnecessary checks are made.
3856  * Note: VM_CHECK_MEMORYSTATUS invoked by the caller.
3857  */
3858 void
vm_page_release_startup(vm_page_t mem)3859 vm_page_release_startup(
3860 	vm_page_t       mem)
3861 {
3862 	vm_page_queue_t queue_free;
3863 
3864 	if (vm_lopage_free_count < vm_lopage_free_limit &&
3865 	    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
3866 		mem->vmp_lopage = TRUE;
3867 		mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q;
3868 		vm_lopage_free_count++;
3869 		queue_free = &vm_lopage_queue_free;
3870 #if CONFIG_SECLUDED_MEMORY
3871 	} else if (vm_page_secluded_count < vm_page_secluded_target) {
3872 		mem->vmp_lopage = FALSE;
3873 		mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
3874 		vm_page_secluded_count++;
3875 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
3876 		vm_page_secluded_count_free++;
3877 		queue_free = &vm_page_queue_secluded;
3878 #endif /* CONFIG_SECLUDED_MEMORY */
3879 	} else {
3880 		mem->vmp_lopage = FALSE;
3881 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
3882 		vm_page_free_count++;
3883 		queue_free = &vm_page_queue_free[VM_PAGE_GET_COLOR(mem)].qhead;
3884 	}
3885 	if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
3886 #if defined(__x86_64__)
3887 		vm_page_queue_enter_clump(queue_free, mem);
3888 #else
3889 		vm_page_queue_enter(queue_free, mem, vmp_pageq);
3890 #endif
3891 	} else {
3892 		vm_page_queue_enter_first(queue_free, mem, vmp_pageq);
3893 	}
3894 }
3895 
3896 /*
3897  *	vm_page_wait:
3898  *
3899  *	Wait for a page to become available.
3900  *	If there are plenty of free pages, then we don't sleep.
3901  *
3902  *	Returns:
3903  *		TRUE:  There may be another page, try again
3904  *		FALSE: We were interrupted out of our wait, don't try again
3905  */
3906 
3907 boolean_t
vm_page_wait(int interruptible)3908 vm_page_wait(
3909 	int     interruptible )
3910 {
3911 	/*
3912 	 *	We can't use vm_page_free_reserved to make this
3913 	 *	determination.  Consider: some thread might
3914 	 *	need to allocate two pages.  The first allocation
3915 	 *	succeeds, the second fails.  After the first page is freed,
3916 	 *	a call to vm_page_wait must really block.
3917 	 */
3918 	kern_return_t   wait_result;
3919 	int             need_wakeup = 0;
3920 	int             is_privileged = current_thread()->options & TH_OPT_VMPRIV;
3921 	event_t         wait_event = NULL;
3922 
3923 	vm_free_page_lock_spin();
3924 
3925 	if (is_privileged && vm_page_free_count) {
3926 		vm_free_page_unlock();
3927 		return TRUE;
3928 	}
3929 
3930 	if (vm_page_free_count >= vm_page_free_target) {
3931 		vm_free_page_unlock();
3932 		return TRUE;
3933 	}
3934 
3935 	if (is_privileged) {
3936 		if (vm_page_free_wanted_privileged++ == 0) {
3937 			need_wakeup = 1;
3938 		}
3939 		wait_event = (event_t)&vm_page_free_wanted_privileged;
3940 #if CONFIG_SECLUDED_MEMORY
3941 	} else if (secluded_for_apps &&
3942 	    task_can_use_secluded_mem(current_task(), FALSE)) {
3943 #if 00
3944 		/* XXX FBDP: need pageq lock for this... */
3945 		/* XXX FBDP: might wait even if pages available, */
3946 		/* XXX FBDP: hopefully not for too long... */
3947 		if (vm_page_secluded_count > 0) {
3948 			vm_free_page_unlock();
3949 			return TRUE;
3950 		}
3951 #endif
3952 		if (vm_page_free_wanted_secluded++ == 0) {
3953 			need_wakeup = 1;
3954 		}
3955 		wait_event = (event_t)&vm_page_free_wanted_secluded;
3956 #endif /* CONFIG_SECLUDED_MEMORY */
3957 	} else {
3958 		if (vm_page_free_wanted++ == 0) {
3959 			need_wakeup = 1;
3960 		}
3961 		wait_event = (event_t)&vm_page_free_count;
3962 	}
3963 
3964 	/*
3965 	 * We don't do a vm_pageout_scan wakeup if we already have
3966 	 * some waiters because vm_pageout_scan checks for waiters
3967 	 * before it returns and does so behind the vm_page_queue_free_lock,
3968 	 * which we own when we bump the waiter counts.
3969 	 */
3970 
3971 	if (vps_dynamic_priority_enabled == TRUE) {
3972 		/*
3973 		 * We are waking up vm_pageout_scan here. If it needs
3974 		 * the vm_page_queue_free_lock before we unlock it
3975 		 * we'll end up just blocking and incur an extra
3976 		 * context switch. Could be a perf. issue.
3977 		 */
3978 
3979 		if (need_wakeup) {
3980 			thread_wakeup((event_t)&vm_page_free_wanted);
3981 		}
3982 
3983 		/*
3984 		 * LD: This event is going to get recorded every time because
3985 		 * we don't get back THREAD_WAITING from lck_mtx_sleep_with_inheritor.
3986 		 * We just block in that routine.
3987 		 */
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 =  lck_mtx_sleep_with_inheritor(&vm_page_queue_free_lock,
3998 		    LCK_SLEEP_UNLOCK,
3999 		    wait_event,
4000 		    vm_pageout_scan_thread,
4001 		    interruptible,
4002 		    0);
4003 	} else {
4004 		wait_result = assert_wait(wait_event, interruptible);
4005 
4006 		vm_free_page_unlock();
4007 
4008 		if (need_wakeup) {
4009 			thread_wakeup((event_t)&vm_page_free_wanted);
4010 		}
4011 
4012 		if (wait_result == THREAD_WAITING) {
4013 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START,
4014 			    vm_page_free_wanted_privileged,
4015 			    vm_page_free_wanted,
4016 #if CONFIG_SECLUDED_MEMORY
4017 			    vm_page_free_wanted_secluded,
4018 #else /* CONFIG_SECLUDED_MEMORY */
4019 			    0,
4020 #endif /* CONFIG_SECLUDED_MEMORY */
4021 			    0);
4022 			wait_result = thread_block(THREAD_CONTINUE_NULL);
4023 			VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block,
4024 			    VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0);
4025 		}
4026 	}
4027 
4028 	return (wait_result == THREAD_AWAKENED) || (wait_result == THREAD_NOT_WAITING);
4029 }
4030 
4031 /*
4032  *	vm_page_alloc:
4033  *
4034  *	Allocate and return a memory cell associated
4035  *	with this VM object/offset pair.
4036  *
4037  *	Object must be locked.
4038  */
4039 
4040 vm_page_t
vm_page_alloc(vm_object_t object,vm_object_offset_t offset)4041 vm_page_alloc(
4042 	vm_object_t             object,
4043 	vm_object_offset_t      offset)
4044 {
4045 	vm_page_t       mem;
4046 	int             grab_options;
4047 
4048 	vm_object_lock_assert_exclusive(object);
4049 	grab_options = 0;
4050 #if CONFIG_SECLUDED_MEMORY
4051 	if (object->can_grab_secluded) {
4052 		grab_options |= VM_PAGE_GRAB_SECLUDED;
4053 	}
4054 #endif /* CONFIG_SECLUDED_MEMORY */
4055 	mem = vm_page_grab_options(grab_options);
4056 	if (mem == VM_PAGE_NULL) {
4057 		return VM_PAGE_NULL;
4058 	}
4059 
4060 	vm_page_insert(mem, object, offset);
4061 
4062 	return mem;
4063 }
4064 
4065 /*
4066  *	vm_page_free_prepare:
4067  *
4068  *	Removes page from any queue it may be on
4069  *	and disassociates it from its VM object.
4070  *
4071  *	Object and page queues must be locked prior to entry.
4072  */
4073 static void
vm_page_free_prepare(vm_page_t mem)4074 vm_page_free_prepare(
4075 	vm_page_t       mem)
4076 {
4077 	vm_page_free_prepare_queues(mem);
4078 	vm_page_free_prepare_object(mem, TRUE);
4079 }
4080 
4081 
4082 void
vm_page_free_prepare_queues(vm_page_t mem)4083 vm_page_free_prepare_queues(
4084 	vm_page_t       mem)
4085 {
4086 	vm_object_t     m_object;
4087 
4088 	VM_PAGE_CHECK(mem);
4089 
4090 	assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
4091 	assert(!mem->vmp_cleaning);
4092 	m_object = VM_PAGE_OBJECT(mem);
4093 
4094 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4095 	if (m_object) {
4096 		vm_object_lock_assert_exclusive(m_object);
4097 	}
4098 	if (mem->vmp_laundry) {
4099 		/*
4100 		 * We may have to free a page while it's being laundered
4101 		 * if we lost its pager (due to a forced unmount, for example).
4102 		 * We need to call vm_pageout_steal_laundry() before removing
4103 		 * the page from its VM object, so that we can remove it
4104 		 * from its pageout queue and adjust the laundry accounting
4105 		 */
4106 		vm_pageout_steal_laundry(mem, TRUE);
4107 	}
4108 
4109 	vm_page_queues_remove(mem, TRUE);
4110 
4111 	if (__improbable(mem->vmp_realtime)) {
4112 		mem->vmp_realtime = false;
4113 		vm_page_realtime_count--;
4114 	}
4115 
4116 	if (VM_PAGE_WIRED(mem)) {
4117 		assert(mem->vmp_wire_count > 0);
4118 
4119 		if (m_object) {
4120 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4121 			VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4122 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4123 
4124 			assert(m_object->resident_page_count >=
4125 			    m_object->wired_page_count);
4126 
4127 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4128 				OSAddAtomic(+1, &vm_page_purgeable_count);
4129 				assert(vm_page_purgeable_wired_count > 0);
4130 				OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4131 			}
4132 			if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4133 			    m_object->purgable == VM_PURGABLE_EMPTY) &&
4134 			    m_object->vo_owner != TASK_NULL) {
4135 				task_t          owner;
4136 				int             ledger_idx_volatile;
4137 				int             ledger_idx_nonvolatile;
4138 				int             ledger_idx_volatile_compressed;
4139 				int             ledger_idx_nonvolatile_compressed;
4140 				boolean_t       do_footprint;
4141 
4142 				owner = VM_OBJECT_OWNER(m_object);
4143 				vm_object_ledger_tag_ledgers(
4144 					m_object,
4145 					&ledger_idx_volatile,
4146 					&ledger_idx_nonvolatile,
4147 					&ledger_idx_volatile_compressed,
4148 					&ledger_idx_nonvolatile_compressed,
4149 					&do_footprint);
4150 				/*
4151 				 * While wired, this page was accounted
4152 				 * as "non-volatile" but it should now
4153 				 * be accounted as "volatile".
4154 				 */
4155 				/* one less "non-volatile"... */
4156 				ledger_debit(owner->ledger,
4157 				    ledger_idx_nonvolatile,
4158 				    PAGE_SIZE);
4159 				if (do_footprint) {
4160 					/* ... and "phys_footprint" */
4161 					ledger_debit(owner->ledger,
4162 					    task_ledgers.phys_footprint,
4163 					    PAGE_SIZE);
4164 				}
4165 				/* one more "volatile" */
4166 				ledger_credit(owner->ledger,
4167 				    ledger_idx_volatile,
4168 				    PAGE_SIZE);
4169 			}
4170 		}
4171 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4172 			vm_page_wire_count--;
4173 		}
4174 
4175 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4176 		mem->vmp_wire_count = 0;
4177 		assert(!mem->vmp_gobbled);
4178 	} else if (mem->vmp_gobbled) {
4179 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4180 			vm_page_wire_count--;
4181 		}
4182 		vm_page_gobble_count--;
4183 	}
4184 }
4185 
4186 
4187 void
vm_page_free_prepare_object(vm_page_t mem,boolean_t remove_from_hash)4188 vm_page_free_prepare_object(
4189 	vm_page_t       mem,
4190 	boolean_t       remove_from_hash)
4191 {
4192 	assert(!mem->vmp_realtime);
4193 	if (mem->vmp_tabled) {
4194 		vm_page_remove(mem, remove_from_hash);  /* clears tabled, object, offset */
4195 	}
4196 	PAGE_WAKEUP(mem);               /* clears wanted */
4197 
4198 	if (mem->vmp_private) {
4199 		mem->vmp_private = FALSE;
4200 		mem->vmp_fictitious = TRUE;
4201 		VM_PAGE_SET_PHYS_PAGE(mem, vm_page_fictitious_addr);
4202 	}
4203 	if (!mem->vmp_fictitious) {
4204 		assert(mem->vmp_pageq.next == 0);
4205 		assert(mem->vmp_pageq.prev == 0);
4206 		assert(mem->vmp_listq.next == 0);
4207 		assert(mem->vmp_listq.prev == 0);
4208 		assert(mem->vmp_specialq.next == 0);
4209 		assert(mem->vmp_specialq.prev == 0);
4210 		assert(mem->vmp_next_m == 0);
4211 		ASSERT_PMAP_FREE(mem);
4212 		{
4213 			vm_page_init(mem, VM_PAGE_GET_PHYS_PAGE(mem), mem->vmp_lopage);
4214 		}
4215 	}
4216 }
4217 
4218 
4219 /*
4220  *	vm_page_free:
4221  *
4222  *	Returns the given page to the free list,
4223  *	disassociating it with any VM object.
4224  *
4225  *	Object and page queues must be locked prior to entry.
4226  */
4227 void
vm_page_free(vm_page_t mem)4228 vm_page_free(
4229 	vm_page_t       mem)
4230 {
4231 	vm_page_free_prepare(mem);
4232 
4233 	if (mem->vmp_fictitious) {
4234 		vm_page_release_fictitious(mem);
4235 	} else {
4236 		vm_page_release(mem, TRUE);  /* page queues are locked */
4237 	}
4238 }
4239 
4240 
4241 void
vm_page_free_unlocked(vm_page_t mem,boolean_t remove_from_hash)4242 vm_page_free_unlocked(
4243 	vm_page_t       mem,
4244 	boolean_t       remove_from_hash)
4245 {
4246 	vm_page_lockspin_queues();
4247 	vm_page_free_prepare_queues(mem);
4248 	vm_page_unlock_queues();
4249 
4250 	vm_page_free_prepare_object(mem, remove_from_hash);
4251 
4252 	if (mem->vmp_fictitious) {
4253 		vm_page_release_fictitious(mem);
4254 	} else {
4255 		vm_page_release(mem, FALSE); /* page queues are not locked */
4256 	}
4257 }
4258 
4259 
4260 /*
4261  * Free a list of pages.  The list can be up to several hundred pages,
4262  * as blocked up by vm_pageout_scan().
4263  * The big win is not having to take the free list lock once
4264  * per page.
4265  *
4266  * The VM page queues lock (vm_page_queue_lock) should NOT be held.
4267  * The VM page free queues lock (vm_page_queue_free_lock) should NOT be held.
4268  */
4269 void
vm_page_free_list(vm_page_t freeq,boolean_t prepare_object)4270 vm_page_free_list(
4271 	vm_page_t       freeq,
4272 	boolean_t       prepare_object)
4273 {
4274 	vm_page_t       mem;
4275 	vm_page_t       nxt;
4276 	vm_page_t       local_freeq;
4277 	int             pg_count;
4278 
4279 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED);
4280 	LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_NOTOWNED);
4281 
4282 	while (freeq) {
4283 		pg_count = 0;
4284 		local_freeq = VM_PAGE_NULL;
4285 		mem = freeq;
4286 
4287 		/*
4288 		 * break up the processing into smaller chunks so
4289 		 * that we can 'pipeline' the pages onto the
4290 		 * free list w/o introducing too much
4291 		 * contention on the global free queue lock
4292 		 */
4293 		while (mem && pg_count < 64) {
4294 			assert((mem->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
4295 			    (mem->vmp_q_state == VM_PAGE_IS_WIRED));
4296 			assert(mem->vmp_specialq.next == 0 &&
4297 			    mem->vmp_specialq.prev == 0);
4298 			/*
4299 			 * &&
4300 			 *   mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
4301 			 */
4302 			nxt = mem->vmp_snext;
4303 			mem->vmp_snext = NULL;
4304 			assert(mem->vmp_pageq.prev == 0);
4305 
4306 			if (vm_page_free_verify && !mem->vmp_fictitious && !mem->vmp_private) {
4307 				ASSERT_PMAP_FREE(mem);
4308 			}
4309 
4310 			if (__improbable(mem->vmp_realtime)) {
4311 				vm_page_lock_queues();
4312 				if (mem->vmp_realtime) {
4313 					mem->vmp_realtime = false;
4314 					vm_page_realtime_count--;
4315 				}
4316 				vm_page_unlock_queues();
4317 			}
4318 
4319 			if (prepare_object == TRUE) {
4320 				vm_page_free_prepare_object(mem, TRUE);
4321 			}
4322 
4323 			if (!mem->vmp_fictitious) {
4324 				assert(mem->vmp_busy);
4325 
4326 				if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) &&
4327 				    vm_lopage_free_count < vm_lopage_free_limit &&
4328 				    VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) {
4329 					vm_page_release(mem, FALSE); /* page queues are not locked */
4330 #if CONFIG_SECLUDED_MEMORY
4331 				} else if (vm_page_secluded_count < vm_page_secluded_target &&
4332 				    num_tasks_can_use_secluded_mem == 0) {
4333 					vm_page_release(mem,
4334 					    FALSE);             /* page queues are not locked */
4335 #endif /* CONFIG_SECLUDED_MEMORY */
4336 				} else {
4337 					/*
4338 					 * IMPORTANT: we can't set the page "free" here
4339 					 * because that would make the page eligible for
4340 					 * a physically-contiguous allocation (see
4341 					 * vm_page_find_contiguous()) right away (we don't
4342 					 * hold the vm_page_queue_free lock).  That would
4343 					 * cause trouble because the page is not actually
4344 					 * in the free queue yet...
4345 					 */
4346 					mem->vmp_snext = local_freeq;
4347 					local_freeq = mem;
4348 					pg_count++;
4349 
4350 					pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem));
4351 				}
4352 			} else {
4353 				assert(VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_fictitious_addr ||
4354 				    VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_guard_addr);
4355 				vm_page_release_fictitious(mem);
4356 			}
4357 			mem = nxt;
4358 		}
4359 		freeq = mem;
4360 
4361 		if ((mem = local_freeq)) {
4362 			unsigned int    avail_free_count;
4363 			unsigned int    need_wakeup = 0;
4364 			unsigned int    need_priv_wakeup = 0;
4365 #if CONFIG_SECLUDED_MEMORY
4366 			unsigned int    need_wakeup_secluded = 0;
4367 #endif /* CONFIG_SECLUDED_MEMORY */
4368 			event_t         priv_wakeup_event, secluded_wakeup_event, normal_wakeup_event;
4369 			boolean_t       priv_wakeup_all, secluded_wakeup_all, normal_wakeup_all;
4370 
4371 			vm_free_page_lock_spin();
4372 
4373 			while (mem) {
4374 				int     color;
4375 
4376 				nxt = mem->vmp_snext;
4377 
4378 				assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
4379 				assert(mem->vmp_busy);
4380 				assert(!mem->vmp_realtime);
4381 				mem->vmp_lopage = FALSE;
4382 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
4383 
4384 				color = VM_PAGE_GET_COLOR(mem);
4385 #if defined(__x86_64__)
4386 				vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
4387 #else
4388 				vm_page_queue_enter(&vm_page_queue_free[color].qhead,
4389 				    mem, vmp_pageq);
4390 #endif
4391 				mem = nxt;
4392 			}
4393 			vm_pageout_vminfo.vm_page_pages_freed += pg_count;
4394 			vm_page_free_count += pg_count;
4395 			avail_free_count = vm_page_free_count;
4396 
4397 			VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, pg_count, 0, 0, 0);
4398 
4399 			if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) {
4400 				if (avail_free_count < vm_page_free_wanted_privileged) {
4401 					need_priv_wakeup = avail_free_count;
4402 					vm_page_free_wanted_privileged -= avail_free_count;
4403 					avail_free_count = 0;
4404 				} else {
4405 					need_priv_wakeup = vm_page_free_wanted_privileged;
4406 					avail_free_count -= vm_page_free_wanted_privileged;
4407 					vm_page_free_wanted_privileged = 0;
4408 				}
4409 			}
4410 #if CONFIG_SECLUDED_MEMORY
4411 			if (vm_page_free_wanted_secluded > 0 &&
4412 			    avail_free_count > vm_page_free_reserved) {
4413 				unsigned int available_pages;
4414 				available_pages = (avail_free_count -
4415 				    vm_page_free_reserved);
4416 				if (available_pages <
4417 				    vm_page_free_wanted_secluded) {
4418 					need_wakeup_secluded = available_pages;
4419 					vm_page_free_wanted_secluded -=
4420 					    available_pages;
4421 					avail_free_count -= available_pages;
4422 				} else {
4423 					need_wakeup_secluded =
4424 					    vm_page_free_wanted_secluded;
4425 					avail_free_count -=
4426 					    vm_page_free_wanted_secluded;
4427 					vm_page_free_wanted_secluded = 0;
4428 				}
4429 			}
4430 #endif /* CONFIG_SECLUDED_MEMORY */
4431 			if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) {
4432 				unsigned int  available_pages;
4433 
4434 				available_pages = avail_free_count - vm_page_free_reserved;
4435 
4436 				if (available_pages >= vm_page_free_wanted) {
4437 					need_wakeup = vm_page_free_wanted;
4438 					vm_page_free_wanted = 0;
4439 				} else {
4440 					need_wakeup = available_pages;
4441 					vm_page_free_wanted -= available_pages;
4442 				}
4443 			}
4444 			vm_free_page_unlock();
4445 
4446 			priv_wakeup_event = NULL;
4447 			secluded_wakeup_event = NULL;
4448 			normal_wakeup_event = NULL;
4449 
4450 			priv_wakeup_all = FALSE;
4451 			secluded_wakeup_all = FALSE;
4452 			normal_wakeup_all = FALSE;
4453 
4454 
4455 			if (need_priv_wakeup != 0) {
4456 				/*
4457 				 * There shouldn't be that many VM-privileged threads,
4458 				 * so let's wake them all up, even if we don't quite
4459 				 * have enough pages to satisfy them all.
4460 				 */
4461 				priv_wakeup_event = (event_t)&vm_page_free_wanted_privileged;
4462 				priv_wakeup_all = TRUE;
4463 			}
4464 #if CONFIG_SECLUDED_MEMORY
4465 			if (need_wakeup_secluded != 0 &&
4466 			    vm_page_free_wanted_secluded == 0) {
4467 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4468 				secluded_wakeup_all = TRUE;
4469 				need_wakeup_secluded = 0;
4470 			} else {
4471 				secluded_wakeup_event = (event_t)&vm_page_free_wanted_secluded;
4472 			}
4473 #endif /* CONFIG_SECLUDED_MEMORY */
4474 			if (need_wakeup != 0 && vm_page_free_wanted == 0) {
4475 				/*
4476 				 * We don't expect to have any more waiters
4477 				 * after this, so let's wake them all up at
4478 				 * once.
4479 				 */
4480 				normal_wakeup_event = (event_t) &vm_page_free_count;
4481 				normal_wakeup_all = TRUE;
4482 				need_wakeup = 0;
4483 			} else {
4484 				normal_wakeup_event = (event_t) &vm_page_free_count;
4485 			}
4486 
4487 			if (priv_wakeup_event ||
4488 #if CONFIG_SECLUDED_MEMORY
4489 			    secluded_wakeup_event ||
4490 #endif /* CONFIG_SECLUDED_MEMORY */
4491 			    normal_wakeup_event) {
4492 				if (vps_dynamic_priority_enabled == TRUE) {
4493 					thread_t thread_woken = NULL;
4494 
4495 					if (priv_wakeup_all == TRUE) {
4496 						wakeup_all_with_inheritor(priv_wakeup_event, THREAD_AWAKENED);
4497 					}
4498 
4499 #if CONFIG_SECLUDED_MEMORY
4500 					if (secluded_wakeup_all == TRUE) {
4501 						wakeup_all_with_inheritor(secluded_wakeup_event, THREAD_AWAKENED);
4502 					}
4503 
4504 					while (need_wakeup_secluded-- != 0) {
4505 						/*
4506 						 * Wake up one waiter per page we just released.
4507 						 */
4508 						wakeup_one_with_inheritor(secluded_wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
4509 						thread_deallocate(thread_woken);
4510 					}
4511 #endif /* CONFIG_SECLUDED_MEMORY */
4512 
4513 					if (normal_wakeup_all == TRUE) {
4514 						wakeup_all_with_inheritor(normal_wakeup_event, THREAD_AWAKENED);
4515 					}
4516 
4517 					while (need_wakeup-- != 0) {
4518 						/*
4519 						 * Wake up one waiter per page we just released.
4520 						 */
4521 						wakeup_one_with_inheritor(normal_wakeup_event, THREAD_AWAKENED, LCK_WAKE_DO_NOT_TRANSFER_PUSH, &thread_woken);
4522 						thread_deallocate(thread_woken);
4523 					}
4524 				} else {
4525 					/*
4526 					 * Non-priority-aware wakeups.
4527 					 */
4528 
4529 					if (priv_wakeup_all == TRUE) {
4530 						thread_wakeup(priv_wakeup_event);
4531 					}
4532 
4533 #if CONFIG_SECLUDED_MEMORY
4534 					if (secluded_wakeup_all == TRUE) {
4535 						thread_wakeup(secluded_wakeup_event);
4536 					}
4537 
4538 					while (need_wakeup_secluded-- != 0) {
4539 						/*
4540 						 * Wake up one waiter per page we just released.
4541 						 */
4542 						thread_wakeup_one(secluded_wakeup_event);
4543 					}
4544 
4545 #endif /* CONFIG_SECLUDED_MEMORY */
4546 					if (normal_wakeup_all == TRUE) {
4547 						thread_wakeup(normal_wakeup_event);
4548 					}
4549 
4550 					while (need_wakeup-- != 0) {
4551 						/*
4552 						 * Wake up one waiter per page we just released.
4553 						 */
4554 						thread_wakeup_one(normal_wakeup_event);
4555 					}
4556 				}
4557 			}
4558 
4559 			VM_CHECK_MEMORYSTATUS;
4560 		}
4561 	}
4562 }
4563 
4564 
4565 /*
4566  *	vm_page_wire:
4567  *
4568  *	Mark this page as wired down by yet
4569  *	another map, removing it from paging queues
4570  *	as necessary.
4571  *
4572  *	The page's object and the page queues must be locked.
4573  */
4574 
4575 
4576 void
vm_page_wire(vm_page_t mem,vm_tag_t tag,boolean_t check_memorystatus)4577 vm_page_wire(
4578 	vm_page_t mem,
4579 	vm_tag_t           tag,
4580 	boolean_t          check_memorystatus)
4581 {
4582 	vm_object_t     m_object;
4583 
4584 	m_object = VM_PAGE_OBJECT(mem);
4585 
4586 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 1);	/* (TEST/DEBUG) */
4587 
4588 	VM_PAGE_CHECK(mem);
4589 	if (m_object) {
4590 		vm_object_lock_assert_exclusive(m_object);
4591 	} else {
4592 		/*
4593 		 * In theory, the page should be in an object before it
4594 		 * gets wired, since we need to hold the object lock
4595 		 * to update some fields in the page structure.
4596 		 * However, some code (i386 pmap, for example) might want
4597 		 * to wire a page before it gets inserted into an object.
4598 		 * That's somewhat OK, as long as nobody else can get to
4599 		 * that page and update it at the same time.
4600 		 */
4601 	}
4602 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4603 	if (!VM_PAGE_WIRED(mem)) {
4604 		if (mem->vmp_laundry) {
4605 			vm_pageout_steal_laundry(mem, TRUE);
4606 		}
4607 
4608 		vm_page_queues_remove(mem, TRUE);
4609 
4610 		assert(mem->vmp_wire_count == 0);
4611 		mem->vmp_q_state = VM_PAGE_IS_WIRED;
4612 
4613 		if (m_object) {
4614 			VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4615 			VM_OBJECT_WIRED_PAGE_ADD(m_object, mem);
4616 			VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, tag);
4617 
4618 			assert(m_object->resident_page_count >=
4619 			    m_object->wired_page_count);
4620 			if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4621 				assert(vm_page_purgeable_count > 0);
4622 				OSAddAtomic(-1, &vm_page_purgeable_count);
4623 				OSAddAtomic(1, &vm_page_purgeable_wired_count);
4624 			}
4625 			if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4626 			    m_object->purgable == VM_PURGABLE_EMPTY) &&
4627 			    m_object->vo_owner != TASK_NULL) {
4628 				task_t          owner;
4629 				int             ledger_idx_volatile;
4630 				int             ledger_idx_nonvolatile;
4631 				int             ledger_idx_volatile_compressed;
4632 				int             ledger_idx_nonvolatile_compressed;
4633 				boolean_t       do_footprint;
4634 
4635 				owner = VM_OBJECT_OWNER(m_object);
4636 				vm_object_ledger_tag_ledgers(
4637 					m_object,
4638 					&ledger_idx_volatile,
4639 					&ledger_idx_nonvolatile,
4640 					&ledger_idx_volatile_compressed,
4641 					&ledger_idx_nonvolatile_compressed,
4642 					&do_footprint);
4643 				/* less volatile bytes */
4644 				ledger_debit(owner->ledger,
4645 				    ledger_idx_volatile,
4646 				    PAGE_SIZE);
4647 				/* more not-quite-volatile bytes */
4648 				ledger_credit(owner->ledger,
4649 				    ledger_idx_nonvolatile,
4650 				    PAGE_SIZE);
4651 				if (do_footprint) {
4652 					/* more footprint */
4653 					ledger_credit(owner->ledger,
4654 					    task_ledgers.phys_footprint,
4655 					    PAGE_SIZE);
4656 				}
4657 			}
4658 			if (m_object->all_reusable) {
4659 				/*
4660 				 * Wired pages are not counted as "re-usable"
4661 				 * in "all_reusable" VM objects, so nothing
4662 				 * to do here.
