1 | /* |
2 | * Copyright (c) 2000-2009 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 Software.Distribution@CS.CMU.EDU |
49 | * School of Computer Science |
50 | * Carnegie Mellon University |
51 | * Pittsburgh PA 15213-3890 |
52 | * |
53 | * any improvements or extensions that they make and grant Carnegie Mellon |
54 | * the rights to redistribute these changes. |
55 | */ |
56 | /* |
57 | */ |
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/counters.h> |
74 | #include <kern/sched_prim.h> |
75 | #include <kern/policy_internal.h> |
76 | #include <kern/task.h> |
77 | #include <kern/thread.h> |
78 | #include <kern/kalloc.h> |
79 | #include <kern/zalloc.h> |
80 | #include <kern/xpr.h> |
81 | #include <kern/ledger.h> |
82 | #include <vm/pmap.h> |
83 | #include <vm/vm_init.h> |
84 | #include <vm/vm_map.h> |
85 | #include <vm/vm_page.h> |
86 | #include <vm/vm_pageout.h> |
87 | #include <vm/vm_kern.h> /* kernel_memory_allocate() */ |
88 | #include <kern/misc_protos.h> |
89 | #include <zone_debug.h> |
90 | #include <mach_debug/zone_info.h> |
91 | #include <vm/cpm.h> |
92 | #include <pexpert/pexpert.h> |
93 | #include <san/kasan.h> |
94 | |
95 | #include <vm/vm_protos.h> |
96 | #include <vm/memory_object.h> |
97 | #include <vm/vm_purgeable_internal.h> |
98 | #include <vm/vm_compressor.h> |
99 | |
100 | #if CONFIG_PHANTOM_CACHE |
101 | #include <vm/vm_phantom_cache.h> |
102 | #endif |
103 | |
104 | #include <IOKit/IOHibernatePrivate.h> |
105 | |
106 | #include <sys/kdebug.h> |
107 | |
108 | |
109 | |
110 | char vm_page_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; |
111 | char vm_page_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; |
112 | char vm_page_non_speculative_pageable_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; |
113 | char vm_page_active_or_inactive_states[VM_PAGE_Q_STATE_ARRAY_SIZE]; |
114 | |
115 | #if CONFIG_SECLUDED_MEMORY |
116 | struct vm_page_secluded_data vm_page_secluded; |
117 | void secluded_suppression_init(void); |
118 | #endif /* CONFIG_SECLUDED_MEMORY */ |
119 | |
120 | boolean_t hibernate_cleaning_in_progress = FALSE; |
121 | boolean_t vm_page_free_verify = TRUE; |
122 | |
123 | uint32_t vm_lopage_free_count = 0; |
124 | uint32_t vm_lopage_free_limit = 0; |
125 | uint32_t vm_lopage_lowater = 0; |
126 | boolean_t vm_lopage_refill = FALSE; |
127 | boolean_t vm_lopage_needed = FALSE; |
128 | |
129 | lck_mtx_ext_t vm_page_queue_lock_ext; |
130 | lck_mtx_ext_t vm_page_queue_free_lock_ext; |
131 | lck_mtx_ext_t vm_purgeable_queue_lock_ext; |
132 | |
133 | int speculative_age_index = 0; |
134 | int speculative_steal_index = 0; |
135 | struct vm_speculative_age_q vm_page_queue_speculative[VM_PAGE_MAX_SPECULATIVE_AGE_Q + 1]; |
136 | |
137 | |
138 | __private_extern__ void vm_page_init_lck_grp(void); |
139 | |
140 | static void vm_page_free_prepare(vm_page_t page); |
141 | static vm_page_t vm_page_grab_fictitious_common(ppnum_t phys_addr); |
142 | |
143 | static void vm_tag_init(void); |
144 | |
145 | uint64_t vm_min_kernel_and_kext_address = VM_MIN_KERNEL_AND_KEXT_ADDRESS; |
146 | uint32_t vm_packed_from_vm_pages_array_mask = VM_PACKED_FROM_VM_PAGES_ARRAY; |
147 | uint32_t vm_packed_pointer_shift = VM_PACKED_POINTER_SHIFT; |
148 | |
149 | /* |
150 | * Associated with page of user-allocatable memory is a |
151 | * page structure. |
152 | */ |
153 | |
154 | /* |
155 | * These variables record the values returned by vm_page_bootstrap, |
156 | * for debugging purposes. The implementation of pmap_steal_memory |
157 | * and pmap_startup here also uses them internally. |
158 | */ |
159 | |
160 | vm_offset_t virtual_space_start; |
161 | vm_offset_t virtual_space_end; |
162 | uint32_t vm_page_pages; |
163 | |
164 | /* |
165 | * The vm_page_lookup() routine, which provides for fast |
166 | * (virtual memory object, offset) to page lookup, employs |
167 | * the following hash table. The vm_page_{insert,remove} |
168 | * routines install and remove associations in the table. |
169 | * [This table is often called the virtual-to-physical, |
170 | * or VP, table.] |
171 | */ |
172 | typedef struct { |
173 | vm_page_packed_t page_list; |
174 | #if MACH_PAGE_HASH_STATS |
175 | int cur_count; /* current count */ |
176 | int hi_count; /* high water mark */ |
177 | #endif /* MACH_PAGE_HASH_STATS */ |
178 | } vm_page_bucket_t; |
179 | |
180 | |
181 | #define BUCKETS_PER_LOCK 16 |
182 | |
183 | vm_page_bucket_t *vm_page_buckets; /* Array of buckets */ |
184 | unsigned int vm_page_bucket_count = 0; /* How big is array? */ |
185 | unsigned int vm_page_hash_mask; /* Mask for hash function */ |
186 | unsigned int vm_page_hash_shift; /* Shift for hash function */ |
187 | uint32_t vm_page_bucket_hash; /* Basic bucket hash */ |
188 | unsigned int vm_page_bucket_lock_count = 0; /* How big is array of locks? */ |
189 | |
190 | #ifndef VM_TAG_ACTIVE_UPDATE |
191 | #error VM_TAG_ACTIVE_UPDATE |
192 | #endif |
193 | #ifndef VM_MAX_TAG_ZONES |
194 | #error VM_MAX_TAG_ZONES |
195 | #endif |
196 | |
197 | boolean_t vm_tag_active_update = VM_TAG_ACTIVE_UPDATE; |
198 | lck_spin_t *vm_page_bucket_locks; |
199 | lck_spin_t vm_objects_wired_lock; |
200 | lck_spin_t vm_allocation_sites_lock; |
201 | |
202 | vm_allocation_site_t vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC + 1]; |
203 | vm_allocation_site_t * vm_allocation_sites[VM_MAX_TAG_VALUE]; |
204 | #if VM_MAX_TAG_ZONES |
205 | vm_allocation_zone_total_t ** vm_allocation_zone_totals; |
206 | #endif /* VM_MAX_TAG_ZONES */ |
207 | |
208 | vm_tag_t vm_allocation_tag_highest; |
209 | |
210 | #if VM_PAGE_BUCKETS_CHECK |
211 | boolean_t vm_page_buckets_check_ready = FALSE; |
212 | #if VM_PAGE_FAKE_BUCKETS |
213 | vm_page_bucket_t *vm_page_fake_buckets; /* decoy buckets */ |
214 | vm_map_offset_t vm_page_fake_buckets_start, vm_page_fake_buckets_end; |
215 | #endif /* VM_PAGE_FAKE_BUCKETS */ |
216 | #endif /* VM_PAGE_BUCKETS_CHECK */ |
217 | |
218 | |
219 | |
220 | #if MACH_PAGE_HASH_STATS |
221 | /* This routine is only for debug. It is intended to be called by |
222 | * hand by a developer using a kernel debugger. This routine prints |
223 | * out vm_page_hash table statistics to the kernel debug console. |
224 | */ |
225 | void |
226 | hash_debug(void) |
227 | { |
228 | int i; |
229 | int numbuckets = 0; |
230 | int highsum = 0; |
231 | int maxdepth = 0; |
232 | |
233 | for (i = 0; i < vm_page_bucket_count; i++) { |
234 | if (vm_page_buckets[i].hi_count) { |
235 | numbuckets++; |
236 | highsum += vm_page_buckets[i].hi_count; |
237 | if (vm_page_buckets[i].hi_count > maxdepth) |
238 | maxdepth = vm_page_buckets[i].hi_count; |
239 | } |
240 | } |
241 | printf("Total number of buckets: %d\n" , vm_page_bucket_count); |
242 | printf("Number used buckets: %d = %d%%\n" , |
243 | numbuckets, 100*numbuckets/vm_page_bucket_count); |
244 | printf("Number unused buckets: %d = %d%%\n" , |
245 | vm_page_bucket_count - numbuckets, |
246 | 100*(vm_page_bucket_count-numbuckets)/vm_page_bucket_count); |
247 | printf("Sum of bucket max depth: %d\n" , highsum); |
248 | printf("Average bucket depth: %d.%2d\n" , |
249 | highsum/vm_page_bucket_count, |
250 | highsum%vm_page_bucket_count); |
251 | printf("Maximum bucket depth: %d\n" , maxdepth); |
252 | } |
253 | #endif /* MACH_PAGE_HASH_STATS */ |
254 | |
255 | /* |
256 | * The virtual page size is currently implemented as a runtime |
257 | * variable, but is constant once initialized using vm_set_page_size. |
258 | * This initialization must be done in the machine-dependent |
259 | * bootstrap sequence, before calling other machine-independent |
260 | * initializations. |
261 | * |
262 | * All references to the virtual page size outside this |
263 | * module must use the PAGE_SIZE, PAGE_MASK and PAGE_SHIFT |
264 | * constants. |
265 | */ |
266 | #if defined(__arm__) || defined(__arm64__) |
267 | vm_size_t page_size; |
268 | vm_size_t page_mask; |
269 | int page_shift; |
270 | #else |
271 | vm_size_t page_size = PAGE_SIZE; |
272 | vm_size_t page_mask = PAGE_MASK; |
273 | int page_shift = PAGE_SHIFT; |
274 | #endif |
275 | |
276 | /* |
277 | * Resident page structures are initialized from |
278 | * a template (see vm_page_alloc). |
279 | * |
280 | * When adding a new field to the virtual memory |
281 | * object structure, be sure to add initialization |
282 | * (see vm_page_bootstrap). |
283 | */ |
284 | struct vm_page vm_page_template; |
285 | |
286 | vm_page_t vm_pages = VM_PAGE_NULL; |
287 | vm_page_t vm_page_array_beginning_addr; |
288 | vm_page_t vm_page_array_ending_addr; |
289 | vm_page_t vm_page_array_boundary; |
290 | |
291 | unsigned int vm_pages_count = 0; |
292 | ppnum_t vm_page_lowest = 0; |
293 | |
294 | /* |
295 | * Resident pages that represent real memory |
296 | * are allocated from a set of free lists, |
297 | * one per color. |
298 | */ |
299 | unsigned int vm_colors; |
300 | unsigned int vm_color_mask; /* mask is == (vm_colors-1) */ |
301 | unsigned int vm_cache_geometry_colors = 0; /* set by hw dependent code during startup */ |
302 | unsigned int vm_free_magazine_refill_limit = 0; |
303 | |
304 | |
305 | struct vm_page_queue_free_head { |
306 | vm_page_queue_head_t qhead; |
307 | } __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
308 | |
309 | struct vm_page_queue_free_head vm_page_queue_free[MAX_COLORS]; |
310 | |
311 | |
312 | unsigned int vm_page_free_wanted; |
313 | unsigned int vm_page_free_wanted_privileged; |
314 | #if CONFIG_SECLUDED_MEMORY |
315 | unsigned int vm_page_free_wanted_secluded; |
316 | #endif /* CONFIG_SECLUDED_MEMORY */ |
317 | unsigned int vm_page_free_count; |
318 | |
319 | /* |
320 | * Occasionally, the virtual memory system uses |
321 | * resident page structures that do not refer to |
322 | * real pages, for example to leave a page with |
323 | * important state information in the VP table. |
324 | * |
325 | * These page structures are allocated the way |
326 | * most other kernel structures are. |
327 | */ |
328 | zone_t vm_page_array_zone; |
329 | zone_t vm_page_zone; |
330 | vm_locks_array_t vm_page_locks; |
331 | decl_lck_mtx_data(,vm_page_alloc_lock) |
332 | lck_mtx_ext_t vm_page_alloc_lock_ext; |
333 | |
334 | unsigned int vm_page_local_q_count = 0; |
335 | unsigned int vm_page_local_q_soft_limit = 250; |
336 | unsigned int vm_page_local_q_hard_limit = 500; |
337 | struct vplq *vm_page_local_q = NULL; |
338 | |
339 | /* N.B. Guard and fictitious pages must not |
340 | * be assigned a zero phys_page value. |
341 | */ |
342 | /* |
343 | * Fictitious pages don't have a physical address, |
344 | * but we must initialize phys_page to something. |
345 | * For debugging, this should be a strange value |
346 | * that the pmap module can recognize in assertions. |
347 | */ |
348 | const ppnum_t vm_page_fictitious_addr = (ppnum_t) -1; |
349 | |
350 | /* |
351 | * Guard pages are not accessible so they don't |
352 | * need a physical address, but we need to enter |
353 | * one in the pmap. |
354 | * Let's make it recognizable and make sure that |
355 | * we don't use a real physical page with that |
356 | * physical address. |
357 | */ |
358 | const ppnum_t vm_page_guard_addr = (ppnum_t) -2; |
359 | |
360 | /* |
361 | * Resident page structures are also chained on |
362 | * queues that are used by the page replacement |
363 | * system (pageout daemon). These queues are |
364 | * defined here, but are shared by the pageout |
365 | * module. The inactive queue is broken into |
366 | * file backed and anonymous for convenience as the |
367 | * pageout daemon often assignes a higher |
368 | * importance to anonymous pages (less likely to pick) |
369 | */ |
370 | vm_page_queue_head_t vm_page_queue_active __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
371 | vm_page_queue_head_t vm_page_queue_inactive __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
372 | #if CONFIG_SECLUDED_MEMORY |
373 | vm_page_queue_head_t vm_page_queue_secluded __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
374 | #endif /* CONFIG_SECLUDED_MEMORY */ |
375 | vm_page_queue_head_t vm_page_queue_anonymous __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); /* inactive memory queue for anonymous pages */ |
376 | vm_page_queue_head_t vm_page_queue_throttled __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
377 | |
378 | queue_head_t vm_objects_wired; |
379 | |
380 | void vm_update_darkwake_mode(boolean_t); |
381 | |
382 | #if CONFIG_BACKGROUND_QUEUE |
383 | vm_page_queue_head_t vm_page_queue_background __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
384 | uint32_t vm_page_background_target; |
385 | uint32_t vm_page_background_target_snapshot; |
386 | uint32_t vm_page_background_count; |
387 | uint64_t vm_page_background_promoted_count; |
388 | |
389 | uint32_t vm_page_background_internal_count; |
390 | uint32_t vm_page_background_external_count; |
391 | |
392 | uint32_t vm_page_background_mode; |
393 | uint32_t vm_page_background_exclude_external; |
394 | #endif |
395 | |
396 | unsigned int vm_page_active_count; |
397 | unsigned int vm_page_inactive_count; |
398 | #if CONFIG_SECLUDED_MEMORY |
399 | unsigned int vm_page_secluded_count; |
400 | unsigned int vm_page_secluded_count_free; |
401 | unsigned int vm_page_secluded_count_inuse; |
402 | #endif /* CONFIG_SECLUDED_MEMORY */ |
403 | unsigned int vm_page_anonymous_count; |
404 | unsigned int vm_page_throttled_count; |
405 | unsigned int vm_page_speculative_count; |
406 | |
407 | unsigned int vm_page_wire_count; |
408 | unsigned int vm_page_wire_count_on_boot = 0; |
409 | unsigned int vm_page_stolen_count; |
410 | unsigned int vm_page_wire_count_initial; |
411 | unsigned int vm_page_pages_initial; |
412 | unsigned int vm_page_gobble_count = 0; |
413 | |
414 | #define VM_PAGE_WIRE_COUNT_WARNING 0 |
415 | #define VM_PAGE_GOBBLE_COUNT_WARNING 0 |
416 | |
417 | unsigned int vm_page_purgeable_count = 0; /* # of pages purgeable now */ |
418 | unsigned int vm_page_purgeable_wired_count = 0; /* # of purgeable pages that are wired now */ |
419 | uint64_t vm_page_purged_count = 0; /* total count of purged pages */ |
420 | |
421 | unsigned int vm_page_xpmapped_external_count = 0; |
422 | unsigned int vm_page_external_count = 0; |
423 | unsigned int vm_page_internal_count = 0; |
424 | unsigned int vm_page_pageable_external_count = 0; |
425 | unsigned int vm_page_pageable_internal_count = 0; |
426 | |
427 | #if DEVELOPMENT || DEBUG |
428 | unsigned int vm_page_speculative_recreated = 0; |
429 | unsigned int vm_page_speculative_created = 0; |
430 | unsigned int vm_page_speculative_used = 0; |
431 | #endif |
432 | |
433 | vm_page_queue_head_t vm_page_queue_cleaned __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
434 | |
435 | unsigned int vm_page_cleaned_count = 0; |
436 | |
437 | uint64_t max_valid_dma_address = 0xffffffffffffffffULL; |
438 | ppnum_t max_valid_low_ppnum = 0xffffffff; |
439 | |
440 | |
441 | /* |
442 | * Several page replacement parameters are also |
443 | * shared with this module, so that page allocation |
444 | * (done here in vm_page_alloc) can trigger the |
445 | * pageout daemon. |
446 | */ |
447 | unsigned int vm_page_free_target = 0; |
448 | unsigned int vm_page_free_min = 0; |
449 | unsigned int vm_page_throttle_limit = 0; |
450 | unsigned int vm_page_inactive_target = 0; |
451 | #if CONFIG_SECLUDED_MEMORY |
452 | unsigned int vm_page_secluded_target = 0; |
453 | #endif /* CONFIG_SECLUDED_MEMORY */ |
454 | unsigned int vm_page_anonymous_min = 0; |
455 | unsigned int vm_page_free_reserved = 0; |
456 | |
457 | |
458 | /* |
459 | * The VM system has a couple of heuristics for deciding |
460 | * that pages are "uninteresting" and should be placed |
461 | * on the inactive queue as likely candidates for replacement. |
462 | * These variables let the heuristics be controlled at run-time |
463 | * to make experimentation easier. |
464 | */ |
465 | |
466 | boolean_t vm_page_deactivate_hint = TRUE; |
467 | |
468 | struct vm_page_stats_reusable vm_page_stats_reusable; |
469 | |
470 | /* |
471 | * vm_set_page_size: |
472 | * |
473 | * Sets the page size, perhaps based upon the memory |
474 | * size. Must be called before any use of page-size |
475 | * dependent functions. |
476 | * |
477 | * Sets page_shift and page_mask from page_size. |
478 | */ |
479 | void |
480 | vm_set_page_size(void) |
481 | { |
482 | page_size = PAGE_SIZE; |
483 | page_mask = PAGE_MASK; |
484 | page_shift = PAGE_SHIFT; |
485 | |
486 | if ((page_mask & page_size) != 0) |
487 | panic("vm_set_page_size: page size not a power of two" ); |
488 | |
489 | for (page_shift = 0; ; page_shift++) |
490 | if ((1U << page_shift) == page_size) |
491 | break; |
492 | } |
493 | |
494 | #if defined (__x86_64__) |
495 | |
496 | #define MAX_CLUMP_SIZE 16 |
497 | #define DEFAULT_CLUMP_SIZE 4 |
498 | |
499 | unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold; |
500 | |
501 | #if DEVELOPMENT || DEBUG |
502 | unsigned long vm_clump_stats[MAX_CLUMP_SIZE+1]; |
503 | unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes; |
504 | |
505 | static inline void vm_clump_update_stats(unsigned int c) { |
506 | assert(c<=vm_clump_size); |
507 | if(c>0 && c<=vm_clump_size) vm_clump_stats[c]+=c; |
508 | vm_clump_allocs+=c; |
509 | } |
510 | #endif /* if DEVELOPMENT || DEBUG */ |
511 | |
512 | /* Called once to setup the VM clump knobs */ |
513 | static void |
514 | vm_page_setup_clump( void ) |
515 | { |
516 | unsigned int override, n; |
517 | |
518 | vm_clump_size = DEFAULT_CLUMP_SIZE; |
519 | if ( PE_parse_boot_argn("clump_size" , &override, sizeof (override)) ) vm_clump_size = override; |
520 | |
521 | if(vm_clump_size > MAX_CLUMP_SIZE) panic("vm_page_setup_clump:: clump_size is too large!" ); |
522 | if(vm_clump_size < 1) panic("vm_page_setup_clump:: clump_size must be >= 1" ); |
523 | if((vm_clump_size & (vm_clump_size-1)) != 0) panic("vm_page_setup_clump:: clump_size must be a power of 2" ); |
524 | |
525 | vm_clump_promote_threshold = vm_clump_size; |
526 | vm_clump_mask = vm_clump_size - 1; |
527 | for(vm_clump_shift=0, n=vm_clump_size; n>1; n>>=1, vm_clump_shift++); |
528 | |
529 | #if DEVELOPMENT || DEBUG |
530 | bzero(vm_clump_stats, sizeof(vm_clump_stats)); |
531 | vm_clump_allocs = vm_clump_inserts = vm_clump_inrange = vm_clump_promotes = 0; |
532 | #endif /* if DEVELOPMENT || DEBUG */ |
533 | } |
534 | |
535 | #endif /* #if defined (__x86_64__) */ |
536 | |
537 | #define COLOR_GROUPS_TO_STEAL 4 |
538 | |
539 | /* Called once during statup, once the cache geometry is known. |
540 | */ |
541 | static void |
542 | vm_page_set_colors( void ) |
543 | { |
544 | unsigned int n, override; |
545 | |
546 | #if defined (__x86_64__) |
547 | /* adjust #colors because we need to color outside the clump boundary */ |
548 | vm_cache_geometry_colors >>= vm_clump_shift; |
549 | #endif |
550 | if ( PE_parse_boot_argn("colors" , &override, sizeof (override)) ) /* colors specified as a boot-arg? */ |
551 | n = override; |
552 | else if ( vm_cache_geometry_colors ) /* do we know what the cache geometry is? */ |
553 | n = vm_cache_geometry_colors; |
554 | else n = DEFAULT_COLORS; /* use default if all else fails */ |
555 | |
556 | if ( n == 0 ) |
557 | n = 1; |
558 | if ( n > MAX_COLORS ) |
559 | n = MAX_COLORS; |
560 | |
561 | /* the count must be a power of 2 */ |
562 | if ( ( n & (n - 1)) != 0 ) |
563 | n = DEFAULT_COLORS; /* use default if all else fails */ |
564 | |
565 | vm_colors = n; |
566 | vm_color_mask = n - 1; |
567 | |
568 | vm_free_magazine_refill_limit = vm_colors * COLOR_GROUPS_TO_STEAL; |
569 | |
570 | #if defined (__x86_64__) |
571 | /* adjust for reduction in colors due to clumping and multiple cores */ |
572 | if (real_ncpus) |
573 | vm_free_magazine_refill_limit *= (vm_clump_size * real_ncpus); |
574 | #endif |
575 | } |
576 | |