4663 				 */
4664 			} else if (mem->vmp_reusable) {
4665 				/*
4666 				 * This page is not "re-usable" when it's
4667 				 * wired, so adjust its state and the
4668 				 * accounting.
4669 				 */
4670 				vm_object_reuse_pages(m_object,
4671 				    mem->vmp_offset,
4672 				    mem->vmp_offset + PAGE_SIZE_64,
4673 				    FALSE);
4674 			}
4675 		}
4676 		assert(!mem->vmp_reusable);
4677 
4678 		if (!mem->vmp_private && !mem->vmp_fictitious && !mem->vmp_gobbled) {
4679 			vm_page_wire_count++;
4680 		}
4681 		if (mem->vmp_gobbled) {
4682 			vm_page_gobble_count--;
4683 		}
4684 		mem->vmp_gobbled = FALSE;
4685 
4686 		if (check_memorystatus == TRUE) {
4687 			VM_CHECK_MEMORYSTATUS;
4688 		}
4689 	}
4690 	assert(!mem->vmp_gobbled);
4691 	assert(mem->vmp_q_state == VM_PAGE_IS_WIRED);
4692 	mem->vmp_wire_count++;
4693 	if (__improbable(mem->vmp_wire_count == 0)) {
4694 		panic("vm_page_wire(%p): wire_count overflow", mem);
4695 	}
4696 	VM_PAGE_CHECK(mem);
4697 }
4698 
4699 /*
4700  *	vm_page_unwire:
4701  *
4702  *	Release one wiring of this page, potentially
4703  *	enabling it to be paged again.
4704  *
4705  *	The page's object and the page queues must be locked.
4706  */
4707 void
vm_page_unwire(vm_page_t mem,boolean_t queueit)4708 vm_page_unwire(
4709 	vm_page_t       mem,
4710 	boolean_t       queueit)
4711 {
4712 	vm_object_t     m_object;
4713 
4714 	m_object = VM_PAGE_OBJECT(mem);
4715 
4716 //	dbgLog(current_thread(), mem->vmp_offset, m_object, 0);	/* (TEST/DEBUG) */
4717 
4718 	VM_PAGE_CHECK(mem);
4719 	assert(VM_PAGE_WIRED(mem));
4720 	assert(mem->vmp_wire_count > 0);
4721 	assert(!mem->vmp_gobbled);
4722 	assert(m_object != VM_OBJECT_NULL);
4723 	vm_object_lock_assert_exclusive(m_object);
4724 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4725 	if (--mem->vmp_wire_count == 0) {
4726 		mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
4727 
4728 		VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object);
4729 		VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem);
4730 		VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag);
4731 		if (!mem->vmp_private && !mem->vmp_fictitious) {
4732 			vm_page_wire_count--;
4733 		}
4734 
4735 		assert(m_object->resident_page_count >=
4736 		    m_object->wired_page_count);
4737 		if (m_object->purgable == VM_PURGABLE_VOLATILE) {
4738 			OSAddAtomic(+1, &vm_page_purgeable_count);
4739 			assert(vm_page_purgeable_wired_count > 0);
4740 			OSAddAtomic(-1, &vm_page_purgeable_wired_count);
4741 		}
4742 		if ((m_object->purgable == VM_PURGABLE_VOLATILE ||
4743 		    m_object->purgable == VM_PURGABLE_EMPTY) &&
4744 		    m_object->vo_owner != TASK_NULL) {
4745 			task_t          owner;
4746 			int             ledger_idx_volatile;
4747 			int             ledger_idx_nonvolatile;
4748 			int             ledger_idx_volatile_compressed;
4749 			int             ledger_idx_nonvolatile_compressed;
4750 			boolean_t       do_footprint;
4751 
4752 			owner = VM_OBJECT_OWNER(m_object);
4753 			vm_object_ledger_tag_ledgers(
4754 				m_object,
4755 				&ledger_idx_volatile,
4756 				&ledger_idx_nonvolatile,
4757 				&ledger_idx_volatile_compressed,
4758 				&ledger_idx_nonvolatile_compressed,
4759 				&do_footprint);
4760 			/* more volatile bytes */
4761 			ledger_credit(owner->ledger,
4762 			    ledger_idx_volatile,
4763 			    PAGE_SIZE);
4764 			/* less not-quite-volatile bytes */
4765 			ledger_debit(owner->ledger,
4766 			    ledger_idx_nonvolatile,
4767 			    PAGE_SIZE);
4768 			if (do_footprint) {
4769 				/* less footprint */
4770 				ledger_debit(owner->ledger,
4771 				    task_ledgers.phys_footprint,
4772 				    PAGE_SIZE);
4773 			}
4774 		}
4775 		assert(m_object != kernel_object);
4776 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
4777 
4778 		if (queueit == TRUE) {
4779 			if (m_object->purgable == VM_PURGABLE_EMPTY) {
4780 				vm_page_deactivate(mem);
4781 			} else {
4782 				vm_page_activate(mem);
4783 			}
4784 		}
4785 
4786 		VM_CHECK_MEMORYSTATUS;
4787 	}
4788 	VM_PAGE_CHECK(mem);
4789 }
4790 
4791 /*
4792  *	vm_page_deactivate:
4793  *
4794  *	Returns the given page to the inactive list,
4795  *	indicating that no physical maps have access
4796  *	to this page.  [Used by the physical mapping system.]
4797  *
4798  *	The page queues must be locked.
4799  */
4800 void
vm_page_deactivate(vm_page_t m)4801 vm_page_deactivate(
4802 	vm_page_t       m)
4803 {
4804 	vm_page_deactivate_internal(m, TRUE);
4805 }
4806 
4807 
4808 void
vm_page_deactivate_internal(vm_page_t m,boolean_t clear_hw_reference)4809 vm_page_deactivate_internal(
4810 	vm_page_t       m,
4811 	boolean_t       clear_hw_reference)
4812 {
4813 	vm_object_t     m_object;
4814 
4815 	m_object = VM_PAGE_OBJECT(m);
4816 
4817 	VM_PAGE_CHECK(m);
4818 	assert(m_object != kernel_object);
4819 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4820 
4821 //	dbgLog(VM_PAGE_GET_PHYS_PAGE(m), vm_page_free_count, vm_page_wire_count, 6);	/* (TEST/DEBUG) */
4822 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4823 	/*
4824 	 *	This page is no longer very interesting.  If it was
4825 	 *	interesting (active or inactive/referenced), then we
4826 	 *	clear the reference bit and (re)enter it in the
4827 	 *	inactive queue.  Note wired pages should not have
4828 	 *	their reference bit cleared.
4829 	 */
4830 	assert( !(m->vmp_absent && !m->vmp_unusual));
4831 
4832 	if (m->vmp_gobbled) {           /* can this happen? */
4833 		assert( !VM_PAGE_WIRED(m));
4834 
4835 		if (!m->vmp_private && !m->vmp_fictitious) {
4836 			vm_page_wire_count--;
4837 		}
4838 		vm_page_gobble_count--;
4839 		m->vmp_gobbled = FALSE;
4840 	}
4841 	/*
4842 	 * if this page is currently on the pageout queue, we can't do the
4843 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4844 	 * and we can't remove it manually since we would need the object lock
4845 	 * (which is not required here) to decrement the activity_in_progress
4846 	 * reference which is held on the object while the page is in the pageout queue...
4847 	 * just let the normal laundry processing proceed
4848 	 */
4849 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4850 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
4851 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
4852 	    VM_PAGE_WIRED(m)) {
4853 		return;
4854 	}
4855 	if (!m->vmp_absent && clear_hw_reference == TRUE) {
4856 		pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(m));
4857 	}
4858 
4859 	m->vmp_reference = FALSE;
4860 	m->vmp_no_cache = FALSE;
4861 
4862 	if (!VM_PAGE_INACTIVE(m)) {
4863 		vm_page_queues_remove(m, FALSE);
4864 
4865 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
4866 		    m->vmp_dirty && m_object->internal &&
4867 		    (m_object->purgable == VM_PURGABLE_DENY ||
4868 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
4869 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
4870 			vm_page_check_pageable_safe(m);
4871 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
4872 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
4873 			vm_page_throttled_count++;
4874 		} else {
4875 			if (m_object->named && m_object->ref_count == 1) {
4876 				vm_page_speculate(m, FALSE);
4877 #if DEVELOPMENT || DEBUG
4878 				vm_page_speculative_recreated++;
4879 #endif
4880 			} else {
4881 				vm_page_enqueue_inactive(m, FALSE);
4882 			}
4883 		}
4884 	}
4885 }
4886 
4887 /*
4888  * vm_page_enqueue_cleaned
4889  *
4890  * Put the page on the cleaned queue, mark it cleaned, etc.
4891  * Being on the cleaned queue (and having m->clean_queue set)
4892  * does ** NOT ** guarantee that the page is clean!
4893  *
4894  * Call with the queues lock held.
4895  */
4896 
4897 void
vm_page_enqueue_cleaned(vm_page_t m)4898 vm_page_enqueue_cleaned(vm_page_t m)
4899 {
4900 	vm_object_t     m_object;
4901 
4902 	m_object = VM_PAGE_OBJECT(m);
4903 
4904 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4905 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4906 	assert( !(m->vmp_absent && !m->vmp_unusual));
4907 
4908 	if (VM_PAGE_WIRED(m)) {
4909 		return;
4910 	}
4911 
4912 	if (m->vmp_gobbled) {
4913 		if (!m->vmp_private && !m->vmp_fictitious) {
4914 			vm_page_wire_count--;
4915 		}
4916 		vm_page_gobble_count--;
4917 		m->vmp_gobbled = FALSE;
4918 	}
4919 	/*
4920 	 * if this page is currently on the pageout queue, we can't do the
4921 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4922 	 * and we can't remove it manually since we would need the object lock
4923 	 * (which is not required here) to decrement the activity_in_progress
4924 	 * reference which is held on the object while the page is in the pageout queue...
4925 	 * just let the normal laundry processing proceed
4926 	 */
4927 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4928 	    (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) ||
4929 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
4930 		return;
4931 	}
4932 	vm_page_queues_remove(m, FALSE);
4933 
4934 	vm_page_check_pageable_safe(m);
4935 	vm_page_queue_enter(&vm_page_queue_cleaned, m, vmp_pageq);
4936 	m->vmp_q_state = VM_PAGE_ON_INACTIVE_CLEANED_Q;
4937 	vm_page_cleaned_count++;
4938 
4939 	vm_page_inactive_count++;
4940 	if (m_object->internal) {
4941 		vm_page_pageable_internal_count++;
4942 	} else {
4943 		vm_page_pageable_external_count++;
4944 	}
4945 	vm_page_add_to_specialq(m, TRUE);
4946 	VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
4947 }
4948 
4949 /*
4950  *	vm_page_activate:
4951  *
4952  *	Put the specified page on the active list (if appropriate).
4953  *
4954  *	The page queues must be locked.
4955  */
4956 
4957 void
vm_page_activate(vm_page_t m)4958 vm_page_activate(
4959 	vm_page_t       m)
4960 {
4961 	vm_object_t     m_object;
4962 
4963 	m_object = VM_PAGE_OBJECT(m);
4964 
4965 	VM_PAGE_CHECK(m);
4966 #ifdef  FIXME_4778297
4967 	assert(m_object != kernel_object);
4968 #endif
4969 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
4970 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
4971 	assert( !(m->vmp_absent && !m->vmp_unusual));
4972 
4973 	if (m->vmp_gobbled) {
4974 		assert( !VM_PAGE_WIRED(m));
4975 		if (!m->vmp_private && !m->vmp_fictitious) {
4976 			vm_page_wire_count--;
4977 		}
4978 		vm_page_gobble_count--;
4979 		m->vmp_gobbled = FALSE;
4980 	}
4981 	/*
4982 	 * if this page is currently on the pageout queue, we can't do the
4983 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
4984 	 * and we can't remove it manually since we would need the object lock
4985 	 * (which is not required here) to decrement the activity_in_progress
4986 	 * reference which is held on the object while the page is in the pageout queue...
4987 	 * just let the normal laundry processing proceed
4988 	 */
4989 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
4990 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
4991 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
4992 		return;
4993 	}
4994 
4995 #if DEBUG
4996 	if (m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q) {
4997 		panic("vm_page_activate: already active");
4998 	}
4999 #endif
5000 
5001 	if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) {
5002 		DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL);
5003 		DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL);
5004 	}
5005 
5006 	/*
5007 	 * A freshly activated page should be promoted in the donation queue.
5008 	 * So we remove it here while preserving its hint and we will enqueue
5009 	 * it again in vm_page_enqueue_active.
5010 	 */
5011 	vm_page_queues_remove(m, ((m->vmp_on_specialq == VM_PAGE_SPECIAL_Q_DONATE) ? TRUE : FALSE));
5012 
5013 	if (!VM_PAGE_WIRED(m)) {
5014 		vm_page_check_pageable_safe(m);
5015 		if (!VM_DYNAMIC_PAGING_ENABLED() &&
5016 		    m->vmp_dirty && m_object->internal &&
5017 		    (m_object->purgable == VM_PURGABLE_DENY ||
5018 		    m_object->purgable == VM_PURGABLE_NONVOLATILE ||
5019 		    m_object->purgable == VM_PURGABLE_VOLATILE)) {
5020 			vm_page_queue_enter(&vm_page_queue_throttled, m, vmp_pageq);
5021 			m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q;
5022 			vm_page_throttled_count++;
5023 		} else {
5024 #if CONFIG_SECLUDED_MEMORY
5025 			if (secluded_for_filecache &&
5026 			    vm_page_secluded_target != 0 &&
5027 			    num_tasks_can_use_secluded_mem == 0 &&
5028 			    m_object->eligible_for_secluded &&
5029 			    !m->vmp_realtime) {
5030 				vm_page_queue_enter(&vm_page_queue_secluded, m, vmp_pageq);
5031 				m->vmp_q_state = VM_PAGE_ON_SECLUDED_Q;
5032 				vm_page_secluded_count++;
5033 				VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
5034 				vm_page_secluded_count_inuse++;
5035 				assert(!m_object->internal);
5036 //				vm_page_pageable_external_count++;
5037 			} else
5038 #endif /* CONFIG_SECLUDED_MEMORY */
5039 			vm_page_enqueue_active(m, FALSE);
5040 		}
5041 		m->vmp_reference = TRUE;
5042 		m->vmp_no_cache = FALSE;
5043 	}
5044 	VM_PAGE_CHECK(m);
5045 }
5046 
5047 
5048 /*
5049  *      vm_page_speculate:
5050  *
5051  *      Put the specified page on the speculative list (if appropriate).
5052  *
5053  *      The page queues must be locked.
5054  */
5055 void
vm_page_speculate(vm_page_t m,boolean_t new)5056 vm_page_speculate(
5057 	vm_page_t       m,
5058 	boolean_t       new)
5059 {
5060 	struct vm_speculative_age_q     *aq;
5061 	vm_object_t     m_object;
5062 
5063 	m_object = VM_PAGE_OBJECT(m);
5064 
5065 	VM_PAGE_CHECK(m);
5066 	vm_page_check_pageable_safe(m);
5067 
5068 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5069 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5070 	assert( !(m->vmp_absent && !m->vmp_unusual));
5071 	assert(m_object->internal == FALSE);
5072 
5073 	/*
5074 	 * if this page is currently on the pageout queue, we can't do the
5075 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5076 	 * and we can't remove it manually since we would need the object lock
5077 	 * (which is not required here) to decrement the activity_in_progress
5078 	 * reference which is held on the object while the page is in the pageout queue...
5079 	 * just let the normal laundry processing proceed
5080 	 */
5081 	if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious ||
5082 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5083 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5084 		return;
5085 	}
5086 
5087 	vm_page_queues_remove(m, FALSE);
5088 
5089 	if (!VM_PAGE_WIRED(m)) {
5090 		mach_timespec_t         ts;
5091 		clock_sec_t sec;
5092 		clock_nsec_t nsec;
5093 
5094 		clock_get_system_nanotime(&sec, &nsec);
5095 		ts.tv_sec = (unsigned int) sec;
5096 		ts.tv_nsec = nsec;
5097 
5098 		if (vm_page_speculative_count == 0) {
5099 			speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5100 			speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5101 
5102 			aq = &vm_page_queue_speculative[speculative_age_index];
5103 
5104 			/*
5105 			 * set the timer to begin a new group
5106 			 */
5107 			aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5108 			aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5109 
5110 			ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5111 		} else {
5112 			aq = &vm_page_queue_speculative[speculative_age_index];
5113 
5114 			if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) {
5115 				speculative_age_index++;
5116 
5117 				if (speculative_age_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) {
5118 					speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5119 				}
5120 				if (speculative_age_index == speculative_steal_index) {
5121 					speculative_steal_index = speculative_age_index + 1;
5122 
5123 					if (speculative_steal_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) {
5124 						speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
5125 					}
5126 				}
5127 				aq = &vm_page_queue_speculative[speculative_age_index];
5128 
5129 				if (!vm_page_queue_empty(&aq->age_q)) {
5130 					vm_page_speculate_ageit(aq);
5131 				}
5132 
5133 				aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000;
5134 				aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC;
5135 
5136 				ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
5137 			}
5138 		}
5139 		vm_page_enqueue_tail(&aq->age_q, &m->vmp_pageq);
5140 		m->vmp_q_state = VM_PAGE_ON_SPECULATIVE_Q;
5141 		vm_page_speculative_count++;
5142 		vm_page_pageable_external_count++;
5143 
5144 		if (new == TRUE) {
5145 			vm_object_lock_assert_exclusive(m_object);
5146 
5147 			m_object->pages_created++;
5148 #if DEVELOPMENT || DEBUG
5149 			vm_page_speculative_created++;
5150 #endif
5151 		}
5152 	}
5153 	VM_PAGE_CHECK(m);
5154 }
5155 
5156 
5157 /*
5158  * move pages from the specified aging bin to
5159  * the speculative bin that pageout_scan claims from
5160  *
5161  *      The page queues must be locked.
5162  */
5163 void
vm_page_speculate_ageit(struct vm_speculative_age_q * aq)5164 vm_page_speculate_ageit(struct vm_speculative_age_q *aq)
5165 {
5166 	struct vm_speculative_age_q     *sq;
5167 	vm_page_t       t;
5168 
5169 	sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q];
5170 
5171 	if (vm_page_queue_empty(&sq->age_q)) {
5172 		sq->age_q.next = aq->age_q.next;
5173 		sq->age_q.prev = aq->age_q.prev;
5174 
5175 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.next);
5176 		t->vmp_pageq.prev = VM_PAGE_PACK_PTR(&sq->age_q);
5177 
5178 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5179 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5180 	} else {
5181 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev);
5182 		t->vmp_pageq.next = aq->age_q.next;
5183 
5184 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.next);
5185 		t->vmp_pageq.prev = sq->age_q.prev;
5186 
5187 		t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.prev);
5188 		t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q);
5189 
5190 		sq->age_q.prev = aq->age_q.prev;
5191 	}
5192 	vm_page_queue_init(&aq->age_q);
5193 }
5194 
5195 
5196 void
vm_page_lru(vm_page_t m)5197 vm_page_lru(
5198 	vm_page_t       m)
5199 {
5200 	VM_PAGE_CHECK(m);
5201 	assert(VM_PAGE_OBJECT(m) != kernel_object);
5202 	assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr);
5203 
5204 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
5205 
5206 	if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q) {
5207 		/*
5208 		 * we don't need to do all the other work that
5209 		 * vm_page_queues_remove and vm_page_enqueue_inactive
5210 		 * bring along for the ride
5211 		 */
5212 		assert(!m->vmp_laundry);
5213 		assert(!m->vmp_private);
5214 
5215 		m->vmp_no_cache = FALSE;
5216 
5217 		vm_page_queue_remove(&vm_page_queue_inactive, m, vmp_pageq);
5218 		vm_page_queue_enter(&vm_page_queue_inactive, m, vmp_pageq);
5219 
5220 		return;
5221 	}
5222 	/*
5223 	 * if this page is currently on the pageout queue, we can't do the
5224 	 * vm_page_queues_remove (which doesn't handle the pageout queue case)
5225 	 * and we can't remove it manually since we would need the object lock
5226 	 * (which is not required here) to decrement the activity_in_progress
5227 	 * reference which is held on the object while the page is in the pageout queue...
5228 	 * just let the normal laundry processing proceed
5229 	 */
5230 	if (m->vmp_laundry || m->vmp_private ||
5231 	    (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) ||
5232 	    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) ||
5233 	    VM_PAGE_WIRED(m)) {
5234 		return;
5235 	}
5236 
5237 	m->vmp_no_cache = FALSE;
5238 
5239 	vm_page_queues_remove(m, FALSE);
5240 
5241 	vm_page_enqueue_inactive(m, FALSE);
5242 }
5243 
5244 
5245 void
vm_page_reactivate_all_throttled(void)5246 vm_page_reactivate_all_throttled(void)
5247 {
5248 	vm_page_t       first_throttled, last_throttled;
5249 	vm_page_t       first_active;
5250 	vm_page_t       m;
5251 	int             extra_active_count;
5252 	int             extra_internal_count, extra_external_count;
5253 	vm_object_t     m_object;
5254 
5255 	if (!VM_DYNAMIC_PAGING_ENABLED()) {
5256 		return;
5257 	}
5258 
5259 	extra_active_count = 0;
5260 	extra_internal_count = 0;
5261 	extra_external_count = 0;
5262 	vm_page_lock_queues();
5263 	if (!vm_page_queue_empty(&vm_page_queue_throttled)) {
5264 		/*
5265 		 * Switch "throttled" pages to "active".
5266 		 */
5267 		vm_page_queue_iterate(&vm_page_queue_throttled, m, vmp_pageq) {
5268 			VM_PAGE_CHECK(m);
5269 			assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
5270 
5271 			m_object = VM_PAGE_OBJECT(m);
5272 
5273 			extra_active_count++;
5274 			if (m_object->internal) {
5275 				extra_internal_count++;
5276 			} else {
5277 				extra_external_count++;
5278 			}
5279 
5280 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5281 			VM_PAGE_CHECK(m);
5282 			vm_page_add_to_specialq(m, FALSE);
5283 		}
5284 
5285 		/*
5286 		 * Transfer the entire throttled queue to a regular LRU page queues.
5287 		 * We insert it at the head of the active queue, so that these pages
5288 		 * get re-evaluated by the LRU algorithm first, since they've been
5289 		 * completely out of it until now.
5290 		 */
5291 		first_throttled = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
5292 		last_throttled = (vm_page_t) vm_page_queue_last(&vm_page_queue_throttled);
5293 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5294 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5295 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5296 		} else {
5297 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled);
5298 		}
5299 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_throttled);
5300 		first_throttled->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5301 		last_throttled->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5302 
5303 #if DEBUG
5304 		printf("reactivated %d throttled pages\n", vm_page_throttled_count);
5305 #endif
5306 		vm_page_queue_init(&vm_page_queue_throttled);
5307 		/*
5308 		 * Adjust the global page counts.
5309 		 */
5310 		vm_page_active_count += extra_active_count;
5311 		vm_page_pageable_internal_count += extra_internal_count;
5312 		vm_page_pageable_external_count += extra_external_count;
5313 		vm_page_throttled_count = 0;
5314 	}
5315 	assert(vm_page_throttled_count == 0);
5316 	assert(vm_page_queue_empty(&vm_page_queue_throttled));
5317 	vm_page_unlock_queues();
5318 }
5319 
5320 
5321 /*
5322  * move pages from the indicated local queue to the global active queue
5323  * its ok to fail if we're below the hard limit and force == FALSE
5324  * the nolocks == TRUE case is to allow this function to be run on
5325  * the hibernate path
5326  */
5327 
5328 void
vm_page_reactivate_local(uint32_t lid,boolean_t force,boolean_t nolocks)5329 vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks)
5330 {
5331 	struct vpl      *lq;
5332 	vm_page_t       first_local, last_local;
5333 	vm_page_t       first_active;
5334 	vm_page_t       m;
5335 	uint32_t        count = 0;
5336 
5337 	if (vm_page_local_q == NULL) {
5338 		return;
5339 	}
5340 
5341 	lq = zpercpu_get_cpu(vm_page_local_q, lid);
5342 
5343 	if (nolocks == FALSE) {
5344 		if (lq->vpl_count < vm_page_local_q_hard_limit && force == FALSE) {
5345 			if (!vm_page_trylockspin_queues()) {
5346 				return;
5347 			}
5348 		} else {
5349 			vm_page_lockspin_queues();
5350 		}
5351 
5352 		VPL_LOCK(&lq->vpl_lock);
5353 	}
5354 	if (lq->vpl_count) {
5355 		/*
5356 		 * Switch "local" pages to "active".
5357 		 */
5358 		assert(!vm_page_queue_empty(&lq->vpl_queue));
5359 
5360 		vm_page_queue_iterate(&lq->vpl_queue, m, vmp_pageq) {
5361 			VM_PAGE_CHECK(m);
5362 			vm_page_check_pageable_safe(m);
5363 			assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q);
5364 			assert(!m->vmp_fictitious);
5365 
5366 			if (m->vmp_local_id != lid) {
5367 				panic("vm_page_reactivate_local: found vm_page_t(%p) with wrong cpuid", m);
5368 			}
5369 
5370 			m->vmp_local_id = 0;
5371 			m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
5372 			VM_PAGE_CHECK(m);
5373 			vm_page_add_to_specialq(m, FALSE);
5374 			count++;
5375 		}
5376 		if (count != lq->vpl_count) {
5377 			panic("vm_page_reactivate_local: count = %d, vm_page_local_count = %d", count, lq->vpl_count);
5378 		}
5379 
5380 		/*
5381 		 * Transfer the entire local queue to a regular LRU page queues.
5382 		 */
5383 		first_local = (vm_page_t) vm_page_queue_first(&lq->vpl_queue);
5384 		last_local = (vm_page_t) vm_page_queue_last(&lq->vpl_queue);
5385 		first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
5386 
5387 		if (vm_page_queue_empty(&vm_page_queue_active)) {
5388 			vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5389 		} else {
5390 			first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local);
5391 		}
5392 		vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_local);
5393 		first_local->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active);
5394 		last_local->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active);
5395 
5396 		vm_page_queue_init(&lq->vpl_queue);
5397 		/*
5398 		 * Adjust the global page counts.
5399 		 */
5400 		vm_page_active_count += lq->vpl_count;
5401 		vm_page_pageable_internal_count += lq->vpl_internal_count;
5402 		vm_page_pageable_external_count += lq->vpl_external_count;
5403 		lq->vpl_count = 0;
5404 		lq->vpl_internal_count = 0;
5405 		lq->vpl_external_count = 0;
5406 	}
5407 	assert(vm_page_queue_empty(&lq->vpl_queue));
5408 
5409 	if (nolocks == FALSE) {
5410 		VPL_UNLOCK(&lq->vpl_lock);
5411 
5412 		vm_page_balance_inactive(count / 4);
5413 		vm_page_unlock_queues();
5414 	}
5415 }
5416 
5417 /*
5418  *	vm_page_part_zero_fill:
5419  *
5420  *	Zero-fill a part of the page.
5421  */
5422 #define PMAP_ZERO_PART_PAGE_IMPLEMENTED
5423 void
vm_page_part_zero_fill(vm_page_t m,vm_offset_t m_pa,vm_size_t len)5424 vm_page_part_zero_fill(
5425 	vm_page_t       m,
5426 	vm_offset_t     m_pa,
5427 	vm_size_t       len)
5428 {
5429 #if 0
5430 	/*
5431 	 * we don't hold the page queue lock
5432 	 * so this check isn't safe to make
5433 	 */
5434 	VM_PAGE_CHECK(m);
5435 #endif
5436 
5437 #ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED
5438 	pmap_zero_part_page(VM_PAGE_GET_PHYS_PAGE(m), m_pa, len);
5439 #else
5440 	vm_page_t       tmp;
5441 	while (1) {
5442 		tmp = vm_page_grab();
5443 		if (tmp == VM_PAGE_NULL) {
5444 			vm_page_wait(THREAD_UNINT);
5445 			continue;
5446 		}
5447 		break;
5448 	}
5449 	vm_page_zero_fill(tmp);
5450 	if (m_pa != 0) {
5451 		vm_page_part_copy(m, 0, tmp, 0, m_pa);
5452 	}
5453 	if ((m_pa + len) < PAGE_SIZE) {
5454 		vm_page_part_copy(m, m_pa + len, tmp,
5455 		    m_pa + len, PAGE_SIZE - (m_pa + len));
5456 	}
5457 	vm_page_copy(tmp, m);
5458 	VM_PAGE_FREE(tmp);
5459 #endif
5460 }
5461 
5462 /*
5463  *	vm_page_zero_fill:
5464  *
5465  *	Zero-fill the specified page.