577 | |
578 | lck_grp_t vm_page_lck_grp_free; |
579 | lck_grp_t vm_page_lck_grp_queue; |
580 | lck_grp_t vm_page_lck_grp_local; |
581 | lck_grp_t vm_page_lck_grp_purge; |
582 | lck_grp_t vm_page_lck_grp_alloc; |
583 | lck_grp_t vm_page_lck_grp_bucket; |
584 | lck_grp_attr_t vm_page_lck_grp_attr; |
585 | lck_attr_t vm_page_lck_attr; |
586 | |
587 | |
588 | __private_extern__ void |
589 | vm_page_init_lck_grp(void) |
590 | { |
591 | /* |
592 | * initialze the vm_page lock world |
593 | */ |
594 | lck_grp_attr_setdefault(&vm_page_lck_grp_attr); |
595 | lck_grp_init(&vm_page_lck_grp_free, "vm_page_free" , &vm_page_lck_grp_attr); |
596 | lck_grp_init(&vm_page_lck_grp_queue, "vm_page_queue" , &vm_page_lck_grp_attr); |
597 | lck_grp_init(&vm_page_lck_grp_local, "vm_page_queue_local" , &vm_page_lck_grp_attr); |
598 | lck_grp_init(&vm_page_lck_grp_purge, "vm_page_purge" , &vm_page_lck_grp_attr); |
599 | lck_grp_init(&vm_page_lck_grp_alloc, "vm_page_alloc" , &vm_page_lck_grp_attr); |
600 | lck_grp_init(&vm_page_lck_grp_bucket, "vm_page_bucket" , &vm_page_lck_grp_attr); |
601 | lck_attr_setdefault(&vm_page_lck_attr); |
602 | lck_mtx_init_ext(&vm_page_alloc_lock, &vm_page_alloc_lock_ext, &vm_page_lck_grp_alloc, &vm_page_lck_attr); |
603 | |
604 | vm_compressor_init_locks(); |
605 | } |
606 | |
607 | #define ROUNDUP_NEXTP2(X) (1U << (32 - __builtin_clz((X) - 1))) |
608 | |
609 | void |
610 | vm_page_init_local_q() |
611 | { |
612 | unsigned int num_cpus; |
613 | unsigned int i; |
614 | struct vplq *t_local_q; |
615 | |
616 | num_cpus = ml_get_max_cpus(); |
617 | |
618 | /* |
619 | * no point in this for a uni-processor system |
620 | */ |
621 | if (num_cpus >= 2) { |
622 | #if KASAN |
623 | /* KASAN breaks the expectation of a size-aligned object by adding a |
624 | * redzone, so explicitly align. */ |
625 | t_local_q = (struct vplq *)kalloc(num_cpus * sizeof(struct vplq) + VM_PACKED_POINTER_ALIGNMENT); |
626 | t_local_q = (void *)(((uintptr_t)t_local_q + (VM_PACKED_POINTER_ALIGNMENT-1)) & ~(VM_PACKED_POINTER_ALIGNMENT-1)); |
627 | #else |
628 | /* round the size up to the nearest power of two */ |
629 | t_local_q = (struct vplq *)kalloc(ROUNDUP_NEXTP2(num_cpus * sizeof(struct vplq))); |
630 | #endif |
631 | |
632 | for (i = 0; i < num_cpus; i++) { |
633 | struct vpl *lq; |
634 | |
635 | lq = &t_local_q[i].vpl_un.vpl; |
636 | VPL_LOCK_INIT(lq, &vm_page_lck_grp_local, &vm_page_lck_attr); |
637 | vm_page_queue_init(&lq->vpl_queue); |
638 | lq->vpl_count = 0; |
639 | lq->vpl_internal_count = 0; |
640 | lq->vpl_external_count = 0; |
641 | } |
642 | vm_page_local_q_count = num_cpus; |
643 | |
644 | vm_page_local_q = (struct vplq *)t_local_q; |
645 | } |
646 | } |
647 | |
648 | /* |
649 | * vm_init_before_launchd |
650 | * |
651 | * This should be called right before launchd is loaded. |
652 | */ |
653 | void |
654 | vm_init_before_launchd() |
655 | { |
656 | vm_page_wire_count_on_boot = vm_page_wire_count; |
657 | } |
658 | |
659 | |
660 | /* |
661 | * vm_page_bootstrap: |
662 | * |
663 | * Initializes the resident memory module. |
664 | * |
665 | * Allocates memory for the page cells, and |
666 | * for the object/offset-to-page hash table headers. |
667 | * Each page cell is initialized and placed on the free list. |
668 | * Returns the range of available kernel virtual memory. |
669 | */ |
670 | |
671 | void |
672 | vm_page_bootstrap( |
673 | vm_offset_t *startp, |
674 | vm_offset_t *endp) |
675 | { |
676 | vm_page_t m; |
677 | unsigned int i; |
678 | unsigned int log1; |
679 | unsigned int log2; |
680 | unsigned int size; |
681 | |
682 | /* |
683 | * Initialize the vm_page template. |
684 | */ |
685 | |
686 | m = &vm_page_template; |
687 | bzero(m, sizeof (*m)); |
688 | |
689 | #if CONFIG_BACKGROUND_QUEUE |
690 | m->vmp_backgroundq.next = 0; |
691 | m->vmp_backgroundq.prev = 0; |
692 | m->vmp_in_background = FALSE; |
693 | m->vmp_on_backgroundq = FALSE; |
694 | #endif |
695 | |
696 | VM_PAGE_ZERO_PAGEQ_ENTRY(m); |
697 | m->vmp_listq.next = 0; |
698 | m->vmp_listq.prev = 0; |
699 | m->vmp_next_m = 0; |
700 | |
701 | m->vmp_object = 0; /* reset later */ |
702 | m->vmp_offset = (vm_object_offset_t) -1; /* reset later */ |
703 | |
704 | m->vmp_wire_count = 0; |
705 | m->vmp_q_state = VM_PAGE_NOT_ON_Q; |
706 | m->vmp_laundry = FALSE; |
707 | m->vmp_reference = FALSE; |
708 | m->vmp_gobbled = FALSE; |
709 | m->vmp_private = FALSE; |
710 | m->vmp_unused_page_bits = 0; |
711 | |
712 | #if !defined(__arm__) && !defined(__arm64__) |
713 | VM_PAGE_SET_PHYS_PAGE(m, 0); /* reset later */ |
714 | #endif |
715 | m->vmp_busy = TRUE; |
716 | m->vmp_wanted = FALSE; |
717 | m->vmp_tabled = FALSE; |
718 | m->vmp_hashed = FALSE; |
719 | m->vmp_fictitious = FALSE; |
720 | m->vmp_pmapped = FALSE; |
721 | m->vmp_wpmapped = FALSE; |
722 | m->vmp_free_when_done = FALSE; |
723 | m->vmp_absent = FALSE; |
724 | m->vmp_error = FALSE; |
725 | m->vmp_dirty = FALSE; |
726 | m->vmp_cleaning = FALSE; |
727 | m->vmp_precious = FALSE; |
728 | m->vmp_clustered = FALSE; |
729 | m->vmp_overwriting = FALSE; |
730 | m->vmp_restart = FALSE; |
731 | m->vmp_unusual = FALSE; |
732 | m->vmp_cs_validated = FALSE; |
733 | m->vmp_cs_tainted = FALSE; |
734 | m->vmp_cs_nx = FALSE; |
735 | m->vmp_no_cache = FALSE; |
736 | m->vmp_reusable = FALSE; |
737 | m->vmp_xpmapped = FALSE; |
738 | m->vmp_written_by_kernel = FALSE; |
739 | m->vmp_unused_object_bits = 0; |
740 | |
741 | /* |
742 | * Initialize the page queues. |
743 | */ |
744 | vm_page_init_lck_grp(); |
745 | |
746 | lck_mtx_init_ext(&vm_page_queue_free_lock, &vm_page_queue_free_lock_ext, &vm_page_lck_grp_free, &vm_page_lck_attr); |
747 | lck_mtx_init_ext(&vm_page_queue_lock, &vm_page_queue_lock_ext, &vm_page_lck_grp_queue, &vm_page_lck_attr); |
748 | lck_mtx_init_ext(&vm_purgeable_queue_lock, &vm_purgeable_queue_lock_ext, &vm_page_lck_grp_purge, &vm_page_lck_attr); |
749 | |
750 | for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) { |
751 | int group; |
752 | |
753 | purgeable_queues[i].token_q_head = 0; |
754 | purgeable_queues[i].token_q_tail = 0; |
755 | for (group = 0; group < NUM_VOLATILE_GROUPS; group++) |
756 | queue_init(&purgeable_queues[i].objq[group]); |
757 | |
758 | purgeable_queues[i].type = i; |
759 | purgeable_queues[i].new_pages = 0; |
760 | #if MACH_ASSERT |
761 | purgeable_queues[i].debug_count_tokens = 0; |
762 | purgeable_queues[i].debug_count_objects = 0; |
763 | #endif |
764 | }; |
765 | purgeable_nonvolatile_count = 0; |
766 | queue_init(&purgeable_nonvolatile_queue); |
767 | |
768 | for (i = 0; i < MAX_COLORS; i++ ) |
769 | vm_page_queue_init(&vm_page_queue_free[i].qhead); |
770 | |
771 | vm_page_queue_init(&vm_lopage_queue_free); |
772 | vm_page_queue_init(&vm_page_queue_active); |
773 | vm_page_queue_init(&vm_page_queue_inactive); |
774 | #if CONFIG_SECLUDED_MEMORY |
775 | vm_page_queue_init(&vm_page_queue_secluded); |
776 | #endif /* CONFIG_SECLUDED_MEMORY */ |
777 | vm_page_queue_init(&vm_page_queue_cleaned); |
778 | vm_page_queue_init(&vm_page_queue_throttled); |
779 | vm_page_queue_init(&vm_page_queue_anonymous); |
780 | queue_init(&vm_objects_wired); |
781 | |
782 | for ( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) { |
783 | vm_page_queue_init(&vm_page_queue_speculative[i].age_q); |
784 | |
785 | vm_page_queue_speculative[i].age_ts.tv_sec = 0; |
786 | vm_page_queue_speculative[i].age_ts.tv_nsec = 0; |
787 | } |
788 | #if CONFIG_BACKGROUND_QUEUE |
789 | vm_page_queue_init(&vm_page_queue_background); |
790 | |
791 | vm_page_background_count = 0; |
792 | vm_page_background_internal_count = 0; |
793 | vm_page_background_external_count = 0; |
794 | vm_page_background_promoted_count = 0; |
795 | |
796 | vm_page_background_target = (unsigned int)(atop_64(max_mem) / 25); |
797 | |
798 | if (vm_page_background_target > VM_PAGE_BACKGROUND_TARGET_MAX) |
799 | vm_page_background_target = VM_PAGE_BACKGROUND_TARGET_MAX; |
800 | |
801 | vm_page_background_mode = VM_PAGE_BG_LEVEL_1; |
802 | vm_page_background_exclude_external = 0; |
803 | |
804 | PE_parse_boot_argn("vm_page_bg_mode" , &vm_page_background_mode, sizeof(vm_page_background_mode)); |
805 | PE_parse_boot_argn("vm_page_bg_exclude_external" , &vm_page_background_exclude_external, sizeof(vm_page_background_exclude_external)); |
806 | PE_parse_boot_argn("vm_page_bg_target" , &vm_page_background_target, sizeof(vm_page_background_target)); |
807 | |
808 | if (vm_page_background_mode > VM_PAGE_BG_LEVEL_1) |
809 | vm_page_background_mode = VM_PAGE_BG_LEVEL_1; |
810 | #endif |
811 | vm_page_free_wanted = 0; |
812 | vm_page_free_wanted_privileged = 0; |
813 | #if CONFIG_SECLUDED_MEMORY |
814 | vm_page_free_wanted_secluded = 0; |
815 | #endif /* CONFIG_SECLUDED_MEMORY */ |
816 | |
817 | #if defined (__x86_64__) |
818 | /* this must be called before vm_page_set_colors() */ |
819 | vm_page_setup_clump(); |
820 | #endif |
821 | |
822 | vm_page_set_colors(); |
823 | |
824 | bzero(vm_page_inactive_states, sizeof(vm_page_inactive_states)); |
825 | vm_page_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; |
826 | vm_page_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; |
827 | vm_page_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; |
828 | |
829 | bzero(vm_page_pageable_states, sizeof(vm_page_pageable_states)); |
830 | vm_page_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; |
831 | vm_page_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; |
832 | vm_page_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; |
833 | vm_page_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1; |
834 | vm_page_pageable_states[VM_PAGE_ON_SPECULATIVE_Q] = 1; |
835 | vm_page_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1; |
836 | #if CONFIG_SECLUDED_MEMORY |
837 | vm_page_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1; |
838 | #endif /* CONFIG_SECLUDED_MEMORY */ |
839 | |
840 | bzero(vm_page_non_speculative_pageable_states, sizeof(vm_page_non_speculative_pageable_states)); |
841 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; |
842 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; |
843 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; |
844 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_ACTIVE_Q] = 1; |
845 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_THROTTLED_Q] = 1; |
846 | #if CONFIG_SECLUDED_MEMORY |
847 | vm_page_non_speculative_pageable_states[VM_PAGE_ON_SECLUDED_Q] = 1; |
848 | #endif /* CONFIG_SECLUDED_MEMORY */ |
849 | |
850 | bzero(vm_page_active_or_inactive_states, sizeof(vm_page_active_or_inactive_states)); |
851 | vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_INTERNAL_Q] = 1; |
852 | vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_EXTERNAL_Q] = 1; |
853 | vm_page_active_or_inactive_states[VM_PAGE_ON_INACTIVE_CLEANED_Q] = 1; |
854 | vm_page_active_or_inactive_states[VM_PAGE_ON_ACTIVE_Q] = 1; |
855 | #if CONFIG_SECLUDED_MEMORY |
856 | vm_page_active_or_inactive_states[VM_PAGE_ON_SECLUDED_Q] = 1; |
857 | #endif /* CONFIG_SECLUDED_MEMORY */ |
858 | |
859 | for (i = 0; i < VM_KERN_MEMORY_FIRST_DYNAMIC; i++) |
860 | { |
861 | vm_allocation_sites_static[i].refcount = 2; |
862 | vm_allocation_sites_static[i].tag = i; |
863 | vm_allocation_sites[i] = &vm_allocation_sites_static[i]; |
864 | } |
865 | vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].refcount = 2; |
866 | vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC].tag = VM_KERN_MEMORY_ANY; |
867 | vm_allocation_sites[VM_KERN_MEMORY_ANY] = &vm_allocation_sites_static[VM_KERN_MEMORY_FIRST_DYNAMIC]; |
868 | |
869 | /* |
870 | * Steal memory for the map and zone subsystems. |
871 | */ |
872 | #if CONFIG_GZALLOC |
873 | gzalloc_configure(); |
874 | #endif |
875 | kernel_debug_string_early("vm_map_steal_memory" ); |
876 | vm_map_steal_memory(); |
877 | |
878 | /* |
879 | * Allocate (and initialize) the virtual-to-physical |
880 | * table hash buckets. |
881 | * |
882 | * The number of buckets should be a power of two to |
883 | * get a good hash function. The following computation |
884 | * chooses the first power of two that is greater |
885 | * than the number of physical pages in the system. |
886 | */ |
887 | |
888 | if (vm_page_bucket_count == 0) { |
889 | unsigned int npages = pmap_free_pages(); |
890 | |
891 | vm_page_bucket_count = 1; |
892 | while (vm_page_bucket_count < npages) |
893 | vm_page_bucket_count <<= 1; |
894 | } |
895 | vm_page_bucket_lock_count = (vm_page_bucket_count + BUCKETS_PER_LOCK - 1) / BUCKETS_PER_LOCK; |
896 | |
897 | vm_page_hash_mask = vm_page_bucket_count - 1; |
898 | |
899 | /* |
900 | * Calculate object shift value for hashing algorithm: |
901 | * O = log2(sizeof(struct vm_object)) |
902 | * B = log2(vm_page_bucket_count) |
903 | * hash shifts the object left by |
904 | * B/2 - O |
905 | */ |
906 | size = vm_page_bucket_count; |
907 | for (log1 = 0; size > 1; log1++) |
908 | size /= 2; |
909 | size = sizeof(struct vm_object); |
910 | for (log2 = 0; size > 1; log2++) |
911 | size /= 2; |
912 | vm_page_hash_shift = log1/2 - log2 + 1; |
913 | |
914 | vm_page_bucket_hash = 1 << ((log1 + 1) >> 1); /* Get (ceiling of sqrt of table size) */ |
915 | vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2); /* Get (ceiling of quadroot of table size) */ |
916 | vm_page_bucket_hash |= 1; /* Set bit and add 1 - always must be 1 to insure unique series */ |
917 | |
918 | if (vm_page_hash_mask & vm_page_bucket_count) |
919 | printf("vm_page_bootstrap: WARNING -- strange page hash\n" ); |
920 | |
921 | #if VM_PAGE_BUCKETS_CHECK |
922 | #if VM_PAGE_FAKE_BUCKETS |
923 | /* |
924 | * Allocate a decoy set of page buckets, to detect |
925 | * any stomping there. |
926 | */ |
927 | vm_page_fake_buckets = (vm_page_bucket_t *) |
928 | pmap_steal_memory(vm_page_bucket_count * |
929 | sizeof(vm_page_bucket_t)); |
930 | vm_page_fake_buckets_start = (vm_map_offset_t) vm_page_fake_buckets; |
931 | vm_page_fake_buckets_end = |
932 | vm_map_round_page((vm_page_fake_buckets_start + |
933 | (vm_page_bucket_count * |
934 | sizeof (vm_page_bucket_t))), |
935 | PAGE_MASK); |
936 | char *cp; |
937 | for (cp = (char *)vm_page_fake_buckets_start; |
938 | cp < (char *)vm_page_fake_buckets_end; |
939 | cp++) { |
940 | *cp = 0x5a; |
941 | } |
942 | #endif /* VM_PAGE_FAKE_BUCKETS */ |
943 | #endif /* VM_PAGE_BUCKETS_CHECK */ |
944 | |
945 | kernel_debug_string_early("vm_page_buckets" ); |
946 | vm_page_buckets = (vm_page_bucket_t *) |
947 | pmap_steal_memory(vm_page_bucket_count * |
948 | sizeof(vm_page_bucket_t)); |
949 | |
950 | kernel_debug_string_early("vm_page_bucket_locks" ); |
951 | vm_page_bucket_locks = (lck_spin_t *) |
952 | pmap_steal_memory(vm_page_bucket_lock_count * |
953 | sizeof(lck_spin_t)); |
954 | |
955 | for (i = 0; i < vm_page_bucket_count; i++) { |
956 | vm_page_bucket_t *bucket = &vm_page_buckets[i]; |
957 | |
958 | bucket->page_list = VM_PAGE_PACK_PTR(VM_PAGE_NULL); |
959 | #if MACH_PAGE_HASH_STATS |
960 | bucket->cur_count = 0; |
961 | bucket->hi_count = 0; |
962 | #endif /* MACH_PAGE_HASH_STATS */ |
963 | } |
964 | |
965 | for (i = 0; i < vm_page_bucket_lock_count; i++) |
966 | lck_spin_init(&vm_page_bucket_locks[i], &vm_page_lck_grp_bucket, &vm_page_lck_attr); |
967 | |
968 | lck_spin_init(&vm_objects_wired_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr); |
969 | lck_spin_init(&vm_allocation_sites_lock, &vm_page_lck_grp_bucket, &vm_page_lck_attr); |
970 | vm_tag_init(); |
971 | |
972 | #if VM_PAGE_BUCKETS_CHECK |
973 | vm_page_buckets_check_ready = TRUE; |
974 | #endif /* VM_PAGE_BUCKETS_CHECK */ |
975 | |
976 | /* |
977 | * Machine-dependent code allocates the resident page table. |
978 | * It uses vm_page_init to initialize the page frames. |
979 | * The code also returns to us the virtual space available |
980 | * to the kernel. We don't trust the pmap module |
981 | * to get the alignment right. |
982 | */ |
983 | |
984 | kernel_debug_string_early("pmap_startup" ); |
985 | pmap_startup(&virtual_space_start, &virtual_space_end); |
986 | virtual_space_start = round_page(virtual_space_start); |
987 | virtual_space_end = trunc_page(virtual_space_end); |
988 | |
989 | *startp = virtual_space_start; |
990 | *endp = virtual_space_end; |
991 | |
992 | /* |
993 | * Compute the initial "wire" count. |
994 | * Up until now, the pages which have been set aside are not under |
995 | * the VM system's control, so although they aren't explicitly |
996 | * wired, they nonetheless can't be moved. At this moment, |
997 | * all VM managed pages are "free", courtesy of pmap_startup. |
998 | */ |
999 | assert((unsigned int) atop_64(max_mem) == atop_64(max_mem)); |
1000 | vm_page_wire_count = ((unsigned int) atop_64(max_mem)) - vm_page_free_count - vm_lopage_free_count; /* initial value */ |
1001 | #if CONFIG_SECLUDED_MEMORY |
1002 | vm_page_wire_count -= vm_page_secluded_count; |
1003 | #endif |
1004 | vm_page_wire_count_initial = vm_page_wire_count; |
1005 | vm_page_pages_initial = vm_page_pages; |
1006 | |
1007 | printf("vm_page_bootstrap: %d free pages and %d wired pages\n" , |
1008 | vm_page_free_count, vm_page_wire_count); |
1009 | |
1010 | kernel_debug_string_early("vm_page_bootstrap complete" ); |
1011 | simple_lock_init(&vm_paging_lock, 0); |
1012 | } |
1013 | |
1014 | #ifndef MACHINE_PAGES |
1015 | /* |
1016 | * We implement pmap_steal_memory and pmap_startup with the help |
1017 | * of two simpler functions, pmap_virtual_space and pmap_next_page. |
1018 | */ |
1019 | |
1020 | void * |
1021 | pmap_steal_memory( |
1022 | vm_size_t size) |
1023 | { |
1024 | kern_return_t kr; |
1025 | vm_offset_t addr, vaddr; |
1026 | ppnum_t phys_page; |
1027 | |
1028 | /* |
1029 | * We round the size to a round multiple. |
1030 | */ |
1031 | |
1032 | size = (size + sizeof (void *) - 1) &~ (sizeof (void *) - 1); |
1033 | |
1034 | /* |
1035 | * If this is the first call to pmap_steal_memory, |
1036 | * we have to initialize ourself. |
1037 | */ |
1038 | |
1039 | if (virtual_space_start == virtual_space_end) { |
1040 | pmap_virtual_space(&virtual_space_start, &virtual_space_end); |
1041 | |
1042 | /* |
1043 | * The initial values must be aligned properly, and |
1044 | * we don't trust the pmap module to do it right. |
1045 | */ |
1046 | |
1047 | virtual_space_start = round_page(virtual_space_start); |
1048 | virtual_space_end = trunc_page(virtual_space_end); |
1049 | } |
1050 | |
1051 | /* |
1052 | * Allocate virtual memory for this request. |
1053 | */ |
1054 | |
1055 | addr = virtual_space_start; |
1056 | virtual_space_start += size; |
1057 | |
1058 | //kprintf("pmap_steal_memory: %08lX - %08lX; size=%08lX\n", (long)addr, (long)virtual_space_start, (long)size); /* (TEST/DEBUG) */ |
1059 | |
1060 | /* |
1061 | * Allocate and map physical pages to back new virtual pages. |
1062 | */ |
1063 | |
1064 | for (vaddr = round_page(addr); |
1065 | vaddr < addr + size; |
1066 | vaddr += PAGE_SIZE) { |
1067 | |
1068 | if (!pmap_next_page_hi(&phys_page)) |
1069 | panic("pmap_steal_memory() size: 0x%llx\n" , (uint64_t)size); |
1070 | |
1071 | /* |
1072 | * XXX Logically, these mappings should be wired, |
1073 | * but some pmap modules barf if they are. |
1074 | */ |
1075 | #if defined(__LP64__) |
1076 | #ifdef __arm64__ |
1077 | /* ARM64_TODO: verify that we really don't need this */ |
1078 | #else |
1079 | pmap_pre_expand(kernel_pmap, vaddr); |
1080 | #endif |
1081 | #endif |
1082 | |
1083 | kr = pmap_enter(kernel_pmap, vaddr, phys_page, |
1084 | VM_PROT_READ|VM_PROT_WRITE, VM_PROT_NONE, |
1085 | VM_WIMG_USE_DEFAULT, FALSE); |
1086 | |
1087 | if (kr != KERN_SUCCESS) { |
1088 | panic("pmap_steal_memory() pmap_enter failed, vaddr=%#lx, phys_page=%u" , |
1089 | (unsigned long)vaddr, phys_page); |
1090 | } |
1091 | |
1092 | /* |
1093 | * Account for newly stolen memory |
1094 | */ |
1095 | vm_page_wire_count++; |