5466  */
5467 void
vm_page_zero_fill(vm_page_t m)5468 vm_page_zero_fill(
5469 	vm_page_t       m)
5470 {
5471 #if 0
5472 	/*
5473 	 * we don't hold the page queue lock
5474 	 * so this check isn't safe to make
5475 	 */
5476 	VM_PAGE_CHECK(m);
5477 #endif
5478 
5479 //	dbgTrace(0xAEAEAEAE, VM_PAGE_GET_PHYS_PAGE(m), 0);		/* (BRINGUP) */
5480 	pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m));
5481 }
5482 
5483 /*
5484  *	vm_page_part_copy:
5485  *
5486  *	copy part of one page to another
5487  */
5488 
5489 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)5490 vm_page_part_copy(
5491 	vm_page_t       src_m,
5492 	vm_offset_t     src_pa,
5493 	vm_page_t       dst_m,
5494 	vm_offset_t     dst_pa,
5495 	vm_size_t       len)
5496 {
5497 #if 0
5498 	/*
5499 	 * we don't hold the page queue lock
5500 	 * so this check isn't safe to make
5501 	 */
5502 	VM_PAGE_CHECK(src_m);
5503 	VM_PAGE_CHECK(dst_m);
5504 #endif
5505 	pmap_copy_part_page(VM_PAGE_GET_PHYS_PAGE(src_m), src_pa,
5506 	    VM_PAGE_GET_PHYS_PAGE(dst_m), dst_pa, len);
5507 }
5508 
5509 /*
5510  *	vm_page_copy:
5511  *
5512  *	Copy one page to another
5513  */
5514 
5515 int vm_page_copy_cs_validations = 0;
5516 int vm_page_copy_cs_tainted = 0;
5517 
5518 void
vm_page_copy(vm_page_t src_m,vm_page_t dest_m)5519 vm_page_copy(
5520 	vm_page_t       src_m,
5521 	vm_page_t       dest_m)
5522 {
5523 	vm_object_t     src_m_object;
5524 
5525 	src_m_object = VM_PAGE_OBJECT(src_m);
5526 
5527 #if 0
5528 	/*
5529 	 * we don't hold the page queue lock
5530 	 * so this check isn't safe to make
5531 	 */
5532 	VM_PAGE_CHECK(src_m);
5533 	VM_PAGE_CHECK(dest_m);
5534 #endif
5535 	vm_object_lock_assert_held(src_m_object);
5536 
5537 	if (src_m_object != VM_OBJECT_NULL &&
5538 	    src_m_object->code_signed) {
5539 		/*
5540 		 * We're copying a page from a code-signed object.
5541 		 * Whoever ends up mapping the copy page might care about
5542 		 * the original page's integrity, so let's validate the
5543 		 * source page now.
5544 		 */
5545 		vm_page_copy_cs_validations++;
5546 		vm_page_validate_cs(src_m, PAGE_SIZE, 0);
5547 #if DEVELOPMENT || DEBUG
5548 		DTRACE_VM4(codesigned_copy,
5549 		    vm_object_t, src_m_object,
5550 		    vm_object_offset_t, src_m->vmp_offset,
5551 		    int, src_m->vmp_cs_validated,
5552 		    int, src_m->vmp_cs_tainted);
5553 #endif /* DEVELOPMENT || DEBUG */
5554 	}
5555 
5556 	/*
5557 	 * Propagate the cs_tainted bit to the copy page. Do not propagate
5558 	 * the cs_validated bit.
5559 	 */
5560 	dest_m->vmp_cs_tainted = src_m->vmp_cs_tainted;
5561 	dest_m->vmp_cs_nx = src_m->vmp_cs_nx;
5562 	if (dest_m->vmp_cs_tainted) {
5563 		vm_page_copy_cs_tainted++;
5564 	}
5565 	dest_m->vmp_error = VMP_ERROR_GET(src_m); /* sliding src_m might have failed... */
5566 	pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(src_m), VM_PAGE_GET_PHYS_PAGE(dest_m));
5567 }
5568 
5569 #if MACH_ASSERT
5570 static void
_vm_page_print(vm_page_t p)5571 _vm_page_print(
5572 	vm_page_t       p)
5573 {
5574 	printf("vm_page %p: \n", p);
5575 	printf("  pageq: next=%p prev=%p\n",
5576 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.next),
5577 	    (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev));
5578 	printf("  listq: next=%p prev=%p\n",
5579 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.next)),
5580 	    (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.prev)));
5581 	printf("  next=%p\n", (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m)));
5582 	printf("  object=%p offset=0x%llx\n", VM_PAGE_OBJECT(p), p->vmp_offset);
5583 	printf("  wire_count=%u\n", p->vmp_wire_count);
5584 	printf("  q_state=%u\n", p->vmp_q_state);
5585 
5586 	printf("  %slaundry, %sref, %sgobbled, %sprivate\n",
5587 	    (p->vmp_laundry ? "" : "!"),
5588 	    (p->vmp_reference ? "" : "!"),
5589 	    (p->vmp_gobbled ? "" : "!"),
5590 	    (p->vmp_private ? "" : "!"));
5591 	printf("  %sbusy, %swanted, %stabled, %sfictitious, %spmapped, %swpmapped\n",
5592 	    (p->vmp_busy ? "" : "!"),
5593 	    (p->vmp_wanted ? "" : "!"),
5594 	    (p->vmp_tabled ? "" : "!"),
5595 	    (p->vmp_fictitious ? "" : "!"),
5596 	    (p->vmp_pmapped ? "" : "!"),
5597 	    (p->vmp_wpmapped ? "" : "!"));
5598 	printf("  %sfree_when_done, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n",
5599 	    (p->vmp_free_when_done ? "" : "!"),
5600 	    (p->vmp_absent ? "" : "!"),
5601 	    (VMP_ERROR_GET(p) ? "" : "!"),
5602 	    (p->vmp_dirty ? "" : "!"),
5603 	    (p->vmp_cleaning ? "" : "!"),
5604 	    (p->vmp_precious ? "" : "!"),
5605 	    (p->vmp_clustered ? "" : "!"));
5606 	printf("  %soverwriting, %srestart, %sunusual\n",
5607 	    (p->vmp_overwriting ? "" : "!"),
5608 	    (p->vmp_restart ? "" : "!"),
5609 	    (p->vmp_unusual ? "" : "!"));
5610 	printf("  cs_validated=%d, cs_tainted=%d, cs_nx=%d, %sno_cache\n",
5611 	    p->vmp_cs_validated,
5612 	    p->vmp_cs_tainted,
5613 	    p->vmp_cs_nx,
5614 	    (p->vmp_no_cache ? "" : "!"));
5615 
5616 	printf("phys_page=0x%x\n", VM_PAGE_GET_PHYS_PAGE(p));
5617 }
5618 
5619 /*
5620  *	Check that the list of pages is ordered by
5621  *	ascending physical address and has no holes.
5622  */
5623 static int
vm_page_verify_contiguous(vm_page_t pages,unsigned int npages)5624 vm_page_verify_contiguous(
5625 	vm_page_t       pages,
5626 	unsigned int    npages)
5627 {
5628 	vm_page_t               m;
5629 	unsigned int            page_count;
5630 	vm_offset_t             prev_addr;
5631 
5632 	prev_addr = VM_PAGE_GET_PHYS_PAGE(pages);
5633 	page_count = 1;
5634 	for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
5635 		if (VM_PAGE_GET_PHYS_PAGE(m) != prev_addr + 1) {
5636 			printf("m %p prev_addr 0x%lx, current addr 0x%x\n",
5637 			    m, (long)prev_addr, VM_PAGE_GET_PHYS_PAGE(m));
5638 			printf("pages %p page_count %d npages %d\n", pages, page_count, npages);
5639 			panic("vm_page_verify_contiguous:  not contiguous!");
5640 		}
5641 		prev_addr = VM_PAGE_GET_PHYS_PAGE(m);
5642 		++page_count;
5643 	}
5644 	if (page_count != npages) {
5645 		printf("pages %p actual count 0x%x but requested 0x%x\n",
5646 		    pages, page_count, npages);
5647 		panic("vm_page_verify_contiguous:  count error");
5648 	}
5649 	return 1;
5650 }
5651 
5652 
5653 /*
5654  *	Check the free lists for proper length etc.
5655  */
5656 static boolean_t vm_page_verify_this_free_list_enabled = FALSE;
5657 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)5658 vm_page_verify_free_list(
5659 	vm_page_queue_head_t    *vm_page_queue,
5660 	unsigned int    color,
5661 	vm_page_t       look_for_page,
5662 	boolean_t       expect_page)
5663 {
5664 	unsigned int    npages;
5665 	vm_page_t       m;
5666 	vm_page_t       prev_m;
5667 	boolean_t       found_page;
5668 
5669 	if (!vm_page_verify_this_free_list_enabled) {
5670 		return 0;
5671 	}
5672 
5673 	found_page = FALSE;
5674 	npages = 0;
5675 	prev_m = (vm_page_t)((uintptr_t)vm_page_queue);
5676 
5677 	vm_page_queue_iterate(vm_page_queue, m, vmp_pageq) {
5678 		if (m == look_for_page) {
5679 			found_page = TRUE;
5680 		}
5681 		if ((vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev) != prev_m) {
5682 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p corrupted prev ptr %p instead of %p",
5683 			    color, npages, m, (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev), prev_m);
5684 		}
5685 		if (!m->vmp_busy) {
5686 			panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy",
5687 			    color, npages, m);
5688 		}
5689 		if (color != (unsigned int) -1) {
5690 			if (VM_PAGE_GET_COLOR(m) != color) {
5691 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u",
5692 				    color, npages, m, VM_PAGE_GET_COLOR(m), color);
5693 			}
5694 			if (m->vmp_q_state != VM_PAGE_ON_FREE_Q) {
5695 				panic("vm_page_verify_free_list(color=%u, npages=%u): page %p - expecting q_state == VM_PAGE_ON_FREE_Q, found %d",
5696 				    color, npages, m, m->vmp_q_state);
5697 			}
5698 		} else {
5699 			if (m->vmp_q_state != VM_PAGE_ON_FREE_LOCAL_Q) {
5700 				panic("vm_page_verify_free_list(npages=%u): local page %p - expecting q_state == VM_PAGE_ON_FREE_LOCAL_Q, found %d",
5701 				    npages, m, m->vmp_q_state);
5702 			}
5703 		}
5704 		++npages;
5705 		prev_m = m;
5706 	}
5707 	if (look_for_page != VM_PAGE_NULL) {
5708 		unsigned int other_color;
5709 
5710 		if (expect_page && !found_page) {
5711 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p not found phys=%u\n",
5712 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5713 			_vm_page_print(look_for_page);
5714 			for (other_color = 0;
5715 			    other_color < vm_colors;
5716 			    other_color++) {
5717 				if (other_color == color) {
5718 					continue;
5719 				}
5720 				vm_page_verify_free_list(&vm_page_queue_free[other_color].qhead,
5721 				    other_color, look_for_page, FALSE);
5722 			}
5723 			if (color == (unsigned int) -1) {
5724 				vm_page_verify_free_list(&vm_lopage_queue_free,
5725 				    (unsigned int) -1, look_for_page, FALSE);
5726 			}
5727 			panic("vm_page_verify_free_list(color=%u)", color);
5728 		}
5729 		if (!expect_page && found_page) {
5730 			printf("vm_page_verify_free_list(color=%u, npages=%u): page %p found phys=%u\n",
5731 			    color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page));
5732 		}
5733 	}
5734 	return npages;
5735 }
5736 
5737 static boolean_t vm_page_verify_all_free_lists_enabled = FALSE;
5738 static void
vm_page_verify_free_lists(void)5739 vm_page_verify_free_lists( void )
5740 {
5741 	unsigned int    color, npages, nlopages;
5742 	boolean_t       toggle = TRUE;
5743 
5744 	if (!vm_page_verify_all_free_lists_enabled) {
5745 		return;
5746 	}
5747 
5748 	npages = 0;
5749 
5750 	vm_free_page_lock();
5751 
5752 	if (vm_page_verify_this_free_list_enabled == TRUE) {
5753 		/*
5754 		 * This variable has been set globally for extra checking of
5755 		 * each free list Q. Since we didn't set it, we don't own it
5756 		 * and we shouldn't toggle it.
5757 		 */
5758 		toggle = FALSE;
5759 	}
5760 
5761 	if (toggle == TRUE) {
5762 		vm_page_verify_this_free_list_enabled = TRUE;
5763 	}
5764 
5765 	for (color = 0; color < vm_colors; color++) {
5766 		npages += vm_page_verify_free_list(&vm_page_queue_free[color].qhead,
5767 		    color, VM_PAGE_NULL, FALSE);
5768 	}
5769 	nlopages = vm_page_verify_free_list(&vm_lopage_queue_free,
5770 	    (unsigned int) -1,
5771 	    VM_PAGE_NULL, FALSE);
5772 	if (npages != vm_page_free_count || nlopages != vm_lopage_free_count) {
5773 		panic("vm_page_verify_free_lists:  "
5774 		    "npages %u free_count %d nlopages %u lo_free_count %u",
5775 		    npages, vm_page_free_count, nlopages, vm_lopage_free_count);
5776 	}
5777 
5778 	if (toggle == TRUE) {
5779 		vm_page_verify_this_free_list_enabled = FALSE;
5780 	}
5781 
5782 	vm_free_page_unlock();
5783 }
5784 
5785 #endif  /* MACH_ASSERT */
5786 
5787 
5788 extern boolean_t(*volatile consider_buffer_cache_collect)(int);
5789 
5790 /*
5791  *	CONTIGUOUS PAGE ALLOCATION
5792  *
5793  *	Find a region large enough to contain at least n pages
5794  *	of contiguous physical memory.
5795  *
5796  *	This is done by traversing the vm_page_t array in a linear fashion
5797  *	we assume that the vm_page_t array has the avaiable physical pages in an
5798  *	ordered, ascending list... this is currently true of all our implementations
5799  *      and must remain so... there can be 'holes' in the array...  we also can
5800  *	no longer tolerate the vm_page_t's in the list being 'freed' and reclaimed
5801  *      which use to happen via 'vm_page_convert'... that function was no longer
5802  *      being called and was removed...
5803  *
5804  *	The basic flow consists of stabilizing some of the interesting state of
5805  *	a vm_page_t behind the vm_page_queue and vm_page_free locks... we start our
5806  *	sweep at the beginning of the array looking for pages that meet our criterea
5807  *	for a 'stealable' page... currently we are pretty conservative... if the page
5808  *	meets this criterea and is physically contiguous to the previous page in the 'run'
5809  *      we keep developing it.  If we hit a page that doesn't fit, we reset our state
5810  *	and start to develop a new run... if at this point we've already considered
5811  *      at least MAX_CONSIDERED_BEFORE_YIELD pages, we'll drop the 2 locks we hold,
5812  *	and mutex_pause (which will yield the processor), to keep the latency low w/r
5813  *	to other threads trying to acquire free pages (or move pages from q to q),
5814  *	and then continue from the spot we left off... we only make 1 pass through the
5815  *	array.  Once we have a 'run' that is long enough, we'll go into the loop which
5816  *      which steals the pages from the queues they're currently on... pages on the free
5817  *	queue can be stolen directly... pages that are on any of the other queues
5818  *	must be removed from the object they are tabled on... this requires taking the
5819  *      object lock... we do this as a 'try' to prevent deadlocks... if the 'try' fails
5820  *	or if the state of the page behind the vm_object lock is no longer viable, we'll
5821  *	dump the pages we've currently stolen back to the free list, and pick up our
5822  *	scan from the point where we aborted the 'current' run.
5823  *
5824  *
5825  *	Requirements:
5826  *		- neither vm_page_queue nor vm_free_list lock can be held on entry
5827  *
5828  *	Returns a pointer to a list of gobbled/wired pages or VM_PAGE_NULL.
5829  *
5830  * Algorithm:
5831  */
5832 
5833 #define MAX_CONSIDERED_BEFORE_YIELD     1000
5834 
5835 
5836 #define RESET_STATE_OF_RUN()    \
5837 	MACRO_BEGIN             \
5838 	prevcontaddr = -2;      \
5839 	start_pnum = -1;        \
5840 	free_considered = 0;    \
5841 	substitute_needed = 0;  \
5842 	npages = 0;             \
5843 	MACRO_END
5844 
5845 /*
5846  * Can we steal in-use (i.e. not free) pages when searching for
5847  * physically-contiguous pages ?
5848  */
5849 #define VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL 1
5850 
5851 static unsigned int vm_page_find_contiguous_last_idx = 0, vm_page_lomem_find_contiguous_last_idx = 0;
5852 #if DEBUG
5853 int vm_page_find_contig_debug = 0;
5854 #endif
5855 
5856 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)5857 vm_page_find_contiguous(
5858 	unsigned int    contig_pages,
5859 	ppnum_t         max_pnum,
5860 	ppnum_t     pnum_mask,
5861 	boolean_t       wire,
5862 	int             flags)
5863 {
5864 	vm_page_t       m = NULL;
5865 	ppnum_t         prevcontaddr = 0;
5866 	ppnum_t         start_pnum = 0;
5867 	unsigned int    npages = 0, considered = 0, scanned = 0;
5868 	unsigned int    page_idx = 0, start_idx = 0, last_idx = 0, orig_last_idx = 0;
5869 	unsigned int    idx_last_contig_page_found = 0;
5870 	int             free_considered = 0, free_available = 0;
5871 	int             substitute_needed = 0;
5872 	int             zone_gc_called = 0;
5873 	boolean_t       wrapped;
5874 	kern_return_t   kr;
5875 #if DEBUG
5876 	clock_sec_t     tv_start_sec = 0, tv_end_sec = 0;
5877 	clock_usec_t    tv_start_usec = 0, tv_end_usec = 0;
5878 #endif
5879 
5880 	int             yielded = 0;
5881 	int             dumped_run = 0;
5882 	int             stolen_pages = 0;
5883 	int             compressed_pages = 0;
5884 
5885 
5886 	if (contig_pages == 0) {
5887 		return VM_PAGE_NULL;
5888 	}
5889 
5890 full_scan_again:
5891 
5892 #if MACH_ASSERT
5893 	vm_page_verify_free_lists();
5894 #endif
5895 #if DEBUG
5896 	clock_get_system_microtime(&tv_start_sec, &tv_start_usec);
5897 #endif
5898 	PAGE_REPLACEMENT_ALLOWED(TRUE);
5899 
5900 	/*
5901 	 * If there are still delayed pages, try to free up some that match.
5902 	 */
5903 	if (__improbable(vm_delayed_count != 0 && contig_pages != 0)) {
5904 		vm_free_delayed_pages_contig(contig_pages, max_pnum, pnum_mask);
5905 	}
5906 
5907 	vm_page_lock_queues();
5908 	vm_free_page_lock();
5909 
5910 	RESET_STATE_OF_RUN();
5911 
5912 	scanned = 0;
5913 	considered = 0;
5914 	free_available = vm_page_free_count - vm_page_free_reserved;
5915 
5916 	wrapped = FALSE;
5917 
5918 	if (flags & KMA_LOMEM) {
5919 		idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx;
5920 	} else {
5921 		idx_last_contig_page_found =  vm_page_find_contiguous_last_idx;
5922 	}
5923 
5924 	orig_last_idx = idx_last_contig_page_found;
5925 	last_idx = orig_last_idx;
5926 
5927 	for (page_idx = last_idx, start_idx = last_idx;
5928 	    npages < contig_pages && page_idx < vm_pages_count;
5929 	    page_idx++) {
5930 retry:
5931 		if (wrapped &&
5932 		    npages == 0 &&
5933 		    page_idx >= orig_last_idx) {
5934 			/*
5935 			 * We're back where we started and we haven't
5936 			 * found any suitable contiguous range.  Let's
5937 			 * give up.
5938 			 */
5939 			break;
5940 		}
5941 		scanned++;
5942 		m = &vm_pages[page_idx];
5943 
5944 		assert(!m->vmp_fictitious);
5945 		assert(!m->vmp_private);
5946 
5947 		if (max_pnum && VM_PAGE_GET_PHYS_PAGE(m) > max_pnum) {
5948 			/* no more low pages... */
5949 			break;
5950 		}
5951 		if (!npages & ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0)) {
5952 			/*
5953 			 * not aligned
5954 			 */
5955 			RESET_STATE_OF_RUN();
5956 		} else if (VM_PAGE_WIRED(m) || m->vmp_gobbled ||
5957 		    m->vmp_laundry || m->vmp_wanted ||
5958 		    m->vmp_cleaning || m->vmp_overwriting || m->vmp_free_when_done) {
5959 			/*
5960 			 * page is in a transient state
5961 			 * or a state we don't want to deal
5962 			 * with, so don't consider it which
5963 			 * means starting a new run
5964 			 */
5965 			RESET_STATE_OF_RUN();
5966 		} else if ((m->vmp_q_state == VM_PAGE_NOT_ON_Q) ||
5967 		    (m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q) ||
5968 		    (m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q) ||
5969 		    (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
5970 			/*
5971 			 * page needs to be on one of our queues (other then the pageout or special free queues)
5972 			 * or it needs to belong to the compressor pool (which is now indicated
5973 			 * by vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR and falls out
5974 			 * from the check for VM_PAGE_NOT_ON_Q)
5975 			 * in order for it to be stable behind the
5976 			 * locks we hold at this point...
5977 			 * if not, don't consider it which
5978 			 * means starting a new run
5979 			 */
5980 			RESET_STATE_OF_RUN();
5981 		} else if ((m->vmp_q_state != VM_PAGE_ON_FREE_Q) && (!m->vmp_tabled || m->vmp_busy)) {
5982 			/*
5983 			 * pages on the free list are always 'busy'
5984 			 * so we couldn't test for 'busy' in the check
5985 			 * for the transient states... pages that are
5986 			 * 'free' are never 'tabled', so we also couldn't
5987 			 * test for 'tabled'.  So we check here to make
5988 			 * sure that a non-free page is not busy and is
5989 			 * tabled on an object...
5990 			 * if not, don't consider it which
5991 			 * means starting a new run
5992 			 */
5993 			RESET_STATE_OF_RUN();
5994 		} else {
5995 			if (VM_PAGE_GET_PHYS_PAGE(m) != prevcontaddr + 1) {
5996 				if ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0) {
5997 					RESET_STATE_OF_RUN();
5998 					goto did_consider;
5999 				} else {
6000 					npages = 1;
6001 					start_idx = page_idx;
6002 					start_pnum = VM_PAGE_GET_PHYS_PAGE(m);
6003 				}
6004 			} else {
6005 				npages++;
6006 			}
6007 			prevcontaddr = VM_PAGE_GET_PHYS_PAGE(m);
6008 
6009 			VM_PAGE_CHECK(m);
6010 			if (m->vmp_q_state == VM_PAGE_ON_FREE_Q) {
6011 				free_considered++;
6012 			} else {
6013 				/*
6014 				 * This page is not free.
6015 				 * If we can't steal used pages,
6016 				 * we have to give up this run
6017 				 * and keep looking.
6018 				 * Otherwise, we might need to
6019 				 * move the contents of this page
6020 				 * into a substitute page.
6021 				 */
6022 #if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
6023 				if (m->vmp_pmapped || m->vmp_dirty || m->vmp_precious) {
6024 					substitute_needed++;
6025 				}
6026 #else
6027 				RESET_STATE_OF_RUN();
6028 #endif
6029 			}
6030 
6031 			if ((free_considered + substitute_needed) > free_available) {
6032 				/*
6033 				 * if we let this run continue
6034 				 * we will end up dropping the vm_page_free_count
6035 				 * below the reserve limit... we need to abort
6036 				 * this run, but we can at least re-consider this
6037 				 * page... thus the jump back to 'retry'
6038 				 */
6039 				RESET_STATE_OF_RUN();
6040 
6041 				if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) {
6042 					considered++;
6043 					goto retry;
6044 				}
6045 				/*
6046 				 * free_available == 0
6047 				 * so can't consider any free pages... if
6048 				 * we went to retry in this case, we'd
6049 				 * get stuck looking at the same page
6050 				 * w/o making any forward progress
6051 				 * we also want to take this path if we've already
6052 				 * reached our limit that controls the lock latency
6053 				 */
6054 			}
6055 		}
6056 did_consider:
6057 		if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) {
6058 			PAGE_REPLACEMENT_ALLOWED(FALSE);
6059 
6060 			vm_free_page_unlock();
6061 			vm_page_unlock_queues();
6062 
6063 			mutex_pause(0);
6064 
6065 			PAGE_REPLACEMENT_ALLOWED(TRUE);
6066 
6067 			vm_page_lock_queues();
6068 			vm_free_page_lock();
6069 
6070 			RESET_STATE_OF_RUN();
6071 			/*
6072 			 * reset our free page limit since we
6073 			 * dropped the lock protecting the vm_page_free_queue
6074 			 */
6075 			free_available = vm_page_free_count - vm_page_free_reserved;
6076 			considered = 0;
6077 
6078 			yielded++;
6079 
6080 			goto retry;
6081 		}
6082 		considered++;
6083 	}
6084 	m = VM_PAGE_NULL;
6085 
6086 	if (npages != contig_pages) {
6087 		if (!wrapped) {
6088 			/*
6089 			 * We didn't find a contiguous range but we didn't
6090 			 * start from the very first page.
6091 			 * Start again from the very first page.
6092 			 */
6093 			RESET_STATE_OF_RUN();
6094 			if (flags & KMA_LOMEM) {
6095 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = 0;
6096 			} else {
6097 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0;
6098 			}
6099 			last_idx = 0;
6100 			page_idx = last_idx;
6101 			wrapped = TRUE;
6102 			goto retry;
6103 		}
6104 		vm_free_page_unlock();
6105 	} else {
6106 		vm_page_t       m1;
6107 		vm_page_t       m2;
6108 		unsigned int    cur_idx;
6109 		unsigned int    tmp_start_idx;
6110 		vm_object_t     locked_object = VM_OBJECT_NULL;
6111 		boolean_t       abort_run = FALSE;
6112 
6113 		assert(page_idx - start_idx == contig_pages);
6114 
6115 		tmp_start_idx = start_idx;
6116 
6117 		/*
6118 		 * first pass through to pull the free pages
6119 		 * off of the free queue so that in case we
6120 		 * need substitute pages, we won't grab any
6121 		 * of the free pages in the run... we'll clear
6122 		 * the 'free' bit in the 2nd pass, and even in
6123 		 * an abort_run case, we'll collect all of the
6124 		 * free pages in this run and return them to the free list
6125 		 */
6126 		while (start_idx < page_idx) {
6127 			m1 = &vm_pages[start_idx++];
6128 
6129 #if !VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL
6130 			assert(m1->vmp_q_state == VM_PAGE_ON_FREE_Q);
6131 #endif
6132 
6133 			if (m1->vmp_q_state == VM_PAGE_ON_FREE_Q) {
6134 				unsigned int color;
6135 
6136 				color = VM_PAGE_GET_COLOR(m1);
6137 #if MACH_ASSERT
6138 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, m1, TRUE);
6139 #endif
6140 				vm_page_queue_remove(&vm_page_queue_free[color].qhead, m1, vmp_pageq);
6141 
6142 				VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6143 #if MACH_ASSERT
6144 				vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, VM_PAGE_NULL, FALSE);
6145 #endif
6146 				/*
6147 				 * Clear the "free" bit so that this page
6148 				 * does not get considered for another
6149 				 * concurrent physically-contiguous allocation.