1096 | vm_page_stolen_count++; |
1097 | } |
1098 | |
1099 | #if KASAN |
1100 | kasan_notify_address(round_page(addr), size); |
1101 | #endif |
1102 | return (void *) addr; |
1103 | } |
1104 | |
1105 | #if CONFIG_SECLUDED_MEMORY |
1106 | /* boot-args to control secluded memory */ |
1107 | unsigned int secluded_mem_mb = 0; /* # of MBs of RAM to seclude */ |
1108 | int secluded_for_iokit = 1; /* IOKit can use secluded memory */ |
1109 | int secluded_for_apps = 1; /* apps can use secluded memory */ |
1110 | int secluded_for_filecache = 2; /* filecache can use seclude memory */ |
1111 | #if 11 |
1112 | int secluded_for_fbdp = 0; |
1113 | #endif |
1114 | uint64_t secluded_shutoff_trigger = 0; |
1115 | #endif /* CONFIG_SECLUDED_MEMORY */ |
1116 | |
1117 | |
1118 | #if defined(__arm__) || defined(__arm64__) |
1119 | extern void patch_low_glo_vm_page_info(void *, void *, uint32_t); |
1120 | unsigned int vm_first_phys_ppnum = 0; |
1121 | #endif |
1122 | |
1123 | |
1124 | void vm_page_release_startup(vm_page_t mem); |
1125 | void |
1126 | pmap_startup( |
1127 | vm_offset_t *startp, |
1128 | vm_offset_t *endp) |
1129 | { |
1130 | unsigned int i, npages, pages_initialized, fill, fillval; |
1131 | ppnum_t phys_page; |
1132 | addr64_t tmpaddr; |
1133 | |
1134 | #if defined(__LP64__) |
1135 | /* |
1136 | * make sure we are aligned on a 64 byte boundary |
1137 | * for VM_PAGE_PACK_PTR (it clips off the low-order |
1138 | * 6 bits of the pointer) |
1139 | */ |
1140 | if (virtual_space_start != virtual_space_end) |
1141 | virtual_space_start = round_page(virtual_space_start); |
1142 | #endif |
1143 | |
1144 | /* |
1145 | * We calculate how many page frames we will have |
1146 | * and then allocate the page structures in one chunk. |
1147 | */ |
1148 | |
1149 | tmpaddr = (addr64_t)pmap_free_pages() * (addr64_t)PAGE_SIZE; /* Get the amount of memory left */ |
1150 | tmpaddr = tmpaddr + (addr64_t)(round_page(virtual_space_start) - virtual_space_start); /* Account for any slop */ |
1151 | npages = (unsigned int)(tmpaddr / (addr64_t)(PAGE_SIZE + sizeof(*vm_pages))); /* Figure size of all vm_page_ts, including enough to hold the vm_page_ts */ |
1152 | |
1153 | vm_pages = (vm_page_t) pmap_steal_memory(npages * sizeof *vm_pages); |
1154 | |
1155 | /* |
1156 | * Initialize the page frames. |
1157 | */ |
1158 | kernel_debug_string_early("Initialize the page frames" ); |
1159 | |
1160 | vm_page_array_beginning_addr = &vm_pages[0]; |
1161 | vm_page_array_ending_addr = &vm_pages[npages]; |
1162 | |
1163 | for (i = 0, pages_initialized = 0; i < npages; i++) { |
1164 | if (!pmap_next_page(&phys_page)) |
1165 | break; |
1166 | #if defined(__arm__) || defined(__arm64__) |
1167 | if (pages_initialized == 0) { |
1168 | vm_first_phys_ppnum = phys_page; |
1169 | patch_low_glo_vm_page_info((void *)vm_page_array_beginning_addr, (void *)vm_page_array_ending_addr, vm_first_phys_ppnum); |
1170 | } |
1171 | assert((i + vm_first_phys_ppnum) == phys_page); |
1172 | #endif |
1173 | if (pages_initialized == 0 || phys_page < vm_page_lowest) |
1174 | vm_page_lowest = phys_page; |
1175 | |
1176 | vm_page_init(&vm_pages[i], phys_page, FALSE); |
1177 | vm_page_pages++; |
1178 | pages_initialized++; |
1179 | } |
1180 | vm_pages_count = pages_initialized; |
1181 | vm_page_array_boundary = &vm_pages[pages_initialized]; |
1182 | |
1183 | #if defined(__LP64__) |
1184 | |
1185 | if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[0]))) != &vm_pages[0]) |
1186 | panic("VM_PAGE_PACK_PTR failed on &vm_pages[0] - %p" , (void *)&vm_pages[0]); |
1187 | |
1188 | if ((vm_page_t)(VM_PAGE_UNPACK_PTR(VM_PAGE_PACK_PTR(&vm_pages[vm_pages_count-1]))) != &vm_pages[vm_pages_count-1]) |
1189 | panic("VM_PAGE_PACK_PTR failed on &vm_pages[vm_pages_count-1] - %p" , (void *)&vm_pages[vm_pages_count-1]); |
1190 | #endif |
1191 | kernel_debug_string_early("page fill/release" ); |
1192 | /* |
1193 | * Check if we want to initialize pages to a known value |
1194 | */ |
1195 | fill = 0; /* Assume no fill */ |
1196 | if (PE_parse_boot_argn("fill" , &fillval, sizeof (fillval))) fill = 1; /* Set fill */ |
1197 | #if DEBUG |
1198 | /* This slows down booting the DEBUG kernel, particularly on |
1199 | * large memory systems, but is worthwhile in deterministically |
1200 | * trapping uninitialized memory usage. |
1201 | */ |
1202 | if (fill == 0) { |
1203 | fill = 1; |
1204 | fillval = 0xDEB8F177; |
1205 | } |
1206 | #endif |
1207 | if (fill) |
1208 | kprintf("Filling vm_pages with pattern: 0x%x\n" , fillval); |
1209 | |
1210 | #if CONFIG_SECLUDED_MEMORY |
1211 | /* default: no secluded mem */ |
1212 | secluded_mem_mb = 0; |
1213 | if (max_mem > 1*1024*1024*1024) { |
1214 | /* default to 90MB for devices with > 1GB of RAM */ |
1215 | secluded_mem_mb = 90; |
1216 | } |
1217 | /* override with value from device tree, if provided */ |
1218 | PE_get_default("kern.secluded_mem_mb" , |
1219 | &secluded_mem_mb, sizeof(secluded_mem_mb)); |
1220 | /* override with value from boot-args, if provided */ |
1221 | PE_parse_boot_argn("secluded_mem_mb" , |
1222 | &secluded_mem_mb, |
1223 | sizeof (secluded_mem_mb)); |
1224 | |
1225 | vm_page_secluded_target = (unsigned int) |
1226 | ((secluded_mem_mb * 1024ULL * 1024ULL) / PAGE_SIZE); |
1227 | PE_parse_boot_argn("secluded_for_iokit" , |
1228 | &secluded_for_iokit, |
1229 | sizeof (secluded_for_iokit)); |
1230 | PE_parse_boot_argn("secluded_for_apps" , |
1231 | &secluded_for_apps, |
1232 | sizeof (secluded_for_apps)); |
1233 | PE_parse_boot_argn("secluded_for_filecache" , |
1234 | &secluded_for_filecache, |
1235 | sizeof (secluded_for_filecache)); |
1236 | #if 11 |
1237 | PE_parse_boot_argn("secluded_for_fbdp" , |
1238 | &secluded_for_fbdp, |
1239 | sizeof (secluded_for_fbdp)); |
1240 | #endif |
1241 | |
1242 | /* |
1243 | * On small devices, allow a large app to effectively suppress |
1244 | * secluded memory until it exits. |
1245 | */ |
1246 | if (max_mem <= 1 * 1024 * 1024 * 1024 && vm_page_secluded_target != 0) { |
1247 | |
1248 | /* |
1249 | * Get an amount from boot-args, else use 500MB. |
1250 | * 500MB was chosen from a Peace daemon tentpole test which used munch |
1251 | * to induce jetsam thrashing of false idle daemons. |
1252 | */ |
1253 | int secluded_shutoff_mb; |
1254 | if (PE_parse_boot_argn("secluded_shutoff_mb" , &secluded_shutoff_mb, |
1255 | sizeof (secluded_shutoff_mb))) |
1256 | secluded_shutoff_trigger = (uint64_t)secluded_shutoff_mb * 1024 * 1024; |
1257 | else |
1258 | secluded_shutoff_trigger = 500 * 1024 * 1024; |
1259 | |
1260 | if (secluded_shutoff_trigger != 0) |
1261 | secluded_suppression_init(); |
1262 | } |
1263 | |
1264 | #endif /* CONFIG_SECLUDED_MEMORY */ |
1265 | |
1266 | /* |
1267 | * By default release pages in reverse order so that physical pages |
1268 | * initially get allocated in ascending addresses. This keeps |
1269 | * the devices (which must address physical memory) happy if |
1270 | * they require several consecutive pages. |
1271 | * |
1272 | * For debugging, you can reverse this ordering and/or fill |
1273 | * all pages with a known value. |
1274 | */ |
1275 | if (vm_himemory_mode == 2) { |
1276 | for (i = 0; i < pages_initialized; i++) { |
1277 | if (fill) |
1278 | fillPage(VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]), fillval); |
1279 | vm_page_release_startup(&vm_pages[i]); |
1280 | } |
1281 | } else { |
1282 | for (i = pages_initialized; i-- > 0; ) { |
1283 | if (fill) |
1284 | fillPage(VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]), fillval); |
1285 | vm_page_release_startup(&vm_pages[i]); |
1286 | } |
1287 | } |
1288 | |
1289 | VM_CHECK_MEMORYSTATUS; |
1290 | |
1291 | #if 0 |
1292 | { |
1293 | vm_page_t xx, xxo, xxl; |
1294 | int i, j, k, l; |
1295 | |
1296 | j = 0; /* (BRINGUP) */ |
1297 | xxl = 0; |
1298 | |
1299 | for( i = 0; i < vm_colors; i++ ) { |
1300 | queue_iterate(&vm_page_queue_free[i].qhead, |
1301 | xx, |
1302 | vm_page_t, |
1303 | vmp_pageq) { /* BRINGUP */ |
1304 | j++; /* (BRINGUP) */ |
1305 | if(j > vm_page_free_count) { /* (BRINGUP) */ |
1306 | panic("pmap_startup: too many pages, xx = %08X, xxl = %08X\n" , xx, xxl); |
1307 | } |
1308 | |
1309 | l = vm_page_free_count - j; /* (BRINGUP) */ |
1310 | k = 0; /* (BRINGUP) */ |
1311 | |
1312 | if(((j - 1) & 0xFFFF) == 0) kprintf("checking number %d of %d\n" , j, vm_page_free_count); |
1313 | |
1314 | for(xxo = xx->pageq.next; xxo != &vm_page_queue_free[i].qhead; xxo = xxo->pageq.next) { /* (BRINGUP) */ |
1315 | k++; |
1316 | if(k > l) panic("pmap_startup: too many in secondary check %d %d\n" , k, l); |
1317 | if((xx->phys_page & 0xFFFFFFFF) == (xxo->phys_page & 0xFFFFFFFF)) { /* (BRINGUP) */ |
1318 | panic("pmap_startup: duplicate physaddr, xx = %08X, xxo = %08X\n" , xx, xxo); |
1319 | } |
1320 | } |
1321 | |
1322 | xxl = xx; |
1323 | } |
1324 | } |
1325 | |
1326 | if(j != vm_page_free_count) { /* (BRINGUP) */ |
1327 | panic("pmap_startup: vm_page_free_count does not match, calc = %d, vm_page_free_count = %08X\n" , j, vm_page_free_count); |
1328 | } |
1329 | } |
1330 | #endif |
1331 | |
1332 | |
1333 | /* |
1334 | * We have to re-align virtual_space_start, |
1335 | * because pmap_steal_memory has been using it. |
1336 | */ |
1337 | |
1338 | virtual_space_start = round_page(virtual_space_start); |
1339 | |
1340 | *startp = virtual_space_start; |
1341 | *endp = virtual_space_end; |
1342 | } |
1343 | #endif /* MACHINE_PAGES */ |
1344 | |
1345 | /* |
1346 | * Routine: vm_page_module_init |
1347 | * Purpose: |
1348 | * Second initialization pass, to be done after |
1349 | * the basic VM system is ready. |
1350 | */ |
1351 | void |
1352 | vm_page_module_init(void) |
1353 | { |
1354 | uint64_t vm_page_zone_pages, vm_page_array_zone_data_size; |
1355 | vm_size_t vm_page_with_ppnum_size; |
1356 | |
1357 | vm_page_array_zone = zinit((vm_size_t) sizeof(struct vm_page), |
1358 | 0, PAGE_SIZE, "vm pages array" ); |
1359 | |
1360 | zone_change(vm_page_array_zone, Z_CALLERACCT, FALSE); |
1361 | zone_change(vm_page_array_zone, Z_EXPAND, FALSE); |
1362 | zone_change(vm_page_array_zone, Z_EXHAUST, TRUE); |
1363 | zone_change(vm_page_array_zone, Z_FOREIGN, TRUE); |
1364 | zone_change(vm_page_array_zone, Z_GZALLOC_EXEMPT, TRUE); |
1365 | /* |
1366 | * Adjust zone statistics to account for the real pages allocated |
1367 | * in vm_page_create(). [Q: is this really what we want?] |
1368 | */ |
1369 | vm_page_array_zone->count += vm_page_pages; |
1370 | vm_page_array_zone->sum_count += vm_page_pages; |
1371 | vm_page_array_zone_data_size = vm_page_pages * vm_page_array_zone->elem_size; |
1372 | vm_page_array_zone->cur_size += vm_page_array_zone_data_size; |
1373 | vm_page_zone_pages = ((round_page(vm_page_array_zone_data_size)) / PAGE_SIZE); |
1374 | OSAddAtomic64(vm_page_zone_pages, &(vm_page_array_zone->page_count)); |
1375 | /* since zone accounts for these, take them out of stolen */ |
1376 | VM_PAGE_MOVE_STOLEN(vm_page_zone_pages); |
1377 | |
1378 | vm_page_with_ppnum_size = (sizeof(struct vm_page_with_ppnum) + (VM_PACKED_POINTER_ALIGNMENT-1)) & ~(VM_PACKED_POINTER_ALIGNMENT - 1); |
1379 | |
1380 | vm_page_zone = zinit(vm_page_with_ppnum_size, |
1381 | 0, PAGE_SIZE, "vm pages" ); |
1382 | |
1383 | zone_change(vm_page_zone, Z_CALLERACCT, FALSE); |
1384 | zone_change(vm_page_zone, Z_EXPAND, FALSE); |
1385 | zone_change(vm_page_zone, Z_EXHAUST, TRUE); |
1386 | zone_change(vm_page_zone, Z_FOREIGN, TRUE); |
1387 | zone_change(vm_page_zone, Z_GZALLOC_EXEMPT, TRUE); |
1388 | zone_change(vm_page_zone, Z_ALIGNMENT_REQUIRED, TRUE); |
1389 | } |
1390 | |
1391 | /* |
1392 | * Routine: vm_page_create |
1393 | * Purpose: |
1394 | * After the VM system is up, machine-dependent code |
1395 | * may stumble across more physical memory. For example, |
1396 | * memory that it was reserving for a frame buffer. |
1397 | * vm_page_create turns this memory into available pages. |
1398 | */ |
1399 | |
1400 | void |
1401 | vm_page_create( |
1402 | ppnum_t start, |
1403 | ppnum_t end) |
1404 | { |
1405 | ppnum_t phys_page; |
1406 | vm_page_t m; |
1407 | |
1408 | for (phys_page = start; |
1409 | phys_page < end; |
1410 | phys_page++) { |
1411 | while ((m = (vm_page_t) vm_page_grab_fictitious_common(phys_page)) |
1412 | == VM_PAGE_NULL) |
1413 | vm_page_more_fictitious(); |
1414 | |
1415 | m->vmp_fictitious = FALSE; |
1416 | pmap_clear_noencrypt(phys_page); |
1417 | |
1418 | vm_page_pages++; |
1419 | vm_page_release(m, FALSE); |
1420 | } |
1421 | } |
1422 | |
1423 | /* |
1424 | * vm_page_hash: |
1425 | * |
1426 | * Distributes the object/offset key pair among hash buckets. |
1427 | * |
1428 | * NOTE: The bucket count must be a power of 2 |
1429 | */ |
1430 | #define vm_page_hash(object, offset) (\ |
1431 | ( (natural_t)((uintptr_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\ |
1432 | & vm_page_hash_mask) |
1433 | |
1434 | |
1435 | /* |
1436 | * vm_page_insert: [ internal use only ] |
1437 | * |
1438 | * Inserts the given mem entry into the object/object-page |
1439 | * table and object list. |
1440 | * |
1441 | * The object must be locked. |
1442 | */ |
1443 | void |
1444 | vm_page_insert( |
1445 | vm_page_t mem, |
1446 | vm_object_t object, |
1447 | vm_object_offset_t offset) |
1448 | { |
1449 | vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, TRUE, FALSE, FALSE, NULL); |
1450 | } |
1451 | |
1452 | void |
1453 | vm_page_insert_wired( |
1454 | vm_page_t mem, |
1455 | vm_object_t object, |
1456 | vm_object_offset_t offset, |
1457 | vm_tag_t tag) |
1458 | { |
1459 | vm_page_insert_internal(mem, object, offset, tag, FALSE, TRUE, FALSE, FALSE, NULL); |
1460 | } |
1461 | |
1462 | void |
1463 | vm_page_insert_internal( |
1464 | vm_page_t mem, |
1465 | vm_object_t object, |
1466 | vm_object_offset_t offset, |
1467 | vm_tag_t tag, |
1468 | boolean_t queues_lock_held, |
1469 | boolean_t insert_in_hash, |
1470 | boolean_t batch_pmap_op, |
1471 | boolean_t batch_accounting, |
1472 | uint64_t *delayed_ledger_update) |
1473 | { |
1474 | vm_page_bucket_t *bucket; |
1475 | lck_spin_t *bucket_lock; |
1476 | int hash_id; |
1477 | task_t owner; |
1478 | int ledger_idx_volatile; |
1479 | int ledger_idx_nonvolatile; |
1480 | int ledger_idx_volatile_compressed; |
1481 | int ledger_idx_nonvolatile_compressed; |
1482 | boolean_t ; |
1483 | |
1484 | XPR(XPR_VM_PAGE, |
1485 | "vm_page_insert, object 0x%X offset 0x%X page 0x%X\n" , |
1486 | object, offset, mem, 0,0); |
1487 | #if 0 |
1488 | /* |
1489 | * we may not hold the page queue lock |
1490 | * so this check isn't safe to make |
1491 | */ |
1492 | VM_PAGE_CHECK(mem); |
1493 | #endif |
1494 | |
1495 | assert(page_aligned(offset)); |
1496 | |
1497 | assert(!VM_PAGE_WIRED(mem) || mem->vmp_private || mem->vmp_fictitious || (tag != VM_KERN_MEMORY_NONE)); |
1498 | |
1499 | /* the vm_submap_object is only a placeholder for submaps */ |
1500 | assert(object != vm_submap_object); |
1501 | |
1502 | vm_object_lock_assert_exclusive(object); |
1503 | LCK_MTX_ASSERT(&vm_page_queue_lock, |
1504 | queues_lock_held ? LCK_MTX_ASSERT_OWNED |
1505 | : LCK_MTX_ASSERT_NOTOWNED); |
1506 | |
1507 | if (queues_lock_held == FALSE) |
1508 | assert(!VM_PAGE_PAGEABLE(mem)); |
1509 | |
1510 | if (insert_in_hash == TRUE) { |
1511 | #if DEBUG || VM_PAGE_CHECK_BUCKETS |
1512 | if (mem->vmp_tabled || mem->vmp_object) |
1513 | panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) " |
1514 | "already in (obj=%p,off=0x%llx)" , |
1515 | mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset); |
1516 | #endif |
1517 | if (object->internal && (offset >= object->vo_size)) { |
1518 | panic("vm_page_insert_internal: (page=%p,obj=%p,off=0x%llx,size=0x%llx) inserted at offset past object bounds" , |
1519 | mem, object, offset, object->vo_size); |
1520 | } |
1521 | |
1522 | assert(vm_page_lookup(object, offset) == VM_PAGE_NULL); |
1523 | |
1524 | /* |
1525 | * Record the object/offset pair in this page |
1526 | */ |
1527 | |
1528 | mem->vmp_object = VM_PAGE_PACK_OBJECT(object); |
1529 | mem->vmp_offset = offset; |
1530 | |
1531 | #if CONFIG_SECLUDED_MEMORY |
1532 | if (object->eligible_for_secluded) { |
1533 | vm_page_secluded.eligible_for_secluded++; |
1534 | } |
1535 | #endif /* CONFIG_SECLUDED_MEMORY */ |
1536 | |
1537 | /* |
1538 | * Insert it into the object_object/offset hash table |
1539 | */ |
1540 | hash_id = vm_page_hash(object, offset); |
1541 | bucket = &vm_page_buckets[hash_id]; |
1542 | bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; |
1543 | |
1544 | lck_spin_lock(bucket_lock); |
1545 | |
1546 | mem->vmp_next_m = bucket->page_list; |
1547 | bucket->page_list = VM_PAGE_PACK_PTR(mem); |
1548 | assert(mem == (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))); |
1549 | |
1550 | #if MACH_PAGE_HASH_STATS |
1551 | if (++bucket->cur_count > bucket->hi_count) |
1552 | bucket->hi_count = bucket->cur_count; |
1553 | #endif /* MACH_PAGE_HASH_STATS */ |
1554 | mem->vmp_hashed = TRUE; |
1555 | lck_spin_unlock(bucket_lock); |
1556 | } |
1557 | |
1558 | { |
1559 | unsigned int cache_attr; |
1560 | |
1561 | cache_attr = object->wimg_bits & VM_WIMG_MASK; |
1562 | |
1563 | if (cache_attr != VM_WIMG_USE_DEFAULT) { |
1564 | PMAP_SET_CACHE_ATTR(mem, object, cache_attr, batch_pmap_op); |
1565 | } |
1566 | } |
1567 | /* |
1568 | * Now link into the object's list of backed pages. |
1569 | */ |
1570 | vm_page_queue_enter(&object->memq, mem, vm_page_t, vmp_listq); |
1571 | object->memq_hint = mem; |
1572 | mem->vmp_tabled = TRUE; |
1573 | |
1574 | /* |
1575 | * Show that the object has one more resident page. |
1576 | */ |
1577 | |
1578 | object->resident_page_count++; |
1579 | if (VM_PAGE_WIRED(mem)) { |
1580 | assert(mem->vmp_wire_count > 0); |
1581 | VM_OBJECT_WIRED_PAGE_UPDATE_START(object); |
1582 | VM_OBJECT_WIRED_PAGE_ADD(object, mem); |
1583 | VM_OBJECT_WIRED_PAGE_UPDATE_END(object, tag); |
1584 | } |
1585 | assert(object->resident_page_count >= object->wired_page_count); |
1586 | |
1587 | if (batch_accounting == FALSE) { |
1588 | if (object->internal) { |
1589 | OSAddAtomic(1, &vm_page_internal_count); |
1590 | } else { |
1591 | OSAddAtomic(1, &vm_page_external_count); |
1592 | } |
1593 | } |
1594 | |
1595 | /* |
1596 | * It wouldn't make sense to insert a "reusable" page in |
1597 | * an object (the page would have been marked "reusable" only |
1598 | * at the time of a madvise(MADV_FREE_REUSABLE) if it was already |
1599 | * in the object at that time). |
1600 | * But a page could be inserted in a "all_reusable" object, if |
1601 | * something faults it in (a vm_read() from another task or a |
1602 | * "use-after-free" issue in user space, for example). It can |
1603 | * also happen if we're relocating a page from that object to |
1604 | * a different physical page during a physically-contiguous |
1605 | * allocation. |
1606 | */ |
1607 | assert(!mem->vmp_reusable); |
1608 | if (object->all_reusable) { |
1609 | OSAddAtomic(+1, &vm_page_stats_reusable.reusable_count); |
1610 | } |
1611 | |
1612 | if (object->purgable == VM_PURGABLE_DENY && |
1613 | ! object->vo_ledger_tag) { |
1614 | owner = TASK_NULL; |
1615 | } else { |
1616 | owner = VM_OBJECT_OWNER(object); |
1617 | vm_object_ledger_tag_ledgers(object, |
1618 | &ledger_idx_volatile, |
1619 | &ledger_idx_nonvolatile, |
1620 | &ledger_idx_volatile_compressed, |
1621 | &ledger_idx_nonvolatile_compressed, |
1622 | &do_footprint); |
1623 | } |
1624 | if (owner && |
1625 | (object->purgable == VM_PURGABLE_NONVOLATILE || |
1626 | object->purgable == VM_PURGABLE_DENY || |
1627 | VM_PAGE_WIRED(mem))) { |
1628 | |
1629 | if (delayed_ledger_update) |
1630 | *delayed_ledger_update += PAGE_SIZE; |
1631 | else { |
1632 | /* more non-volatile bytes */ |
1633 | ledger_credit(owner->ledger, |
1634 | ledger_idx_nonvolatile, |
1635 | PAGE_SIZE); |
1636 | if (do_footprint) { |