6150 				 */
6151 				m1->vmp_q_state = VM_PAGE_NOT_ON_Q;
6152 				assert(m1->vmp_busy);
6153 
6154 				vm_page_free_count--;
6155 			}
6156 		}
6157 		if (flags & KMA_LOMEM) {
6158 			vm_page_lomem_find_contiguous_last_idx = page_idx;
6159 		} else {
6160 			vm_page_find_contiguous_last_idx = page_idx;
6161 		}
6162 
6163 		/*
6164 		 * we can drop the free queue lock at this point since
6165 		 * we've pulled any 'free' candidates off of the list
6166 		 * we need it dropped so that we can do a vm_page_grab
6167 		 * when substituing for pmapped/dirty pages
6168 		 */
6169 		vm_free_page_unlock();
6170 
6171 		start_idx = tmp_start_idx;
6172 		cur_idx = page_idx - 1;
6173 
6174 		while (start_idx++ < page_idx) {
6175 			/*
6176 			 * must go through the list from back to front
6177 			 * so that the page list is created in the
6178 			 * correct order - low -> high phys addresses
6179 			 */
6180 			m1 = &vm_pages[cur_idx--];
6181 
6182 			if (m1->vmp_object == 0) {
6183 				/*
6184 				 * page has already been removed from
6185 				 * the free list in the 1st pass
6186 				 */
6187 				assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6188 				assert(m1->vmp_offset == (vm_object_offset_t) -1);
6189 				assert(m1->vmp_busy);
6190 				assert(!m1->vmp_wanted);
6191 				assert(!m1->vmp_laundry);
6192 			} else {
6193 				vm_object_t object;
6194 				int refmod;
6195 				boolean_t disconnected, reusable;
6196 
6197 				if (abort_run == TRUE) {
6198 					continue;
6199 				}
6200 
6201 				assert(m1->vmp_q_state != VM_PAGE_NOT_ON_Q);
6202 
6203 				object = VM_PAGE_OBJECT(m1);
6204 
6205 				if (object != locked_object) {
6206 					if (locked_object) {
6207 						vm_object_unlock(locked_object);
6208 						locked_object = VM_OBJECT_NULL;
6209 					}
6210 					if (vm_object_lock_try(object)) {
6211 						locked_object = object;
6212 					}
6213 				}
6214 				if (locked_object == VM_OBJECT_NULL ||
6215 				    (VM_PAGE_WIRED(m1) || m1->vmp_gobbled ||
6216 				    m1->vmp_laundry || m1->vmp_wanted ||
6217 				    m1->vmp_cleaning || m1->vmp_overwriting || m1->vmp_free_when_done || m1->vmp_busy) ||
6218 				    (m1->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) {
6219 					if (locked_object) {
6220 						vm_object_unlock(locked_object);
6221 						locked_object = VM_OBJECT_NULL;
6222 					}
6223 					tmp_start_idx = cur_idx;
6224 					abort_run = TRUE;
6225 					continue;
6226 				}
6227 
6228 				disconnected = FALSE;
6229 				reusable = FALSE;
6230 
6231 				if ((m1->vmp_reusable ||
6232 				    object->all_reusable) &&
6233 				    (m1->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q) &&
6234 				    !m1->vmp_dirty &&
6235 				    !m1->vmp_reference) {
6236 					/* reusable page... */
6237 					refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6238 					disconnected = TRUE;
6239 					if (refmod == 0) {
6240 						/*
6241 						 * ... not reused: can steal
6242 						 * without relocating contents.
6243 						 */
6244 						reusable = TRUE;
6245 					}
6246 				}
6247 
6248 				if ((m1->vmp_pmapped &&
6249 				    !reusable) ||
6250 				    m1->vmp_dirty ||
6251 				    m1->vmp_precious) {
6252 					vm_object_offset_t offset;
6253 
6254 					m2 = vm_page_grab_options(VM_PAGE_GRAB_Q_LOCK_HELD);
6255 
6256 					if (m2 == VM_PAGE_NULL) {
6257 						if (locked_object) {
6258 							vm_object_unlock(locked_object);
6259 							locked_object = VM_OBJECT_NULL;
6260 						}
6261 						tmp_start_idx = cur_idx;
6262 						abort_run = TRUE;
6263 						continue;
6264 					}
6265 					if (!disconnected) {
6266 						if (m1->vmp_pmapped) {
6267 							refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1));
6268 						} else {
6269 							refmod = 0;
6270 						}
6271 					}
6272 
6273 					/* copy the page's contents */
6274 					pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(m1), VM_PAGE_GET_PHYS_PAGE(m2));
6275 					/* copy the page's state */
6276 					assert(!VM_PAGE_WIRED(m1));
6277 					assert(m1->vmp_q_state != VM_PAGE_ON_FREE_Q);
6278 					assert(m1->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q);
6279 					assert(!m1->vmp_laundry);
6280 					m2->vmp_reference       = m1->vmp_reference;
6281 					assert(!m1->vmp_gobbled);
6282 					assert(!m1->vmp_private);
6283 					m2->vmp_no_cache        = m1->vmp_no_cache;
6284 					m2->vmp_xpmapped        = 0;
6285 					assert(!m1->vmp_busy);
6286 					assert(!m1->vmp_wanted);
6287 					assert(!m1->vmp_fictitious);
6288 					m2->vmp_pmapped = m1->vmp_pmapped; /* should flush cache ? */
6289 					m2->vmp_wpmapped        = m1->vmp_wpmapped;
6290 					assert(!m1->vmp_free_when_done);
6291 					m2->vmp_absent  = m1->vmp_absent;
6292 					m2->vmp_error   = VMP_ERROR_GET(m1);
6293 					m2->vmp_dirty   = m1->vmp_dirty;
6294 					assert(!m1->vmp_cleaning);
6295 					m2->vmp_precious        = m1->vmp_precious;
6296 					m2->vmp_clustered       = m1->vmp_clustered;
6297 					assert(!m1->vmp_overwriting);
6298 					m2->vmp_restart = m1->vmp_restart;
6299 					m2->vmp_unusual = m1->vmp_unusual;
6300 					m2->vmp_cs_validated = m1->vmp_cs_validated;
6301 					m2->vmp_cs_tainted      = m1->vmp_cs_tainted;
6302 					m2->vmp_cs_nx   = m1->vmp_cs_nx;
6303 
6304 					m2->vmp_realtime = m1->vmp_realtime;
6305 					m1->vmp_realtime = false;
6306 
6307 					/*
6308 					 * If m1 had really been reusable,
6309 					 * we would have just stolen it, so
6310 					 * let's not propagate it's "reusable"
6311 					 * bit and assert that m2 is not
6312 					 * marked as "reusable".
6313 					 */
6314 					// m2->vmp_reusable	= m1->vmp_reusable;
6315 					assert(!m2->vmp_reusable);
6316 
6317 					// assert(!m1->vmp_lopage);
6318 
6319 					if (m1->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6320 						m2->vmp_q_state = VM_PAGE_USED_BY_COMPRESSOR;
6321 						/*
6322 						 * We just grabbed m2 up above and so it isn't
6323 						 * going to be on any special Q as yet and so
6324 						 * we don't need to 'remove' it from the special
6325 						 * queues. Just resetting the state should be enough.
6326 						 */
6327 						m2->vmp_on_specialq = VM_PAGE_SPECIAL_Q_EMPTY;
6328 					}
6329 
6330 					/*
6331 					 * page may need to be flushed if
6332 					 * it is marshalled into a UPL
6333 					 * that is going to be used by a device
6334 					 * that doesn't support coherency
6335 					 */
6336 					m2->vmp_written_by_kernel = TRUE;
6337 
6338 					/*
6339 					 * make sure we clear the ref/mod state
6340 					 * from the pmap layer... else we risk
6341 					 * inheriting state from the last time
6342 					 * this page was used...
6343 					 */
6344 					pmap_clear_refmod(VM_PAGE_GET_PHYS_PAGE(m2), VM_MEM_MODIFIED | VM_MEM_REFERENCED);
6345 
6346 					if (refmod & VM_MEM_REFERENCED) {
6347 						m2->vmp_reference = TRUE;
6348 					}
6349 					if (refmod & VM_MEM_MODIFIED) {
6350 						SET_PAGE_DIRTY(m2, TRUE);
6351 					}
6352 					offset = m1->vmp_offset;
6353 
6354 					/*
6355 					 * completely cleans up the state
6356 					 * of the page so that it is ready
6357 					 * to be put onto the free list, or
6358 					 * for this purpose it looks like it
6359 					 * just came off of the free list
6360 					 */
6361 					vm_page_free_prepare(m1);
6362 
6363 					/*
6364 					 * now put the substitute page
6365 					 * on the object
6366 					 */
6367 					vm_page_insert_internal(m2, locked_object, offset, VM_KERN_MEMORY_NONE, TRUE, TRUE, FALSE, FALSE, NULL);
6368 
6369 					if (m2->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
6370 						m2->vmp_pmapped = TRUE;
6371 						m2->vmp_wpmapped = TRUE;
6372 
6373 						PMAP_ENTER(kernel_pmap, (vm_map_offset_t)m2->vmp_offset, m2,
6374 						    VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE, kr);
6375 
6376 						assert(kr == KERN_SUCCESS);
6377 
6378 						compressed_pages++;
6379 					} else {
6380 						if (m2->vmp_reference) {
6381 							vm_page_activate(m2);
6382 						} else {
6383 							vm_page_deactivate(m2);
6384 						}
6385 					}
6386 					PAGE_WAKEUP_DONE(m2);
6387 				} else {
6388 					assert(m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR);
6389 
6390 					/*
6391 					 * completely cleans up the state
6392 					 * of the page so that it is ready
6393 					 * to be put onto the free list, or
6394 					 * for this purpose it looks like it
6395 					 * just came off of the free list
6396 					 */
6397 					vm_page_free_prepare(m1);
6398 				}
6399 
6400 				stolen_pages++;
6401 			}
6402 			if (m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR) {
6403 				/*
6404 				 * The Q state is preserved on m1 because vm_page_queues_remove doesn't
6405 				 * change it for pages marked as used-by-compressor.
6406 				 */
6407 				vm_page_assign_special_state(m1, VM_PAGE_SPECIAL_Q_BG);
6408 			}
6409 			VM_PAGE_ZERO_PAGEQ_ENTRY(m1);
6410 			m1->vmp_snext = m;
6411 			m = m1;
6412 		}
6413 		if (locked_object) {
6414 			vm_object_unlock(locked_object);
6415 			locked_object = VM_OBJECT_NULL;
6416 		}
6417 
6418 		if (abort_run == TRUE) {
6419 			/*
6420 			 * want the index of the last
6421 			 * page in this run that was
6422 			 * successfully 'stolen', so back
6423 			 * it up 1 for the auto-decrement on use
6424 			 * and 1 more to bump back over this page
6425 			 */
6426 			page_idx = tmp_start_idx + 2;
6427 			if (page_idx >= vm_pages_count) {
6428 				if (wrapped) {
6429 					if (m != VM_PAGE_NULL) {
6430 						vm_page_unlock_queues();
6431 						vm_page_free_list(m, FALSE);
6432 						vm_page_lock_queues();
6433 						m = VM_PAGE_NULL;
6434 					}
6435 					dumped_run++;
6436 					goto done_scanning;
6437 				}
6438 				page_idx = last_idx = 0;
6439 				wrapped = TRUE;
6440 			}
6441 			abort_run = FALSE;
6442 
6443 			/*
6444 			 * We didn't find a contiguous range but we didn't
6445 			 * start from the very first page.
6446 			 * Start again from the very first page.
6447 			 */
6448 			RESET_STATE_OF_RUN();
6449 
6450 			if (flags & KMA_LOMEM) {
6451 				idx_last_contig_page_found  = vm_page_lomem_find_contiguous_last_idx = page_idx;
6452 			} else {
6453 				idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx;
6454 			}
6455 
6456 			last_idx = page_idx;
6457 
6458 			if (m != VM_PAGE_NULL) {
6459 				vm_page_unlock_queues();
6460 				vm_page_free_list(m, FALSE);
6461 				vm_page_lock_queues();
6462 				m = VM_PAGE_NULL;
6463 			}
6464 			dumped_run++;
6465 
6466 			vm_free_page_lock();
6467 			/*
6468 			 * reset our free page limit since we
6469 			 * dropped the lock protecting the vm_page_free_queue
6470 			 */
6471 			free_available = vm_page_free_count - vm_page_free_reserved;
6472 			goto retry;
6473 		}
6474 
6475 		for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
6476 			assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q);
6477 			assert(m1->vmp_wire_count == 0);
6478 
6479 			if (wire == TRUE) {
6480 				m1->vmp_wire_count++;
6481 				m1->vmp_q_state = VM_PAGE_IS_WIRED;
6482 			} else {
6483 				m1->vmp_gobbled = TRUE;
6484 			}
6485 		}
6486 		if (wire == FALSE) {
6487 			vm_page_gobble_count += npages;
6488 		}
6489 
6490 		/*
6491 		 * gobbled pages are also counted as wired pages
6492 		 */
6493 		vm_page_wire_count += npages;
6494 
6495 		assert(vm_page_verify_contiguous(m, npages));
6496 	}
6497 done_scanning:
6498 	PAGE_REPLACEMENT_ALLOWED(FALSE);
6499 
6500 	vm_page_unlock_queues();
6501 
6502 #if DEBUG
6503 	clock_get_system_microtime(&tv_end_sec, &tv_end_usec);
6504 
6505 	tv_end_sec -= tv_start_sec;
6506 	if (tv_end_usec < tv_start_usec) {
6507 		tv_end_sec--;
6508 		tv_end_usec += 1000000;
6509 	}
6510 	tv_end_usec -= tv_start_usec;
6511 	if (tv_end_usec >= 1000000) {
6512 		tv_end_sec++;
6513 		tv_end_sec -= 1000000;
6514 	}
6515 	if (vm_page_find_contig_debug) {
6516 		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",
6517 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6518 		    (long)tv_end_sec, tv_end_usec, orig_last_idx,
6519 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages);
6520 	}
6521 
6522 #endif
6523 #if MACH_ASSERT
6524 	vm_page_verify_free_lists();
6525 #endif
6526 	if (m == NULL && zone_gc_called < 2) {
6527 		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",
6528 		    __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT,
6529 		        scanned, yielded, dumped_run, stolen_pages, compressed_pages, vm_page_wire_count);
6530 
6531 		if (consider_buffer_cache_collect != NULL) {
6532 			(void)(*consider_buffer_cache_collect)(1);
6533 		}
6534 
6535 		zone_gc(zone_gc_called ? ZONE_GC_DRAIN : ZONE_GC_TRIM);
6536 
6537 		zone_gc_called++;
6538 
6539 		printf("vm_page_find_contiguous: zone_gc called... wired count is %d\n", vm_page_wire_count);
6540 		goto full_scan_again;
6541 	}
6542 
6543 	return m;
6544 }
6545 
6546 /*
6547  *	Allocate a list of contiguous, wired pages.
6548  */
6549 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)6550 cpm_allocate(
6551 	vm_size_t       size,
6552 	vm_page_t       *list,
6553 	ppnum_t         max_pnum,
6554 	ppnum_t         pnum_mask,
6555 	boolean_t       wire,
6556 	int             flags)
6557 {
6558 	vm_page_t               pages;
6559 	unsigned int            npages;
6560 
6561 	if (size % PAGE_SIZE != 0) {
6562 		return KERN_INVALID_ARGUMENT;
6563 	}
6564 
6565 	npages = (unsigned int) (size / PAGE_SIZE);
6566 	if (npages != size / PAGE_SIZE) {
6567 		/* 32-bit overflow */
6568 		return KERN_INVALID_ARGUMENT;
6569 	}
6570 
6571 	/*
6572 	 *	Obtain a pointer to a subset of the free
6573 	 *	list large enough to satisfy the request;
6574 	 *	the region will be physically contiguous.
6575 	 */
6576 	pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags);
6577 
6578 	if (pages == VM_PAGE_NULL) {
6579 		return KERN_NO_SPACE;
6580 	}
6581 	/*
6582 	 * determine need for wakeups
6583 	 */
6584 	if (vm_page_free_count < vm_page_free_min) {
6585 		vm_free_page_lock();
6586 		if (vm_pageout_running == FALSE) {
6587 			vm_free_page_unlock();
6588 			thread_wakeup((event_t) &vm_page_free_wanted);
6589 		} else {
6590 			vm_free_page_unlock();
6591 		}
6592 	}
6593 
6594 	VM_CHECK_MEMORYSTATUS;
6595 
6596 	/*
6597 	 *	The CPM pages should now be available and
6598 	 *	ordered by ascending physical address.
6599 	 */
6600 	assert(vm_page_verify_contiguous(pages, npages));
6601 
6602 	*list = pages;
6603 	return KERN_SUCCESS;
6604 }
6605 
6606 
6607 unsigned int vm_max_delayed_work_limit = DEFAULT_DELAYED_WORK_LIMIT;
6608 
6609 /*
6610  * when working on a 'run' of pages, it is necessary to hold
6611  * the vm_page_queue_lock (a hot global lock) for certain operations
6612  * on the page... however, the majority of the work can be done
6613  * while merely holding the object lock... in fact there are certain
6614  * collections of pages that don't require any work brokered by the
6615  * vm_page_queue_lock... to mitigate the time spent behind the global
6616  * lock, go to a 2 pass algorithm... collect pages up to DELAYED_WORK_LIMIT
6617  * while doing all of the work that doesn't require the vm_page_queue_lock...
6618  * then call vm_page_do_delayed_work to acquire the vm_page_queue_lock and do the
6619  * necessary work for each page... we will grab the busy bit on the page
6620  * if it's not already held so that vm_page_do_delayed_work can drop the object lock
6621  * if it can't immediately take the vm_page_queue_lock in order to compete
6622  * for the locks in the same order that vm_pageout_scan takes them.
6623  * the operation names are modeled after the names of the routines that
6624  * need to be called in order to make the changes very obvious in the
6625  * original loop
6626  */
6627 
6628 void
vm_page_do_delayed_work(vm_object_t object,vm_tag_t tag,struct vm_page_delayed_work * dwp,int dw_count)6629 vm_page_do_delayed_work(
6630 	vm_object_t     object,
6631 	vm_tag_t        tag,
6632 	struct vm_page_delayed_work *dwp,
6633 	int             dw_count)
6634 {
6635 	int             j;
6636 	vm_page_t       m;
6637 	vm_page_t       local_free_q = VM_PAGE_NULL;
6638 
6639 	/*
6640 	 * pageout_scan takes the vm_page_lock_queues first
6641 	 * then tries for the object lock... to avoid what
6642 	 * is effectively a lock inversion, we'll go to the
6643 	 * trouble of taking them in that same order... otherwise
6644 	 * if this object contains the majority of the pages resident
6645 	 * in the UBC (or a small set of large objects actively being
6646 	 * worked on contain the majority of the pages), we could
6647 	 * cause the pageout_scan thread to 'starve' in its attempt
6648 	 * to find pages to move to the free queue, since it has to
6649 	 * successfully acquire the object lock of any candidate page
6650 	 * before it can steal/clean it.
6651 	 */
6652 	if (!vm_page_trylockspin_queues()) {
6653 		vm_object_unlock(object);
6654 
6655 		/*
6656 		 * "Turnstile enabled vm_pageout_scan" can be runnable
6657 		 * for a very long time without getting on a core.
6658 		 * If this is a higher priority thread it could be
6659 		 * waiting here for a very long time respecting the fact
6660 		 * that pageout_scan would like its object after VPS does
6661 		 * a mutex_pause(0).
6662 		 * So we cap the number of yields in the vm_object_lock_avoid()
6663 		 * case to a single mutex_pause(0) which will give vm_pageout_scan
6664 		 * 10us to run and grab the object if needed.
6665 		 */
6666 		vm_page_lockspin_queues();
6667 
6668 		for (j = 0;; j++) {
6669 			if ((!vm_object_lock_avoid(object) ||
6670 			    (vps_dynamic_priority_enabled && (j > 0))) &&
6671 			    _vm_object_lock_try(object)) {
6672 				break;
6673 			}
6674 			vm_page_unlock_queues();
6675 			mutex_pause(j);
6676 			vm_page_lockspin_queues();
6677 		}
6678 	}
6679 	for (j = 0; j < dw_count; j++, dwp++) {
6680 		m = dwp->dw_m;
6681 
6682 		if (dwp->dw_mask & DW_vm_pageout_throttle_up) {
6683 			vm_pageout_throttle_up(m);
6684 		}
6685 #if CONFIG_PHANTOM_CACHE
6686 		if (dwp->dw_mask & DW_vm_phantom_cache_update) {
6687 			vm_phantom_cache_update(m);
6688 		}
6689 #endif
6690 		if (dwp->dw_mask & DW_vm_page_wire) {
6691 			vm_page_wire(m, tag, FALSE);
6692 		} else if (dwp->dw_mask & DW_vm_page_unwire) {
6693 			boolean_t       queueit;
6694 
6695 			queueit = (dwp->dw_mask & (DW_vm_page_free | DW_vm_page_deactivate_internal)) ? FALSE : TRUE;
6696 
6697 			vm_page_unwire(m, queueit);
6698 		}
6699 		if (dwp->dw_mask & DW_vm_page_free) {
6700 			vm_page_free_prepare_queues(m);
6701 
6702 			assert(m->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0);
6703 			/*
6704 			 * Add this page to our list of reclaimed pages,
6705 			 * to be freed later.
6706 			 */
6707 			m->vmp_snext = local_free_q;
6708 			local_free_q = m;
6709 		} else {
6710 			if (dwp->dw_mask & DW_vm_page_deactivate_internal) {
6711 				vm_page_deactivate_internal(m, FALSE);
6712 			} else if (dwp->dw_mask & DW_vm_page_activate) {
6713 				if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6714 					vm_page_activate(m);
6715 				}
6716 			} else if (dwp->dw_mask & DW_vm_page_speculate) {
6717 				vm_page_speculate(m, TRUE);
6718 			} else if (dwp->dw_mask & DW_enqueue_cleaned) {
6719 				/*
6720 				 * if we didn't hold the object lock and did this,
6721 				 * we might disconnect the page, then someone might
6722 				 * soft fault it back in, then we would put it on the
6723 				 * cleaned queue, and so we would have a referenced (maybe even dirty)
6724 				 * page on that queue, which we don't want
6725 				 */
6726 				int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
6727 
6728 				if ((refmod_state & VM_MEM_REFERENCED)) {
6729 					/*
6730 					 * this page has been touched since it got cleaned; let's activate it
6731 					 * if it hasn't already been
6732 					 */
6733 					VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1);
6734 					VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1);
6735 
6736 					if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) {
6737 						vm_page_activate(m);
6738 					}
6739 				} else {
6740 					m->vmp_reference = FALSE;
6741 					vm_page_enqueue_cleaned(m);
6742 				}
6743 			} else if (dwp->dw_mask & DW_vm_page_lru) {
6744 				vm_page_lru(m);
6745 			} else if (dwp->dw_mask & DW_VM_PAGE_QUEUES_REMOVE) {
6746 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
6747 					vm_page_queues_remove(m, TRUE);
6748 				}
6749 			}
6750 			if (dwp->dw_mask & DW_set_reference) {
6751 				m->vmp_reference = TRUE;
6752 			} else if (dwp->dw_mask & DW_clear_reference) {
6753 				m->vmp_reference = FALSE;
6754 			}
6755 
6756 			if (dwp->dw_mask & DW_move_page) {
6757 				if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) {
6758 					vm_page_queues_remove(m, FALSE);
6759 
6760 					assert(VM_PAGE_OBJECT(m) != kernel_object);
6761 
6762 					vm_page_enqueue_inactive(m, FALSE);
6763 				}
6764 			}
6765 			if (dwp->dw_mask & DW_clear_busy) {
6766 				m->vmp_busy = FALSE;
6767 			}
6768 
6769 			if (dwp->dw_mask & DW_PAGE_WAKEUP) {
6770 				PAGE_WAKEUP(m);
6771 			}
6772 		}
6773 	}
6774 	vm_page_unlock_queues();
6775 
6776 	if (local_free_q) {
6777 		vm_page_free_list(local_free_q, TRUE);
6778 	}
6779 
6780 	VM_CHECK_MEMORYSTATUS;
6781 }
6782 
6783 __abortlike
6784 static void
__vm_page_alloc_list_failed_panic(vm_size_t page_count,kma_flags_t flags,kern_return_t kr)6785 __vm_page_alloc_list_failed_panic(
6786 	vm_size_t       page_count,
6787 	kma_flags_t     flags,
6788 	kern_return_t   kr)
6789 {
6790 	panic("vm_page_alloc_list(%zd, 0x%x) failed unexpectedly with %d",
6791 	    (size_t)page_count, flags, kr);
6792 }
6793 
6794 kern_return_t
vm_page_alloc_list(vm_size_t page_count,kma_flags_t flags,vm_page_t * list)6795 vm_page_alloc_list(
6796 	vm_size_t   page_count,
6797 	kma_flags_t flags,
6798 	vm_page_t  *list)
6799 {
6800 	vm_page_t       page_list = VM_PAGE_NULL;
6801 	vm_page_t       mem;
6802 	kern_return_t   kr = KERN_SUCCESS;
6803 	int             page_grab_count = 0;
6804 #if DEVELOPMENT || DEBUG
6805 	task_t          task;
6806 #endif /* DEVELOPMENT || DEBUG */
6807 
6808 	for (vm_size_t i = 0; i < page_count; i++) {
6809 		for (;;) {
6810 			if (flags & KMA_LOMEM) {
6811 				mem = vm_page_grablo();
6812 			} else {
6813 				mem = vm_page_grab();
6814 			}
6815 
6816 			if (mem != VM_PAGE_NULL) {
6817 				break;
6818 			}
6819 
6820 			if (flags & KMA_NOPAGEWAIT) {
6821 				kr = KERN_RESOURCE_SHORTAGE;
6822 				goto out;
6823 			}
6824 			if ((flags & KMA_LOMEM) && (vm_lopage_needed == TRUE)) {
6825 				kr = KERN_RESOURCE_SHORTAGE;
6826 				goto out;
6827 			}
6828 
6829 			/* VM privileged threads should have waited in vm_page_grab() and not get here. */
6830 			assert(!(current_thread()->options & TH_OPT_VMPRIV));
6831 
6832 			if ((flags & KMA_NOFAIL) == 0) {
6833 				uint64_t unavailable = ptoa_64(vm_page_wire_count + vm_page_free_target);
6834 				if (unavailable > max_mem || ptoa_64(page_count) > (max_mem - unavailable)) {
6835 					kr = KERN_RESOURCE_SHORTAGE;
6836 					goto out;
6837 				}
6838 			}
6839 			VM_PAGE_WAIT();
6840 		}
6841 
6842 		page_grab_count++;
6843 		mem->vmp_snext = page_list;
6844 		page_list = mem;
6845 	}
6846 
6847 	if ((KMA_ZERO | KMA_NOENCRYPT) & flags) {
6848 		for (mem = page_list; mem; mem = mem->vmp_snext) {
6849 			vm_page_zero_fill(mem);
6850 		}
6851 	}
6852 
6853 out:
6854 #if DEBUG || DEVELOPMENT
6855 	task = current_task_early();
6856 	if (task != NULL) {
6857 		ledger_credit(task->ledger, task_ledgers.pages_grabbed_kern, page_grab_count);
6858 	}
6859 #endif
6860 
6861 	if (kr == KERN_SUCCESS) {
6862 		*list = page_list;
6863 	} else if (flags & KMA_NOFAIL) {
6864 		__vm_page_alloc_list_failed_panic(page_count, flags, kr);
6865 	} else {
6866 		vm_page_free_list(page_list, FALSE);
6867 	}
6868 
6869 	return kr;
6870 }
6871 
6872 void
vm_page_set_offset(vm_page_t page,vm_object_offset_t offset)6873 vm_page_set_offset(vm_page_t page, vm_object_offset_t offset)
6874 {
6875 	page->vmp_offset = offset;
6876 }
6877 
6878 vm_page_t
vm_page_get_next(vm_page_t page)6879 vm_page_get_next(vm_page_t page)
6880 {
6881 	return page->vmp_snext;
6882 }
6883 
6884 vm_object_offset_t
vm_page_get_offset(vm_page_t page)6885 vm_page_get_offset(vm_page_t page)
6886 {
6887 	return page->vmp_offset;
6888 }
6889 
6890 ppnum_t
vm_page_get_phys_page(vm_page_t page)6891 vm_page_get_phys_page(vm_page_t page)
6892 {
6893 	return VM_PAGE_GET_PHYS_PAGE(page);
6894 }
6895 
6896 
6897 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
6898 
6899 #if HIBERNATION
6900 
6901 static vm_page_t hibernate_gobble_queue;
6902 
6903 static int  hibernate_drain_pageout_queue(struct vm_pageout_queue *);
6904 static int  hibernate_flush_dirty_pages(int);
6905 static int  hibernate_flush_queue(vm_page_queue_head_t *, int);
6906 
6907 void hibernate_flush_wait(void);
6908 void hibernate_mark_in_progress(void);
6909 void hibernate_clear_in_progress(void);
6910 
6911 void            hibernate_free_range(int, int);
6912 void            hibernate_hash_insert_page(vm_page_t);
6913 uint32_t        hibernate_mark_as_unneeded(addr64_t, addr64_t, hibernate_page_list_t *, hibernate_page_list_t *);
6914 uint32_t        hibernate_teardown_vm_structs(hibernate_page_list_t *, hibernate_page_list_t *);
6915 ppnum_t         hibernate_lookup_paddr(unsigned int);
6916 
6917 struct hibernate_statistics {
6918 	int hibernate_considered;
6919 	int hibernate_reentered_on_q;
6920 	int hibernate_found_dirty;
6921 	int hibernate_skipped_cleaning;
6922 	int hibernate_skipped_transient;
6923 	int hibernate_skipped_precious;
6924 	int hibernate_skipped_external;
6925 	int hibernate_queue_nolock;
6926 	int hibernate_queue_paused;
6927 	int hibernate_throttled;
6928 	int hibernate_throttle_timeout;
6929 	int hibernate_drained;
6930 	int hibernate_drain_timeout;
6931 	int cd_lock_failed;
6932 	int cd_found_precious;
6933 	int cd_found_wired;
6934 	int cd_found_busy;
6935 	int cd_found_unusual;
6936 	int cd_found_cleaning;
6937 	int cd_found_laundry;
6938 	int cd_found_dirty;
6939 	int cd_found_xpmapped;
6940 	int cd_skipped_xpmapped;
6941 	int cd_local_free;
6942 	int cd_total_free;
6943 	int cd_vm_page_wire_count;
6944 	int cd_vm_struct_pages_unneeded;
6945 	int cd_pages;
6946 	int cd_discarded;
6947 	int cd_count_wire;
6948 } hibernate_stats;
6949 
6950 
6951 /*
6952  * clamp the number of 'xpmapped' pages we'll sweep into the hibernation image
6953  * so that we don't overrun the estimated image size, which would
6954  * result in a hibernation failure.