1637 | /* more footprint */ |
1638 | ledger_credit(owner->ledger, |
1639 | task_ledgers.phys_footprint, |
1640 | PAGE_SIZE); |
1641 | } |
1642 | } |
1643 | |
1644 | } else if (owner && |
1645 | (object->purgable == VM_PURGABLE_VOLATILE || |
1646 | object->purgable == VM_PURGABLE_EMPTY)) { |
1647 | assert(! VM_PAGE_WIRED(mem)); |
1648 | /* more volatile bytes */ |
1649 | ledger_credit(owner->ledger, |
1650 | ledger_idx_volatile, |
1651 | PAGE_SIZE); |
1652 | } |
1653 | |
1654 | if (object->purgable == VM_PURGABLE_VOLATILE) { |
1655 | if (VM_PAGE_WIRED(mem)) { |
1656 | OSAddAtomic(+1, &vm_page_purgeable_wired_count); |
1657 | } else { |
1658 | OSAddAtomic(+1, &vm_page_purgeable_count); |
1659 | } |
1660 | } else if (object->purgable == VM_PURGABLE_EMPTY && |
1661 | mem->vmp_q_state == VM_PAGE_ON_THROTTLED_Q) { |
1662 | /* |
1663 | * This page belongs to a purged VM object but hasn't |
1664 | * been purged (because it was "busy"). |
1665 | * It's in the "throttled" queue and hence not |
1666 | * visible to vm_pageout_scan(). Move it to a pageable |
1667 | * queue, so that it can eventually be reclaimed, instead |
1668 | * of lingering in the "empty" object. |
1669 | */ |
1670 | if (queues_lock_held == FALSE) |
1671 | vm_page_lockspin_queues(); |
1672 | vm_page_deactivate(mem); |
1673 | if (queues_lock_held == FALSE) |
1674 | vm_page_unlock_queues(); |
1675 | } |
1676 | |
1677 | #if VM_OBJECT_TRACKING_OP_MODIFIED |
1678 | if (vm_object_tracking_inited && |
1679 | object->internal && |
1680 | object->resident_page_count == 0 && |
1681 | object->pager == NULL && |
1682 | object->shadow != NULL && |
1683 | object->shadow->copy == object) { |
1684 | void *bt[VM_OBJECT_TRACKING_BTDEPTH]; |
1685 | int numsaved = 0; |
1686 | |
1687 | numsaved =OSBacktrace(bt, VM_OBJECT_TRACKING_BTDEPTH); |
1688 | btlog_add_entry(vm_object_tracking_btlog, |
1689 | object, |
1690 | VM_OBJECT_TRACKING_OP_MODIFIED, |
1691 | bt, |
1692 | numsaved); |
1693 | } |
1694 | #endif /* VM_OBJECT_TRACKING_OP_MODIFIED */ |
1695 | } |
1696 | |
1697 | /* |
1698 | * vm_page_replace: |
1699 | * |
1700 | * Exactly like vm_page_insert, except that we first |
1701 | * remove any existing page at the given offset in object. |
1702 | * |
1703 | * The object must be locked. |
1704 | */ |
1705 | void |
1706 | vm_page_replace( |
1707 | vm_page_t mem, |
1708 | vm_object_t object, |
1709 | vm_object_offset_t offset) |
1710 | { |
1711 | vm_page_bucket_t *bucket; |
1712 | vm_page_t found_m = VM_PAGE_NULL; |
1713 | lck_spin_t *bucket_lock; |
1714 | int hash_id; |
1715 | |
1716 | #if 0 |
1717 | /* |
1718 | * we don't hold the page queue lock |
1719 | * so this check isn't safe to make |
1720 | */ |
1721 | VM_PAGE_CHECK(mem); |
1722 | #endif |
1723 | vm_object_lock_assert_exclusive(object); |
1724 | #if DEBUG || VM_PAGE_CHECK_BUCKETS |
1725 | if (mem->vmp_tabled || mem->vmp_object) |
1726 | panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) " |
1727 | "already in (obj=%p,off=0x%llx)" , |
1728 | mem, object, offset, VM_PAGE_OBJECT(mem), mem->vmp_offset); |
1729 | #endif |
1730 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); |
1731 | |
1732 | assert(!VM_PAGE_PAGEABLE(mem)); |
1733 | |
1734 | /* |
1735 | * Record the object/offset pair in this page |
1736 | */ |
1737 | mem->vmp_object = VM_PAGE_PACK_OBJECT(object); |
1738 | mem->vmp_offset = offset; |
1739 | |
1740 | /* |
1741 | * Insert it into the object_object/offset hash table, |
1742 | * replacing any page that might have been there. |
1743 | */ |
1744 | |
1745 | hash_id = vm_page_hash(object, offset); |
1746 | bucket = &vm_page_buckets[hash_id]; |
1747 | bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; |
1748 | |
1749 | lck_spin_lock(bucket_lock); |
1750 | |
1751 | if (bucket->page_list) { |
1752 | vm_page_packed_t *mp = &bucket->page_list; |
1753 | vm_page_t m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp)); |
1754 | |
1755 | do { |
1756 | /* |
1757 | * compare packed object pointers |
1758 | */ |
1759 | if (m->vmp_object == mem->vmp_object && m->vmp_offset == offset) { |
1760 | /* |
1761 | * Remove old page from hash list |
1762 | */ |
1763 | *mp = m->vmp_next_m; |
1764 | m->vmp_hashed = FALSE; |
1765 | m->vmp_next_m = VM_PAGE_PACK_PTR(NULL); |
1766 | |
1767 | found_m = m; |
1768 | break; |
1769 | } |
1770 | mp = &m->vmp_next_m; |
1771 | } while ((m = (vm_page_t)(VM_PAGE_UNPACK_PTR(*mp)))); |
1772 | |
1773 | mem->vmp_next_m = bucket->page_list; |
1774 | } else { |
1775 | mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL); |
1776 | } |
1777 | /* |
1778 | * insert new page at head of hash list |
1779 | */ |
1780 | bucket->page_list = VM_PAGE_PACK_PTR(mem); |
1781 | mem->vmp_hashed = TRUE; |
1782 | |
1783 | lck_spin_unlock(bucket_lock); |
1784 | |
1785 | if (found_m) { |
1786 | /* |
1787 | * there was already a page at the specified |
1788 | * offset for this object... remove it from |
1789 | * the object and free it back to the free list |
1790 | */ |
1791 | vm_page_free_unlocked(found_m, FALSE); |
1792 | } |
1793 | vm_page_insert_internal(mem, object, offset, VM_KERN_MEMORY_NONE, FALSE, FALSE, FALSE, FALSE, NULL); |
1794 | } |
1795 | |
1796 | /* |
1797 | * vm_page_remove: [ internal use only ] |
1798 | * |
1799 | * Removes the given mem entry from the object/offset-page |
1800 | * table and the object page list. |
1801 | * |
1802 | * The object must be locked. |
1803 | */ |
1804 | |
1805 | void |
1806 | vm_page_remove( |
1807 | vm_page_t mem, |
1808 | boolean_t remove_from_hash) |
1809 | { |
1810 | vm_page_bucket_t *bucket; |
1811 | vm_page_t this; |
1812 | lck_spin_t *bucket_lock; |
1813 | int hash_id; |
1814 | task_t owner; |
1815 | vm_object_t m_object; |
1816 | int ledger_idx_volatile; |
1817 | int ledger_idx_nonvolatile; |
1818 | int ledger_idx_volatile_compressed; |
1819 | int ledger_idx_nonvolatile_compressed; |
1820 | int ; |
1821 | |
1822 | m_object = VM_PAGE_OBJECT(mem); |
1823 | |
1824 | XPR(XPR_VM_PAGE, |
1825 | "vm_page_remove, object 0x%X offset 0x%X page 0x%X\n" , |
1826 | m_object, mem->vmp_offset, |
1827 | mem, 0,0); |
1828 | |
1829 | vm_object_lock_assert_exclusive(m_object); |
1830 | assert(mem->vmp_tabled); |
1831 | assert(!mem->vmp_cleaning); |
1832 | assert(!mem->vmp_laundry); |
1833 | |
1834 | if (VM_PAGE_PAGEABLE(mem)) { |
1835 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
1836 | } |
1837 | #if 0 |
1838 | /* |
1839 | * we don't hold the page queue lock |
1840 | * so this check isn't safe to make |
1841 | */ |
1842 | VM_PAGE_CHECK(mem); |
1843 | #endif |
1844 | if (remove_from_hash == TRUE) { |
1845 | /* |
1846 | * Remove from the object_object/offset hash table |
1847 | */ |
1848 | hash_id = vm_page_hash(m_object, mem->vmp_offset); |
1849 | bucket = &vm_page_buckets[hash_id]; |
1850 | bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; |
1851 | |
1852 | lck_spin_lock(bucket_lock); |
1853 | |
1854 | if ((this = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list))) == mem) { |
1855 | /* optimize for common case */ |
1856 | |
1857 | bucket->page_list = mem->vmp_next_m; |
1858 | } else { |
1859 | vm_page_packed_t *prev; |
1860 | |
1861 | for (prev = &this->vmp_next_m; |
1862 | (this = (vm_page_t)(VM_PAGE_UNPACK_PTR(*prev))) != mem; |
1863 | prev = &this->vmp_next_m) |
1864 | continue; |
1865 | *prev = this->vmp_next_m; |
1866 | } |
1867 | #if MACH_PAGE_HASH_STATS |
1868 | bucket->cur_count--; |
1869 | #endif /* MACH_PAGE_HASH_STATS */ |
1870 | mem->vmp_hashed = FALSE; |
1871 | this->vmp_next_m = VM_PAGE_PACK_PTR(NULL); |
1872 | lck_spin_unlock(bucket_lock); |
1873 | } |
1874 | /* |
1875 | * Now remove from the object's list of backed pages. |
1876 | */ |
1877 | |
1878 | vm_page_remove_internal(mem); |
1879 | |
1880 | /* |
1881 | * And show that the object has one fewer resident |
1882 | * page. |
1883 | */ |
1884 | |
1885 | assert(m_object->resident_page_count > 0); |
1886 | m_object->resident_page_count--; |
1887 | |
1888 | if (m_object->internal) { |
1889 | #if DEBUG |
1890 | assert(vm_page_internal_count); |
1891 | #endif /* DEBUG */ |
1892 | |
1893 | OSAddAtomic(-1, &vm_page_internal_count); |
1894 | } else { |
1895 | assert(vm_page_external_count); |
1896 | OSAddAtomic(-1, &vm_page_external_count); |
1897 | |
1898 | if (mem->vmp_xpmapped) { |
1899 | assert(vm_page_xpmapped_external_count); |
1900 | OSAddAtomic(-1, &vm_page_xpmapped_external_count); |
1901 | } |
1902 | } |
1903 | if (!m_object->internal && |
1904 | m_object->cached_list.next && |
1905 | m_object->cached_list.prev) { |
1906 | if (m_object->resident_page_count == 0) |
1907 | vm_object_cache_remove(m_object); |
1908 | } |
1909 | |
1910 | if (VM_PAGE_WIRED(mem)) { |
1911 | assert(mem->vmp_wire_count > 0); |
1912 | VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); |
1913 | VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); |
1914 | VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); |
1915 | } |
1916 | assert(m_object->resident_page_count >= |
1917 | m_object->wired_page_count); |
1918 | if (mem->vmp_reusable) { |
1919 | assert(m_object->reusable_page_count > 0); |
1920 | m_object->reusable_page_count--; |
1921 | assert(m_object->reusable_page_count <= |
1922 | m_object->resident_page_count); |
1923 | mem->vmp_reusable = FALSE; |
1924 | OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count); |
1925 | vm_page_stats_reusable.reused_remove++; |
1926 | } else if (m_object->all_reusable) { |
1927 | OSAddAtomic(-1, &vm_page_stats_reusable.reusable_count); |
1928 | vm_page_stats_reusable.reused_remove++; |
1929 | } |
1930 | |
1931 | if (m_object->purgable == VM_PURGABLE_DENY && |
1932 | ! m_object->vo_ledger_tag) { |
1933 | owner = TASK_NULL; |
1934 | } else { |
1935 | owner = VM_OBJECT_OWNER(m_object); |
1936 | vm_object_ledger_tag_ledgers(m_object, |
1937 | &ledger_idx_volatile, |
1938 | &ledger_idx_nonvolatile, |
1939 | &ledger_idx_volatile_compressed, |
1940 | &ledger_idx_nonvolatile_compressed, |
1941 | &do_footprint); |
1942 | } |
1943 | if (owner && |
1944 | (m_object->purgable == VM_PURGABLE_NONVOLATILE || |
1945 | m_object->purgable == VM_PURGABLE_DENY || |
1946 | VM_PAGE_WIRED(mem))) { |
1947 | /* less non-volatile bytes */ |
1948 | ledger_debit(owner->ledger, |
1949 | ledger_idx_nonvolatile, |
1950 | PAGE_SIZE); |
1951 | if (do_footprint) { |
1952 | /* less footprint */ |
1953 | ledger_debit(owner->ledger, |
1954 | task_ledgers.phys_footprint, |
1955 | PAGE_SIZE); |
1956 | } |
1957 | } else if (owner && |
1958 | (m_object->purgable == VM_PURGABLE_VOLATILE || |
1959 | m_object->purgable == VM_PURGABLE_EMPTY)) { |
1960 | assert(! VM_PAGE_WIRED(mem)); |
1961 | /* less volatile bytes */ |
1962 | ledger_debit(owner->ledger, |
1963 | ledger_idx_volatile, |
1964 | PAGE_SIZE); |
1965 | } |
1966 | if (m_object->purgable == VM_PURGABLE_VOLATILE) { |
1967 | if (VM_PAGE_WIRED(mem)) { |
1968 | assert(vm_page_purgeable_wired_count > 0); |
1969 | OSAddAtomic(-1, &vm_page_purgeable_wired_count); |
1970 | } else { |
1971 | assert(vm_page_purgeable_count > 0); |
1972 | OSAddAtomic(-1, &vm_page_purgeable_count); |
1973 | } |
1974 | } |
1975 | |
1976 | if (m_object->set_cache_attr == TRUE) |
1977 | pmap_set_cache_attributes(VM_PAGE_GET_PHYS_PAGE(mem), 0); |
1978 | |
1979 | mem->vmp_tabled = FALSE; |
1980 | mem->vmp_object = 0; |
1981 | mem->vmp_offset = (vm_object_offset_t) -1; |
1982 | } |
1983 | |
1984 | |
1985 | /* |
1986 | * vm_page_lookup: |
1987 | * |
1988 | * Returns the page associated with the object/offset |
1989 | * pair specified; if none is found, VM_PAGE_NULL is returned. |
1990 | * |
1991 | * The object must be locked. No side effects. |
1992 | */ |
1993 | |
1994 | #define VM_PAGE_HASH_LOOKUP_THRESHOLD 10 |
1995 | |
1996 | #if DEBUG_VM_PAGE_LOOKUP |
1997 | |
1998 | struct { |
1999 | uint64_t vpl_total; |
2000 | uint64_t vpl_empty_obj; |
2001 | uint64_t vpl_bucket_NULL; |
2002 | uint64_t vpl_hit_hint; |
2003 | uint64_t vpl_hit_hint_next; |
2004 | uint64_t vpl_hit_hint_prev; |
2005 | uint64_t vpl_fast; |
2006 | uint64_t vpl_slow; |
2007 | uint64_t vpl_hit; |
2008 | uint64_t vpl_miss; |
2009 | |
2010 | uint64_t vpl_fast_elapsed; |
2011 | uint64_t vpl_slow_elapsed; |
2012 | } vm_page_lookup_stats __attribute__((aligned(8))); |
2013 | |
2014 | #endif |
2015 | |
2016 | #define KDP_VM_PAGE_WALK_MAX 1000 |
2017 | |
2018 | vm_page_t |
2019 | kdp_vm_page_lookup( |
2020 | vm_object_t object, |
2021 | vm_object_offset_t offset) |
2022 | { |
2023 | vm_page_t cur_page; |
2024 | int num_traversed = 0; |
2025 | |
2026 | if (not_in_kdp) { |
2027 | panic("panic: kdp_vm_page_lookup done outside of kernel debugger" ); |
2028 | } |
2029 | |
2030 | vm_page_queue_iterate(&object->memq, cur_page, vm_page_t, vmp_listq) { |
2031 | if (cur_page->vmp_offset == offset) { |
2032 | return cur_page; |
2033 | } |
2034 | num_traversed++; |
2035 | |
2036 | if (num_traversed >= KDP_VM_PAGE_WALK_MAX) { |
2037 | return VM_PAGE_NULL; |
2038 | } |
2039 | } |
2040 | |
2041 | return VM_PAGE_NULL; |
2042 | } |
2043 | |
2044 | vm_page_t |
2045 | vm_page_lookup( |
2046 | vm_object_t object, |
2047 | vm_object_offset_t offset) |
2048 | { |
2049 | vm_page_t mem; |
2050 | vm_page_bucket_t *bucket; |
2051 | vm_page_queue_entry_t qe; |
2052 | lck_spin_t *bucket_lock = NULL; |
2053 | int hash_id; |
2054 | #if DEBUG_VM_PAGE_LOOKUP |
2055 | uint64_t start, elapsed; |
2056 | |
2057 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_total); |
2058 | #endif |
2059 | vm_object_lock_assert_held(object); |
2060 | |
2061 | if (object->resident_page_count == 0) { |
2062 | #if DEBUG_VM_PAGE_LOOKUP |
2063 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_empty_obj); |
2064 | #endif |
2065 | return (VM_PAGE_NULL); |
2066 | } |
2067 | |
2068 | mem = object->memq_hint; |
2069 | |
2070 | if (mem != VM_PAGE_NULL) { |
2071 | assert(VM_PAGE_OBJECT(mem) == object); |
2072 | |
2073 | if (mem->vmp_offset == offset) { |
2074 | #if DEBUG_VM_PAGE_LOOKUP |
2075 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint); |
2076 | #endif |
2077 | return (mem); |
2078 | } |
2079 | qe = (vm_page_queue_entry_t)vm_page_queue_next(&mem->vmp_listq); |
2080 | |
2081 | if (! vm_page_queue_end(&object->memq, qe)) { |
2082 | vm_page_t next_page; |
2083 | |
2084 | next_page = (vm_page_t)((uintptr_t)qe); |
2085 | assert(VM_PAGE_OBJECT(next_page) == object); |
2086 | |
2087 | if (next_page->vmp_offset == offset) { |
2088 | object->memq_hint = next_page; /* new hint */ |
2089 | #if DEBUG_VM_PAGE_LOOKUP |
2090 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_next); |
2091 | #endif |
2092 | return (next_page); |
2093 | } |
2094 | } |
2095 | qe = (vm_page_queue_entry_t)vm_page_queue_prev(&mem->vmp_listq); |
2096 | |
2097 | if (! vm_page_queue_end(&object->memq, qe)) { |
2098 | vm_page_t prev_page; |
2099 | |
2100 | prev_page = (vm_page_t)((uintptr_t)qe); |
2101 | assert(VM_PAGE_OBJECT(prev_page) == object); |
2102 | |
2103 | if (prev_page->vmp_offset == offset) { |
2104 | object->memq_hint = prev_page; /* new hint */ |
2105 | #if DEBUG_VM_PAGE_LOOKUP |
2106 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit_hint_prev); |
2107 | #endif |
2108 | return (prev_page); |
2109 | } |
2110 | } |
2111 | } |
2112 | /* |
2113 | * Search the hash table for this object/offset pair |
2114 | */ |
2115 | hash_id = vm_page_hash(object, offset); |
2116 | bucket = &vm_page_buckets[hash_id]; |
2117 | |
2118 | /* |
2119 | * since we hold the object lock, we are guaranteed that no |
2120 | * new pages can be inserted into this object... this in turn |
2121 | * guarantess that the page we're looking for can't exist |
2122 | * if the bucket it hashes to is currently NULL even when looked |
2123 | * at outside the scope of the hash bucket lock... this is a |
2124 | * really cheap optimiztion to avoid taking the lock |
2125 | */ |
2126 | if (!bucket->page_list) { |
2127 | #if DEBUG_VM_PAGE_LOOKUP |
2128 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_bucket_NULL); |
2129 | #endif |
2130 | return (VM_PAGE_NULL); |
2131 | } |
2132 | |
2133 | #if DEBUG_VM_PAGE_LOOKUP |
2134 | start = mach_absolute_time(); |
2135 | #endif |
2136 | if (object->resident_page_count <= VM_PAGE_HASH_LOOKUP_THRESHOLD) { |
2137 | /* |
2138 | * on average, it's roughly 3 times faster to run a short memq list |
2139 | * than to take the spin lock and go through the hash list |
2140 | */ |
2141 | mem = (vm_page_t)vm_page_queue_first(&object->memq); |
2142 | |
2143 | while (!vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) { |
2144 | |
2145 | if (mem->vmp_offset == offset) |
2146 | break; |
2147 | |
2148 | mem = (vm_page_t)vm_page_queue_next(&mem->vmp_listq); |
2149 | } |
2150 | if (vm_page_queue_end(&object->memq, (vm_page_queue_entry_t)mem)) |
2151 | mem = NULL; |
2152 | } else { |
2153 | vm_page_object_t packed_object; |
2154 | |
2155 | packed_object = VM_PAGE_PACK_OBJECT(object); |
2156 | |
2157 | bucket_lock = &vm_page_bucket_locks[hash_id / BUCKETS_PER_LOCK]; |
2158 | |
2159 | lck_spin_lock(bucket_lock); |
2160 | |
2161 | for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); |
2162 | mem != VM_PAGE_NULL; |
2163 | mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m))) { |
2164 | #if 0 |
2165 | /* |
2166 | * we don't hold the page queue lock |
2167 | * so this check isn't safe to make |
2168 | */ |
2169 | VM_PAGE_CHECK(mem); |
2170 | #endif |
2171 | if ((mem->vmp_object == packed_object) && (mem->vmp_offset == offset)) |
2172 | break; |
2173 | } |
2174 | lck_spin_unlock(bucket_lock); |
2175 | } |
2176 | |
2177 | #if DEBUG_VM_PAGE_LOOKUP |
2178 | elapsed = mach_absolute_time() - start; |
2179 | |
2180 | if (bucket_lock) { |
2181 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_slow); |
2182 | OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_slow_elapsed); |
2183 | } else { |
2184 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_fast); |
2185 | OSAddAtomic64(elapsed, &vm_page_lookup_stats.vpl_fast_elapsed); |
2186 | } |
2187 | if (mem != VM_PAGE_NULL) |
2188 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_hit); |
2189 | else |
2190 | OSAddAtomic64(1, &vm_page_lookup_stats.vpl_miss); |
2191 | #endif |
2192 | if (mem != VM_PAGE_NULL) { |
2193 | assert(VM_PAGE_OBJECT(mem) == object); |
2194 | |
2195 | object->memq_hint = mem; |
2196 | } |
2197 | return (mem); |
2198 | } |
2199 | |
2200 | |
2201 | /* |
2202 | * vm_page_rename: |
2203 | * |
2204 | * Move the given memory entry from its |
2205 | * current object to the specified target object/offset. |
2206 | * |
2207 | * The object must be locked. |
2208 | */ |
2209 | void |
2210 | vm_page_rename( |
2211 | vm_page_t mem, |
2212 | vm_object_t new_object, |
2213 | vm_object_offset_t new_offset) |
2214 | { |
2215 | boolean_t internal_to_external, external_to_internal; |
2216 | vm_tag_t tag; |
2217 | vm_object_t m_object; |
2218 | |
2219 | m_object = VM_PAGE_OBJECT(mem); |
2220 | |
2221 | assert(m_object != new_object); |
2222 | assert(m_object); |
2223 | |
2224 | XPR(XPR_VM_PAGE, |
2225 | "vm_page_rename, new object 0x%X, offset 0x%X page 0x%X\n" , |
2226 | new_object, new_offset, |
2227 | mem, 0,0); |
2228 | |
2229 | /* |
2230 | * Changes to mem->vmp_object require the page lock because |
2231 | * the pageout daemon uses that lock to get the object. |
2232 | */ |
2233 | vm_page_lockspin_queues(); |
2234 | |
2235 | internal_to_external = FALSE; |
2236 | external_to_internal = FALSE; |
2237 | |
2238 | if (mem->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q) { |
2239 | /* |
2240 | * it's much easier to get the vm_page_pageable_xxx accounting correct |
2241 | * if we first move the page to the active queue... it's going to end |
2242 | * up there anyway, and we don't do vm_page_rename's frequently enough |
2243 | * for this to matter. |
2244 | */ |
2245 | vm_page_queues_remove(mem, FALSE); |
2246 | vm_page_activate(mem); |
2247 | } |
2248 | if (VM_PAGE_PAGEABLE(mem)) { |
2249 | if (m_object->internal && !new_object->internal) { |
2250 | internal_to_external = TRUE; |
2251 | } |
2252 | if (!m_object->internal && new_object->internal) { |
2253 | external_to_internal = TRUE; |
2254 | } |
2255 | } |
2256 | |
2257 | tag = m_object->wire_tag; |
2258 | vm_page_remove(mem, TRUE); |
2259 | vm_page_insert_internal(mem, new_object, new_offset, tag, TRUE, TRUE, FALSE, FALSE, NULL); |
2260 | |
2261 | if (internal_to_external) { |
2262 | vm_page_pageable_internal_count--; |
2263 | vm_page_pageable_external_count++; |
2264 | } else if (external_to_internal) { |
2265 | vm_page_pageable_external_count--; |
2266 | vm_page_pageable_internal_count++; |
2267 | } |
2268 | |
2269 | vm_page_unlock_queues(); |
2270 | } |
2271 | |
2272 | /* |
2273 | * vm_page_init: |
2274 | * |
2275 | * Initialize the fields in a new page. |
2276 | * This takes a structure with random values and initializes it |
2277 | * so that it can be given to vm_page_release or vm_page_insert. |
2278 | */ |
2279 | void |
2280 | vm_page_init( |
2281 | vm_page_t mem, |
2282 | ppnum_t phys_page, |
2283 | boolean_t lopage) |
2284 | { |
2285 | assert(phys_page); |
2286 | |
2287 | #if DEBUG |
2288 | if ((phys_page != vm_page_fictitious_addr) && (phys_page != vm_page_guard_addr)) { |
2289 | if (!(pmap_valid_page(phys_page))) { |
2290 | panic("vm_page_init: non-DRAM phys_page 0x%x\n" , phys_page); |
2291 | } |
2292 | } |
2293 | #endif |
2294 | *mem = vm_page_template; |
2295 | |
2296 | VM_PAGE_SET_PHYS_PAGE(mem, phys_page); |
2297 | #if 0 |
2298 | /* |
2299 | * we're leaving this turned off for now... currently pages |
2300 | * come off the free list and are either immediately dirtied/referenced |
2301 | * due to zero-fill or COW faults, or are used to read or write files... |
2302 | * in the file I/O case, the UPL mechanism takes care of clearing |
2303 | * the state of the HW ref/mod bits in a somewhat fragile way. |
2304 | * Since we may change the way this works in the future (to toughen it up), |
2305 | * I'm leaving this as a reminder of where these bits could get cleared |
2306 | */ |
2307 | |
2308 | /* |
2309 | * make sure both the h/w referenced and modified bits are |
2310 | * clear at this point... we are especially dependent on |
2311 | * not finding a 'stale' h/w modified in a number of spots |
2312 | * once this page goes back into use |
2313 | */ |
2314 | pmap_clear_refmod(phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED); |
2315 | #endif |
2316 | mem->vmp_lopage = lopage; |
2317 | } |
2318 | |
2319 | /* |
2320 | * vm_page_grab_fictitious: |
2321 | * |
2322 | * Remove a fictitious page from the free list. |
2323 | * Returns VM_PAGE_NULL if there are no free pages. |
2324 | */ |
2325 | int c_vm_page_grab_fictitious = 0; |
2326 | int c_vm_page_grab_fictitious_failed = 0; |
2327 | int c_vm_page_release_fictitious = 0; |
2328 | int c_vm_page_more_fictitious = 0; |
2329 | |
2330 | vm_page_t |
2331 | vm_page_grab_fictitious_common( |
2332 | ppnum_t phys_addr) |
2333 | { |
2334 | vm_page_t m; |
2335 | |
2336 | if ((m = (vm_page_t)zget(vm_page_zone))) { |
2337 | |
2338 | vm_page_init(m, phys_addr, FALSE); |
2339 | m->vmp_fictitious = TRUE; |
2340 | |
2341 | c_vm_page_grab_fictitious++; |
2342 | } else |
2343 | c_vm_page_grab_fictitious_failed++; |
2344 | |
2345 | return m; |
2346 | } |
2347 | |
2348 | vm_page_t |
2349 | vm_page_grab_fictitious(void) |
2350 | { |
2351 | return vm_page_grab_fictitious_common(vm_page_fictitious_addr); |
2352 | } |
2353 | |
2354 | int vm_guard_count; |
2355 | |
2356 | |
2357 | vm_page_t |
2358 | vm_page_grab_guard(void) |
2359 | { |
2360 | vm_page_t page; |
2361 | page = vm_page_grab_fictitious_common(vm_page_guard_addr); |
2362 | if (page) OSAddAtomic(1, &vm_guard_count); |
2363 | return page; |
2364 | } |
2365 | |
2366 | |
2367 | /* |
2368 | * vm_page_release_fictitious: |
2369 | * |
2370 | * Release a fictitious page to the zone pool |
2371 | */ |
2372 | void |
2373 | vm_page_release_fictitious( |
2374 | vm_page_t m) |
2375 | { |
2376 | assert((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || (m->vmp_q_state == VM_PAGE_IS_WIRED)); |
2377 | assert(m->vmp_fictitious); |
2378 | assert(VM_PAGE_GET_PHYS_PAGE(m) == vm_page_fictitious_addr || |
2379 | VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr); |
2380 | |
2381 | |
2382 | if (VM_PAGE_GET_PHYS_PAGE(m) == vm_page_guard_addr) OSAddAtomic(-1, &vm_guard_count); |
2383 | |
2384 | c_vm_page_release_fictitious++; |
2385 | |
2386 | zfree(vm_page_zone, m); |
2387 | } |
2388 | |
2389 | /* |
2390 | * vm_page_more_fictitious: |
2391 | * |
2392 | * Add more fictitious pages to the zone. |
2393 | * Allowed to block. This routine is way intimate |
2394 | * with the zones code, for several reasons: |
2395 | * 1. we need to carve some page structures out of physical |
2396 | * memory before zones work, so they _cannot_ come from |
2397 | * the zone_map. |
2398 | * 2. the zone needs to be collectable in order to prevent |
2399 | * growth without bound. These structures are used by |
2400 | * the device pager (by the hundreds and thousands), as |
2401 | * private pages for pageout, and as blocking pages for |
2402 | * pagein. Temporary bursts in demand should not result in |
2403 | * permanent allocation of a resource. |
2404 | * 3. To smooth allocation humps, we allocate single pages |
2405 | * with kernel_memory_allocate(), and cram them into the |
2406 | * zone. |
2407 | */ |
2408 | |
2409 | void vm_page_more_fictitious(void) |
2410 | { |
2411 | vm_offset_t addr; |
2412 | kern_return_t retval; |
2413 | |
2414 | c_vm_page_more_fictitious++; |
2415 | |
2416 | /* |
2417 | * Allocate a single page from the zone_map. Do not wait if no physical |
2418 | * pages are immediately available, and do not zero the space. We need |
2419 | * our own blocking lock here to prevent having multiple, |
2420 | * simultaneous requests from piling up on the zone_map lock. Exactly |
2421 | * one (of our) threads should be potentially waiting on the map lock. |
2422 | * If winner is not vm-privileged, then the page allocation will fail, |
2423 | * and it will temporarily block here in the vm_page_wait(). |
2424 | */ |
2425 | lck_mtx_lock(&vm_page_alloc_lock); |
2426 | /* |
2427 | * If another thread allocated space, just bail out now. |
2428 | */ |
2429 | if (zone_free_count(vm_page_zone) > 5) { |
2430 | /* |
2431 | * The number "5" is a small number that is larger than the |
2432 | * number of fictitious pages that any single caller will |
2433 | * attempt to allocate. Otherwise, a thread will attempt to |
2434 | * acquire a fictitious page (vm_page_grab_fictitious), fail, |
2435 | * release all of the resources and locks already acquired, |
2436 | * and then call this routine. This routine finds the pages |
2437 | * that the caller released, so fails to allocate new space. |
2438 | * The process repeats infinitely. The largest known number |
2439 | * of fictitious pages required in this manner is 2. 5 is |
2440 | * simply a somewhat larger number. |
2441 | */ |
2442 | lck_mtx_unlock(&vm_page_alloc_lock); |
2443 | return; |
2444 | } |
2445 | |
2446 | retval = kernel_memory_allocate(zone_map, |
2447 | &addr, PAGE_SIZE, 0, |
2448 | KMA_KOBJECT|KMA_NOPAGEWAIT, VM_KERN_MEMORY_ZONE); |
2449 | if (retval != KERN_SUCCESS) { |
2450 | /* |
2451 | * No page was available. Drop the |
2452 | * lock to give another thread a chance at it, and |
2453 | * wait for the pageout daemon to make progress. |
2454 | */ |
2455 | lck_mtx_unlock(&vm_page_alloc_lock); |
2456 | vm_page_wait(THREAD_UNINT); |
2457 | return; |
2458 | } |
2459 | |
2460 | zcram(vm_page_zone, addr, PAGE_SIZE); |
2461 | |
2462 | lck_mtx_unlock(&vm_page_alloc_lock); |
2463 | } |
2464 | |
2465 | |
2466 | /* |
2467 | * vm_pool_low(): |
2468 | * |
2469 | * Return true if it is not likely that a non-vm_privileged thread |
2470 | * can get memory without blocking. Advisory only, since the |
2471 | * situation may change under us. |
2472 | */ |
2473 | int |
2474 | vm_pool_low(void) |
2475 | { |
2476 | /* No locking, at worst we will fib. */ |
2477 | return( vm_page_free_count <= vm_page_free_reserved ); |
2478 | } |
2479 | |
2480 | boolean_t vm_darkwake_mode = FALSE; |
2481 | |
2482 | /* |
2483 | * vm_update_darkwake_mode(): |
2484 | * |
2485 | * Tells the VM that the system is in / out of darkwake. |
2486 | * |
2487 | * Today, the VM only lowers/raises the background queue target |
2488 | * so as to favor consuming more/less background pages when |
2489 | * darwake is ON/OFF. |
2490 | * |
2491 | * We might need to do more things in the future. |
2492 | */ |
2493 | |
2494 | void |
2495 | vm_update_darkwake_mode(boolean_t darkwake_mode) |
2496 | { |
2497 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); |
2498 | |
2499 | vm_page_lockspin_queues(); |
2500 | |
2501 | if (vm_darkwake_mode == darkwake_mode) { |
2502 | /* |
2503 | * No change. |
2504 | */ |
2505 | vm_page_unlock_queues(); |
2506 | return; |
2507 | } |
2508 | |
2509 | vm_darkwake_mode = darkwake_mode; |
2510 | |
2511 | if (vm_darkwake_mode == TRUE) { |
2512 | #if CONFIG_BACKGROUND_QUEUE |
2513 | |
2514 | /* save background target to restore later */ |
2515 | vm_page_background_target_snapshot = vm_page_background_target; |
2516 | |
2517 | /* target is set to 0...no protection for background pages */ |
2518 | vm_page_background_target = 0; |
2519 | |
2520 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
2521 | |
2522 | } else if (vm_darkwake_mode == FALSE) { |
2523 | #if CONFIG_BACKGROUND_QUEUE |
2524 | |
2525 | if (vm_page_background_target_snapshot) { |
2526 | vm_page_background_target = vm_page_background_target_snapshot; |
2527 | } |
2528 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
2529 | } |
2530 | vm_page_unlock_queues(); |
2531 | } |
2532 | |
2533 | #if CONFIG_BACKGROUND_QUEUE |
2534 | |
2535 | void |
2536 | vm_page_update_background_state(vm_page_t mem) |
2537 | { |
2538 | if (vm_page_background_mode == VM_PAGE_BG_DISABLED) |
2539 | return; |
2540 | |
2541 | if (mem->vmp_in_background == FALSE) |
2542 | return; |
2543 | |
2544 | task_t my_task = current_task(); |
2545 | |
2546 | if (my_task) { |
2547 | if (task_get_darkwake_mode(my_task)) { |
2548 | return; |
2549 | } |
2550 | } |
2551 | |
2552 | #if BACKGROUNDQ_BASED_ON_QOS |
2553 | if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_QOS) <= THREAD_QOS_LEGACY) |
2554 | return; |
2555 | #else |
2556 | if (my_task) { |
2557 | if (proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG)) |
2558 | return; |
2559 | } |
2560 | #endif |
2561 | vm_page_lockspin_queues(); |
2562 | |
2563 | mem->vmp_in_background = FALSE; |
2564 | vm_page_background_promoted_count++; |
2565 | |
2566 | vm_page_remove_from_backgroundq(mem); |
2567 | |
2568 | vm_page_unlock_queues(); |
2569 | } |
2570 | |
2571 | |
2572 | void |
2573 | vm_page_assign_background_state(vm_page_t mem) |
2574 | { |
2575 | if (vm_page_background_mode == VM_PAGE_BG_DISABLED) |
2576 | return; |
2577 | |
2578 | task_t my_task = current_task(); |
2579 | |
2580 | if (my_task) { |
2581 | if (task_get_darkwake_mode(my_task)) { |
2582 | mem->vmp_in_background = TRUE; |
2583 | return; |
2584 | } |
2585 | } |
2586 | |
2587 | #if BACKGROUNDQ_BASED_ON_QOS |
2588 | if (proc_get_effective_thread_policy(current_thread(), TASK_POLICY_QOS) <= THREAD_QOS_LEGACY) |
2589 | mem->vmp_in_background = TRUE; |
2590 | else |
2591 | mem->vmp_in_background = FALSE; |
2592 | #else |
2593 | if (my_task) |
2594 | mem->vmp_in_background = proc_get_effective_task_policy(my_task, TASK_POLICY_DARWIN_BG); |
2595 | #endif |
2596 | } |
2597 | |
2598 | |
2599 | void |
2600 | vm_page_remove_from_backgroundq( |
2601 | vm_page_t mem) |
2602 | { |
2603 | vm_object_t m_object; |
2604 | |
2605 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
2606 | |
2607 | if (mem->vmp_on_backgroundq) { |
2608 | vm_page_queue_remove(&vm_page_queue_background, mem, vm_page_t, vmp_backgroundq); |
2609 | |
2610 | mem->vmp_backgroundq.next = 0; |
2611 | mem->vmp_backgroundq.prev = 0; |
2612 | mem->vmp_on_backgroundq = FALSE; |
2613 | |
2614 | vm_page_background_count--; |
2615 | |
2616 | m_object = VM_PAGE_OBJECT(mem); |
2617 | |
2618 | if (m_object->internal) |
2619 | vm_page_background_internal_count--; |
2620 | else |
2621 | vm_page_background_external_count--; |
2622 | } else { |
2623 | assert(VM_PAGE_UNPACK_PTR(mem->vmp_backgroundq.next) == (uintptr_t)NULL && |
2624 | VM_PAGE_UNPACK_PTR(mem->vmp_backgroundq.prev) == (uintptr_t)NULL); |
2625 | } |
2626 | } |
2627 | |
2628 | |
2629 | void |
2630 | vm_page_add_to_backgroundq( |
2631 | vm_page_t mem, |
2632 | boolean_t first) |
2633 | { |
2634 | vm_object_t m_object; |
2635 | |
2636 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
2637 | |
2638 | if (vm_page_background_mode == VM_PAGE_BG_DISABLED) |
2639 | return; |
2640 | |
2641 | if (mem->vmp_on_backgroundq == FALSE) { |
2642 | |
2643 | m_object = VM_PAGE_OBJECT(mem); |
2644 | |
2645 | if (vm_page_background_exclude_external && !m_object->internal) |
2646 | return; |
2647 | |
2648 | if (first == TRUE) |
2649 | vm_page_queue_enter_first(&vm_page_queue_background, mem, vm_page_t, vmp_backgroundq); |
2650 | else |
2651 | vm_page_queue_enter(&vm_page_queue_background, mem, vm_page_t, vmp_backgroundq); |
2652 | mem->vmp_on_backgroundq = TRUE; |
2653 | |
2654 | vm_page_background_count++; |
2655 | |
2656 | if (m_object->internal) |
2657 | vm_page_background_internal_count++; |
2658 | else |
2659 | vm_page_background_external_count++; |
2660 | } |
2661 | } |
2662 | |
2663 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
2664 | |
2665 | /* |
2666 | * this is an interface to support bring-up of drivers |
2667 | * on platforms with physical memory > 4G... |
2668 | */ |
2669 | int vm_himemory_mode = 2; |
2670 | |
2671 | |
2672 | /* |
2673 | * this interface exists to support hardware controllers |
2674 | * incapable of generating DMAs with more than 32 bits |
2675 | * of address on platforms with physical memory > 4G... |
2676 | */ |
2677 | unsigned int vm_lopages_allocated_q = 0; |
2678 | unsigned int vm_lopages_allocated_cpm_success = 0; |
2679 | unsigned int vm_lopages_allocated_cpm_failed = 0; |
2680 | vm_page_queue_head_t vm_lopage_queue_free __attribute__((aligned(VM_PACKED_POINTER_ALIGNMENT))); |
2681 | |
2682 | vm_page_t |
2683 | vm_page_grablo(void) |
2684 | { |
2685 | vm_page_t mem; |
2686 | |
2687 | if (vm_lopage_needed == FALSE) |
2688 | return (vm_page_grab()); |
2689 | |
2690 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
2691 | |
2692 | if ( !vm_page_queue_empty(&vm_lopage_queue_free)) { |
2693 | vm_page_queue_remove_first(&vm_lopage_queue_free, |
2694 | mem, |
2695 | vm_page_t, |
2696 | vmp_pageq); |
2697 | assert(vm_lopage_free_count); |
2698 | assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q); |
2699 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
2700 | |
2701 | vm_lopage_free_count--; |
2702 | vm_lopages_allocated_q++; |
2703 | |
2704 | if (vm_lopage_free_count < vm_lopage_lowater) |
2705 | vm_lopage_refill = TRUE; |
2706 | |
2707 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2708 | |
2709 | #if CONFIG_BACKGROUND_QUEUE |
2710 | vm_page_assign_background_state(mem); |
2711 | #endif |
2712 | } else { |
2713 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2714 | |
2715 | if (cpm_allocate(PAGE_SIZE, &mem, atop(0xffffffff), 0, FALSE, KMA_LOMEM) != KERN_SUCCESS) { |
2716 | |
2717 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
2718 | vm_lopages_allocated_cpm_failed++; |
2719 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2720 | |
2721 | return (VM_PAGE_NULL); |
2722 | } |
2723 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
2724 | |
2725 | mem->vmp_busy = TRUE; |
2726 | |
2727 | vm_page_lockspin_queues(); |
2728 | |
2729 | mem->vmp_gobbled = FALSE; |
2730 | vm_page_gobble_count--; |
2731 | vm_page_wire_count--; |
2732 | |
2733 | vm_lopages_allocated_cpm_success++; |
2734 | vm_page_unlock_queues(); |
2735 | } |
2736 | assert(mem->vmp_busy); |
2737 | assert(!mem->vmp_pmapped); |
2738 | assert(!mem->vmp_wpmapped); |
2739 | assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); |
2740 | |
2741 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
2742 | |
2743 | disable_preemption(); |
2744 | PROCESSOR_DATA(current_processor(), page_grab_count) += 1; |
2745 | VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, 0, 1, 0, 0); |
2746 | enable_preemption(); |
2747 | |
2748 | return (mem); |
2749 | } |
2750 | |
2751 | |
2752 | /* |
2753 | * vm_page_grab: |
2754 | * |
2755 | * first try to grab a page from the per-cpu free list... |
2756 | * this must be done while pre-emption is disabled... if |
2757 | * a page is available, we're done... |
2758 | * if no page is available, grab the vm_page_queue_free_lock |
2759 | * and see if current number of free pages would allow us |
2760 | * to grab at least 1... if not, return VM_PAGE_NULL as before... |
2761 | * if there are pages available, disable preemption and |
2762 | * recheck the state of the per-cpu free list... we could |
2763 | * have been preempted and moved to a different cpu, or |
2764 | * some other thread could have re-filled it... if still |
2765 | * empty, figure out how many pages we can steal from the |
2766 | * global free queue and move to the per-cpu queue... |
2767 | * return 1 of these pages when done... only wakeup the |
2768 | * pageout_scan thread if we moved pages from the global |
2769 | * list... no need for the wakeup if we've satisfied the |
2770 | * request from the per-cpu queue. |
2771 | */ |
2772 | |
2773 | #if CONFIG_SECLUDED_MEMORY |
2774 | vm_page_t vm_page_grab_secluded(void); |
2775 | #endif /* CONFIG_SECLUDED_MEMORY */ |
2776 | |
2777 | vm_page_t |
2778 | vm_page_grab(void) |
2779 | { |
2780 | return vm_page_grab_options(0); |
2781 | } |
2782 | |
2783 | #if HIBERNATION |
2784 | boolean_t hibernate_rebuild_needed = FALSE; |
2785 | #endif /* HIBERNATION */ |
2786 | |
2787 | vm_page_t |
2788 | vm_page_grab_options( |
2789 | int grab_options) |
2790 | { |
2791 | vm_page_t mem; |
2792 | |
2793 | disable_preemption(); |
2794 | |
2795 | if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) { |
2796 | return_page_from_cpu_list: |
2797 | assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); |
2798 | |
2799 | #if HIBERNATION |
2800 | if (hibernate_rebuild_needed) { |
2801 | panic("%s:%d should not modify cpu->free_pages while hibernating" , __FUNCTION__, __LINE__); |
2802 | } |
2803 | #endif /* HIBERNATION */ |
2804 | PROCESSOR_DATA(current_processor(), page_grab_count) += 1; |
2805 | PROCESSOR_DATA(current_processor(), free_pages) = mem->vmp_snext; |
2806 | VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); |
2807 | |
2808 | enable_preemption(); |
2809 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
2810 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
2811 | |
2812 | assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); |
2813 | assert(mem->vmp_tabled == FALSE); |
2814 | assert(mem->vmp_object == 0); |
2815 | assert(!mem->vmp_laundry); |
2816 | assertf(pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem)), "page = 0x%llx" , (uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)); |
2817 | assert(mem->vmp_busy); |
2818 | assert(!mem->vmp_pmapped); |
2819 | assert(!mem->vmp_wpmapped); |
2820 | assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); |
2821 | |
2822 | #if CONFIG_BACKGROUND_QUEUE |
2823 | vm_page_assign_background_state(mem); |
2824 | #endif |
2825 | return mem; |
2826 | } |
2827 | enable_preemption(); |
2828 | |
2829 | |
2830 | /* |
2831 | * Optionally produce warnings if the wire or gobble |