6955  *
6956  * We use a size value instead of pages because we don't want to take up more space
6957  * on disk if the system has a 16K page size vs 4K. Also, we are not guaranteed
6958  * to have that additional space available.
6959  *
6960  * Since this was set at 40000 pages on X86 we are going to use 160MB as our
6961  * xpmapped size.
6962  */
6963 #define HIBERNATE_XPMAPPED_LIMIT        ((160 * 1024 * 1024ULL) / PAGE_SIZE)
6964 
6965 
6966 static int
hibernate_drain_pageout_queue(struct vm_pageout_queue * q)6967 hibernate_drain_pageout_queue(struct vm_pageout_queue *q)
6968 {
6969 	wait_result_t   wait_result;
6970 
6971 	vm_page_lock_queues();
6972 
6973 	while (!vm_page_queue_empty(&q->pgo_pending)) {
6974 		q->pgo_draining = TRUE;
6975 
6976 		assert_wait_timeout((event_t) (&q->pgo_laundry + 1), THREAD_INTERRUPTIBLE, 5000, 1000 * NSEC_PER_USEC);
6977 
6978 		vm_page_unlock_queues();
6979 
6980 		wait_result = thread_block(THREAD_CONTINUE_NULL);
6981 
6982 		if (wait_result == THREAD_TIMED_OUT && !vm_page_queue_empty(&q->pgo_pending)) {
6983 			hibernate_stats.hibernate_drain_timeout++;
6984 
6985 			if (q == &vm_pageout_queue_external) {
6986 				return 0;
6987 			}
6988 
6989 			return 1;
6990 		}
6991 		vm_page_lock_queues();
6992 
6993 		hibernate_stats.hibernate_drained++;
6994 	}
6995 	vm_page_unlock_queues();
6996 
6997 	return 0;
6998 }
6999 
7000 
7001 boolean_t hibernate_skip_external = FALSE;
7002 
7003 static int
hibernate_flush_queue(vm_page_queue_head_t * q,int qcount)7004 hibernate_flush_queue(vm_page_queue_head_t *q, int qcount)
7005 {
7006 	vm_page_t       m;
7007 	vm_object_t     l_object = NULL;
7008 	vm_object_t     m_object = NULL;
7009 	int             refmod_state = 0;
7010 	int             try_failed_count = 0;
7011 	int             retval = 0;
7012 	int             current_run = 0;
7013 	struct  vm_pageout_queue *iq;
7014 	struct  vm_pageout_queue *eq;
7015 	struct  vm_pageout_queue *tq;
7016 
7017 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START,
7018 	    VM_KERNEL_UNSLIDE_OR_PERM(q), qcount);
7019 
7020 	iq = &vm_pageout_queue_internal;
7021 	eq = &vm_pageout_queue_external;
7022 
7023 	vm_page_lock_queues();
7024 
7025 	while (qcount && !vm_page_queue_empty(q)) {
7026 		if (current_run++ == 1000) {
7027 			if (hibernate_should_abort()) {
7028 				retval = 1;
7029 				break;
7030 			}
7031 			current_run = 0;
7032 		}
7033 
7034 		m = (vm_page_t) vm_page_queue_first(q);
7035 		m_object = VM_PAGE_OBJECT(m);
7036 
7037 		/*
7038 		 * check to see if we currently are working
7039 		 * with the same object... if so, we've
7040 		 * already got the lock
7041 		 */
7042 		if (m_object != l_object) {
7043 			/*
7044 			 * the object associated with candidate page is
7045 			 * different from the one we were just working
7046 			 * with... dump the lock if we still own it
7047 			 */
7048 			if (l_object != NULL) {
7049 				vm_object_unlock(l_object);
7050 				l_object = NULL;
7051 			}
7052 			/*
7053 			 * Try to lock object; since we've alread got the
7054 			 * page queues lock, we can only 'try' for this one.
7055 			 * if the 'try' fails, we need to do a mutex_pause
7056 			 * to allow the owner of the object lock a chance to
7057 			 * run...
7058 			 */
7059 			if (!vm_object_lock_try_scan(m_object)) {
7060 				if (try_failed_count > 20) {
7061 					hibernate_stats.hibernate_queue_nolock++;
7062 
7063 					goto reenter_pg_on_q;
7064 				}
7065 
7066 				vm_page_unlock_queues();
7067 				mutex_pause(try_failed_count++);
7068 				vm_page_lock_queues();
7069 
7070 				hibernate_stats.hibernate_queue_paused++;
7071 				continue;
7072 			} else {
7073 				l_object = m_object;
7074 			}
7075 		}
7076 		if (!m_object->alive || m->vmp_cleaning || m->vmp_laundry || m->vmp_busy || m->vmp_absent || VMP_ERROR_GET(m)) {
7077 			/*
7078 			 * page is not to be cleaned
7079 			 * put it back on the head of its queue
7080 			 */
7081 			if (m->vmp_cleaning) {
7082 				hibernate_stats.hibernate_skipped_cleaning++;
7083 			} else {
7084 				hibernate_stats.hibernate_skipped_transient++;
7085 			}
7086 
7087 			goto reenter_pg_on_q;
7088 		}
7089 		if (m_object->copy == VM_OBJECT_NULL) {
7090 			if (m_object->purgable == VM_PURGABLE_VOLATILE || m_object->purgable == VM_PURGABLE_EMPTY) {
7091 				/*
7092 				 * let the normal hibernate image path
7093 				 * deal with these
7094 				 */
7095 				goto reenter_pg_on_q;
7096 			}
7097 		}
7098 		if (!m->vmp_dirty && m->vmp_pmapped) {
7099 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7100 
7101 			if ((refmod_state & VM_MEM_MODIFIED)) {
7102 				SET_PAGE_DIRTY(m, FALSE);
7103 			}
7104 		} else {
7105 			refmod_state = 0;
7106 		}
7107 
7108 		if (!m->vmp_dirty) {
7109 			/*
7110 			 * page is not to be cleaned
7111 			 * put it back on the head of its queue
7112 			 */
7113 			if (m->vmp_precious) {
7114 				hibernate_stats.hibernate_skipped_precious++;
7115 			}
7116 
7117 			goto reenter_pg_on_q;
7118 		}
7119 
7120 		if (hibernate_skip_external == TRUE && !m_object->internal) {
7121 			hibernate_stats.hibernate_skipped_external++;
7122 
7123 			goto reenter_pg_on_q;
7124 		}
7125 		tq = NULL;
7126 
7127 		if (m_object->internal) {
7128 			if (VM_PAGE_Q_THROTTLED(iq)) {
7129 				tq = iq;
7130 			}
7131 		} else if (VM_PAGE_Q_THROTTLED(eq)) {
7132 			tq = eq;
7133 		}
7134 
7135 		if (tq != NULL) {
7136 			wait_result_t   wait_result;
7137 			int             wait_count = 5;
7138 
7139 			if (l_object != NULL) {
7140 				vm_object_unlock(l_object);
7141 				l_object = NULL;
7142 			}
7143 
7144 			while (retval == 0) {
7145 				tq->pgo_throttled = TRUE;
7146 
7147 				assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000 * NSEC_PER_USEC);
7148 
7149 				vm_page_unlock_queues();
7150 
7151 				wait_result = thread_block(THREAD_CONTINUE_NULL);
7152 
7153 				vm_page_lock_queues();
7154 
7155 				if (wait_result != THREAD_TIMED_OUT) {
7156 					break;
7157 				}
7158 				if (!VM_PAGE_Q_THROTTLED(tq)) {
7159 					break;
7160 				}
7161 
7162 				if (hibernate_should_abort()) {
7163 					retval = 1;
7164 				}
7165 
7166 				if (--wait_count == 0) {
7167 					hibernate_stats.hibernate_throttle_timeout++;
7168 
7169 					if (tq == eq) {
7170 						hibernate_skip_external = TRUE;
7171 						break;
7172 					}
7173 					retval = 1;
7174 				}
7175 			}
7176 			if (retval) {
7177 				break;
7178 			}
7179 
7180 			hibernate_stats.hibernate_throttled++;
7181 
7182 			continue;
7183 		}
7184 		/*
7185 		 * we've already factored out pages in the laundry which
7186 		 * means this page can't be on the pageout queue so it's
7187 		 * safe to do the vm_page_queues_remove
7188 		 */
7189 		vm_page_queues_remove(m, TRUE);
7190 
7191 		if (m_object->internal == TRUE) {
7192 			pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(m), PMAP_OPTIONS_COMPRESSOR, NULL);
7193 		}
7194 
7195 		vm_pageout_cluster(m);
7196 
7197 		hibernate_stats.hibernate_found_dirty++;
7198 
7199 		goto next_pg;
7200 
7201 reenter_pg_on_q:
7202 		vm_page_queue_remove(q, m, vmp_pageq);
7203 		vm_page_queue_enter(q, m, vmp_pageq);
7204 
7205 		hibernate_stats.hibernate_reentered_on_q++;
7206 next_pg:
7207 		hibernate_stats.hibernate_considered++;
7208 
7209 		qcount--;
7210 		try_failed_count = 0;
7211 	}
7212 	if (l_object != NULL) {
7213 		vm_object_unlock(l_object);
7214 		l_object = NULL;
7215 	}
7216 
7217 	vm_page_unlock_queues();
7218 
7219 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_END, hibernate_stats.hibernate_found_dirty, retval, 0, 0, 0);
7220 
7221 	return retval;
7222 }
7223 
7224 
7225 static int
hibernate_flush_dirty_pages(int pass)7226 hibernate_flush_dirty_pages(int pass)
7227 {
7228 	struct vm_speculative_age_q     *aq;
7229 	uint32_t        i;
7230 
7231 	if (vm_page_local_q) {
7232 		zpercpu_foreach_cpu(lid) {
7233 			vm_page_reactivate_local(lid, TRUE, FALSE);
7234 		}
7235 	}
7236 
7237 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
7238 		int             qcount;
7239 		vm_page_t       m;
7240 
7241 		aq = &vm_page_queue_speculative[i];
7242 
7243 		if (vm_page_queue_empty(&aq->age_q)) {
7244 			continue;
7245 		}
7246 		qcount = 0;
7247 
7248 		vm_page_lockspin_queues();
7249 
7250 		vm_page_queue_iterate(&aq->age_q, m, vmp_pageq) {
7251 			qcount++;
7252 		}
7253 		vm_page_unlock_queues();
7254 
7255 		if (qcount) {
7256 			if (hibernate_flush_queue(&aq->age_q, qcount)) {
7257 				return 1;
7258 			}
7259 		}
7260 	}
7261 	if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_page_anonymous_count - vm_page_cleaned_count)) {
7262 		return 1;
7263 	}
7264 	/* XXX FBDP TODO: flush secluded queue */
7265 	if (hibernate_flush_queue(&vm_page_queue_anonymous, vm_page_anonymous_count)) {
7266 		return 1;
7267 	}
7268 	if (hibernate_flush_queue(&vm_page_queue_cleaned, vm_page_cleaned_count)) {
7269 		return 1;
7270 	}
7271 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7272 		return 1;
7273 	}
7274 
7275 	if (pass == 1) {
7276 		vm_compressor_record_warmup_start();
7277 	}
7278 
7279 	if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) {
7280 		if (pass == 1) {
7281 			vm_compressor_record_warmup_end();
7282 		}
7283 		return 1;
7284 	}
7285 	if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) {
7286 		if (pass == 1) {
7287 			vm_compressor_record_warmup_end();
7288 		}
7289 		return 1;
7290 	}
7291 	if (pass == 1) {
7292 		vm_compressor_record_warmup_end();
7293 	}
7294 
7295 	if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external)) {
7296 		return 1;
7297 	}
7298 
7299 	return 0;
7300 }
7301 
7302 
7303 void
hibernate_reset_stats()7304 hibernate_reset_stats()
7305 {
7306 	bzero(&hibernate_stats, sizeof(struct hibernate_statistics));
7307 }
7308 
7309 
7310 int
hibernate_flush_memory()7311 hibernate_flush_memory()
7312 {
7313 	int     retval;
7314 
7315 	assert(VM_CONFIG_COMPRESSOR_IS_PRESENT);
7316 
7317 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0);
7318 
7319 	hibernate_cleaning_in_progress = TRUE;
7320 	hibernate_skip_external = FALSE;
7321 
7322 	if ((retval = hibernate_flush_dirty_pages(1)) == 0) {
7323 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_START, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7324 
7325 		vm_compressor_flush();
7326 
7327 		KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0);
7328 
7329 		if (consider_buffer_cache_collect != NULL) {
7330 			unsigned int orig_wire_count;
7331 
7332 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, 0, 0, 0, 0, 0);
7333 			orig_wire_count = vm_page_wire_count;
7334 
7335 			(void)(*consider_buffer_cache_collect)(1);
7336 			zone_gc(ZONE_GC_DRAIN);
7337 
7338 			HIBLOG("hibernate_flush_memory: buffer_cache_gc freed up %d wired pages\n", orig_wire_count - vm_page_wire_count);
7339 
7340 			KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0);
7341 		}
7342 	}
7343 	hibernate_cleaning_in_progress = FALSE;
7344 
7345 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0);
7346 
7347 	if (retval) {
7348 		HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n", VM_PAGE_COMPRESSOR_COUNT);
7349 	}
7350 
7351 
7352 	HIBPRINT("hibernate_flush_memory() considered(%d) reentered_on_q(%d) found_dirty(%d)\n",
7353 	    hibernate_stats.hibernate_considered,
7354 	    hibernate_stats.hibernate_reentered_on_q,
7355 	    hibernate_stats.hibernate_found_dirty);
7356 	HIBPRINT("   skipped_cleaning(%d) skipped_transient(%d) skipped_precious(%d) skipped_external(%d) queue_nolock(%d)\n",
7357 	    hibernate_stats.hibernate_skipped_cleaning,
7358 	    hibernate_stats.hibernate_skipped_transient,
7359 	    hibernate_stats.hibernate_skipped_precious,
7360 	    hibernate_stats.hibernate_skipped_external,
7361 	    hibernate_stats.hibernate_queue_nolock);
7362 	HIBPRINT("   queue_paused(%d) throttled(%d) throttle_timeout(%d) drained(%d) drain_timeout(%d)\n",
7363 	    hibernate_stats.hibernate_queue_paused,
7364 	    hibernate_stats.hibernate_throttled,
7365 	    hibernate_stats.hibernate_throttle_timeout,
7366 	    hibernate_stats.hibernate_drained,
7367 	    hibernate_stats.hibernate_drain_timeout);
7368 
7369 	return retval;
7370 }
7371 
7372 
7373 static void
hibernate_page_list_zero(hibernate_page_list_t * list)7374 hibernate_page_list_zero(hibernate_page_list_t *list)
7375 {
7376 	uint32_t             bank;
7377 	hibernate_bitmap_t * bitmap;
7378 
7379 	bitmap = &list->bank_bitmap[0];
7380 	for (bank = 0; bank < list->bank_count; bank++) {
7381 		uint32_t last_bit;
7382 
7383 		bzero((void *) &bitmap->bitmap[0], bitmap->bitmapwords << 2);
7384 		// set out-of-bound bits at end of bitmap.
7385 		last_bit = ((bitmap->last_page - bitmap->first_page + 1) & 31);
7386 		if (last_bit) {
7387 			bitmap->bitmap[bitmap->bitmapwords - 1] = (0xFFFFFFFF >> last_bit);
7388 		}
7389 
7390 		bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords];
7391 	}
7392 }
7393 
7394 void
hibernate_free_gobble_pages(void)7395 hibernate_free_gobble_pages(void)
7396 {
7397 	vm_page_t m, next;
7398 	uint32_t  count = 0;
7399 
7400 	m = (vm_page_t) hibernate_gobble_queue;
7401 	while (m) {
7402 		next = m->vmp_snext;
7403 		vm_page_free(m);
7404 		count++;
7405 		m = next;
7406 	}
7407 	hibernate_gobble_queue = VM_PAGE_NULL;
7408 
7409 	if (count) {
7410 		HIBLOG("Freed %d pages\n", count);
7411 	}
7412 }
7413 
7414 static boolean_t
hibernate_consider_discard(vm_page_t m,boolean_t preflight)7415 hibernate_consider_discard(vm_page_t m, boolean_t preflight)
7416 {
7417 	vm_object_t object = NULL;
7418 	int                  refmod_state;
7419 	boolean_t            discard = FALSE;
7420 
7421 	do{
7422 		if (m->vmp_private) {
7423 			panic("hibernate_consider_discard: private");
7424 		}
7425 
7426 		object = VM_PAGE_OBJECT(m);
7427 
7428 		if (!vm_object_lock_try(object)) {
7429 			object = NULL;
7430 			if (!preflight) {
7431 				hibernate_stats.cd_lock_failed++;
7432 			}
7433 			break;
7434 		}
7435 		if (VM_PAGE_WIRED(m)) {
7436 			if (!preflight) {
7437 				hibernate_stats.cd_found_wired++;
7438 			}
7439 			break;
7440 		}
7441 		if (m->vmp_precious) {
7442 			if (!preflight) {
7443 				hibernate_stats.cd_found_precious++;
7444 			}
7445 			break;
7446 		}
7447 		if (m->vmp_busy || !object->alive) {
7448 			/*
7449 			 *	Somebody is playing with this page.
7450 			 */
7451 			if (!preflight) {
7452 				hibernate_stats.cd_found_busy++;
7453 			}
7454 			break;
7455 		}
7456 		if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7457 			/*
7458 			 * If it's unusual in anyway, ignore it
7459 			 */
7460 			if (!preflight) {
7461 				hibernate_stats.cd_found_unusual++;
7462 			}
7463 			break;
7464 		}
7465 		if (m->vmp_cleaning) {
7466 			if (!preflight) {
7467 				hibernate_stats.cd_found_cleaning++;
7468 			}
7469 			break;
7470 		}
7471 		if (m->vmp_laundry) {
7472 			if (!preflight) {
7473 				hibernate_stats.cd_found_laundry++;
7474 			}
7475 			break;
7476 		}
7477 		if (!m->vmp_dirty) {
7478 			refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m));
7479 
7480 			if (refmod_state & VM_MEM_REFERENCED) {
7481 				m->vmp_reference = TRUE;
7482 			}
7483 			if (refmod_state & VM_MEM_MODIFIED) {
7484 				SET_PAGE_DIRTY(m, FALSE);
7485 			}
7486 		}
7487 
7488 		/*
7489 		 * If it's clean or purgeable we can discard the page on wakeup.
7490 		 */
7491 		discard = (!m->vmp_dirty)
7492 		    || (VM_PURGABLE_VOLATILE == object->purgable)
7493 		    || (VM_PURGABLE_EMPTY == object->purgable);
7494 
7495 
7496 		if (discard == FALSE) {
7497 			if (!preflight) {
7498 				hibernate_stats.cd_found_dirty++;
7499 			}
7500 		} else if (m->vmp_xpmapped && m->vmp_reference && !object->internal) {
7501 			if (hibernate_stats.cd_found_xpmapped < HIBERNATE_XPMAPPED_LIMIT) {
7502 				if (!preflight) {
7503 					hibernate_stats.cd_found_xpmapped++;
7504 				}
7505 				discard = FALSE;
7506 			} else {
7507 				if (!preflight) {
7508 					hibernate_stats.cd_skipped_xpmapped++;
7509 				}
7510 			}
7511 		}
7512 	}while (FALSE);
7513 
7514 	if (object) {
7515 		vm_object_unlock(object);
7516 	}
7517 
7518 	return discard;
7519 }
7520 
7521 
7522 static void
hibernate_discard_page(vm_page_t m)7523 hibernate_discard_page(vm_page_t m)
7524 {
7525 	vm_object_t m_object;
7526 
7527 	if (m->vmp_absent || m->vmp_unusual || VMP_ERROR_GET(m)) {
7528 		/*
7529 		 * If it's unusual in anyway, ignore
7530 		 */
7531 		return;
7532 	}
7533 
7534 	m_object = VM_PAGE_OBJECT(m);
7535 
7536 #if MACH_ASSERT || DEBUG
7537 	if (!vm_object_lock_try(m_object)) {
7538 		panic("hibernate_discard_page(%p) !vm_object_lock_try", m);
7539 	}
7540 #else
7541 	/* No need to lock page queue for token delete, hibernate_vm_unlock()
7542 	 *  makes sure these locks are uncontended before sleep */
7543 #endif /* MACH_ASSERT || DEBUG */
7544 
7545 	if (m->vmp_pmapped == TRUE) {
7546 		__unused int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m));
7547 	}
7548 
7549 	if (m->vmp_laundry) {
7550 		panic("hibernate_discard_page(%p) laundry", m);
7551 	}
7552 	if (m->vmp_private) {
7553 		panic("hibernate_discard_page(%p) private", m);
7554 	}
7555 	if (m->vmp_fictitious) {
7556 		panic("hibernate_discard_page(%p) fictitious", m);
7557 	}
7558 
7559 	if (VM_PURGABLE_VOLATILE == m_object->purgable) {
7560 		/* object should be on a queue */
7561 		assert((m_object->objq.next != NULL) && (m_object->objq.prev != NULL));
7562 		purgeable_q_t old_queue = vm_purgeable_object_remove(m_object);
7563 		assert(old_queue);
7564 		if (m_object->purgeable_when_ripe) {
7565 			vm_purgeable_token_delete_first(old_queue);
7566 		}
7567 		vm_object_lock_assert_exclusive(m_object);
7568 		m_object->purgable = VM_PURGABLE_EMPTY;
7569 
7570 		/*
7571 		 * Purgeable ledgers:  pages of VOLATILE and EMPTY objects are
7572 		 * accounted in the "volatile" ledger, so no change here.
7573 		 * We have to update vm_page_purgeable_count, though, since we're
7574 		 * effectively purging this object.
7575 		 */
7576 		unsigned int delta;
7577 		assert(m_object->resident_page_count >= m_object->wired_page_count);
7578 		delta = (m_object->resident_page_count - m_object->wired_page_count);
7579 		assert(vm_page_purgeable_count >= delta);
7580 		assert(delta > 0);
7581 		OSAddAtomic(-delta, (SInt32 *)&vm_page_purgeable_count);
7582 	}
7583 
7584 	vm_page_free(m);
7585 
7586 #if MACH_ASSERT || DEBUG
7587 	vm_object_unlock(m_object);
7588 #endif  /* MACH_ASSERT || DEBUG */
7589 }
7590 
7591 /*
7592  *  Grab locks for hibernate_page_list_setall()
7593  */
7594 void
hibernate_vm_lock_queues(void)7595 hibernate_vm_lock_queues(void)
7596 {
7597 	vm_object_lock(compressor_object);
7598 	vm_page_lock_queues();
7599 	vm_free_page_lock();
7600 	lck_mtx_lock(&vm_purgeable_queue_lock);
7601 
7602 	if (vm_page_local_q) {
7603 		zpercpu_foreach(lq, vm_page_local_q) {
7604 			VPL_LOCK(&lq->vpl_lock);
7605 		}
7606 	}
7607 }
7608 
7609 void
hibernate_vm_unlock_queues(void)7610 hibernate_vm_unlock_queues(void)
7611 {
7612 	if (vm_page_local_q) {
7613 		zpercpu_foreach(lq, vm_page_local_q) {
7614 			VPL_UNLOCK(&lq->vpl_lock);
7615 		}
7616 	}
7617 	lck_mtx_unlock(&vm_purgeable_queue_lock);
7618 	vm_free_page_unlock();
7619 	vm_page_unlock_queues();
7620 	vm_object_unlock(compressor_object);
7621 }
7622 
7623 /*
7624  *  Bits zero in the bitmaps => page needs to be saved. All pages default to be saved,
7625  *  pages known to VM to not need saving are subtracted.
7626  *  Wired pages to be saved are present in page_list_wired, pageable in page_list.
7627  */
7628 
7629 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)7630 hibernate_page_list_setall(hibernate_page_list_t * page_list,
7631     hibernate_page_list_t * page_list_wired,
7632     hibernate_page_list_t * page_list_pal,
7633     boolean_t preflight,
7634     boolean_t will_discard,
7635     uint32_t * pagesOut)
7636 {
7637 	uint64_t start, end, nsec;
7638 	vm_page_t m;
7639 	vm_page_t next;
7640 	uint32_t pages = page_list->page_count;
7641 	uint32_t count_anonymous = 0, count_throttled = 0, count_compressor = 0;
7642 	uint32_t count_inactive = 0, count_active = 0, count_speculative = 0, count_cleaned = 0;
7643 	uint32_t count_wire = pages;
7644 	uint32_t count_discard_active    = 0;
7645 	uint32_t count_discard_inactive  = 0;
7646 	uint32_t count_retired = 0;
7647 	uint32_t count_discard_cleaned   = 0;
7648 	uint32_t count_discard_purgeable = 0;
7649 	uint32_t count_discard_speculative = 0;
7650 	uint32_t count_discard_vm_struct_pages = 0;
7651 	uint32_t i;
7652 	uint32_t             bank;
7653 	hibernate_bitmap_t * bitmap;
7654 	hibernate_bitmap_t * bitmap_wired;
7655 	boolean_t                    discard_all;
7656 	boolean_t            discard = FALSE;
7657 
7658 	HIBLOG("hibernate_page_list_setall(preflight %d) start\n", preflight);
7659 
7660 	if (preflight) {
7661 		page_list       = NULL;
7662 		page_list_wired = NULL;
7663 		page_list_pal   = NULL;
7664 		discard_all     = FALSE;
7665 	} else {
7666 		discard_all     = will_discard;
7667 	}
7668 
7669 #if MACH_ASSERT || DEBUG
7670 	if (!preflight) {
7671 		assert(hibernate_vm_locks_are_safe());
7672 		vm_page_lock_queues();
7673 		if (vm_page_local_q) {
7674 			zpercpu_foreach(lq, vm_page_local_q) {
7675 				VPL_LOCK(&lq->vpl_lock);
7676 			}
7677 		}
7678 	}
7679 #endif  /* MACH_ASSERT || DEBUG */
7680 
7681 
7682 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0);
7683 
7684 	clock_get_uptime(&start);
7685 
7686 	if (!preflight) {
7687 		hibernate_page_list_zero(page_list);
7688 		hibernate_page_list_zero(page_list_wired);
7689 		hibernate_page_list_zero(page_list_pal);
7690 
7691 		hibernate_stats.cd_vm_page_wire_count = vm_page_wire_count;
7692 		hibernate_stats.cd_pages = pages;
7693 	}
7694 
7695 	if (vm_page_local_q) {
7696 		zpercpu_foreach_cpu(lid) {
7697 			vm_page_reactivate_local(lid, TRUE, !preflight);
7698 		}
7699 	}
7700 
7701 	if (preflight) {
7702 		vm_object_lock(compressor_object);
7703 		vm_page_lock_queues();
7704 		vm_free_page_lock();
7705 	}
7706 
7707 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
7708 
7709 	hibernation_vmqueues_inspection = TRUE;
7710 
7711 	m = (vm_page_t) hibernate_gobble_queue;
7712 	while (m) {
7713 		pages--;
7714 		count_wire--;
7715 		if (!preflight) {
7716 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7717 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7718 		}
7719 		m = m->vmp_snext;
7720 	}
7721 
7722 	if (!preflight) {
7723 		percpu_foreach(free_pages_head, free_pages) {
7724 			for (m = *free_pages_head; m; m = m->vmp_snext) {
7725 				assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q);
7726 
7727 				pages--;
7728 				count_wire--;
7729 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7730 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7731 
7732 				hibernate_stats.cd_local_free++;
7733 				hibernate_stats.cd_total_free++;
7734 			}
7735 		}
7736 	}
7737 
7738 	for (i = 0; i < vm_colors; i++) {
7739 		vm_page_queue_iterate(&vm_page_queue_free[i].qhead, m, vmp_pageq) {
7740 			assert(m->vmp_q_state == VM_PAGE_ON_FREE_Q);
7741 
7742 			pages--;
7743 			count_wire--;
7744 			if (!preflight) {
7745 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7746 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7747 
7748 				hibernate_stats.cd_total_free++;
7749 			}
7750 		}
7751 	}
7752 
7753 	vm_page_queue_iterate(&vm_lopage_queue_free, m, vmp_pageq) {
7754 		assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q);
7755 
7756 		pages--;
7757 		count_wire--;
7758 		if (!preflight) {
7759 			hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7760 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7761 
7762 			hibernate_stats.cd_total_free++;
7763 		}
7764 	}
7765 
7766 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled);
7767 	while (m && !vm_page_queue_end(&vm_page_queue_throttled, (vm_page_queue_entry_t)m)) {
7768 		assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q);
7769 
7770 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7771 		discard = FALSE;
7772 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode)
7773 		    && hibernate_consider_discard(m, preflight)) {
7774 			if (!preflight) {
7775 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7776 			}
7777 			count_discard_inactive++;
7778 			discard = discard_all;
7779 		} else {
7780 			count_throttled++;
7781 		}
7782 		count_wire--;
7783 		if (!preflight) {
7784 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7785 		}
7786 
7787 		if (discard) {
7788 			hibernate_discard_page(m);
7789 		}
7790 		m = next;
7791 	}
7792 
7793 	m = (vm_page_t)vm_page_queue_first(&vm_page_queue_anonymous);
7794 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
7795 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
7796 
7797 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7798 		discard = FALSE;
7799 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7800 		    hibernate_consider_discard(m, preflight)) {
7801 			if (!preflight) {
7802 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7803 			}
7804 			if (m->vmp_dirty) {
7805 				count_discard_purgeable++;
7806 			} else {
7807 				count_discard_inactive++;
7808 			}
7809 			discard = discard_all;
7810 		} else {
7811 			count_anonymous++;
7812 		}
7813 		count_wire--;
7814 		if (!preflight) {
7815 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7816 		}
7817 		if (discard) {
7818 			hibernate_discard_page(m);
7819 		}
7820 		m = next;
7821 	}
7822 
7823 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
7824 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
7825 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
7826 
7827 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7828 		discard = FALSE;
7829 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7830 		    hibernate_consider_discard(m, preflight)) {
7831 			if (!preflight) {
7832 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7833 			}
7834 			if (m->vmp_dirty) {
7835 				count_discard_purgeable++;
7836 			} else {
7837 				count_discard_cleaned++;
7838 			}
7839 			discard = discard_all;
7840 		} else {
7841 			count_cleaned++;
7842 		}
7843 		count_wire--;
7844 		if (!preflight) {
7845 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7846 		}
7847 		if (discard) {
7848 			hibernate_discard_page(m);
7849 		}
7850 		m = next;
7851 	}
7852 
7853 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
7854 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
7855 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
7856 
7857 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7858 		discard = FALSE;
7859 		if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) &&
7860 		    hibernate_consider_discard(m, preflight)) {
7861 			if (!preflight) {
7862 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7863 			}
7864 			if (m->vmp_dirty) {
7865 				count_discard_purgeable++;
7866 			} else {
7867 				count_discard_active++;
7868 			}
7869 			discard = discard_all;
7870 		} else {
7871 			count_active++;
7872 		}
7873 		count_wire--;
7874 		if (!preflight) {
7875 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7876 		}
7877 		if (discard) {
7878 			hibernate_discard_page(m);
7879 		}
7880 		m = next;
7881 	}
7882 
7883 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
7884 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
7885 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
7886 
7887 		next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7888 		discard = FALSE;
7889 		if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7890 		    hibernate_consider_discard(m, preflight)) {
7891 			if (!preflight) {
7892 				hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7893 			}
7894 			if (m->vmp_dirty) {
7895 				count_discard_purgeable++;
7896 			} else {
7897 				count_discard_inactive++;
7898 			}
7899 			discard = discard_all;
7900 		} else {
7901 			count_inactive++;
7902 		}
7903 		count_wire--;
7904 		if (!preflight) {
7905 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7906 		}
7907 		if (discard) {
7908 			hibernate_discard_page(m);
7909 		}
7910 		m = next;
7911 	}
7912 	/* XXX FBDP TODO: secluded queue */
7913 
7914 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
7915 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
7916 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
7917 			assertf(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q,
7918 			    "Bad page: %p (0x%x:0x%x) on queue %d has state: %d (Discard: %d, Preflight: %d)",
7919 			    m, m->vmp_pageq.next, m->vmp_pageq.prev, i, m->vmp_q_state, discard, preflight);
7920 
7921 			next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
7922 			discard = FALSE;
7923 			if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) &&
7924 			    hibernate_consider_discard(m, preflight)) {
7925 				if (!preflight) {
7926 					hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7927 				}
7928 				count_discard_speculative++;
7929 				discard = discard_all;
7930 			} else {
7931 				count_speculative++;
7932 			}
7933 			count_wire--;
7934 			if (!preflight) {
7935 				hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7936 			}
7937 			if (discard) {
7938 				hibernate_discard_page(m);
7939 			}
7940 			m = next;
7941 		}
7942 	}
7943 
7944 	vm_page_queue_iterate(&compressor_object->memq, m, vmp_listq) {
7945 		assert(m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR);
7946 
7947 		count_compressor++;
7948 		count_wire--;
7949 		if (!preflight) {
7950 			hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m));
7951 		}
7952 	}
7953 
7954 
7955 	if (preflight == FALSE && discard_all == TRUE) {
7956 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START);
7957 
7958 		HIBLOG("hibernate_teardown started\n");
7959 		count_discard_vm_struct_pages = hibernate_teardown_vm_structs(page_list, page_list_wired);
7960 		HIBLOG("hibernate_teardown completed - discarded %d\n", count_discard_vm_struct_pages);
7961 
7962 		pages -= count_discard_vm_struct_pages;
7963 		count_wire -= count_discard_vm_struct_pages;
7964 
7965 		hibernate_stats.cd_vm_struct_pages_unneeded = count_discard_vm_struct_pages;
7966 
7967 		KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_END);
7968 	}
7969 
7970 	if (!preflight) {
7971 		// pull wired from hibernate_bitmap
7972 		bitmap = &page_list->bank_bitmap[0];
7973 		bitmap_wired = &page_list_wired->bank_bitmap[0];
7974 		for (bank = 0; bank < page_list->bank_count; bank++) {
7975 			for (i = 0; i < bitmap->bitmapwords; i++) {
7976 				bitmap->bitmap[i] = bitmap->bitmap[i] | ~bitmap_wired->bitmap[i];
7977 			}
7978 			bitmap = (hibernate_bitmap_t *)&bitmap->bitmap[bitmap->bitmapwords];
7979 			bitmap_wired = (hibernate_bitmap_t *) &bitmap_wired->bitmap[bitmap_wired->bitmapwords];
7980 		}
7981 	}
7982 
7983 	// machine dependent adjustments
7984 	hibernate_page_list_setall_machine(page_list, page_list_wired, preflight, &pages);
7985 
7986 	if (!preflight) {
7987 		hibernate_stats.cd_count_wire = count_wire;
7988 		hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable +
7989 		    count_discard_speculative + count_discard_cleaned + count_discard_vm_struct_pages;
7990 	}
7991 
7992 	clock_get_uptime(&end);
7993 	absolutetime_to_nanoseconds(end - start, &nsec);
7994 	HIBLOG("hibernate_page_list_setall time: %qd ms\n", nsec / 1000000ULL);
7995 
7996 	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",
7997 	    pages, count_wire, count_active, count_inactive, count_cleaned, count_speculative, count_anonymous, count_throttled, count_compressor, hibernate_stats.cd_found_xpmapped,
7998 	    discard_all ? "did" : "could",
7999 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned, count_retired);
8000 
8001 	if (hibernate_stats.cd_skipped_xpmapped) {
8002 		HIBLOG("WARNING: hibernate_page_list_setall skipped %d xpmapped pages\n", hibernate_stats.cd_skipped_xpmapped);
8003 	}
8004 
8005 	*pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative - count_discard_cleaned - count_retired;
8006 
8007 	if (preflight && will_discard) {
8008 		*pagesOut -= count_compressor + count_throttled + count_anonymous + count_inactive + count_cleaned + count_speculative + count_active;
8009 		/*
8010 		 * We try to keep max HIBERNATE_XPMAPPED_LIMIT pages around in the hibernation image
8011 		 * even if these are clean and so we need to size the hibernation image accordingly.
8012 		 *
8013 		 * NB: We have to assume all HIBERNATE_XPMAPPED_LIMIT pages might show up because 'dirty'
8014 		 * xpmapped pages aren't distinguishable from other 'dirty' pages in preflight. So we might
8015 		 * only see part of the xpmapped pages if we look at 'cd_found_xpmapped' which solely tracks
8016 		 * clean xpmapped pages.
8017 		 *
8018 		 * Since these pages are all cleaned by the time we are in the post-preflight phase, we might
8019 		 * see a much larger number in 'cd_found_xpmapped' now than we did in the preflight phase
8020 		 */
8021 		*pagesOut +=  HIBERNATE_XPMAPPED_LIMIT;
8022 	}
8023 
8024 	hibernation_vmqueues_inspection = FALSE;
8025 
8026 #if MACH_ASSERT || DEBUG
8027 	if (!preflight) {
8028 		if (vm_page_local_q) {
8029 			zpercpu_foreach(lq, vm_page_local_q) {
8030 				VPL_UNLOCK(&lq->vpl_lock);
8031 			}
8032 		}
8033 		vm_page_unlock_queues();
8034 	}
8035 #endif  /* MACH_ASSERT || DEBUG */
8036 
8037 	if (preflight) {
8038 		vm_free_page_unlock();
8039 		vm_page_unlock_queues();
8040 		vm_object_unlock(compressor_object);
8041 	}
8042 
8043 	KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_END, count_wire, *pagesOut, 0, 0, 0);
8044 }
8045 
8046 void
hibernate_page_list_discard(hibernate_page_list_t * page_list)8047 hibernate_page_list_discard(hibernate_page_list_t * page_list)
8048 {
8049 	uint64_t  start, end, nsec;
8050 	vm_page_t m;
8051 	vm_page_t next;
8052 	uint32_t  i;
8053 	uint32_t  count_discard_active    = 0;
8054 	uint32_t  count_discard_inactive  = 0;
8055 	uint32_t  count_discard_purgeable = 0;
8056 	uint32_t  count_discard_cleaned   = 0;
8057 	uint32_t  count_discard_speculative = 0;
8058 
8059 
8060 #if MACH_ASSERT || DEBUG
8061 	vm_page_lock_queues();
8062 	if (vm_page_local_q) {
8063 		zpercpu_foreach(lq, vm_page_local_q) {
8064 			VPL_LOCK(&lq->vpl_lock);
8065 		}
8066 	}
8067 #endif  /* MACH_ASSERT || DEBUG */
8068 
8069 	clock_get_uptime(&start);
8070 
8071 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_anonymous);
8072 	while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) {
8073 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q);
8074 
8075 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8076 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8077 			if (m->vmp_dirty) {
8078 				count_discard_purgeable++;
8079 			} else {
8080 				count_discard_inactive++;
8081 			}
8082 			hibernate_discard_page(m);
8083 		}
8084 		m = next;
8085 	}
8086 
8087 	for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) {
8088 		m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q);
8089 		while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) {
8090 			assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_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 				count_discard_speculative++;
8095 				hibernate_discard_page(m);
8096 			}
8097 			m = next;
8098 		}
8099 	}
8100 
8101 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive);
8102 	while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) {
8103 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q);
8104 
8105 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8106 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8107 			if (m->vmp_dirty) {
8108 				count_discard_purgeable++;
8109 			} else {
8110 				count_discard_inactive++;
8111 			}
8112 			hibernate_discard_page(m);
8113 		}
8114 		m = next;
8115 	}
8116 	/* XXX FBDP TODO: secluded queue */
8117 
8118 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active);
8119 	while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) {
8120 		assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q);
8121 
8122 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8123 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8124 			if (m->vmp_dirty) {
8125 				count_discard_purgeable++;
8126 			} else {
8127 				count_discard_active++;
8128 			}
8129 			hibernate_discard_page(m);
8130 		}
8131 		m = next;
8132 	}
8133 
8134 	m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned);
8135 	while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) {
8136 		assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q);
8137 
8138 		next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next);
8139 		if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) {
8140 			if (m->vmp_dirty) {
8141 				count_discard_purgeable++;
8142 			} else {
8143 				count_discard_cleaned++;
8144 			}
8145 			hibernate_discard_page(m);
8146 		}
8147 		m = next;
8148 	}
8149 
8150 #if MACH_ASSERT || DEBUG
8151 	if (vm_page_local_q) {
8152 		zpercpu_foreach(lq, vm_page_local_q) {
8153 			VPL_UNLOCK(&lq->vpl_lock);
8154 		}
8155 	}
8156 	vm_page_unlock_queues();
8157 #endif  /* MACH_ASSERT || DEBUG */
8158 
8159 	clock_get_uptime(&end);
8160 	absolutetime_to_nanoseconds(end - start, &nsec);
8161 	HIBLOG("hibernate_page_list_discard time: %qd ms, discarded act %d inact %d purgeable %d spec %d cleaned %d\n",
8162 	    nsec / 1000000ULL,
8163 	    count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned);
8164 }
8165 
8166 boolean_t       hibernate_paddr_map_inited = FALSE;
8167 unsigned int    hibernate_teardown_last_valid_compact_indx = -1;
8168 vm_page_t       hibernate_rebuild_hash_list = NULL;
8169 
8170 unsigned int    hibernate_teardown_found_tabled_pages = 0;
8171 unsigned int    hibernate_teardown_found_created_pages = 0;
8172 unsigned int    hibernate_teardown_found_free_pages = 0;
8173 unsigned int    hibernate_teardown_vm_page_free_count;
8174 
8175 
8176 struct ppnum_mapping {
8177 	struct ppnum_mapping    *ppnm_next;
8178 	ppnum_t                 ppnm_base_paddr;
8179 	unsigned int            ppnm_sindx;
8180 	unsigned int            ppnm_eindx;
8181 };
8182 
8183 struct ppnum_mapping    *ppnm_head;
8184 struct ppnum_mapping    *ppnm_last_found = NULL;
8185 
8186 
8187 void
hibernate_create_paddr_map(void)8188 hibernate_create_paddr_map(void)
8189 {
8190 	unsigned int    i;
8191 	ppnum_t         next_ppnum_in_run = 0;
8192 	struct ppnum_mapping *ppnm = NULL;
8193 
8194 	if (hibernate_paddr_map_inited == FALSE) {
8195 		for (i = 0; i < vm_pages_count; i++) {
8196 			if (ppnm) {
8197 				ppnm->ppnm_eindx = i;
8198 			}
8199 
8200 			if (ppnm == NULL || VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) != next_ppnum_in_run) {
8201 				ppnm = zalloc_permanent_type(struct ppnum_mapping);
8202 
8203 				ppnm->ppnm_next = ppnm_head;
8204 				ppnm_head = ppnm;
8205 
8206 				ppnm->ppnm_sindx = i;
8207 				ppnm->ppnm_base_paddr = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]);
8208 			}
8209 			next_ppnum_in_run = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) + 1;
8210 		}
8211 		ppnm->ppnm_eindx = vm_pages_count;
8212 
8213 		hibernate_paddr_map_inited = TRUE;
8214 	}
8215 }
8216 
8217 ppnum_t
hibernate_lookup_paddr(unsigned int indx)8218 hibernate_lookup_paddr(unsigned int indx)
8219 {
8220 	struct ppnum_mapping *ppnm = NULL;
8221 
8222 	ppnm = ppnm_last_found;
8223 
8224 	if (ppnm) {
8225 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8226 			goto done;
8227 		}
8228 	}
8229 	for (ppnm = ppnm_head; ppnm; ppnm = ppnm->ppnm_next) {
8230 		if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) {
8231 			ppnm_last_found = ppnm;
8232 			break;
8233 		}
8234 	}
8235 	if (ppnm == NULL) {
8236 		panic("hibernate_lookup_paddr of %d failed", indx);
8237 	}
8238 done:
8239 	return ppnm->ppnm_base_paddr + (indx - ppnm->ppnm_sindx);
8240 }
8241 
8242 
8243 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)8244 hibernate_mark_as_unneeded(addr64_t saddr, addr64_t eaddr, hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8245 {
8246 	addr64_t        saddr_aligned;
8247 	addr64_t        eaddr_aligned;
8248 	addr64_t        addr;
8249 	ppnum_t         paddr;
8250 	unsigned int    mark_as_unneeded_pages = 0;
8251 
8252 	saddr_aligned = (saddr + PAGE_MASK_64) & ~PAGE_MASK_64;
8253 	eaddr_aligned = eaddr & ~PAGE_MASK_64;
8254 
8255 	for (addr = saddr_aligned; addr < eaddr_aligned; addr += PAGE_SIZE_64) {
8256 		paddr = pmap_find_phys(kernel_pmap, addr);
8257 
8258 		assert(paddr);
8259 
8260 		hibernate_page_bitset(page_list, TRUE, paddr);
8261 		hibernate_page_bitset(page_list_wired, TRUE, paddr);
8262 
8263 		mark_as_unneeded_pages++;
8264 	}
8265 	return mark_as_unneeded_pages;
8266 }
8267 
8268 
8269 void
hibernate_hash_insert_page(vm_page_t mem)8270 hibernate_hash_insert_page(vm_page_t mem)
8271 {
8272 	vm_page_bucket_t *bucket;
8273 	int             hash_id;
8274 	vm_object_t     m_object;
8275 
8276 	m_object = VM_PAGE_OBJECT(mem);
8277 
8278 	assert(mem->vmp_hashed);
8279 	assert(m_object);
8280 	assert(mem->vmp_offset != (vm_object_offset_t) -1);
8281 
8282 	/*
8283 	 *	Insert it into the object_object/offset hash table
8284 	 */
8285 	hash_id = vm_page_hash(m_object, mem->vmp_offset);
8286 	bucket = &vm_page_buckets[hash_id];
8287 
8288 	mem->vmp_next_m = bucket->page_list;
8289 	bucket->page_list = VM_PAGE_PACK_PTR(mem);
8290 }
8291 
8292 
8293 void
hibernate_free_range(int sindx,int eindx)8294 hibernate_free_range(int sindx, int eindx)
8295 {
8296 	vm_page_t       mem;
8297 	unsigned int    color;
8298 
8299 	while (sindx < eindx) {
8300 		mem = &vm_pages[sindx];
8301 
8302 		vm_page_init(mem, hibernate_lookup_paddr(sindx), FALSE);
8303 
8304 		mem->vmp_lopage = FALSE;
8305 		mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8306 
8307 		color = VM_PAGE_GET_COLOR(mem);
8308 #if defined(__x86_64__)
8309 		vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, mem);
8310 #else
8311 		vm_page_queue_enter(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8312 #endif
8313 		vm_page_free_count++;
8314 
8315 		sindx++;
8316 	}
8317 }
8318 
8319 void
hibernate_rebuild_vm_structs(void)8320 hibernate_rebuild_vm_structs(void)
8321 {
8322 	int             i, cindx, sindx, eindx;
8323 	vm_page_t       mem, tmem, mem_next;
8324 	AbsoluteTime    startTime, endTime;
8325 	uint64_t        nsec;
8326 
8327 	if (hibernate_rebuild_needed == FALSE) {
8328 		return;
8329 	}
8330 
8331 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START);
8332 	HIBLOG("hibernate_rebuild started\n");
8333 
8334 	clock_get_uptime(&startTime);
8335 
8336 	pal_hib_rebuild_pmap_structs();
8337 
8338 	bzero(&vm_page_buckets[0], vm_page_bucket_count * sizeof(vm_page_bucket_t));
8339 	eindx = vm_pages_count;
8340 
8341 	/*
8342 	 * Mark all the vm_pages[] that have not been initialized yet as being
8343 	 * transient. This is needed to ensure that buddy page search is corrrect.
8344 	 * Without this random data in these vm_pages[] can trip the buddy search
8345 	 */
8346 	for (i = hibernate_teardown_last_valid_compact_indx + 1; i < eindx; ++i) {
8347 		vm_pages[i].vmp_q_state = VM_PAGE_NOT_ON_Q;
8348 	}
8349 
8350 	for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) {
8351 		mem = &vm_pages[cindx];
8352 		assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q);
8353 		/*
8354 		 * hibernate_teardown_vm_structs leaves the location where
8355 		 * this vm_page_t must be located in "next".
8356 		 */
8357 		tmem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8358 		mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL);
8359 
8360 		sindx = (int)(tmem - &vm_pages[0]);
8361 
8362 		if (mem != tmem) {
8363 			/*
8364 			 * this vm_page_t was moved by hibernate_teardown_vm_structs,
8365 			 * so move it back to its real location
8366 			 */
8367 			*tmem = *mem;
8368 			mem = tmem;
8369 		}
8370 		if (mem->vmp_hashed) {
8371 			hibernate_hash_insert_page(mem);
8372 		}
8373 		/*
8374 		 * the 'hole' between this vm_page_t and the previous
8375 		 * vm_page_t we moved needs to be initialized as
8376 		 * a range of free vm_page_t's
8377 		 */
8378 		hibernate_free_range(sindx + 1, eindx);
8379 
8380 		eindx = sindx;
8381 	}
8382 	if (sindx) {
8383 		hibernate_free_range(0, sindx);
8384 	}
8385 
8386 	assert(vm_page_free_count == hibernate_teardown_vm_page_free_count);
8387 
8388 	/*
8389 	 * process the list of vm_page_t's that were entered in the hash,
8390 	 * but were not located in the vm_pages arrary... these are
8391 	 * vm_page_t's that were created on the fly (i.e. fictitious)
8392 	 */
8393 	for (mem = hibernate_rebuild_hash_list; mem; mem = mem_next) {
8394 		mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8395 
8396 		mem->vmp_next_m = 0;
8397 		hibernate_hash_insert_page(mem);
8398 	}
8399 	hibernate_rebuild_hash_list = NULL;
8400 
8401 	clock_get_uptime(&endTime);
8402 	SUB_ABSOLUTETIME(&endTime, &startTime);
8403 	absolutetime_to_nanoseconds(endTime, &nsec);
8404 
8405 	HIBLOG("hibernate_rebuild completed - took %qd msecs\n", nsec / 1000000ULL);
8406 
8407 	hibernate_rebuild_needed = FALSE;
8408 
8409 	KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END);
8410 }
8411 
8412 uint32_t
hibernate_teardown_vm_structs(hibernate_page_list_t * page_list,hibernate_page_list_t * page_list_wired)8413 hibernate_teardown_vm_structs(hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired)
8414 {
8415 	unsigned int    i;
8416 	unsigned int    compact_target_indx;
8417 	vm_page_t       mem, mem_next;
8418 	vm_page_bucket_t *bucket;
8419 	unsigned int    mark_as_unneeded_pages = 0;
8420 	unsigned int    unneeded_vm_page_bucket_pages = 0;
8421 	unsigned int    unneeded_vm_pages_pages = 0;
8422 	unsigned int    unneeded_pmap_pages = 0;
8423 	addr64_t        start_of_unneeded = 0;
8424 	addr64_t        end_of_unneeded = 0;
8425 
8426 
8427 	if (hibernate_should_abort()) {
8428 		return 0;
8429 	}
8430 
8431 	hibernate_rebuild_needed = TRUE;
8432 
8433 	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",
8434 	    vm_page_wire_count, vm_page_free_count, vm_page_active_count, vm_page_inactive_count, vm_page_speculative_count,
8435 	    vm_page_cleaned_count, compressor_object->resident_page_count);
8436 
8437 	for (i = 0; i < vm_page_bucket_count; i++) {
8438 		bucket = &vm_page_buckets[i];
8439 
8440 		for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); mem != VM_PAGE_NULL; mem = mem_next) {
8441 			assert(mem->vmp_hashed);
8442 
8443 			mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m));
8444 
8445 			if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) {
8446 				mem->vmp_next_m = VM_PAGE_PACK_PTR(hibernate_rebuild_hash_list);
8447 				hibernate_rebuild_hash_list = mem;
8448 			}
8449 		}
8450 	}
8451 	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);
8452 	mark_as_unneeded_pages += unneeded_vm_page_bucket_pages;
8453 
8454 	hibernate_teardown_vm_page_free_count = vm_page_free_count;
8455 
8456 	compact_target_indx = 0;
8457 
8458 	for (i = 0; i < vm_pages_count; i++) {
8459 		mem = &vm_pages[i];
8460 
8461 		if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) {
8462 			unsigned int color;
8463 
8464 			assert(mem->vmp_busy);
8465 			assert(!mem->vmp_lopage);
8466 
8467 			color = VM_PAGE_GET_COLOR(mem);
8468 
8469 			vm_page_queue_remove(&vm_page_queue_free[color].qhead, mem, vmp_pageq);
8470 
8471 			VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
8472 
8473 			vm_page_free_count--;
8474 
8475 			hibernate_teardown_found_free_pages++;
8476 
8477 			if (vm_pages[compact_target_indx].vmp_q_state != VM_PAGE_ON_FREE_Q) {
8478 				compact_target_indx = i;
8479 			}
8480 		} else {
8481 			/*
8482 			 * record this vm_page_t's original location
8483 			 * we need this even if it doesn't get moved
8484 			 * as an indicator to the rebuild function that
8485 			 * we don't have to move it
8486 			 */
8487 			mem->vmp_next_m = VM_PAGE_PACK_PTR(mem);
8488 
8489 			if (vm_pages[compact_target_indx].vmp_q_state == VM_PAGE_ON_FREE_Q) {
8490 				/*
8491 				 * we've got a hole to fill, so
8492 				 * move this vm_page_t to it's new home
8493 				 */
8494 				vm_pages[compact_target_indx] = *mem;
8495 				mem->vmp_q_state = VM_PAGE_ON_FREE_Q;
8496 
8497 				hibernate_teardown_last_valid_compact_indx = compact_target_indx;
8498 				compact_target_indx++;
8499 			} else {
8500 				hibernate_teardown_last_valid_compact_indx = i;
8501 			}
8502 		}
8503 	}
8504 	unneeded_vm_pages_pages = hibernate_mark_as_unneeded((addr64_t)&vm_pages[hibernate_teardown_last_valid_compact_indx + 1],
8505 	    (addr64_t)&vm_pages[vm_pages_count - 1], page_list, page_list_wired);
8506 	mark_as_unneeded_pages += unneeded_vm_pages_pages;
8507 
8508 	pal_hib_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded);
8509 
8510 	if (start_of_unneeded) {
8511 		unneeded_pmap_pages = hibernate_mark_as_unneeded(start_of_unneeded, end_of_unneeded, page_list, page_list_wired);
8512 		mark_as_unneeded_pages += unneeded_pmap_pages;
8513 	}
8514 	HIBLOG("hibernate_teardown: mark_as_unneeded_pages %d, %d, %d\n", unneeded_vm_page_bucket_pages, unneeded_vm_pages_pages, unneeded_pmap_pages);
8515 
8516 	return mark_as_unneeded_pages;
8517 }
8518 
8519 
8520 #endif /* HIBERNATION */
8521 
8522 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
8523 
8524 #include <mach_vm_debug.h>
8525 #if     MACH_VM_DEBUG
8526 
8527 #include <mach_debug/hash_info.h>
8528 #include <vm/vm_debug.h>
8529 
8530 /*
8531  *	Routine:	vm_page_info
8532  *	Purpose:
8533  *		Return information about the global VP table.
8534  *		Fills the buffer with as much information as possible
8535  *		and returns the desired size of the buffer.
8536  *	Conditions:
8537  *		Nothing locked.  The caller should provide
8538  *		possibly-pageable memory.
8539  */
8540 
8541 unsigned int
vm_page_info(hash_info_bucket_t * info,unsigned int count)8542 vm_page_info(
8543 	hash_info_bucket_t *info,
8544 	unsigned int count)
8545 {
8546 	unsigned int i;
8547 	lck_spin_t      *bucket_lock;
8548 
8549 	if (vm_page_bucket_count < count) {
8550 		count = vm_page_bucket_count;
8551 	}
8552 
8553 	for (i = 0; i < count; i++) {
8554 		vm_page_bucket_t *bucket = &vm_page_buckets[i];
8555 		unsigned int bucket_count = 0;
8556 		vm_page_t m;
8557 
8558 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8559 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8560 
8561 		for (m = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8562 		    m != VM_PAGE_NULL;
8563 		    m = (vm_page_t)(VM_PAGE_UNPACK_PTR(m->vmp_next_m))) {
8564 			bucket_count++;
8565 		}
8566 
8567 		lck_spin_unlock(bucket_lock);
8568 
8569 		/* don't touch pageable memory while holding locks */
8570 		info[i].hib_count = bucket_count;
8571 	}
8572 
8573 	return vm_page_bucket_count;
8574 }
8575 #endif  /* MACH_VM_DEBUG */
8576 
8577 #if VM_PAGE_BUCKETS_CHECK
8578 void
vm_page_buckets_check(void)8579 vm_page_buckets_check(void)
8580 {
8581 	unsigned int i;
8582 	vm_page_t p;
8583 	unsigned int p_hash;
8584 	vm_page_bucket_t *bucket;
8585 	lck_spin_t      *bucket_lock;
8586 
8587 	if (!vm_page_buckets_check_ready) {
8588 		return;
8589 	}
8590 
8591 #if HIBERNATION
8592 	if (hibernate_rebuild_needed ||
8593 	    hibernate_rebuild_hash_list) {
8594 		panic("BUCKET_CHECK: hibernation in progress: "
8595 		    "rebuild_needed=%d rebuild_hash_list=%p\n",
8596 		    hibernate_rebuild_needed,
8597 		    hibernate_rebuild_hash_list);
8598 	}
8599 #endif /* HIBERNATION */
8600 
8601 #if VM_PAGE_FAKE_BUCKETS
8602 	char *cp;
8603 	for (cp = (char *) vm_page_fake_buckets_start;
8604 	    cp < (char *) vm_page_fake_buckets_end;
8605 	    cp++) {
8606 		if (*cp != 0x5a) {
8607 			panic("BUCKET_CHECK: corruption at %p in fake buckets "
8608 			    "[0x%llx:0x%llx]\n",
8609 			    cp,
8610 			    (uint64_t) vm_page_fake_buckets_start,
8611 			    (uint64_t) vm_page_fake_buckets_end);
8612 		}
8613 	}
8614 #endif /* VM_PAGE_FAKE_BUCKETS */
8615 
8616 	for (i = 0; i < vm_page_bucket_count; i++) {
8617 		vm_object_t     p_object;
8618 
8619 		bucket = &vm_page_buckets[i];
8620 		if (!bucket->page_list) {
8621 			continue;
8622 		}
8623 
8624 		bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK];
8625 		lck_spin_lock_grp(bucket_lock, &vm_page_lck_grp_bucket);
8626 		p = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list));
8627 
8628 		while (p != VM_PAGE_NULL) {
8629 			p_object = VM_PAGE_OBJECT(p);
8630 
8631 			if (!p->vmp_hashed) {
8632 				panic("BUCKET_CHECK: page %p (%p,0x%llx) "
8633 				    "hash %d in bucket %d at %p "
8634 				    "is not hashed\n",
8635 				    p, p_object, p->vmp_offset,
8636 				    p_hash, i, bucket);
8637 			}
8638 			p_hash = vm_page_hash(p_object, p->vmp_offset);
8639 			if (p_hash != i) {
8640 				panic("BUCKET_CHECK: corruption in bucket %d "
8641 				    "at %p: page %p object %p offset 0x%llx "
8642 				    "hash %d\n",
8643 				    i, bucket, p, p_object, p->vmp_offset,
8644 				    p_hash);
8645 			}
8646 			p = (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m));
8647 		}
8648 		lck_spin_unlock(bucket_lock);
8649 	}
8650 
8651 //	printf("BUCKET_CHECK: checked buckets\n");
8652 }
8653 #endif /* VM_PAGE_BUCKETS_CHECK */
8654 
8655 /*
8656  * 'vm_fault_enter' will place newly created pages (zero-fill and COW) onto the
8657  * local queues if they exist... its the only spot in the system where we add pages
8658  * to those queues...  once on those queues, those pages can only move to one of the
8659  * global page queues or the free queues... they NEVER move from local q to local q.