2832 | * counts exceed some threshold. |
2833 | */ |
2834 | #if VM_PAGE_WIRE_COUNT_WARNING |
2835 | if (vm_page_wire_count >= VM_PAGE_WIRE_COUNT_WARNING) { |
2836 | printf("mk: vm_page_grab(): high wired page count of %d\n" , |
2837 | vm_page_wire_count); |
2838 | } |
2839 | #endif |
2840 | #if VM_PAGE_GOBBLE_COUNT_WARNING |
2841 | if (vm_page_gobble_count >= VM_PAGE_GOBBLE_COUNT_WARNING) { |
2842 | printf("mk: vm_page_grab(): high gobbled page count of %d\n" , |
2843 | vm_page_gobble_count); |
2844 | } |
2845 | #endif |
2846 | |
2847 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
2848 | |
2849 | /* |
2850 | * Only let privileged threads (involved in pageout) |
2851 | * dip into the reserved pool. |
2852 | */ |
2853 | if ((vm_page_free_count < vm_page_free_reserved) && |
2854 | !(current_thread()->options & TH_OPT_VMPRIV)) { |
2855 | /* no page for us in the free queue... */ |
2856 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2857 | mem = VM_PAGE_NULL; |
2858 | |
2859 | #if CONFIG_SECLUDED_MEMORY |
2860 | /* ... but can we try and grab from the secluded queue? */ |
2861 | if (vm_page_secluded_count > 0 && |
2862 | ((grab_options & VM_PAGE_GRAB_SECLUDED) || |
2863 | task_can_use_secluded_mem(current_task(), TRUE))) { |
2864 | mem = vm_page_grab_secluded(); |
2865 | if (grab_options & VM_PAGE_GRAB_SECLUDED) { |
2866 | vm_page_secluded.grab_for_iokit++; |
2867 | if (mem) { |
2868 | vm_page_secluded.grab_for_iokit_success++; |
2869 | } |
2870 | } |
2871 | if (mem) { |
2872 | VM_CHECK_MEMORYSTATUS; |
2873 | |
2874 | disable_preemption(); |
2875 | PROCESSOR_DATA(current_processor(), page_grab_count) += 1; |
2876 | VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); |
2877 | enable_preemption(); |
2878 | |
2879 | return mem; |
2880 | } |
2881 | } |
2882 | #else /* CONFIG_SECLUDED_MEMORY */ |
2883 | (void) grab_options; |
2884 | #endif /* CONFIG_SECLUDED_MEMORY */ |
2885 | } |
2886 | else { |
2887 | vm_page_t head; |
2888 | vm_page_t tail; |
2889 | unsigned int pages_to_steal; |
2890 | unsigned int color; |
2891 | unsigned int clump_end, sub_count; |
2892 | |
2893 | while ( vm_page_free_count == 0 ) { |
2894 | |
2895 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2896 | /* |
2897 | * must be a privileged thread to be |
2898 | * in this state since a non-privileged |
2899 | * thread would have bailed if we were |
2900 | * under the vm_page_free_reserved mark |
2901 | */ |
2902 | VM_PAGE_WAIT(); |
2903 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
2904 | } |
2905 | |
2906 | disable_preemption(); |
2907 | |
2908 | if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) { |
2909 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2910 | |
2911 | /* |
2912 | * we got preempted and moved to another processor |
2913 | * or we got preempted and someone else ran and filled the cache |
2914 | */ |
2915 | goto return_page_from_cpu_list; |
2916 | } |
2917 | if (vm_page_free_count <= vm_page_free_reserved) |
2918 | pages_to_steal = 1; |
2919 | else { |
2920 | if (vm_free_magazine_refill_limit <= (vm_page_free_count - vm_page_free_reserved)) |
2921 | pages_to_steal = vm_free_magazine_refill_limit; |
2922 | else |
2923 | pages_to_steal = (vm_page_free_count - vm_page_free_reserved); |
2924 | } |
2925 | color = PROCESSOR_DATA(current_processor(), start_color); |
2926 | head = tail = NULL; |
2927 | |
2928 | vm_page_free_count -= pages_to_steal; |
2929 | clump_end = sub_count = 0; |
2930 | |
2931 | while (pages_to_steal--) { |
2932 | |
2933 | while (vm_page_queue_empty(&vm_page_queue_free[color].qhead)) |
2934 | color = (color + 1) & vm_color_mask; |
2935 | #if defined(__x86_64__) |
2936 | vm_page_queue_remove_first_with_clump(&vm_page_queue_free[color].qhead, |
2937 | mem, |
2938 | vm_page_t, |
2939 | vmp_pageq, |
2940 | clump_end); |
2941 | #else |
2942 | vm_page_queue_remove_first(&vm_page_queue_free[color].qhead, |
2943 | mem, |
2944 | vm_page_t, |
2945 | vmp_pageq); |
2946 | #endif |
2947 | |
2948 | assert(mem->vmp_q_state == VM_PAGE_ON_FREE_Q); |
2949 | |
2950 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
2951 | |
2952 | #if defined(__arm__) || defined(__arm64__) |
2953 | color = (color + 1) & vm_color_mask; |
2954 | #else |
2955 | |
2956 | #if DEVELOPMENT || DEBUG |
2957 | |
2958 | sub_count++; |
2959 | if (clump_end) { |
2960 | vm_clump_update_stats(sub_count); |
2961 | sub_count = 0; |
2962 | color = (color + 1) & vm_color_mask; |
2963 | } |
2964 | #else |
2965 | if (clump_end) color = (color + 1) & vm_color_mask; |
2966 | |
2967 | #endif /* if DEVELOPMENT || DEBUG */ |
2968 | |
2969 | #endif /* if defined(__arm__) || defined(__arm64__) */ |
2970 | |
2971 | if (head == NULL) |
2972 | head = mem; |
2973 | else |
2974 | tail->vmp_snext = mem; |
2975 | tail = mem; |
2976 | |
2977 | assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); |
2978 | assert(mem->vmp_tabled == FALSE); |
2979 | assert(mem->vmp_object == 0); |
2980 | assert(!mem->vmp_laundry); |
2981 | |
2982 | mem->vmp_q_state = VM_PAGE_ON_FREE_LOCAL_Q; |
2983 | |
2984 | assertf(pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem)), "page = 0x%llx" , (uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)); |
2985 | assert(mem->vmp_busy); |
2986 | assert(!mem->vmp_pmapped); |
2987 | assert(!mem->vmp_wpmapped); |
2988 | assert(!pmap_is_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem))); |
2989 | } |
2990 | #if defined (__x86_64__) && (DEVELOPMENT || DEBUG) |
2991 | vm_clump_update_stats(sub_count); |
2992 | #endif |
2993 | lck_mtx_unlock(&vm_page_queue_free_lock); |
2994 | |
2995 | #if HIBERNATION |
2996 | if (hibernate_rebuild_needed) { |
2997 | panic("%s:%d should not modify cpu->free_pages while hibernating" , __FUNCTION__, __LINE__); |
2998 | } |
2999 | #endif /* HIBERNATION */ |
3000 | PROCESSOR_DATA(current_processor(), free_pages) = head->vmp_snext; |
3001 | PROCESSOR_DATA(current_processor(), start_color) = color; |
3002 | |
3003 | /* |
3004 | * satisfy this request |
3005 | */ |
3006 | PROCESSOR_DATA(current_processor(), page_grab_count) += 1; |
3007 | VM_DEBUG_EVENT(vm_page_grab, VM_PAGE_GRAB, DBG_FUNC_NONE, grab_options, 0, 0, 0); |
3008 | mem = head; |
3009 | assert(mem->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); |
3010 | |
3011 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
3012 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
3013 | |
3014 | enable_preemption(); |
3015 | } |
3016 | /* |
3017 | * Decide if we should poke the pageout daemon. |
3018 | * We do this if the free count is less than the low |
3019 | * water mark, or if the free count is less than the high |
3020 | * water mark (but above the low water mark) and the inactive |
3021 | * count is less than its target. |
3022 | * |
3023 | * We don't have the counts locked ... if they change a little, |
3024 | * it doesn't really matter. |
3025 | */ |
3026 | if (vm_page_free_count < vm_page_free_min) |
3027 | thread_wakeup((event_t) &vm_page_free_wanted); |
3028 | |
3029 | VM_CHECK_MEMORYSTATUS; |
3030 | |
3031 | if (mem) { |
3032 | // dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 4); /* (TEST/DEBUG) */ |
3033 | |
3034 | #if CONFIG_BACKGROUND_QUEUE |
3035 | vm_page_assign_background_state(mem); |
3036 | #endif |
3037 | } |
3038 | return mem; |
3039 | } |
3040 | |
3041 | #if CONFIG_SECLUDED_MEMORY |
3042 | vm_page_t |
3043 | vm_page_grab_secluded(void) |
3044 | { |
3045 | vm_page_t mem; |
3046 | vm_object_t object; |
3047 | int refmod_state; |
3048 | |
3049 | if (vm_page_secluded_count == 0) { |
3050 | /* no secluded pages to grab... */ |
3051 | return VM_PAGE_NULL; |
3052 | } |
3053 | |
3054 | /* secluded queue is protected by the VM page queue lock */ |
3055 | vm_page_lock_queues(); |
3056 | |
3057 | if (vm_page_secluded_count == 0) { |
3058 | /* no secluded pages to grab... */ |
3059 | vm_page_unlock_queues(); |
3060 | return VM_PAGE_NULL; |
3061 | } |
3062 | |
3063 | #if 00 |
3064 | /* can we grab from the secluded queue? */ |
3065 | if (vm_page_secluded_count > vm_page_secluded_target || |
3066 | (vm_page_secluded_count > 0 && |
3067 | task_can_use_secluded_mem(current_task(), TRUE))) { |
3068 | /* OK */ |
3069 | } else { |
3070 | /* can't grab from secluded queue... */ |
3071 | vm_page_unlock_queues(); |
3072 | return VM_PAGE_NULL; |
3073 | } |
3074 | #endif |
3075 | |
3076 | /* we can grab a page from secluded queue! */ |
3077 | assert((vm_page_secluded_count_free + |
3078 | vm_page_secluded_count_inuse) == |
3079 | vm_page_secluded_count); |
3080 | if (current_task()->task_can_use_secluded_mem) { |
3081 | assert(num_tasks_can_use_secluded_mem > 0); |
3082 | } |
3083 | assert(!vm_page_queue_empty(&vm_page_queue_secluded)); |
3084 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
3085 | mem = (vm_page_t)vm_page_queue_first(&vm_page_queue_secluded); |
3086 | assert(mem->vmp_q_state == VM_PAGE_ON_SECLUDED_Q); |
3087 | vm_page_queues_remove(mem, TRUE); |
3088 | |
3089 | object = VM_PAGE_OBJECT(mem); |
3090 | |
3091 | assert(!mem->vmp_fictitious); |
3092 | assert(!VM_PAGE_WIRED(mem)); |
3093 | if (object == VM_OBJECT_NULL) { |
3094 | /* free for grab! */ |
3095 | vm_page_unlock_queues(); |
3096 | vm_page_secluded.grab_success_free++; |
3097 | |
3098 | assert(mem->vmp_busy); |
3099 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
3100 | assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL); |
3101 | assert(mem->vmp_pageq.next == 0); |
3102 | assert(mem->vmp_pageq.prev == 0); |
3103 | assert(mem->vmp_listq.next == 0); |
3104 | assert(mem->vmp_listq.prev == 0); |
3105 | #if CONFIG_BACKGROUND_QUEUE |
3106 | assert(mem->vmp_on_backgroundq == 0); |
3107 | assert(mem->vmp_backgroundq.next == 0); |
3108 | assert(mem->vmp_backgroundq.prev == 0); |
3109 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
3110 | return mem; |
3111 | } |
3112 | |
3113 | assert(!object->internal); |
3114 | // vm_page_pageable_external_count--; |
3115 | |
3116 | if (!vm_object_lock_try(object)) { |
3117 | // printf("SECLUDED: page %p: object %p locked\n", mem, object); |
3118 | vm_page_secluded.grab_failure_locked++; |
3119 | reactivate_secluded_page: |
3120 | vm_page_activate(mem); |
3121 | vm_page_unlock_queues(); |
3122 | return VM_PAGE_NULL; |
3123 | } |
3124 | if (mem->vmp_busy || |
3125 | mem->vmp_cleaning || |
3126 | mem->vmp_laundry) { |
3127 | /* can't steal page in this state... */ |
3128 | vm_object_unlock(object); |
3129 | vm_page_secluded.grab_failure_state++; |
3130 | goto reactivate_secluded_page; |
3131 | } |
3132 | |
3133 | mem->vmp_busy = TRUE; |
3134 | refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem)); |
3135 | if (refmod_state & VM_MEM_REFERENCED) { |
3136 | mem->vmp_reference = TRUE; |
3137 | } |
3138 | if (refmod_state & VM_MEM_MODIFIED) { |
3139 | SET_PAGE_DIRTY(mem, FALSE); |
3140 | } |
3141 | if (mem->vmp_dirty || mem->vmp_precious) { |
3142 | /* can't grab a dirty page; re-activate */ |
3143 | // printf("SECLUDED: dirty page %p\n", mem); |
3144 | PAGE_WAKEUP_DONE(mem); |
3145 | vm_page_secluded.grab_failure_dirty++; |
3146 | vm_object_unlock(object); |
3147 | goto reactivate_secluded_page; |
3148 | } |
3149 | if (mem->vmp_reference) { |
3150 | /* it's been used but we do need to grab a page... */ |
3151 | } |
3152 | |
3153 | vm_page_unlock_queues(); |
3154 | |
3155 | /* finish what vm_page_free() would have done... */ |
3156 | vm_page_free_prepare_object(mem, TRUE); |
3157 | vm_object_unlock(object); |
3158 | object = VM_OBJECT_NULL; |
3159 | if (vm_page_free_verify) { |
3160 | assertf(pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem)), "page = 0x%llx" , (uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)); |
3161 | } |
3162 | pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); |
3163 | vm_page_secluded.grab_success_other++; |
3164 | |
3165 | assert(mem->vmp_busy); |
3166 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
3167 | assert(VM_PAGE_OBJECT(mem) == VM_OBJECT_NULL); |
3168 | assert(mem->vmp_pageq.next == 0); |
3169 | assert(mem->vmp_pageq.prev == 0); |
3170 | assert(mem->vmp_listq.next == 0); |
3171 | assert(mem->vmp_listq.prev == 0); |
3172 | #if CONFIG_BACKGROUND_QUEUE |
3173 | assert(mem->vmp_on_backgroundq == 0); |
3174 | assert(mem->vmp_backgroundq.next == 0); |
3175 | assert(mem->vmp_backgroundq.prev == 0); |
3176 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
3177 | |
3178 | return mem; |
3179 | } |
3180 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3181 | |
3182 | /* |
3183 | * vm_page_release: |
3184 | * |
3185 | * Return a page to the free list. |
3186 | */ |
3187 | |
3188 | void |
3189 | vm_page_release( |
3190 | vm_page_t mem, |
3191 | boolean_t page_queues_locked) |
3192 | { |
3193 | unsigned int color; |
3194 | int need_wakeup = 0; |
3195 | int need_priv_wakeup = 0; |
3196 | #if CONFIG_SECLUDED_MEMORY |
3197 | int need_secluded_wakeup = 0; |
3198 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3199 | |
3200 | if (page_queues_locked) { |
3201 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
3202 | } else { |
3203 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); |
3204 | } |
3205 | |
3206 | assert(!mem->vmp_private && !mem->vmp_fictitious); |
3207 | if (vm_page_free_verify) { |
3208 | assertf(pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem)), "page = 0x%llx" , (uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)); |
3209 | } |
3210 | // dbgLog(VM_PAGE_GET_PHYS_PAGE(mem), vm_page_free_count, vm_page_wire_count, 5); /* (TEST/DEBUG) */ |
3211 | |
3212 | pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); |
3213 | |
3214 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
3215 | |
3216 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
3217 | assert(mem->vmp_busy); |
3218 | assert(!mem->vmp_laundry); |
3219 | assert(mem->vmp_object == 0); |
3220 | assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); |
3221 | assert(mem->vmp_listq.next == 0 && mem->vmp_listq.prev == 0); |
3222 | #if CONFIG_BACKGROUND_QUEUE |
3223 | assert(mem->vmp_backgroundq.next == 0 && |
3224 | mem->vmp_backgroundq.prev == 0 && |
3225 | mem->vmp_on_backgroundq == FALSE); |
3226 | #endif |
3227 | if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) && |
3228 | vm_lopage_free_count < vm_lopage_free_limit && |
3229 | VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) { |
3230 | /* |
3231 | * this exists to support hardware controllers |
3232 | * incapable of generating DMAs with more than 32 bits |
3233 | * of address on platforms with physical memory > 4G... |
3234 | */ |
3235 | vm_page_queue_enter_first(&vm_lopage_queue_free, |
3236 | mem, |
3237 | vm_page_t, |
3238 | vmp_pageq); |
3239 | vm_lopage_free_count++; |
3240 | |
3241 | if (vm_lopage_free_count >= vm_lopage_free_limit) |
3242 | vm_lopage_refill = FALSE; |
3243 | |
3244 | mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q; |
3245 | mem->vmp_lopage = TRUE; |
3246 | #if CONFIG_SECLUDED_MEMORY |
3247 | } else if (vm_page_free_count > vm_page_free_reserved && |
3248 | vm_page_secluded_count < vm_page_secluded_target && |
3249 | num_tasks_can_use_secluded_mem == 0) { |
3250 | /* |
3251 | * XXX FBDP TODO: also avoid refilling secluded queue |
3252 | * when some IOKit objects are already grabbing from it... |
3253 | */ |
3254 | if (!page_queues_locked) { |
3255 | if (!vm_page_trylock_queues()) { |
3256 | /* take locks in right order */ |
3257 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3258 | vm_page_lock_queues(); |
3259 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
3260 | } |
3261 | } |
3262 | mem->vmp_lopage = FALSE; |
3263 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
3264 | vm_page_queue_enter_first(&vm_page_queue_secluded, |
3265 | mem, |
3266 | vm_page_t, |
3267 | vmp_pageq); |
3268 | mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; |
3269 | vm_page_secluded_count++; |
3270 | vm_page_secluded_count_free++; |
3271 | if (!page_queues_locked) { |
3272 | vm_page_unlock_queues(); |
3273 | } |
3274 | LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_OWNED); |
3275 | if (vm_page_free_wanted_secluded > 0) { |
3276 | vm_page_free_wanted_secluded--; |
3277 | need_secluded_wakeup = 1; |
3278 | } |
3279 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3280 | } else { |
3281 | mem->vmp_lopage = FALSE; |
3282 | mem->vmp_q_state = VM_PAGE_ON_FREE_Q; |
3283 | |
3284 | color = VM_PAGE_GET_COLOR(mem); |
3285 | #if defined(__x86_64__) |
3286 | vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, |
3287 | mem, |
3288 | vm_page_t, |
3289 | vmp_pageq); |
3290 | #else |
3291 | vm_page_queue_enter(&vm_page_queue_free[color].qhead, |
3292 | mem, |
3293 | vm_page_t, |
3294 | vmp_pageq); |
3295 | #endif |
3296 | vm_page_free_count++; |
3297 | /* |
3298 | * Check if we should wake up someone waiting for page. |
3299 | * But don't bother waking them unless they can allocate. |
3300 | * |
3301 | * We wakeup only one thread, to prevent starvation. |
3302 | * Because the scheduling system handles wait queues FIFO, |
3303 | * if we wakeup all waiting threads, one greedy thread |
3304 | * can starve multiple niceguy threads. When the threads |
3305 | * all wakeup, the greedy threads runs first, grabs the page, |
3306 | * and waits for another page. It will be the first to run |
3307 | * when the next page is freed. |
3308 | * |
3309 | * However, there is a slight danger here. |
3310 | * The thread we wake might not use the free page. |
3311 | * Then the other threads could wait indefinitely |
3312 | * while the page goes unused. To forestall this, |
3313 | * the pageout daemon will keep making free pages |
3314 | * as long as vm_page_free_wanted is non-zero. |
3315 | */ |
3316 | |
3317 | assert(vm_page_free_count > 0); |
3318 | if (vm_page_free_wanted_privileged > 0) { |
3319 | vm_page_free_wanted_privileged--; |
3320 | need_priv_wakeup = 1; |
3321 | #if CONFIG_SECLUDED_MEMORY |
3322 | } else if (vm_page_free_wanted_secluded > 0 && |
3323 | vm_page_free_count > vm_page_free_reserved) { |
3324 | vm_page_free_wanted_secluded--; |
3325 | need_secluded_wakeup = 1; |
3326 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3327 | } else if (vm_page_free_wanted > 0 && |
3328 | vm_page_free_count > vm_page_free_reserved) { |
3329 | vm_page_free_wanted--; |
3330 | need_wakeup = 1; |
3331 | } |
3332 | } |
3333 | vm_pageout_vminfo.vm_page_pages_freed++; |
3334 | |
3335 | VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, 1, 0, 0, 0); |
3336 | |
3337 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3338 | |
3339 | if (need_priv_wakeup) |
3340 | thread_wakeup_one((event_t) &vm_page_free_wanted_privileged); |
3341 | #if CONFIG_SECLUDED_MEMORY |
3342 | else if (need_secluded_wakeup) |
3343 | thread_wakeup_one((event_t) &vm_page_free_wanted_secluded); |
3344 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3345 | else if (need_wakeup) |
3346 | thread_wakeup_one((event_t) &vm_page_free_count); |
3347 | |
3348 | VM_CHECK_MEMORYSTATUS; |
3349 | } |
3350 | |
3351 | /* |
3352 | * This version of vm_page_release() is used only at startup |
3353 | * when we are single-threaded and pages are being released |
3354 | * for the first time. Hence, no locking or unnecessary checks are made. |
3355 | * Note: VM_CHECK_MEMORYSTATUS invoked by the caller. |
3356 | */ |
3357 | void |
3358 | vm_page_release_startup( |
3359 | vm_page_t mem) |
3360 | { |
3361 | vm_page_queue_t queue_free; |
3362 | |
3363 | if (vm_lopage_free_count < vm_lopage_free_limit && |
3364 | VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) { |
3365 | mem->vmp_lopage = TRUE; |
3366 | mem->vmp_q_state = VM_PAGE_ON_FREE_LOPAGE_Q; |
3367 | vm_lopage_free_count++; |
3368 | queue_free = &vm_lopage_queue_free; |
3369 | #if CONFIG_SECLUDED_MEMORY |
3370 | } else if (vm_page_secluded_count < vm_page_secluded_target) { |