8660  * the 'local' state is stable when vm_page_queues_remove is called since we're behind
8661  * the global vm_page_queue_lock at this point...  we still need to take the local lock
8662  * in case this operation is being run on a different CPU then the local queue's identity,
8663  * but we don't have to worry about the page moving to a global queue or becoming wired
8664  * while we're grabbing the local lock since those operations would require the global
8665  * vm_page_queue_lock to be held, and we already own it.
8666  *
8667  * this is why its safe to utilze the wire_count field in the vm_page_t as the local_id...
8668  * 'wired' and local are ALWAYS mutually exclusive conditions.
8669  */
8670 
8671 void
vm_page_queues_remove(vm_page_t mem,boolean_t remove_from_specialq)8672 vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_specialq)
8673 {
8674 	boolean_t       was_pageable = TRUE;
8675 	vm_object_t     m_object;
8676 
8677 	m_object = VM_PAGE_OBJECT(mem);
8678 
8679 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8680 
8681 	if (mem->vmp_q_state == VM_PAGE_NOT_ON_Q) {
8682 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8683 		if (remove_from_specialq == TRUE) {
8684 			vm_page_remove_from_specialq(mem);
8685 		}
8686 		/*if (mem->vmp_on_specialq != VM_PAGE_SPECIAL_Q_EMPTY) {
8687 		 *       assert(mem->vmp_specialq.next != 0);
8688 		 *       assert(mem->vmp_specialq.prev != 0);
8689 		 *  } else {*/
8690 		if (mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY) {
8691 			assert(mem->vmp_specialq.next == 0);
8692 			assert(mem->vmp_specialq.prev == 0);
8693 		}
8694 		return;
8695 	}
8696 
8697 	if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) {
8698 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8699 		assert(mem->vmp_specialq.next == 0 &&
8700 		    mem->vmp_specialq.prev == 0 &&
8701 		    mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
8702 		return;
8703 	}
8704 	if (mem->vmp_q_state == VM_PAGE_IS_WIRED) {
8705 		/*
8706 		 * might put these guys on a list for debugging purposes
8707 		 * if we do, we'll need to remove this assert
8708 		 */
8709 		assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0);
8710 		assert(mem->vmp_specialq.next == 0 &&
8711 		    mem->vmp_specialq.prev == 0);
8712 		/*
8713 		 * Recall that vmp_on_specialq also means a request to put
8714 		 * it on the special Q. So we don't want to reset that bit
8715 		 * just because a wiring request came in. We might want to
8716 		 * put it on the special queue post-unwiring.
8717 		 *
8718 		 * &&
8719 		 * mem->vmp_on_specialq == VM_PAGE_SPECIAL_Q_EMPTY);
8720 		 */
8721 		return;
8722 	}
8723 
8724 	assert(m_object != compressor_object);
8725 	assert(m_object != kernel_object);
8726 	assert(!mem->vmp_fictitious);
8727 
8728 	switch (mem->vmp_q_state) {
8729 	case VM_PAGE_ON_ACTIVE_LOCAL_Q:
8730 	{
8731 		struct vpl      *lq;
8732 
8733 		lq = zpercpu_get_cpu(vm_page_local_q, mem->vmp_local_id);
8734 		VPL_LOCK(&lq->vpl_lock);
8735 		vm_page_queue_remove(&lq->vpl_queue, mem, vmp_pageq);
8736 		mem->vmp_local_id = 0;
8737 		lq->vpl_count--;
8738 		if (m_object->internal) {
8739 			lq->vpl_internal_count--;
8740 		} else {
8741 			lq->vpl_external_count--;
8742 		}
8743 		VPL_UNLOCK(&lq->vpl_lock);
8744 		was_pageable = FALSE;
8745 		break;
8746 	}
8747 	case VM_PAGE_ON_ACTIVE_Q:
8748 	{
8749 		vm_page_queue_remove(&vm_page_queue_active, mem, vmp_pageq);
8750 		vm_page_active_count--;
8751 		break;
8752 	}
8753 
8754 	case VM_PAGE_ON_INACTIVE_INTERNAL_Q:
8755 	{
8756 		assert(m_object->internal == TRUE);
8757 
8758 		vm_page_inactive_count--;
8759 		vm_page_queue_remove(&vm_page_queue_anonymous, mem, vmp_pageq);
8760 		vm_page_anonymous_count--;
8761 
8762 		vm_purgeable_q_advance_all();
8763 		vm_page_balance_inactive(3);
8764 		break;
8765 	}
8766 
8767 	case VM_PAGE_ON_INACTIVE_EXTERNAL_Q:
8768 	{
8769 		assert(m_object->internal == FALSE);
8770 
8771 		vm_page_inactive_count--;
8772 		vm_page_queue_remove(&vm_page_queue_inactive, mem, vmp_pageq);
8773 		vm_purgeable_q_advance_all();
8774 		vm_page_balance_inactive(3);
8775 		break;
8776 	}
8777 
8778 	case VM_PAGE_ON_INACTIVE_CLEANED_Q:
8779 	{
8780 		assert(m_object->internal == FALSE);
8781 
8782 		vm_page_inactive_count--;
8783 		vm_page_queue_remove(&vm_page_queue_cleaned, mem, vmp_pageq);
8784 		vm_page_cleaned_count--;
8785 		vm_page_balance_inactive(3);
8786 		break;
8787 	}
8788 
8789 	case VM_PAGE_ON_THROTTLED_Q:
8790 	{
8791 		assert(m_object->internal == TRUE);
8792 
8793 		vm_page_queue_remove(&vm_page_queue_throttled, mem, vmp_pageq);
8794 		vm_page_throttled_count--;
8795 		was_pageable = FALSE;
8796 		break;
8797 	}
8798 
8799 	case VM_PAGE_ON_SPECULATIVE_Q:
8800 	{
8801 		assert(m_object->internal == FALSE);
8802 
8803 		vm_page_remque(&mem->vmp_pageq);
8804 		vm_page_speculative_count--;
8805 		vm_page_balance_inactive(3);
8806 		break;
8807 	}
8808 
8809 #if CONFIG_SECLUDED_MEMORY
8810 	case VM_PAGE_ON_SECLUDED_Q:
8811 	{
8812 		vm_page_queue_remove(&vm_page_queue_secluded, mem, vmp_pageq);
8813 		vm_page_secluded_count--;
8814 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
8815 		if (m_object == VM_OBJECT_NULL) {
8816 			vm_page_secluded_count_free--;
8817 			was_pageable = FALSE;
8818 		} else {
8819 			assert(!m_object->internal);
8820 			vm_page_secluded_count_inuse--;
8821 			was_pageable = FALSE;
8822 //			was_pageable = TRUE;
8823 		}
8824 		break;
8825 	}
8826 #endif /* CONFIG_SECLUDED_MEMORY */
8827 
8828 	default:
8829 	{
8830 		/*
8831 		 *	if (mem->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)
8832 		 *              NOTE: vm_page_queues_remove does not deal with removing pages from the pageout queue...
8833 		 *              the caller is responsible for determing if the page is on that queue, and if so, must
8834 		 *              either first remove it (it needs both the page queues lock and the object lock to do
8835 		 *              this via vm_pageout_steal_laundry), or avoid the call to vm_page_queues_remove
8836 		 *
8837 		 *	we also don't expect to encounter VM_PAGE_ON_FREE_Q, VM_PAGE_ON_FREE_LOCAL_Q, VM_PAGE_ON_FREE_LOPAGE_Q
8838 		 *	or any of the undefined states
8839 		 */
8840 		panic("vm_page_queues_remove - bad page q_state (%p, %d)", mem, mem->vmp_q_state);
8841 		break;
8842 	}
8843 	}
8844 	VM_PAGE_ZERO_PAGEQ_ENTRY(mem);
8845 	mem->vmp_q_state = VM_PAGE_NOT_ON_Q;
8846 
8847 	if (remove_from_specialq == TRUE) {
8848 		vm_page_remove_from_specialq(mem);
8849 	}
8850 	if (was_pageable) {
8851 		if (m_object->internal) {
8852 			vm_page_pageable_internal_count--;
8853 		} else {
8854 			vm_page_pageable_external_count--;
8855 		}
8856 	}
8857 }
8858 
8859 void
vm_page_remove_internal(vm_page_t page)8860 vm_page_remove_internal(vm_page_t page)
8861 {
8862 	vm_object_t __object = VM_PAGE_OBJECT(page);
8863 	if (page == __object->memq_hint) {
8864 		vm_page_t       __new_hint;
8865 		vm_page_queue_entry_t   __qe;
8866 		__qe = (vm_page_queue_entry_t)vm_page_queue_next(&page->vmp_listq);
8867 		if (vm_page_queue_end(&__object->memq, __qe)) {
8868 			__qe = (vm_page_queue_entry_t)vm_page_queue_prev(&page->vmp_listq);
8869 			if (vm_page_queue_end(&__object->memq, __qe)) {
8870 				__qe = NULL;
8871 			}
8872 		}
8873 		__new_hint = (vm_page_t)((uintptr_t) __qe);
8874 		__object->memq_hint = __new_hint;
8875 	}
8876 	vm_page_queue_remove(&__object->memq, page, vmp_listq);
8877 #if CONFIG_SECLUDED_MEMORY
8878 	if (__object->eligible_for_secluded) {
8879 		vm_page_secluded.eligible_for_secluded--;
8880 	}
8881 #endif /* CONFIG_SECLUDED_MEMORY */
8882 }
8883 
8884 void
vm_page_enqueue_inactive(vm_page_t mem,boolean_t first)8885 vm_page_enqueue_inactive(vm_page_t mem, boolean_t first)
8886 {
8887 	vm_object_t     m_object;
8888 
8889 	m_object = VM_PAGE_OBJECT(mem);
8890 
8891 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8892 	assert(!mem->vmp_fictitious);
8893 	assert(!mem->vmp_laundry);
8894 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
8895 	vm_page_check_pageable_safe(mem);
8896 
8897 	if (m_object->internal) {
8898 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_INTERNAL_Q;
8899 
8900 		if (first == TRUE) {
8901 			vm_page_queue_enter_first(&vm_page_queue_anonymous, mem, vmp_pageq);
8902 		} else {
8903 			vm_page_queue_enter(&vm_page_queue_anonymous, mem, vmp_pageq);
8904 		}
8905 
8906 		vm_page_anonymous_count++;
8907 		vm_page_pageable_internal_count++;
8908 	} else {
8909 		mem->vmp_q_state = VM_PAGE_ON_INACTIVE_EXTERNAL_Q;
8910 
8911 		if (first == TRUE) {
8912 			vm_page_queue_enter_first(&vm_page_queue_inactive, mem, vmp_pageq);
8913 		} else {
8914 			vm_page_queue_enter(&vm_page_queue_inactive, mem, vmp_pageq);
8915 		}
8916 
8917 		vm_page_pageable_external_count++;
8918 	}
8919 	vm_page_inactive_count++;
8920 	token_new_pagecount++;
8921 
8922 	vm_page_add_to_specialq(mem, FALSE);
8923 }
8924 
8925 void
vm_page_enqueue_active(vm_page_t mem,boolean_t first)8926 vm_page_enqueue_active(vm_page_t mem, boolean_t first)
8927 {
8928 	vm_object_t     m_object;
8929 
8930 	m_object = VM_PAGE_OBJECT(mem);
8931 
8932 	LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED);
8933 	assert(!mem->vmp_fictitious);
8934 	assert(!mem->vmp_laundry);
8935 	assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q);
8936 	vm_page_check_pageable_safe(mem);
8937 
8938 	mem->vmp_q_state = VM_PAGE_ON_ACTIVE_Q;
8939 	if (first == TRUE) {
8940 		vm_page_queue_enter_first(&vm_page_queue_active, mem, vmp_pageq);
8941 	} else {
8942 		vm_page_queue_enter(&vm_page_queue_active, mem, vmp_pageq);
8943 	}
8944 	vm_page_active_count++;
8945 
8946 	if (m_object->internal) {
8947 		vm_page_pageable_internal_count++;
8948 	} else {
8949 		vm_page_pageable_external_count++;
8950 	}
8951 
8952 	vm_page_add_to_specialq(mem, FALSE);
8953 	vm_page_balance_inactive(3);
8954 }
8955 
8956 /*
8957  * Pages from special kernel objects shouldn't
8958  * be placed on pageable queues.
8959  */
8960 void
vm_page_check_pageable_safe(vm_page_t page)8961 vm_page_check_pageable_safe(vm_page_t page)
8962 {
8963 	vm_object_t     page_object;
8964 
8965 	page_object = VM_PAGE_OBJECT(page);
8966 
8967 	if (page_object == kernel_object) {
8968 		panic("vm_page_check_pageable_safe: trying to add page"
8969 		    "from kernel object (%p) to pageable queue", kernel_object);
8970 	}
8971 
8972 	if (page_object == compressor_object) {
8973 		panic("vm_page_check_pageable_safe: trying to add page"
8974 		    "from compressor object (%p) to pageable queue", compressor_object);
8975 	}
8976 }
8977 
8978 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
8979 * wired page diagnose
8980 * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
8981 
8982 #include <libkern/OSKextLibPrivate.h>
8983 
8984 #define KA_SIZE(namelen, subtotalscount)        \
8985 	(sizeof(struct vm_allocation_site) + (namelen) + 1 + ((subtotalscount) * sizeof(struct vm_allocation_total)))
8986 
8987 #define KA_NAME(alloc)  \
8988 	((char *)(&(alloc)->subtotals[(alloc->subtotalscount)]))
8989 
8990 #define KA_NAME_LEN(alloc)      \
8991     (VM_TAG_NAME_LEN_MAX & (alloc->flags >> VM_TAG_NAME_LEN_SHIFT))
8992 
8993 vm_tag_t
vm_tag_bt(void)8994 vm_tag_bt(void)
8995 {
8996 	uintptr_t* frameptr;
8997 	uintptr_t* frameptr_next;
8998 	uintptr_t retaddr;
8999 	uintptr_t kstackb, kstackt;
9000 	const vm_allocation_site_t * site;
9001 	thread_t cthread;
9002 	kern_allocation_name_t name;
9003 
9004 	cthread = current_thread();
9005 	if (__improbable(cthread == NULL)) {
9006 		return VM_KERN_MEMORY_OSFMK;
9007 	}
9008 
9009 	if ((name = thread_get_kernel_state(cthread)->allocation_name)) {
9010 		if (!name->tag) {
9011 			vm_tag_alloc(name);
9012 		}
9013 		return name->tag;
9014 	}
9015 
9016 	kstackb = cthread->kernel_stack;
9017 	kstackt = kstackb + kernel_stack_size;
9018 
9019 	/* Load stack frame pointer (EBP on x86) into frameptr */
9020 	frameptr = __builtin_frame_address(0);
9021 	site = NULL;
9022 	while (frameptr != NULL) {
9023 		/* Verify thread stack bounds */
9024 		if (((uintptr_t)(frameptr + 2) > kstackt) || ((uintptr_t)frameptr < kstackb)) {
9025 			break;
9026 		}
9027 
9028 		/* Next frame pointer is pointed to by the previous one */
9029 		frameptr_next = (uintptr_t*) *frameptr;
9030 
9031 		/* Pull return address from one spot above the frame pointer */
9032 		retaddr = *(frameptr + 1);
9033 
9034 #if defined(HAS_APPLE_PAC)
9035 		retaddr = (uintptr_t) ptrauth_strip((void *)retaddr, ptrauth_key_return_address);
9036 #endif
9037 
9038 		if (((retaddr < vm_kernel_builtinkmod_text_end) && (retaddr >= vm_kernel_builtinkmod_text))
9039 		    || (retaddr < vm_kernel_stext) || (retaddr > vm_kernel_top)) {
9040 			site = OSKextGetAllocationSiteForCaller(retaddr);
9041 			break;
9042 		}
9043 		frameptr = frameptr_next;
9044 	}
9045 
9046 	return site ? site->tag : VM_KERN_MEMORY_NONE;
9047 }
9048 
9049 static uint64_t free_tag_bits[VM_MAX_TAG_VALUE / 64];
9050 
9051 void
vm_tag_alloc_locked(vm_allocation_site_t * site,vm_allocation_site_t ** releasesiteP)9052 vm_tag_alloc_locked(vm_allocation_site_t * site, vm_allocation_site_t ** releasesiteP)
9053 {
9054 	vm_tag_t tag;
9055 	uint64_t avail;
9056 	uint32_t idx;
9057 	vm_allocation_site_t * prev;
9058 
9059 	if (site->tag) {
9060 		return;
9061 	}
9062 
9063 	idx = 0;
9064 	while (TRUE) {
9065 		avail = free_tag_bits[idx];
9066 		if (avail) {
9067 			tag = (vm_tag_t)__builtin_clzll(avail);
9068 			avail &= ~(1ULL << (63 - tag));
9069 			free_tag_bits[idx] = avail;
9070 			tag += (idx << 6);
9071 			break;
9072 		}
9073 		idx++;
9074 		if (idx >= ARRAY_COUNT(free_tag_bits)) {
9075 			for (idx = 0; idx < ARRAY_COUNT(vm_allocation_sites); idx++) {
9076 				prev = vm_allocation_sites[idx];
9077 				if (!prev) {
9078 					continue;
9079 				}
9080 				if (!KA_NAME_LEN(prev)) {
9081 					continue;
9082 				}
9083 				if (!prev->tag) {
9084 					continue;
9085 				}
9086 				if (prev->total) {
9087 					continue;
9088 				}
9089 				if (1 != prev->refcount) {
9090 					continue;
9091 				}
9092 
9093 				assert(idx == prev->tag);
9094 				tag = (vm_tag_t)idx;
9095 				prev->tag = VM_KERN_MEMORY_NONE;
9096 				*releasesiteP = prev;
9097 				break;
9098 			}
9099 			if (idx >= ARRAY_COUNT(vm_allocation_sites)) {
9100 				tag = VM_KERN_MEMORY_ANY;
9101 			}
9102 			break;
9103 		}
9104 	}
9105 	site->tag = tag;
9106 
9107 	OSAddAtomic16(1, &site->refcount);
9108 
9109 	if (VM_KERN_MEMORY_ANY != tag) {
9110 		vm_allocation_sites[tag] = site;
9111 	}
9112 
9113 	if (tag > vm_allocation_tag_highest) {
9114 		vm_allocation_tag_highest = tag;
9115 	}
9116 }
9117 
9118 static void
vm_tag_free_locked(vm_tag_t tag)9119 vm_tag_free_locked(vm_tag_t tag)
9120 {
9121 	uint64_t avail;
9122 	uint32_t idx;
9123 	uint64_t bit;
9124 
9125 	if (VM_KERN_MEMORY_ANY == tag) {
9126 		return;
9127 	}
9128 
9129 	idx = (tag >> 6);
9130 	avail = free_tag_bits[idx];
9131 	tag &= 63;
9132 	bit = (1ULL << (63 - tag));
9133 	assert(!(avail & bit));
9134 	free_tag_bits[idx] = (avail | bit);
9135 }
9136 
9137 static void
vm_tag_init(void)9138 vm_tag_init(void)
9139 {
9140 	vm_tag_t tag;
9141 	for (tag = VM_KERN_MEMORY_FIRST_DYNAMIC; tag < VM_KERN_MEMORY_ANY; tag++) {
9142 		vm_tag_free_locked(tag);
9143 	}
9144 
9145 	for (tag = VM_KERN_MEMORY_ANY + 1; tag < VM_MAX_TAG_VALUE; tag++) {
9146 		vm_tag_free_locked(tag);
9147 	}
9148 }
9149 
9150 vm_tag_t
vm_tag_alloc(vm_allocation_site_t * site)9151 vm_tag_alloc(vm_allocation_site_t * site)
9152 {
9153 	vm_allocation_site_t * releasesite;
9154 
9155 	if (!site->tag) {
9156 		releasesite = NULL;
9157 		lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9158 		vm_tag_alloc_locked(site, &releasesite);
9159 		lck_ticket_unlock(&vm_allocation_sites_lock);
9160 		if (releasesite) {
9161 			kern_allocation_name_release(releasesite);
9162 		}
9163 	}
9164 
9165 	return site->tag;
9166 }
9167 
9168 void
vm_tag_update_size(vm_tag_t tag,int64_t delta)9169 vm_tag_update_size(vm_tag_t tag, int64_t delta)
9170 {
9171 	vm_allocation_site_t * allocation;
9172 	uint64_t value;
9173 
9174 	assert(VM_KERN_MEMORY_NONE != tag);
9175 	assert(tag < VM_MAX_TAG_VALUE);
9176 
9177 	allocation = vm_allocation_sites[tag];
9178 	assert(allocation);
9179 
9180 	value = os_atomic_add(&allocation->total, delta, relaxed);
9181 	if (delta < 0) {
9182 		assertf(value + (uint64_t)-delta > value,
9183 		    "tag %d, site %p", tag, allocation);
9184 	}
9185 
9186 #if DEBUG || DEVELOPMENT
9187 	if (value > allocation->peak) {
9188 		os_atomic_max(&allocation->peak, value, relaxed);
9189 	}
9190 #endif /* DEBUG || DEVELOPMENT */
9191 
9192 	if (tag < VM_KERN_MEMORY_FIRST_DYNAMIC) {
9193 		return;
9194 	}
9195 
9196 	if (value == (uint64_t)delta && !allocation->tag) {
9197 		vm_tag_alloc(allocation);
9198 	}
9199 }
9200 
9201 uint64_t
vm_tag_get_size(vm_tag_t tag)9202 vm_tag_get_size(vm_tag_t tag)
9203 {
9204 	vm_allocation_site_t * allocation;
9205 
9206 	assert(VM_KERN_MEMORY_NONE != tag);
9207 	assert(tag < VM_MAX_TAG_VALUE);
9208 
9209 	allocation = vm_allocation_sites[tag];
9210 	return allocation ? os_atomic_load(&allocation->total, relaxed) : 0;
9211 }
9212 
9213 void
kern_allocation_update_size(kern_allocation_name_t allocation,int64_t delta)9214 kern_allocation_update_size(kern_allocation_name_t allocation, int64_t delta)
9215 {
9216 	uint64_t prior;
9217 
9218 	if (delta < 0) {
9219 		assertf(allocation->total >= ((uint64_t)-delta), "name %p", allocation);
9220 	}
9221 	prior = OSAddAtomic64(delta, &allocation->total);
9222 
9223 #if DEBUG || DEVELOPMENT
9224 
9225 	uint64_t new, peak;
9226 	new = prior + delta;
9227 	do{
9228 		peak = allocation->peak;
9229 		if (new <= peak) {
9230 			break;
9231 		}
9232 	}while (!OSCompareAndSwap64(peak, new, &allocation->peak));
9233 
9234 #endif /* DEBUG || DEVELOPMENT */
9235 
9236 	if (!prior && !allocation->tag) {
9237 		vm_tag_alloc(allocation);
9238 	}
9239 }
9240 
9241 #if VM_TAG_SIZECLASSES
9242 
9243 void
vm_allocation_zones_init(void)9244 vm_allocation_zones_init(void)
9245 {
9246 	vm_offset_t   addr;
9247 	vm_size_t     size;
9248 
9249 	const vm_tag_t early_tags[] = {
9250 		VM_KERN_MEMORY_DIAG,
9251 		VM_KERN_MEMORY_KALLOC,
9252 		VM_KERN_MEMORY_KALLOC_DATA,
9253 		VM_KERN_MEMORY_KALLOC_TYPE,
9254 		VM_KERN_MEMORY_LIBKERN,
9255 		VM_KERN_MEMORY_OSFMK,
9256 		VM_KERN_MEMORY_RECOUNT,
9257 	};
9258 
9259 	size = VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *)
9260 	    + ARRAY_COUNT(early_tags) * VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9261 
9262 	kmem_alloc(kernel_map, &addr, round_page(size),
9263 	    KMA_NOFAIL | KMA_KOBJECT | KMA_ZERO | KMA_PERMANENT,
9264 	    VM_KERN_MEMORY_DIAG);
9265 
9266 	vm_allocation_zone_totals = (vm_allocation_zone_total_t **) addr;
9267 	addr += VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t * *);
9268 
9269 	// prepopulate early tag ranges so allocations
9270 	// in vm_tag_update_zone_size() and early boot won't recurse
9271 	for (size_t i = 0; i < ARRAY_COUNT(early_tags); i++) {
9272 		vm_allocation_zone_totals[early_tags[i]] = (vm_allocation_zone_total_t *)addr;
9273 		addr += VM_TAG_SIZECLASSES * sizeof(vm_allocation_zone_total_t);
9274 	}
9275 }
9276 
9277 __attribute__((noinline))
9278 static vm_tag_t
vm_tag_zone_stats_alloc(vm_tag_t tag,zalloc_flags_t flags)9279 vm_tag_zone_stats_alloc(vm_tag_t tag, zalloc_flags_t flags)
9280 {
9281 	vm_allocation_zone_total_t *stats;
9282 	vm_size_t size = sizeof(*stats) * VM_TAG_SIZECLASSES;
9283 
9284 	flags = Z_VM_TAG(Z_ZERO | flags, VM_KERN_MEMORY_DIAG);
9285 	stats = kalloc_data(size, flags);
9286 	if (!stats) {
9287 		return VM_KERN_MEMORY_NONE;
9288 	}
9289 	if (!os_atomic_cmpxchg(&vm_allocation_zone_totals[tag], NULL, stats, release)) {
9290 		kfree_data(stats, size);
9291 	}
9292 	return tag;
9293 }
9294 
9295 vm_tag_t
vm_tag_will_update_zone(vm_tag_t tag,uint32_t zidx,uint32_t zflags)9296 vm_tag_will_update_zone(vm_tag_t tag, uint32_t zidx, uint32_t zflags)
9297 {
9298 	assert(VM_KERN_MEMORY_NONE != tag);
9299 	assert(tag < VM_MAX_TAG_VALUE);
9300 
9301 	if (zidx >= VM_TAG_SIZECLASSES) {
9302 		return VM_KERN_MEMORY_NONE;
9303 	}
9304 
9305 	if (__probable(vm_allocation_zone_totals[tag])) {
9306 		return tag;
9307 	}
9308 	return vm_tag_zone_stats_alloc(tag, zflags);
9309 }
9310 
9311 void
vm_tag_update_zone_size(vm_tag_t tag,uint32_t zidx,long delta)9312 vm_tag_update_zone_size(vm_tag_t tag, uint32_t zidx, long delta)
9313 {
9314 	vm_allocation_zone_total_t *stats;
9315 	vm_size_t value;
9316 
9317 	assert(VM_KERN_MEMORY_NONE != tag);
9318 	assert(tag < VM_MAX_TAG_VALUE);
9319 
9320 	if (zidx >= VM_TAG_SIZECLASSES) {
9321 		return;
9322 	}
9323 
9324 	stats = vm_allocation_zone_totals[tag];
9325 	assert(stats);
9326 	stats += zidx;
9327 
9328 	value = os_atomic_add(&stats->vazt_total, delta, relaxed);
9329 	if (delta < 0) {
9330 		assertf((long)value >= 0, "zidx %d, tag %d, %p", zidx, tag, stats);
9331 		return;
9332 	} else if (os_atomic_load(&stats->vazt_peak, relaxed) < value) {
9333 		os_atomic_max(&stats->vazt_peak, value, relaxed);
9334 	}
9335 }
9336 
9337 #endif /* VM_TAG_SIZECLASSES */
9338 
9339 void
kern_allocation_update_subtotal(kern_allocation_name_t allocation,uint32_t subtag,int64_t delta)9340 kern_allocation_update_subtotal(kern_allocation_name_t allocation, uint32_t subtag, int64_t delta)
9341 {
9342 	kern_allocation_name_t other;
9343 	struct vm_allocation_total * total;
9344 	uint32_t subidx;
9345 
9346 	subidx = 0;
9347 	assert(VM_KERN_MEMORY_NONE != subtag);
9348 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9349 	for (; subidx < allocation->subtotalscount; subidx++) {
9350 		if (VM_KERN_MEMORY_NONE == allocation->subtotals[subidx].tag) {
9351 			allocation->subtotals[subidx].tag = (vm_tag_t)subtag;
9352 			break;
9353 		}
9354 		if (subtag == allocation->subtotals[subidx].tag) {
9355 			break;
9356 		}
9357 	}
9358 	lck_ticket_unlock(&vm_allocation_sites_lock);
9359 	assert(subidx < allocation->subtotalscount);
9360 	if (subidx >= allocation->subtotalscount) {
9361 		return;
9362 	}
9363 
9364 	total = &allocation->subtotals[subidx];
9365 	other = vm_allocation_sites[subtag];
9366 	assert(other);
9367 
9368 	if (delta < 0) {
9369 		assertf(total->total >= ((uint64_t)-delta), "name %p", allocation);
9370 		assertf(other->mapped >= ((uint64_t)-delta), "other %p", other);
9371 	}
9372 	OSAddAtomic64(delta, &other->mapped);
9373 	OSAddAtomic64(delta, &total->total);
9374 }
9375 
9376 const char *
kern_allocation_get_name(kern_allocation_name_t allocation)9377 kern_allocation_get_name(kern_allocation_name_t allocation)
9378 {
9379 	return KA_NAME(allocation);
9380 }
9381 
9382 kern_allocation_name_t
kern_allocation_name_allocate(const char * name,uint16_t subtotalscount)9383 kern_allocation_name_allocate(const char * name, uint16_t subtotalscount)
9384 {
9385 	kern_allocation_name_t allocation;
9386 	uint16_t namelen;
9387 
9388 	namelen = (uint16_t)strnlen(name, MACH_MEMORY_INFO_NAME_MAX_LEN - 1);
9389 
9390 	allocation = kalloc_data(KA_SIZE(namelen, subtotalscount), Z_WAITOK | Z_ZERO);
9391 	allocation->refcount       = 1;
9392 	allocation->subtotalscount = subtotalscount;
9393 	allocation->flags          = (uint16_t)(namelen << VM_TAG_NAME_LEN_SHIFT);
9394 	strlcpy(KA_NAME(allocation), name, namelen + 1);
9395 
9396 	vm_tag_alloc(allocation);
9397 	return allocation;
9398 }
9399 
9400 void
kern_allocation_name_release(kern_allocation_name_t allocation)9401 kern_allocation_name_release(kern_allocation_name_t allocation)
9402 {
9403 	assert(allocation->refcount > 0);
9404 	if (1 == OSAddAtomic16(-1, &allocation->refcount)) {
9405 		kfree_data(allocation,
9406 		    KA_SIZE(KA_NAME_LEN(allocation), allocation->subtotalscount));
9407 	}
9408 }
9409 
9410 vm_tag_t
kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation)9411 kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation)
9412 {
9413 	return vm_tag_alloc(allocation);
9414 }
9415 
9416 #if !VM_TAG_ACTIVE_UPDATE
9417 static void
vm_page_count_object(mach_memory_info_t * info,unsigned int __unused num_info,vm_object_t object)9418 vm_page_count_object(mach_memory_info_t * info, unsigned int __unused num_info, vm_object_t object)
9419 {
9420 	if (!object->wired_page_count) {
9421 		return;
9422 	}
9423 	if (object != kernel_object) {
9424 		assert(object->wire_tag < num_info);
9425 		info[object->wire_tag].size += ptoa_64(object->wired_page_count);
9426 	}
9427 }
9428 
9429 typedef void (*vm_page_iterate_proc)(mach_memory_info_t * info,
9430     unsigned int num_info, vm_object_t object);
9431 
9432 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)9433 vm_page_iterate_purgeable_objects(mach_memory_info_t * info, unsigned int num_info,
9434     vm_page_iterate_proc proc, purgeable_q_t queue,
9435     int group)
9436 {
9437 	vm_object_t object;
9438 
9439 	for (object = (vm_object_t) queue_first(&queue->objq[group]);
9440 	    !queue_end(&queue->objq[group], (queue_entry_t) object);
9441 	    object = (vm_object_t) queue_next(&object->objq)) {
9442 		proc(info, num_info, object);
9443 	}
9444 }
9445 
9446 static void
vm_page_iterate_objects(mach_memory_info_t * info,unsigned int num_info,vm_page_iterate_proc proc)9447 vm_page_iterate_objects(mach_memory_info_t * info, unsigned int num_info,
9448     vm_page_iterate_proc proc)
9449 {
9450 	vm_object_t     object;
9451 
9452 	lck_spin_lock_grp(&vm_objects_wired_lock, &vm_page_lck_grp_bucket);
9453 	queue_iterate(&vm_objects_wired,
9454 	    object,
9455 	    vm_object_t,
9456 	    wired_objq)
9457 	{
9458 		proc(info, num_info, object);
9459 	}
9460 	lck_spin_unlock(&vm_objects_wired_lock);
9461 }
9462 #endif /* ! VM_TAG_ACTIVE_UPDATE */
9463 
9464 static uint64_t
process_account(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes,boolean_t iterated)9465 process_account(mach_memory_info_t * info, unsigned int num_info,
9466     uint64_t zones_collectable_bytes, boolean_t iterated)
9467 {
9468 	size_t                 namelen;
9469 	unsigned int           idx, count, nextinfo;
9470 	vm_allocation_site_t * site;
9471 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9472 
9473 	for (idx = 0; idx <= vm_allocation_tag_highest; idx++) {
9474 		site = vm_allocation_sites[idx];
9475 		if (!site) {
9476 			continue;
9477 		}
9478 		info[idx].mapped = site->mapped;
9479 		info[idx].tag    = site->tag;
9480 		if (!iterated) {
9481 			info[idx].size = site->total;
9482 #if DEBUG || DEVELOPMENT
9483 			info[idx].peak = site->peak;
9484 #endif /* DEBUG || DEVELOPMENT */
9485 		} else {
9486 			if (!site->subtotalscount && (site->total != info[idx].size)) {
9487 				printf("tag mismatch[%d] 0x%qx, iter 0x%qx\n", idx, site->total, info[idx].size);
9488 				info[idx].size = site->total;
9489 			}
9490 		}
9491 		info[idx].flags |= VM_KERN_SITE_WIRED;
9492 		if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC) {
9493 			info[idx].site   = idx;
9494 			info[idx].flags |= VM_KERN_SITE_TAG;
9495 			if (VM_KERN_MEMORY_ZONE == idx) {
9496 				info[idx].flags |= VM_KERN_SITE_HIDE;
9497 				info[idx].flags &= ~VM_KERN_SITE_WIRED;
9498 				info[idx].collectable_bytes = zones_collectable_bytes;
9499 			}
9500 		} else if ((namelen = (VM_TAG_NAME_LEN_MAX & (site->flags >> VM_TAG_NAME_LEN_SHIFT)))) {
9501 			info[idx].site   = 0;
9502 			info[idx].flags |= VM_KERN_SITE_NAMED;
9503 			if (namelen > sizeof(info[idx].name)) {
9504 				namelen = sizeof(info[idx].name);
9505 			}
9506 			strncpy(&info[idx].name[0], KA_NAME(site), namelen);
9507 		} else if (VM_TAG_KMOD & site->flags) {
9508 			info[idx].site   = OSKextGetKmodIDForSite(site, NULL, 0);
9509 			info[idx].flags |= VM_KERN_SITE_KMOD;
9510 		} else {
9511 			info[idx].site   = VM_KERNEL_UNSLIDE(site);
9512 			info[idx].flags |= VM_KERN_SITE_KERNEL;
9513 		}
9514 	}
9515 
9516 	nextinfo = (vm_allocation_tag_highest + 1);
9517 	count    = nextinfo;
9518 	if (count >= num_info) {
9519 		count = num_info;
9520 	}
9521 
9522 	for (idx = 0; idx < count; idx++) {
9523 		site = vm_allocation_sites[idx];
9524 		if (!site) {
9525 			continue;
9526 		}
9527 #if VM_TAG_SIZECLASSES
9528 		vm_allocation_zone_total_t * zone;
9529 		unsigned int                 zidx;
9530 
9531 		if (vm_allocation_zone_totals
9532 		    && (zone = vm_allocation_zone_totals[idx])
9533 		    && (nextinfo < num_info)) {
9534 			for (zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
9535 				if (!zone[zidx].vazt_peak) {
9536 					continue;
9537 				}
9538 				info[nextinfo]        = info[idx];
9539 				info[nextinfo].zone   = (uint16_t)zone_index_from_tag_index(zidx);
9540 				info[nextinfo].flags  &= ~VM_KERN_SITE_WIRED;
9541 				info[nextinfo].flags  |= VM_KERN_SITE_ZONE;
9542 				info[nextinfo].flags  |= VM_KERN_SITE_KALLOC;
9543 				info[nextinfo].size   = zone[zidx].vazt_total;
9544 				info[nextinfo].peak   = zone[zidx].vazt_peak;
9545 				info[nextinfo].mapped = 0;
9546 				nextinfo++;
9547 			}
9548 		}
9549 #endif /* VM_TAG_SIZECLASSES */
9550 		if (site->subtotalscount) {
9551 			uint64_t mapped, mapcost, take;
9552 			uint32_t sub;
9553 			vm_tag_t alloctag;
9554 
9555 			info[idx].size = site->total;
9556 			mapped = info[idx].size;
9557 			info[idx].mapped = mapped;
9558 			mapcost = 0;
9559 			for (sub = 0; sub < site->subtotalscount; sub++) {
9560 				alloctag = site->subtotals[sub].tag;
9561 				assert(alloctag < num_info);
9562 				if (info[alloctag].name[0]) {
9563 					continue;
9564 				}
9565 				take = site->subtotals[sub].total;
9566 				if (take > info[alloctag].size) {
9567 					take = info[alloctag].size;
9568 				}
9569 				if (take > mapped) {
9570 					take = mapped;
9571 				}
9572 				info[alloctag].mapped  -= take;
9573 				info[alloctag].size    -= take;
9574 				mapped                 -= take;
9575 				mapcost                += take;
9576 			}
9577 			info[idx].size = mapcost;
9578 		}
9579 	}
9580 	lck_ticket_unlock(&vm_allocation_sites_lock);
9581 
9582 	return 0;
9583 }
9584 
9585 uint32_t
vm_page_diagnose_estimate(void)9586 vm_page_diagnose_estimate(void)
9587 {
9588 	vm_allocation_site_t * site;
9589 	uint32_t               count = zone_view_count;
9590 	uint32_t               idx;
9591 
9592 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9593 	for (idx = 0; idx < VM_MAX_TAG_VALUE; idx++) {
9594 		site = vm_allocation_sites[idx];
9595 		if (!site) {
9596 			continue;
9597 		}
9598 		count++;
9599 #if VM_TAG_SIZECLASSES
9600 		if (vm_allocation_zone_totals) {
9601 			vm_allocation_zone_total_t * zone;
9602 			zone = vm_allocation_zone_totals[idx];
9603 			if (!zone) {
9604 				continue;
9605 			}
9606 			for (uint32_t zidx = 0; zidx < VM_TAG_SIZECLASSES; zidx++) {
9607 				count += (zone[zidx].vazt_peak != 0);
9608 			}
9609 		}
9610 #endif
9611 	}
9612 	lck_ticket_unlock(&vm_allocation_sites_lock);
9613 
9614 	/* some slop for new tags created */
9615 	count += 8;
9616 	count += VM_KERN_COUNTER_COUNT;
9617 
9618 	return count;
9619 }
9620 
9621 static void
vm_page_diagnose_zone_stats(mach_memory_info_t * info,zone_stats_t zstats,bool percpu)9622 vm_page_diagnose_zone_stats(mach_memory_info_t *info, zone_stats_t zstats,
9623     bool percpu)
9624 {
9625 	zpercpu_foreach(zs, zstats) {
9626 		info->size += zs->zs_mem_allocated - zs->zs_mem_freed;
9627 	}
9628 	if (percpu) {
9629 		info->size *= zpercpu_count();
9630 	}
9631 	info->flags |= VM_KERN_SITE_NAMED | VM_KERN_SITE_ZONE_VIEW;
9632 }
9633 
9634 static void
vm_page_diagnose_zone(mach_memory_info_t * info,zone_t z)9635 vm_page_diagnose_zone(mach_memory_info_t *info, zone_t z)
9636 {
9637 	vm_page_diagnose_zone_stats(info, z->z_stats, z->z_percpu);
9638 	snprintf(info->name, sizeof(info->name),
9639 	    "%s%s[raw]", zone_heap_name(z), z->z_name);
9640 }
9641 
9642 static int
vm_page_diagnose_heap(mach_memory_info_t * info,kalloc_heap_t kheap)9643 vm_page_diagnose_heap(mach_memory_info_t *info, kalloc_heap_t kheap)
9644 {
9645 	struct kalloc_heap *kh = kheap->kh_views;
9646 	int i = 0;
9647 
9648 	for (; i < KHEAP_NUM_ZONES; i++) {
9649 		vm_page_diagnose_zone(info + i, zone_by_id(kheap->kh_zstart + i));
9650 	}
9651 
9652 	while (kh) {
9653 		vm_page_diagnose_zone_stats(info + i, kh->kh_stats, false);
9654 		snprintf(info[i].name, sizeof(info[i].name),
9655 		    "%skalloc[%s]", kheap->kh_name, kh->kh_name);
9656 		kh = kh->kh_views;
9657 		i++;
9658 	}
9659 
9660 	return i;
9661 }
9662 
9663 static int
vm_page_diagnose_kt_heaps(mach_memory_info_t * info)9664 vm_page_diagnose_kt_heaps(mach_memory_info_t *info)
9665 {
9666 	uint32_t idx = 0;
9667 	vm_page_diagnose_zone_stats(info + idx, KHEAP_KT_VAR->kh_stats, false);
9668 	snprintf(info[idx].name, sizeof(info[idx].name),
9669 	    "%s[raw]", KHEAP_KT_VAR->kh_name);
9670 	idx++;
9671 
9672 	for (uint32_t i = 0; i < KT_VAR_MAX_HEAPS; i++) {
9673 		struct kheap_info heap = kalloc_type_heap_array[i];
9674 
9675 		for (kalloc_type_var_view_t ktv = heap.kt_views; ktv;
9676 		    ktv = (kalloc_type_var_view_t) ktv->kt_next) {
9677 			if (ktv->kt_stats && ktv->kt_stats != KHEAP_KT_VAR->kh_stats) {
9678 				vm_page_diagnose_zone_stats(info + idx, ktv->kt_stats, false);
9679 				snprintf(info[idx].name, sizeof(info[idx].name),
9680 				    "%s[%s]", KHEAP_KT_VAR->kh_name, ktv->kt_name);
9681 				idx++;
9682 			}
9683 		}
9684 	}
9685 
9686 	return idx;
9687 }
9688 
9689 kern_return_t
vm_page_diagnose(mach_memory_info_t * info,unsigned int num_info,uint64_t zones_collectable_bytes)9690 vm_page_diagnose(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes)
9691 {
9692 	uint64_t                 wired_size;
9693 	uint64_t                 wired_managed_size;
9694 	uint64_t                 wired_reserved_size;
9695 	boolean_t                iterate;
9696 	mach_memory_info_t     * counts;
9697 	uint32_t                 i;
9698 
9699 	bzero(info, num_info * sizeof(mach_memory_info_t));
9700 
9701 	if (!vm_page_wire_count_initial) {
9702 		return KERN_ABORTED;
9703 	}
9704 
9705 #if !XNU_TARGET_OS_OSX
9706 	wired_size          = ptoa_64(vm_page_wire_count);
9707 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count);
9708 #else /* !XNU_TARGET_OS_OSX */
9709 	wired_size          = ptoa_64(vm_page_wire_count + vm_lopage_free_count + vm_page_throttled_count);
9710 	wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count + vm_page_throttled_count);
9711 #endif /* !XNU_TARGET_OS_OSX */
9712 	wired_managed_size  = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial);
9713 
9714 	wired_size += booter_size;
9715 
9716 	assert(num_info >= VM_KERN_COUNTER_COUNT);
9717 	num_info -= VM_KERN_COUNTER_COUNT;
9718 	counts = &info[num_info];
9719 
9720 #define SET_COUNT(xcount, xsize, xflags)                        \
9721     counts[xcount].tag   = VM_MAX_TAG_VALUE + xcount;   \
9722     counts[xcount].site  = (xcount);                            \
9723     counts[xcount].size  = (xsize);                                 \
9724     counts[xcount].mapped  = (xsize);                           \
9725     counts[xcount].flags = VM_KERN_SITE_COUNTER | xflags;
9726 
9727 	SET_COUNT(VM_KERN_COUNT_MANAGED, ptoa_64(vm_page_pages), 0);
9728 	SET_COUNT(VM_KERN_COUNT_WIRED, wired_size, 0);
9729 	SET_COUNT(VM_KERN_COUNT_WIRED_MANAGED, wired_managed_size, 0);
9730 	SET_COUNT(VM_KERN_COUNT_RESERVED, wired_reserved_size, VM_KERN_SITE_WIRED);
9731 	SET_COUNT(VM_KERN_COUNT_STOLEN, ptoa_64(vm_page_stolen_count), VM_KERN_SITE_WIRED);
9732 	SET_COUNT(VM_KERN_COUNT_LOPAGE, ptoa_64(vm_lopage_free_count), VM_KERN_SITE_WIRED);
9733 	SET_COUNT(VM_KERN_COUNT_WIRED_BOOT, ptoa_64(vm_page_wire_count_on_boot), 0);
9734 	SET_COUNT(VM_KERN_COUNT_BOOT_STOLEN, booter_size, VM_KERN_SITE_WIRED);
9735 	SET_COUNT(VM_KERN_COUNT_WIRED_STATIC_KERNELCACHE, ptoa_64(vm_page_kernelcache_count), 0);
9736 
9737 #define SET_MAP(xcount, xsize, xfree, xlargest) \
9738     counts[xcount].site    = (xcount);                  \
9739     counts[xcount].size    = (xsize);                   \
9740     counts[xcount].mapped  = (xsize);                   \
9741     counts[xcount].free    = (xfree);                   \
9742     counts[xcount].largest = (xlargest);                \
9743     counts[xcount].flags   = VM_KERN_SITE_COUNTER;
9744 
9745 	vm_map_size_t map_size, map_free, map_largest;
9746 
9747 	vm_map_sizes(kernel_map, &map_size, &map_free, &map_largest);
9748 	SET_MAP(VM_KERN_COUNT_MAP_KERNEL, map_size, map_free, map_largest);
9749 
9750 	zone_map_sizes(&map_size, &map_free, &map_largest);
9751 	SET_MAP(VM_KERN_COUNT_MAP_ZONE, map_size, map_free, map_largest);
9752 
9753 	assert(num_info >= zone_view_count);
9754 	num_info -= zone_view_count;
9755 	counts = &info[num_info];
9756 	i = 0;
9757 
9758 	i += vm_page_diagnose_heap(counts + i, KHEAP_DEFAULT);
9759 	if (KHEAP_DATA_BUFFERS->kh_heap_id == KHEAP_ID_DATA_BUFFERS) {
9760 		i += vm_page_diagnose_heap(counts + i, KHEAP_DATA_BUFFERS);
9761 	}
9762 	if (KHEAP_KT_VAR->kh_heap_id == KHEAP_ID_KT_VAR) {
9763 		i += vm_page_diagnose_kt_heaps(counts + i);
9764 	}
9765 	assert(i <= zone_view_count);
9766 
9767 	zone_index_foreach(zidx) {
9768 		zone_t z = &zone_array[zidx];
9769 		zone_security_flags_t zsflags = zone_security_array[zidx];
9770 		zone_view_t zv = z->z_views;
9771 
9772 		if (zv == NULL) {
9773 			continue;
9774 		}
9775 
9776 		zone_stats_t zv_stats_head = z->z_stats;
9777 		bool has_raw_view = false;
9778 
9779 		for (; zv; zv = zv->zv_next) {
9780 			/*
9781 			 * kalloc_types that allocate from the same zone are linked
9782 			 * as views. Only print the ones that have their own stats.
9783 			 */
9784 			if (zv->zv_stats == zv_stats_head) {
9785 				continue;
9786 			}
9787 			has_raw_view = true;
9788 			vm_page_diagnose_zone_stats(counts + i, zv->zv_stats,
9789 			    z->z_percpu);
9790 			snprintf(counts[i].name, sizeof(counts[i].name), "%s%s[%s]",
9791 			    zone_heap_name(z), z->z_name, zv->zv_name);
9792 			i++;
9793 			assert(i <= zone_view_count);
9794 		}
9795 
9796 		/*
9797 		 * Print raw views for non kalloc or kalloc_type zones
9798 		 */
9799 		bool kalloc_type = zsflags.z_kalloc_type;
9800 		if ((zsflags.z_kheap_id == KHEAP_ID_NONE && !kalloc_type) ||
9801 		    (kalloc_type && has_raw_view)) {
9802 			vm_page_diagnose_zone(counts + i, z);
9803 			i++;
9804 			assert(i <= zone_view_count);
9805 		}
9806 	}
9807 
9808 	iterate = !VM_TAG_ACTIVE_UPDATE;
9809 	if (iterate) {
9810 		enum                       { kMaxKernelDepth = 1 };
9811 		vm_map_t                     maps[kMaxKernelDepth];
9812 		vm_map_entry_t               entries[kMaxKernelDepth];
9813 		vm_map_t                     map;
9814 		vm_map_entry_t               entry;
9815 		vm_object_offset_t           offset;
9816 		vm_page_t                    page;
9817 		int                          stackIdx, count;
9818 
9819 #if !VM_TAG_ACTIVE_UPDATE
9820 		vm_page_iterate_objects(info, num_info, &vm_page_count_object);
9821 #endif /* ! VM_TAG_ACTIVE_UPDATE */
9822 
9823 		map = kernel_map;
9824 		stackIdx = 0;
9825 		while (map) {
9826 			vm_map_lock(map);
9827 			for (entry = map->hdr.links.next; map; entry = entry->links.next) {
9828 				if (entry->is_sub_map) {
9829 					assert(stackIdx < kMaxKernelDepth);
9830 					maps[stackIdx] = map;
9831 					entries[stackIdx] = entry;
9832 					stackIdx++;
9833 					map = VME_SUBMAP(entry);
9834 					entry = NULL;
9835 					break;
9836 				}
9837 				if (VME_OBJECT(entry) == kernel_object) {
9838 					count = 0;
9839 					vm_object_lock(VME_OBJECT(entry));
9840 					for (offset = entry->links.start; offset < entry->links.end; offset += page_size) {
9841 						page = vm_page_lookup(VME_OBJECT(entry), offset);
9842 						if (page && VM_PAGE_WIRED(page)) {
9843 							count++;
9844 						}
9845 					}
9846 					vm_object_unlock(VME_OBJECT(entry));
9847 
9848 					if (count) {
9849 						assert(VME_ALIAS(entry) != VM_KERN_MEMORY_NONE);
9850 						assert(VME_ALIAS(entry) < num_info);
9851 						info[VME_ALIAS(entry)].size += ptoa_64(count);
9852 					}
9853 				}
9854 				while (map && (entry == vm_map_last_entry(map))) {
9855 					vm_map_unlock(map);
9856 					if (!stackIdx) {
9857 						map = NULL;
9858 					} else {
9859 						--stackIdx;
9860 						map = maps[stackIdx];
9861 						entry = entries[stackIdx];
9862 					}
9863 				}
9864 			}
9865 		}
9866 	}
9867 
9868 	process_account(info, num_info, zones_collectable_bytes, iterate);
9869 
9870 	return KERN_SUCCESS;
9871 }
9872 
9873 #if DEBUG || DEVELOPMENT
9874 
9875 kern_return_t
vm_kern_allocation_info(uintptr_t addr,vm_size_t * size,vm_tag_t * tag,vm_size_t * zone_size)9876 vm_kern_allocation_info(uintptr_t addr, vm_size_t * size, vm_tag_t * tag, vm_size_t * zone_size)
9877 {
9878 	kern_return_t  ret;
9879 	vm_size_t      zsize;
9880 	vm_map_t       map;
9881 	vm_map_entry_t entry;
9882 
9883 	zsize = zone_element_info((void *) addr, tag);
9884 	if (zsize) {
9885 		*zone_size = *size = zsize;
9886 		return KERN_SUCCESS;
9887 	}
9888 
9889 	*zone_size = 0;
9890 	ret = KERN_INVALID_ADDRESS;
9891 	for (map = kernel_map; map;) {
9892 		vm_map_lock(map);
9893 		if (!vm_map_lookup_entry_allow_pgz(map, addr, &entry)) {
9894 			break;
9895 		}
9896 		if (entry->is_sub_map) {
9897 			if (map != kernel_map) {
9898 				break;
9899 			}
9900 			map = VME_SUBMAP(entry);
9901 			continue;
9902 		}
9903 		if (entry->vme_start != addr) {
9904 			break;
9905 		}
9906 		*tag = (vm_tag_t)VME_ALIAS(entry);
9907 		*size = (entry->vme_end - addr);
9908 		ret = KERN_SUCCESS;
9909 		break;
9910 	}
9911 	if (map != kernel_map) {
9912 		vm_map_unlock(map);
9913 	}
9914 	vm_map_unlock(kernel_map);
9915 
9916 	return ret;
9917 }
9918 
9919 #endif /* DEBUG || DEVELOPMENT */
9920 
9921 uint32_t
vm_tag_get_kext(vm_tag_t tag,char * name,vm_size_t namelen)9922 vm_tag_get_kext(vm_tag_t tag, char * name, vm_size_t namelen)
9923 {
9924 	vm_allocation_site_t * site;
9925 	uint32_t               kmodId;
9926 
9927 	kmodId = 0;
9928 	lck_ticket_lock(&vm_allocation_sites_lock, LCK_GRP_NULL);
9929 	if ((site = vm_allocation_sites[tag])) {
9930 		if (VM_TAG_KMOD & site->flags) {
9931 			kmodId = OSKextGetKmodIDForSite(site, name, namelen);
9932 		}
9933 	}
9934 	lck_ticket_unlock(&vm_allocation_sites_lock);
9935 
9936 	return kmodId;
9937 }
9938 
9939 
9940 #if CONFIG_SECLUDED_MEMORY
9941 /*
9942  * Note that there's no locking around other accesses to vm_page_secluded_target.
9943  * That should be OK, since these are the only place where it can be changed after
9944  * initialization. Other users (like vm_pageout) may see the wrong value briefly,
9945  * but will eventually get the correct value. This brief mismatch is OK as pageout
9946  * and page freeing will auto-adjust the vm_page_secluded_count to match the target
9947  * over time.
9948  */
9949 unsigned int vm_page_secluded_suppress_cnt = 0;
9950 unsigned int vm_page_secluded_save_target;
9951 
9952 LCK_GRP_DECLARE(secluded_suppress_slock_grp, "secluded_suppress_slock");
9953 LCK_SPIN_DECLARE(secluded_suppress_slock, &secluded_suppress_slock_grp);
9954 
9955 void
start_secluded_suppression(task_t task)9956 start_secluded_suppression(task_t task)
9957 {
9958 	if (task->task_suppressed_secluded) {
9959 		return;
9960 	}
9961 	lck_spin_lock(&secluded_suppress_slock);
9962 	if (!task->task_suppressed_secluded && vm_page_secluded_suppress_cnt++ == 0) {
9963 		task->task_suppressed_secluded = TRUE;
9964 		vm_page_secluded_save_target = vm_page_secluded_target;
9965 		vm_page_secluded_target = 0;
9966 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
9967 	}
9968 	lck_spin_unlock(&secluded_suppress_slock);
9969 }
9970 
9971 void
stop_secluded_suppression(task_t task)9972 stop_secluded_suppression(task_t task)
9973 {
9974 	lck_spin_lock(&secluded_suppress_slock);
9975 	if (task->task_suppressed_secluded && --vm_page_secluded_suppress_cnt == 0) {
9976 		task->task_suppressed_secluded = FALSE;
9977 		vm_page_secluded_target = vm_page_secluded_save_target;
9978 		VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE();
9979 	}
9980 	lck_spin_unlock(&secluded_suppress_slock);
9981 }
9982 
9983 #endif /* CONFIG_SECLUDED_MEMORY */
9984 
9985 /*
9986  * Move the list of retired pages on the vm_page_queue_retired to
9987  * their final resting place on retired_pages_object.
9988  */
9989 void
vm_retire_boot_pages(void)9990 vm_retire_boot_pages(void)
9991 {
9992 }
9993 
9994 /*
9995  * This holds the reported physical address if an ECC error leads to a panic.
9996  * SMC will store it in PMU SRAM under the 'sECC' key.
9997  */
9998 uint64_t ecc_panic_physical_address = 0;
9999 
10000