3371 | mem->vmp_lopage = FALSE; |
3372 | mem->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; |
3373 | vm_page_secluded_count++; |
3374 | vm_page_secluded_count_free++; |
3375 | queue_free = &vm_page_queue_secluded; |
3376 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3377 | } else { |
3378 | mem->vmp_lopage = FALSE; |
3379 | mem->vmp_q_state = VM_PAGE_ON_FREE_Q; |
3380 | vm_page_free_count++; |
3381 | queue_free = &vm_page_queue_free[VM_PAGE_GET_COLOR(mem)].qhead; |
3382 | } |
3383 | if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) { |
3384 | #if defined(__x86_64__) |
3385 | vm_page_queue_enter_clump(queue_free, mem, vm_page_t, vmp_pageq); |
3386 | #else |
3387 | vm_page_queue_enter(queue_free, mem, vm_page_t, vmp_pageq); |
3388 | #endif |
3389 | } else |
3390 | vm_page_queue_enter_first(queue_free, mem, vm_page_t, vmp_pageq); |
3391 | } |
3392 | |
3393 | /* |
3394 | * vm_page_wait: |
3395 | * |
3396 | * Wait for a page to become available. |
3397 | * If there are plenty of free pages, then we don't sleep. |
3398 | * |
3399 | * Returns: |
3400 | * TRUE: There may be another page, try again |
3401 | * FALSE: We were interrupted out of our wait, don't try again |
3402 | */ |
3403 | |
3404 | boolean_t |
3405 | vm_page_wait( |
3406 | int interruptible ) |
3407 | { |
3408 | /* |
3409 | * We can't use vm_page_free_reserved to make this |
3410 | * determination. Consider: some thread might |
3411 | * need to allocate two pages. The first allocation |
3412 | * succeeds, the second fails. After the first page is freed, |
3413 | * a call to vm_page_wait must really block. |
3414 | */ |
3415 | kern_return_t wait_result; |
3416 | int need_wakeup = 0; |
3417 | int is_privileged = current_thread()->options & TH_OPT_VMPRIV; |
3418 | |
3419 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
3420 | |
3421 | if (is_privileged && vm_page_free_count) { |
3422 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3423 | return TRUE; |
3424 | } |
3425 | |
3426 | if (vm_page_free_count >= vm_page_free_target) { |
3427 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3428 | return TRUE; |
3429 | } |
3430 | |
3431 | if (is_privileged) { |
3432 | if (vm_page_free_wanted_privileged++ == 0) |
3433 | need_wakeup = 1; |
3434 | wait_result = assert_wait((event_t)&vm_page_free_wanted_privileged, interruptible); |
3435 | #if CONFIG_SECLUDED_MEMORY |
3436 | } else if (secluded_for_apps && |
3437 | task_can_use_secluded_mem(current_task(), FALSE)) { |
3438 | #if 00 |
3439 | /* XXX FBDP: need pageq lock for this... */ |
3440 | /* XXX FBDP: might wait even if pages available, */ |
3441 | /* XXX FBDP: hopefully not for too long... */ |
3442 | if (vm_page_secluded_count > 0) { |
3443 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3444 | return TRUE; |
3445 | } |
3446 | #endif |
3447 | if (vm_page_free_wanted_secluded++ == 0) { |
3448 | need_wakeup = 1; |
3449 | } |
3450 | wait_result = assert_wait( |
3451 | (event_t)&vm_page_free_wanted_secluded, |
3452 | interruptible); |
3453 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3454 | } else { |
3455 | if (vm_page_free_wanted++ == 0) |
3456 | need_wakeup = 1; |
3457 | wait_result = assert_wait((event_t)&vm_page_free_count, |
3458 | interruptible); |
3459 | } |
3460 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3461 | counter(c_vm_page_wait_block++); |
3462 | |
3463 | if (need_wakeup) |
3464 | thread_wakeup((event_t)&vm_page_free_wanted); |
3465 | |
3466 | if (wait_result == THREAD_WAITING) { |
3467 | VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, VM_PAGE_WAIT_BLOCK, DBG_FUNC_START, |
3468 | vm_page_free_wanted_privileged, |
3469 | vm_page_free_wanted, |
3470 | #if CONFIG_SECLUDED_MEMORY |
3471 | vm_page_free_wanted_secluded, |
3472 | #else /* CONFIG_SECLUDED_MEMORY */ |
3473 | 0, |
3474 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3475 | 0); |
3476 | wait_result = thread_block(THREAD_CONTINUE_NULL); |
3477 | VM_DEBUG_CONSTANT_EVENT(vm_page_wait_block, |
3478 | VM_PAGE_WAIT_BLOCK, DBG_FUNC_END, 0, 0, 0, 0); |
3479 | } |
3480 | |
3481 | return (wait_result == THREAD_AWAKENED); |
3482 | } |
3483 | |
3484 | /* |
3485 | * vm_page_alloc: |
3486 | * |
3487 | * Allocate and return a memory cell associated |
3488 | * with this VM object/offset pair. |
3489 | * |
3490 | * Object must be locked. |
3491 | */ |
3492 | |
3493 | vm_page_t |
3494 | vm_page_alloc( |
3495 | vm_object_t object, |
3496 | vm_object_offset_t offset) |
3497 | { |
3498 | vm_page_t mem; |
3499 | int grab_options; |
3500 | |
3501 | vm_object_lock_assert_exclusive(object); |
3502 | grab_options = 0; |
3503 | #if CONFIG_SECLUDED_MEMORY |
3504 | if (object->can_grab_secluded) { |
3505 | grab_options |= VM_PAGE_GRAB_SECLUDED; |
3506 | } |
3507 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3508 | mem = vm_page_grab_options(grab_options); |
3509 | if (mem == VM_PAGE_NULL) |
3510 | return VM_PAGE_NULL; |
3511 | |
3512 | vm_page_insert(mem, object, offset); |
3513 | |
3514 | return(mem); |
3515 | } |
3516 | |
3517 | /* |
3518 | * vm_page_alloc_guard: |
3519 | * |
3520 | * Allocate a fictitious page which will be used |
3521 | * as a guard page. The page will be inserted into |
3522 | * the object and returned to the caller. |
3523 | */ |
3524 | |
3525 | vm_page_t |
3526 | vm_page_alloc_guard( |
3527 | vm_object_t object, |
3528 | vm_object_offset_t offset) |
3529 | { |
3530 | vm_page_t mem; |
3531 | |
3532 | vm_object_lock_assert_exclusive(object); |
3533 | mem = vm_page_grab_guard(); |
3534 | if (mem == VM_PAGE_NULL) |
3535 | return VM_PAGE_NULL; |
3536 | |
3537 | vm_page_insert(mem, object, offset); |
3538 | |
3539 | return(mem); |
3540 | } |
3541 | |
3542 | |
3543 | counter(unsigned int c_laundry_pages_freed = 0;) |
3544 | |
3545 | /* |
3546 | * vm_page_free_prepare: |
3547 | * |
3548 | * Removes page from any queue it may be on |
3549 | * and disassociates it from its VM object. |
3550 | * |
3551 | * Object and page queues must be locked prior to entry. |
3552 | */ |
3553 | static void |
3554 | vm_page_free_prepare( |
3555 | vm_page_t mem) |
3556 | { |
3557 | vm_page_free_prepare_queues(mem); |
3558 | vm_page_free_prepare_object(mem, TRUE); |
3559 | } |
3560 | |
3561 | |
3562 | void |
3563 | vm_page_free_prepare_queues( |
3564 | vm_page_t mem) |
3565 | { |
3566 | vm_object_t m_object; |
3567 | |
3568 | VM_PAGE_CHECK(mem); |
3569 | |
3570 | assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q); |
3571 | assert(!mem->vmp_cleaning); |
3572 | m_object = VM_PAGE_OBJECT(mem); |
3573 | |
3574 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
3575 | if (m_object) { |
3576 | vm_object_lock_assert_exclusive(m_object); |
3577 | } |
3578 | if (mem->vmp_laundry) { |
3579 | /* |
3580 | * We may have to free a page while it's being laundered |
3581 | * if we lost its pager (due to a forced unmount, for example). |
3582 | * We need to call vm_pageout_steal_laundry() before removing |
3583 | * the page from its VM object, so that we can remove it |
3584 | * from its pageout queue and adjust the laundry accounting |
3585 | */ |
3586 | vm_pageout_steal_laundry(mem, TRUE); |
3587 | counter(++c_laundry_pages_freed); |
3588 | } |
3589 | |
3590 | vm_page_queues_remove(mem, TRUE); |
3591 | |
3592 | if (VM_PAGE_WIRED(mem)) { |
3593 | assert(mem->vmp_wire_count > 0); |
3594 | |
3595 | if (m_object) { |
3596 | |
3597 | VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); |
3598 | VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); |
3599 | VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); |
3600 | |
3601 | assert(m_object->resident_page_count >= |
3602 | m_object->wired_page_count); |
3603 | |
3604 | if (m_object->purgable == VM_PURGABLE_VOLATILE) { |
3605 | OSAddAtomic(+1, &vm_page_purgeable_count); |
3606 | assert(vm_page_purgeable_wired_count > 0); |
3607 | OSAddAtomic(-1, &vm_page_purgeable_wired_count); |
3608 | } |
3609 | if ((m_object->purgable == VM_PURGABLE_VOLATILE || |
3610 | m_object->purgable == VM_PURGABLE_EMPTY) && |
3611 | m_object->vo_owner != TASK_NULL) { |
3612 | task_t owner; |
3613 | int ledger_idx_volatile; |
3614 | int ledger_idx_nonvolatile; |
3615 | int ledger_idx_volatile_compressed; |
3616 | int ledger_idx_nonvolatile_compressed; |
3617 | boolean_t ; |
3618 | |
3619 | owner = VM_OBJECT_OWNER(m_object); |
3620 | vm_object_ledger_tag_ledgers( |
3621 | m_object, |
3622 | &ledger_idx_volatile, |
3623 | &ledger_idx_nonvolatile, |
3624 | &ledger_idx_volatile_compressed, |
3625 | &ledger_idx_nonvolatile_compressed, |
3626 | &do_footprint); |
3627 | /* |
3628 | * While wired, this page was accounted |
3629 | * as "non-volatile" but it should now |
3630 | * be accounted as "volatile". |
3631 | */ |
3632 | /* one less "non-volatile"... */ |
3633 | ledger_debit(owner->ledger, |
3634 | ledger_idx_nonvolatile, |
3635 | PAGE_SIZE); |
3636 | if (do_footprint) { |
3637 | /* ... and "phys_footprint" */ |
3638 | ledger_debit(owner->ledger, |
3639 | task_ledgers.phys_footprint, |
3640 | PAGE_SIZE); |
3641 | } |
3642 | /* one more "volatile" */ |
3643 | ledger_credit(owner->ledger, |
3644 | ledger_idx_volatile, |
3645 | PAGE_SIZE); |
3646 | } |
3647 | } |
3648 | if (!mem->vmp_private && !mem->vmp_fictitious) |
3649 | vm_page_wire_count--; |
3650 | |
3651 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
3652 | mem->vmp_wire_count = 0; |
3653 | assert(!mem->vmp_gobbled); |
3654 | } else if (mem->vmp_gobbled) { |
3655 | if (!mem->vmp_private && !mem->vmp_fictitious) |
3656 | vm_page_wire_count--; |
3657 | vm_page_gobble_count--; |
3658 | } |
3659 | } |
3660 | |
3661 | |
3662 | void |
3663 | vm_page_free_prepare_object( |
3664 | vm_page_t mem, |
3665 | boolean_t remove_from_hash) |
3666 | { |
3667 | if (mem->vmp_tabled) |
3668 | vm_page_remove(mem, remove_from_hash); /* clears tabled, object, offset */ |
3669 | |
3670 | PAGE_WAKEUP(mem); /* clears wanted */ |
3671 | |
3672 | if (mem->vmp_private) { |
3673 | mem->vmp_private = FALSE; |
3674 | mem->vmp_fictitious = TRUE; |
3675 | VM_PAGE_SET_PHYS_PAGE(mem, vm_page_fictitious_addr); |
3676 | } |
3677 | if ( !mem->vmp_fictitious) { |
3678 | assert(mem->vmp_pageq.next == 0); |
3679 | assert(mem->vmp_pageq.prev == 0); |
3680 | assert(mem->vmp_listq.next == 0); |
3681 | assert(mem->vmp_listq.prev == 0); |
3682 | #if CONFIG_BACKGROUND_QUEUE |
3683 | assert(mem->vmp_backgroundq.next == 0); |
3684 | assert(mem->vmp_backgroundq.prev == 0); |
3685 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
3686 | assert(mem->vmp_next_m == 0); |
3687 | vm_page_init(mem, VM_PAGE_GET_PHYS_PAGE(mem), mem->vmp_lopage); |
3688 | } |
3689 | } |
3690 | |
3691 | |
3692 | /* |
3693 | * vm_page_free: |
3694 | * |
3695 | * Returns the given page to the free list, |
3696 | * disassociating it with any VM object. |
3697 | * |
3698 | * Object and page queues must be locked prior to entry. |
3699 | */ |
3700 | void |
3701 | vm_page_free( |
3702 | vm_page_t mem) |
3703 | { |
3704 | vm_page_free_prepare(mem); |
3705 | |
3706 | if (mem->vmp_fictitious) { |
3707 | vm_page_release_fictitious(mem); |
3708 | } else { |
3709 | vm_page_release(mem, |
3710 | TRUE); /* page queues are locked */ |
3711 | } |
3712 | } |
3713 | |
3714 | |
3715 | void |
3716 | vm_page_free_unlocked( |
3717 | vm_page_t mem, |
3718 | boolean_t remove_from_hash) |
3719 | { |
3720 | vm_page_lockspin_queues(); |
3721 | vm_page_free_prepare_queues(mem); |
3722 | vm_page_unlock_queues(); |
3723 | |
3724 | vm_page_free_prepare_object(mem, remove_from_hash); |
3725 | |
3726 | if (mem->vmp_fictitious) { |
3727 | vm_page_release_fictitious(mem); |
3728 | } else { |
3729 | vm_page_release(mem, FALSE); /* page queues are not locked */ |
3730 | } |
3731 | } |
3732 | |
3733 | |
3734 | /* |
3735 | * Free a list of pages. The list can be up to several hundred pages, |
3736 | * as blocked up by vm_pageout_scan(). |
3737 | * The big win is not having to take the free list lock once |
3738 | * per page. |
3739 | * |
3740 | * The VM page queues lock (vm_page_queue_lock) should NOT be held. |
3741 | * The VM page free queues lock (vm_page_queue_free_lock) should NOT be held. |
3742 | */ |
3743 | void |
3744 | vm_page_free_list( |
3745 | vm_page_t freeq, |
3746 | boolean_t prepare_object) |
3747 | { |
3748 | vm_page_t mem; |
3749 | vm_page_t nxt; |
3750 | vm_page_t local_freeq; |
3751 | int pg_count; |
3752 | |
3753 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_NOTOWNED); |
3754 | LCK_MTX_ASSERT(&vm_page_queue_free_lock, LCK_MTX_ASSERT_NOTOWNED); |
3755 | |
3756 | while (freeq) { |
3757 | |
3758 | pg_count = 0; |
3759 | local_freeq = VM_PAGE_NULL; |
3760 | mem = freeq; |
3761 | |
3762 | /* |
3763 | * break up the processing into smaller chunks so |
3764 | * that we can 'pipeline' the pages onto the |
3765 | * free list w/o introducing too much |
3766 | * contention on the global free queue lock |
3767 | */ |
3768 | while (mem && pg_count < 64) { |
3769 | |
3770 | assert((mem->vmp_q_state == VM_PAGE_NOT_ON_Q) || |
3771 | (mem->vmp_q_state == VM_PAGE_IS_WIRED)); |
3772 | #if CONFIG_BACKGROUND_QUEUE |
3773 | assert(mem->vmp_backgroundq.next == 0 && |
3774 | mem->vmp_backgroundq.prev == 0 && |
3775 | mem->vmp_on_backgroundq == FALSE); |
3776 | #endif |
3777 | nxt = mem->vmp_snext; |
3778 | mem->vmp_snext = NULL; |
3779 | assert(mem->vmp_pageq.prev == 0); |
3780 | |
3781 | if (vm_page_free_verify && !mem->vmp_fictitious && !mem->vmp_private) { |
3782 | assertf(pmap_verify_free(VM_PAGE_GET_PHYS_PAGE(mem)), "page = 0x%llx" , (uint64_t)VM_PAGE_GET_PHYS_PAGE(mem)); |
3783 | } |
3784 | if (prepare_object == TRUE) |
3785 | vm_page_free_prepare_object(mem, TRUE); |
3786 | |
3787 | if (!mem->vmp_fictitious) { |
3788 | assert(mem->vmp_busy); |
3789 | |
3790 | if ((mem->vmp_lopage == TRUE || vm_lopage_refill == TRUE) && |
3791 | vm_lopage_free_count < vm_lopage_free_limit && |
3792 | VM_PAGE_GET_PHYS_PAGE(mem) < max_valid_low_ppnum) { |
3793 | vm_page_release(mem, FALSE); /* page queues are not locked */ |
3794 | #if CONFIG_SECLUDED_MEMORY |
3795 | } else if (vm_page_secluded_count < vm_page_secluded_target && |
3796 | num_tasks_can_use_secluded_mem == 0) { |
3797 | vm_page_release(mem, |
3798 | FALSE); /* page queues are not locked */ |
3799 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3800 | } else { |
3801 | /* |
3802 | * IMPORTANT: we can't set the page "free" here |
3803 | * because that would make the page eligible for |
3804 | * a physically-contiguous allocation (see |
3805 | * vm_page_find_contiguous()) right away (we don't |
3806 | * hold the vm_page_queue_free lock). That would |
3807 | * cause trouble because the page is not actually |
3808 | * in the free queue yet... |
3809 | */ |
3810 | mem->vmp_snext = local_freeq; |
3811 | local_freeq = mem; |
3812 | pg_count++; |
3813 | |
3814 | pmap_clear_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); |
3815 | } |
3816 | } else { |
3817 | assert(VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_fictitious_addr || |
3818 | VM_PAGE_GET_PHYS_PAGE(mem) == vm_page_guard_addr); |
3819 | vm_page_release_fictitious(mem); |
3820 | } |
3821 | mem = nxt; |
3822 | } |
3823 | freeq = mem; |
3824 | |
3825 | if ( (mem = local_freeq) ) { |
3826 | unsigned int avail_free_count; |
3827 | unsigned int need_wakeup = 0; |
3828 | unsigned int need_priv_wakeup = 0; |
3829 | #if CONFIG_SECLUDED_MEMORY |
3830 | unsigned int need_wakeup_secluded = 0; |
3831 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3832 | |
3833 | lck_mtx_lock_spin(&vm_page_queue_free_lock); |
3834 | |
3835 | while (mem) { |
3836 | int color; |
3837 | |
3838 | nxt = mem->vmp_snext; |
3839 | |
3840 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
3841 | assert(mem->vmp_busy); |
3842 | mem->vmp_lopage = FALSE; |
3843 | mem->vmp_q_state = VM_PAGE_ON_FREE_Q; |
3844 | |
3845 | color = VM_PAGE_GET_COLOR(mem); |
3846 | #if defined(__x86_64__) |
3847 | vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, |
3848 | mem, |
3849 | vm_page_t, |
3850 | vmp_pageq); |
3851 | #else |
3852 | vm_page_queue_enter(&vm_page_queue_free[color].qhead, |
3853 | mem, |
3854 | vm_page_t, |
3855 | vmp_pageq); |
3856 | #endif |
3857 | mem = nxt; |
3858 | } |
3859 | vm_pageout_vminfo.vm_page_pages_freed += pg_count; |
3860 | vm_page_free_count += pg_count; |
3861 | avail_free_count = vm_page_free_count; |
3862 | |
3863 | VM_DEBUG_CONSTANT_EVENT(vm_page_release, VM_PAGE_RELEASE, DBG_FUNC_NONE, pg_count, 0, 0, 0); |
3864 | |
3865 | if (vm_page_free_wanted_privileged > 0 && avail_free_count > 0) { |
3866 | |
3867 | if (avail_free_count < vm_page_free_wanted_privileged) { |
3868 | need_priv_wakeup = avail_free_count; |
3869 | vm_page_free_wanted_privileged -= avail_free_count; |
3870 | avail_free_count = 0; |
3871 | } else { |
3872 | need_priv_wakeup = vm_page_free_wanted_privileged; |
3873 | avail_free_count -= vm_page_free_wanted_privileged; |
3874 | vm_page_free_wanted_privileged = 0; |
3875 | } |
3876 | } |
3877 | #if CONFIG_SECLUDED_MEMORY |
3878 | if (vm_page_free_wanted_secluded > 0 && |
3879 | avail_free_count > vm_page_free_reserved) { |
3880 | unsigned int available_pages; |
3881 | available_pages = (avail_free_count - |
3882 | vm_page_free_reserved); |
3883 | if (available_pages < |
3884 | vm_page_free_wanted_secluded) { |
3885 | need_wakeup_secluded = available_pages; |
3886 | vm_page_free_wanted_secluded -= |
3887 | available_pages; |
3888 | avail_free_count -= available_pages; |
3889 | } else { |
3890 | need_wakeup_secluded = |
3891 | vm_page_free_wanted_secluded; |
3892 | avail_free_count -= |
3893 | vm_page_free_wanted_secluded; |
3894 | vm_page_free_wanted_secluded = 0; |
3895 | } |
3896 | } |
3897 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3898 | if (vm_page_free_wanted > 0 && avail_free_count > vm_page_free_reserved) { |
3899 | unsigned int available_pages; |
3900 | |
3901 | available_pages = avail_free_count - vm_page_free_reserved; |
3902 | |
3903 | if (available_pages >= vm_page_free_wanted) { |
3904 | need_wakeup = vm_page_free_wanted; |
3905 | vm_page_free_wanted = 0; |
3906 | } else { |
3907 | need_wakeup = available_pages; |
3908 | vm_page_free_wanted -= available_pages; |
3909 | } |
3910 | } |
3911 | lck_mtx_unlock(&vm_page_queue_free_lock); |
3912 | |
3913 | if (need_priv_wakeup != 0) { |
3914 | /* |
3915 | * There shouldn't be that many VM-privileged threads, |
3916 | * so let's wake them all up, even if we don't quite |
3917 | * have enough pages to satisfy them all. |
3918 | */ |
3919 | thread_wakeup((event_t)&vm_page_free_wanted_privileged); |
3920 | } |
3921 | #if CONFIG_SECLUDED_MEMORY |
3922 | if (need_wakeup_secluded != 0 && |
3923 | vm_page_free_wanted_secluded == 0) { |
3924 | thread_wakeup((event_t) |
3925 | &vm_page_free_wanted_secluded); |
3926 | } else { |
3927 | for (; |
3928 | need_wakeup_secluded != 0; |
3929 | need_wakeup_secluded--) { |
3930 | thread_wakeup_one( |
3931 | (event_t) |
3932 | &vm_page_free_wanted_secluded); |
3933 | } |
3934 | } |
3935 | #endif /* CONFIG_SECLUDED_MEMORY */ |
3936 | if (need_wakeup != 0 && vm_page_free_wanted == 0) { |
3937 | /* |
3938 | * We don't expect to have any more waiters |
3939 | * after this, so let's wake them all up at |
3940 | * once. |
3941 | */ |
3942 | thread_wakeup((event_t) &vm_page_free_count); |
3943 | } else for (; need_wakeup != 0; need_wakeup--) { |
3944 | /* |
3945 | * Wake up one waiter per page we just released. |
3946 | */ |
3947 | thread_wakeup_one((event_t) &vm_page_free_count); |
3948 | } |
3949 | |
3950 | VM_CHECK_MEMORYSTATUS; |
3951 | } |
3952 | } |
3953 | } |
3954 | |
3955 | |
3956 | /* |
3957 | * vm_page_wire: |
3958 | * |
3959 | * Mark this page as wired down by yet |
3960 | * another map, removing it from paging queues |
3961 | * as necessary. |
3962 | * |
3963 | * The page's object and the page queues must be locked. |
3964 | */ |
3965 | |
3966 | |
3967 | void |
3968 | vm_page_wire( |
3969 | vm_page_t mem, |
3970 | vm_tag_t tag, |
3971 | boolean_t check_memorystatus) |
3972 | { |
3973 | vm_object_t m_object; |
3974 | |
3975 | m_object = VM_PAGE_OBJECT(mem); |
3976 | |
3977 | // dbgLog(current_thread(), mem->vmp_offset, m_object, 1); /* (TEST/DEBUG) */ |
3978 | |
3979 | VM_PAGE_CHECK(mem); |
3980 | if (m_object) { |
3981 | vm_object_lock_assert_exclusive(m_object); |
3982 | } else { |
3983 | /* |
3984 | * In theory, the page should be in an object before it |
3985 | * gets wired, since we need to hold the object lock |
3986 | * to update some fields in the page structure. |
3987 | * However, some code (i386 pmap, for example) might want |
3988 | * to wire a page before it gets inserted into an object. |
3989 | * That's somewhat OK, as long as nobody else can get to |
3990 | * that page and update it at the same time. |
3991 | */ |
3992 | } |
3993 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
3994 | if ( !VM_PAGE_WIRED(mem)) { |
3995 | |
3996 | if (mem->vmp_laundry) |
3997 | vm_pageout_steal_laundry(mem, TRUE); |
3998 | |
3999 | vm_page_queues_remove(mem, TRUE); |
4000 | |
4001 | assert(mem->vmp_wire_count == 0); |
4002 | mem->vmp_q_state = VM_PAGE_IS_WIRED; |
4003 | |
4004 | if (m_object) { |
4005 | |
4006 | VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); |
4007 | VM_OBJECT_WIRED_PAGE_ADD(m_object, mem); |
4008 | VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, tag); |
4009 | |
4010 | assert(m_object->resident_page_count >= |
4011 | m_object->wired_page_count); |
4012 | if (m_object->purgable == VM_PURGABLE_VOLATILE) { |
4013 | assert(vm_page_purgeable_count > 0); |
4014 | OSAddAtomic(-1, &vm_page_purgeable_count); |
4015 | OSAddAtomic(1, &vm_page_purgeable_wired_count); |
4016 | } |
4017 | if ((m_object->purgable == VM_PURGABLE_VOLATILE || |
4018 | m_object->purgable == VM_PURGABLE_EMPTY) && |
4019 | m_object->vo_owner != TASK_NULL) { |
4020 | task_t owner; |
4021 | int ledger_idx_volatile; |
4022 | int ledger_idx_nonvolatile; |
4023 | int ledger_idx_volatile_compressed; |
4024 | int ledger_idx_nonvolatile_compressed; |
4025 | boolean_t ; |
4026 | |
4027 | owner = VM_OBJECT_OWNER(m_object); |
4028 | vm_object_ledger_tag_ledgers( |
4029 | m_object, |
4030 | &ledger_idx_volatile, |
4031 | &ledger_idx_nonvolatile, |
4032 | &ledger_idx_volatile_compressed, |
4033 | &ledger_idx_nonvolatile_compressed, |
4034 | &do_footprint); |
4035 | /* less volatile bytes */ |
4036 | ledger_debit(owner->ledger, |
4037 | ledger_idx_volatile, |
4038 | PAGE_SIZE); |
4039 | /* more not-quite-volatile bytes */ |
4040 | ledger_credit(owner->ledger, |
4041 | ledger_idx_nonvolatile, |
4042 | PAGE_SIZE); |
4043 | if (do_footprint) { |
4044 | /* more footprint */ |
4045 | ledger_credit(owner->ledger, |
4046 | task_ledgers.phys_footprint, |
4047 | PAGE_SIZE); |
4048 | } |
4049 | } |
4050 | if (m_object->all_reusable) { |
4051 | /* |
4052 | * Wired pages are not counted as "re-usable" |
4053 | * in "all_reusable" VM objects, so nothing |
4054 | * to do here. |
4055 | */ |
4056 | } else if (mem->vmp_reusable) { |
4057 | /* |
4058 | * This page is not "re-usable" when it's |
4059 | * wired, so adjust its state and the |
4060 | * accounting. |
4061 | */ |
4062 | vm_object_reuse_pages(m_object, |
4063 | mem->vmp_offset, |
4064 | mem->vmp_offset+PAGE_SIZE_64, |
4065 | FALSE); |
4066 | } |
4067 | } |
4068 | assert(!mem->vmp_reusable); |
4069 | |
4070 | if (!mem->vmp_private && !mem->vmp_fictitious && !mem->vmp_gobbled) |
4071 | vm_page_wire_count++; |
4072 | if (mem->vmp_gobbled) |
4073 | vm_page_gobble_count--; |
4074 | mem->vmp_gobbled = FALSE; |
4075 | |
4076 | if (check_memorystatus == TRUE) { |
4077 | VM_CHECK_MEMORYSTATUS; |
4078 | } |
4079 | } |
4080 | assert(!mem->vmp_gobbled); |
4081 | assert(mem->vmp_q_state == VM_PAGE_IS_WIRED); |
4082 | mem->vmp_wire_count++; |
4083 | if (__improbable(mem->vmp_wire_count == 0)) { |
4084 | panic("vm_page_wire(%p): wire_count overflow" , mem); |
4085 | } |
4086 | VM_PAGE_CHECK(mem); |
4087 | } |
4088 | |
4089 | /* |
4090 | * vm_page_unwire: |
4091 | * |
4092 | * Release one wiring of this page, potentially |
4093 | * enabling it to be paged again. |
4094 | * |
4095 | * The page's object and the page queues must be locked. |
4096 | */ |
4097 | void |
4098 | vm_page_unwire( |
4099 | vm_page_t mem, |
4100 | boolean_t queueit) |
4101 | { |
4102 | vm_object_t m_object; |
4103 | |
4104 | m_object = VM_PAGE_OBJECT(mem); |
4105 | |
4106 | // dbgLog(current_thread(), mem->vmp_offset, m_object, 0); /* (TEST/DEBUG) */ |
4107 | |
4108 | VM_PAGE_CHECK(mem); |
4109 | assert(VM_PAGE_WIRED(mem)); |
4110 | assert(mem->vmp_wire_count > 0); |
4111 | assert(!mem->vmp_gobbled); |
4112 | assert(m_object != VM_OBJECT_NULL); |
4113 | vm_object_lock_assert_exclusive(m_object); |
4114 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4115 | if (--mem->vmp_wire_count == 0) { |
4116 | |
4117 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
4118 | |
4119 | VM_OBJECT_WIRED_PAGE_UPDATE_START(m_object); |
4120 | VM_OBJECT_WIRED_PAGE_REMOVE(m_object, mem); |
4121 | VM_OBJECT_WIRED_PAGE_UPDATE_END(m_object, m_object->wire_tag); |
4122 | if (!mem->vmp_private && !mem->vmp_fictitious) { |
4123 | vm_page_wire_count--; |
4124 | } |
4125 | |
4126 | assert(m_object->resident_page_count >= |
4127 | m_object->wired_page_count); |
4128 | if (m_object->purgable == VM_PURGABLE_VOLATILE) { |
4129 | OSAddAtomic(+1, &vm_page_purgeable_count); |
4130 | assert(vm_page_purgeable_wired_count > 0); |
4131 | OSAddAtomic(-1, &vm_page_purgeable_wired_count); |
4132 | } |
4133 | if ((m_object->purgable == VM_PURGABLE_VOLATILE || |
4134 | m_object->purgable == VM_PURGABLE_EMPTY) && |
4135 | m_object->vo_owner != TASK_NULL) { |
4136 | task_t owner; |
4137 | int ledger_idx_volatile; |
4138 | int ledger_idx_nonvolatile; |
4139 | int ledger_idx_volatile_compressed; |
4140 | int ledger_idx_nonvolatile_compressed; |
4141 | boolean_t ; |
4142 | |
4143 | owner = VM_OBJECT_OWNER(m_object); |
4144 | vm_object_ledger_tag_ledgers( |
4145 | m_object, |
4146 | &ledger_idx_volatile, |
4147 | &ledger_idx_nonvolatile, |
4148 | &ledger_idx_volatile_compressed, |
4149 | &ledger_idx_nonvolatile_compressed, |
4150 | &do_footprint); |
4151 | /* more volatile bytes */ |
4152 | ledger_credit(owner->ledger, |
4153 | ledger_idx_volatile, |
4154 | PAGE_SIZE); |
4155 | /* less not-quite-volatile bytes */ |
4156 | ledger_debit(owner->ledger, |
4157 | ledger_idx_nonvolatile, |
4158 | PAGE_SIZE); |
4159 | if (do_footprint) { |
4160 | /* less footprint */ |
4161 | ledger_debit(owner->ledger, |
4162 | task_ledgers.phys_footprint, |
4163 | PAGE_SIZE); |
4164 | } |
4165 | } |
4166 | assert(m_object != kernel_object); |
4167 | assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); |
4168 | |
4169 | if (queueit == TRUE) { |
4170 | if (m_object->purgable == VM_PURGABLE_EMPTY) { |
4171 | vm_page_deactivate(mem); |
4172 | } else { |
4173 | vm_page_activate(mem); |
4174 | } |
4175 | } |
4176 | |
4177 | VM_CHECK_MEMORYSTATUS; |
4178 | |
4179 | } |
4180 | VM_PAGE_CHECK(mem); |
4181 | } |
4182 | |
4183 | /* |
4184 | * vm_page_deactivate: |
4185 | * |
4186 | * Returns the given page to the inactive list, |
4187 | * indicating that no physical maps have access |
4188 | * to this page. [Used by the physical mapping system.] |
4189 | * |
4190 | * The page queues must be locked. |
4191 | */ |
4192 | void |
4193 | vm_page_deactivate( |
4194 | vm_page_t m) |
4195 | { |
4196 | vm_page_deactivate_internal(m, TRUE); |
4197 | } |
4198 | |
4199 | |
4200 | void |
4201 | vm_page_deactivate_internal( |
4202 | vm_page_t m, |
4203 | boolean_t clear_hw_reference) |
4204 | { |
4205 | vm_object_t m_object; |
4206 | |
4207 | m_object = VM_PAGE_OBJECT(m); |
4208 | |
4209 | VM_PAGE_CHECK(m); |
4210 | assert(m_object != kernel_object); |
4211 | assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); |
4212 | |
4213 | // dbgLog(VM_PAGE_GET_PHYS_PAGE(m), vm_page_free_count, vm_page_wire_count, 6); /* (TEST/DEBUG) */ |
4214 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4215 | /* |
4216 | * This page is no longer very interesting. If it was |
4217 | * interesting (active or inactive/referenced), then we |
4218 | * clear the reference bit and (re)enter it in the |
4219 | * inactive queue. Note wired pages should not have |
4220 | * their reference bit cleared. |
4221 | */ |
4222 | assert ( !(m->vmp_absent && !m->vmp_unusual)); |
4223 | |
4224 | if (m->vmp_gobbled) { /* can this happen? */ |
4225 | assert( !VM_PAGE_WIRED(m)); |
4226 | |
4227 | if (!m->vmp_private && !m->vmp_fictitious) |
4228 | vm_page_wire_count--; |
4229 | vm_page_gobble_count--; |
4230 | m->vmp_gobbled = FALSE; |
4231 | } |
4232 | /* |
4233 | * if this page is currently on the pageout queue, we can't do the |
4234 | * vm_page_queues_remove (which doesn't handle the pageout queue case) |
4235 | * and we can't remove it manually since we would need the object lock |
4236 | * (which is not required here) to decrement the activity_in_progress |
4237 | * reference which is held on the object while the page is in the pageout queue... |
4238 | * just let the normal laundry processing proceed |
4239 | */ |
4240 | if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious || |
4241 | (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || |
4242 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) || |
4243 | VM_PAGE_WIRED(m)) { |
4244 | return; |
4245 | } |
4246 | if (!m->vmp_absent && clear_hw_reference == TRUE) |
4247 | pmap_clear_reference(VM_PAGE_GET_PHYS_PAGE(m)); |
4248 | |
4249 | m->vmp_reference = FALSE; |
4250 | m->vmp_no_cache = FALSE; |
4251 | |
4252 | if ( !VM_PAGE_INACTIVE(m)) { |
4253 | vm_page_queues_remove(m, FALSE); |
4254 | |
4255 | if (!VM_DYNAMIC_PAGING_ENABLED() && |
4256 | m->vmp_dirty && m_object->internal && |
4257 | (m_object->purgable == VM_PURGABLE_DENY || |
4258 | m_object->purgable == VM_PURGABLE_NONVOLATILE || |
4259 | m_object->purgable == VM_PURGABLE_VOLATILE)) { |
4260 | vm_page_check_pageable_safe(m); |
4261 | vm_page_queue_enter(&vm_page_queue_throttled, m, vm_page_t, vmp_pageq); |
4262 | m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q; |
4263 | vm_page_throttled_count++; |
4264 | } else { |
4265 | if (m_object->named && m_object->ref_count == 1) { |
4266 | vm_page_speculate(m, FALSE); |
4267 | #if DEVELOPMENT || DEBUG |
4268 | vm_page_speculative_recreated++; |
4269 | #endif |
4270 | } else { |
4271 | vm_page_enqueue_inactive(m, FALSE); |
4272 | } |
4273 | } |
4274 | } |
4275 | } |
4276 | |
4277 | /* |
4278 | * vm_page_enqueue_cleaned |
4279 | * |
4280 | * Put the page on the cleaned queue, mark it cleaned, etc. |
4281 | * Being on the cleaned queue (and having m->clean_queue set) |
4282 | * does ** NOT ** guarantee that the page is clean! |
4283 | * |
4284 | * Call with the queues lock held. |
4285 | */ |
4286 | |
4287 | void vm_page_enqueue_cleaned(vm_page_t m) |
4288 | { |
4289 | vm_object_t m_object; |
4290 | |
4291 | m_object = VM_PAGE_OBJECT(m); |
4292 | |
4293 | assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); |
4294 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4295 | assert( !(m->vmp_absent && !m->vmp_unusual)); |
4296 | |
4297 | if (VM_PAGE_WIRED(m)) { |
4298 | return; |
4299 | } |
4300 | |
4301 | if (m->vmp_gobbled) { |
4302 | if (!m->vmp_private && !m->vmp_fictitious) |
4303 | vm_page_wire_count--; |
4304 | vm_page_gobble_count--; |
4305 | m->vmp_gobbled = FALSE; |
4306 | } |
4307 | /* |
4308 | * if this page is currently on the pageout queue, we can't do the |
4309 | * vm_page_queues_remove (which doesn't handle the pageout queue case) |
4310 | * and we can't remove it manually since we would need the object lock |
4311 | * (which is not required here) to decrement the activity_in_progress |
4312 | * reference which is held on the object while the page is in the pageout queue... |
4313 | * just let the normal laundry processing proceed |
4314 | */ |
4315 | if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious || |
4316 | (m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q) || |
4317 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { |
4318 | return; |
4319 | } |
4320 | vm_page_queues_remove(m, FALSE); |
4321 | |
4322 | vm_page_check_pageable_safe(m); |
4323 | vm_page_queue_enter(&vm_page_queue_cleaned, m, vm_page_t, vmp_pageq); |
4324 | m->vmp_q_state = VM_PAGE_ON_INACTIVE_CLEANED_Q; |
4325 | vm_page_cleaned_count++; |
4326 | |
4327 | vm_page_inactive_count++; |
4328 | if (m_object->internal) { |
4329 | vm_page_pageable_internal_count++; |
4330 | } else { |
4331 | vm_page_pageable_external_count++; |
4332 | } |
4333 | #if CONFIG_BACKGROUND_QUEUE |
4334 | if (m->vmp_in_background) |
4335 | vm_page_add_to_backgroundq(m, TRUE); |
4336 | #endif |
4337 | VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1); |
4338 | } |
4339 | |
4340 | /* |
4341 | * vm_page_activate: |
4342 | * |
4343 | * Put the specified page on the active list (if appropriate). |
4344 | * |
4345 | * The page queues must be locked. |
4346 | */ |
4347 | |
4348 | void |
4349 | vm_page_activate( |
4350 | vm_page_t m) |
4351 | { |
4352 | vm_object_t m_object; |
4353 | |
4354 | m_object = VM_PAGE_OBJECT(m); |
4355 | |
4356 | VM_PAGE_CHECK(m); |
4357 | #ifdef FIXME_4778297 |
4358 | assert(m_object != kernel_object); |
4359 | #endif |
4360 | assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); |
4361 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4362 | assert( !(m->vmp_absent && !m->vmp_unusual)); |
4363 | |
4364 | if (m->vmp_gobbled) { |
4365 | assert( !VM_PAGE_WIRED(m)); |
4366 | if (!m->vmp_private && !m->vmp_fictitious) |
4367 | vm_page_wire_count--; |
4368 | vm_page_gobble_count--; |
4369 | m->vmp_gobbled = FALSE; |
4370 | } |
4371 | /* |
4372 | * if this page is currently on the pageout queue, we can't do the |
4373 | * vm_page_queues_remove (which doesn't handle the pageout queue case) |
4374 | * and we can't remove it manually since we would need the object lock |
4375 | * (which is not required here) to decrement the activity_in_progress |
4376 | * reference which is held on the object while the page is in the pageout queue... |
4377 | * just let the normal laundry processing proceed |
4378 | */ |
4379 | if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious || |
4380 | (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || |
4381 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) |
4382 | return; |
4383 | |
4384 | #if DEBUG |
4385 | if (m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q) |
4386 | panic("vm_page_activate: already active" ); |
4387 | #endif |
4388 | |
4389 | if (m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q) { |
4390 | DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL); |
4391 | DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL); |
4392 | } |
4393 | |
4394 | vm_page_queues_remove(m, FALSE); |
4395 | |
4396 | if ( !VM_PAGE_WIRED(m)) { |
4397 | vm_page_check_pageable_safe(m); |
4398 | if (!VM_DYNAMIC_PAGING_ENABLED() && |
4399 | m->vmp_dirty && m_object->internal && |
4400 | (m_object->purgable == VM_PURGABLE_DENY || |
4401 | m_object->purgable == VM_PURGABLE_NONVOLATILE || |
4402 | m_object->purgable == VM_PURGABLE_VOLATILE)) { |
4403 | vm_page_queue_enter(&vm_page_queue_throttled, m, vm_page_t, vmp_pageq); |
4404 | m->vmp_q_state = VM_PAGE_ON_THROTTLED_Q; |
4405 | vm_page_throttled_count++; |
4406 | } else { |
4407 | #if CONFIG_SECLUDED_MEMORY |
4408 | if (secluded_for_filecache && |
4409 | vm_page_secluded_target != 0 && |
4410 | num_tasks_can_use_secluded_mem == 0 && |
4411 | m_object->eligible_for_secluded) { |
4412 | vm_page_queue_enter(&vm_page_queue_secluded, m, |
4413 | vm_page_t, vmp_pageq); |
4414 | m->vmp_q_state = VM_PAGE_ON_SECLUDED_Q; |
4415 | vm_page_secluded_count++; |
4416 | vm_page_secluded_count_inuse++; |
4417 | assert(!m_object->internal); |
4418 | // vm_page_pageable_external_count++; |
4419 | } else |
4420 | #endif /* CONFIG_SECLUDED_MEMORY */ |
4421 | vm_page_enqueue_active(m, FALSE); |
4422 | } |
4423 | m->vmp_reference = TRUE; |
4424 | m->vmp_no_cache = FALSE; |
4425 | } |
4426 | VM_PAGE_CHECK(m); |
4427 | } |
4428 | |
4429 | |
4430 | /* |
4431 | * vm_page_speculate: |
4432 | * |
4433 | * Put the specified page on the speculative list (if appropriate). |
4434 | * |
4435 | * The page queues must be locked. |
4436 | */ |
4437 | void |
4438 | vm_page_speculate( |
4439 | vm_page_t m, |
4440 | boolean_t new) |
4441 | { |
4442 | struct vm_speculative_age_q *aq; |
4443 | vm_object_t m_object; |
4444 | |
4445 | m_object = VM_PAGE_OBJECT(m); |
4446 | |
4447 | VM_PAGE_CHECK(m); |
4448 | vm_page_check_pageable_safe(m); |
4449 | |
4450 | assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); |
4451 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4452 | assert( !(m->vmp_absent && !m->vmp_unusual)); |
4453 | assert(m_object->internal == FALSE); |
4454 | |
4455 | /* |
4456 | * if this page is currently on the pageout queue, we can't do the |
4457 | * vm_page_queues_remove (which doesn't handle the pageout queue case) |
4458 | * and we can't remove it manually since we would need the object lock |
4459 | * (which is not required here) to decrement the activity_in_progress |
4460 | * reference which is held on the object while the page is in the pageout queue... |
4461 | * just let the normal laundry processing proceed |
4462 | */ |
4463 | if (m->vmp_laundry || m->vmp_private || m->vmp_fictitious || |
4464 | (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || |
4465 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) |
4466 | return; |
4467 | |
4468 | vm_page_queues_remove(m, FALSE); |
4469 | |
4470 | if ( !VM_PAGE_WIRED(m)) { |
4471 | mach_timespec_t ts; |
4472 | clock_sec_t sec; |
4473 | clock_nsec_t nsec; |
4474 | |
4475 | clock_get_system_nanotime(&sec, &nsec); |
4476 | ts.tv_sec = (unsigned int) sec; |
4477 | ts.tv_nsec = nsec; |
4478 | |
4479 | if (vm_page_speculative_count == 0) { |
4480 | |
4481 | speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q; |
4482 | speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q; |
4483 | |
4484 | aq = &vm_page_queue_speculative[speculative_age_index]; |
4485 | |
4486 | /* |
4487 | * set the timer to begin a new group |
4488 | */ |
4489 | aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000; |
4490 | aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC; |
4491 | ADD_MACH_TIMESPEC(&aq->age_ts, &ts); |
4492 | } else { |
4493 | aq = &vm_page_queue_speculative[speculative_age_index]; |
4494 | |
4495 | if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) { |
4496 | |
4497 | speculative_age_index++; |
4498 | |
4499 | if (speculative_age_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) |
4500 | speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q; |
4501 | if (speculative_age_index == speculative_steal_index) { |
4502 | speculative_steal_index = speculative_age_index + 1; |
4503 | |
4504 | if (speculative_steal_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q) |
4505 | speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q; |
4506 | } |
4507 | aq = &vm_page_queue_speculative[speculative_age_index]; |
4508 | |
4509 | if (!vm_page_queue_empty(&aq->age_q)) |
4510 | vm_page_speculate_ageit(aq); |
4511 | |
4512 | aq->age_ts.tv_sec = vm_pageout_state.vm_page_speculative_q_age_ms / 1000; |
4513 | aq->age_ts.tv_nsec = (vm_pageout_state.vm_page_speculative_q_age_ms % 1000) * 1000 * NSEC_PER_USEC; |
4514 | ADD_MACH_TIMESPEC(&aq->age_ts, &ts); |
4515 | } |
4516 | } |
4517 | vm_page_enqueue_tail(&aq->age_q, &m->vmp_pageq); |
4518 | m->vmp_q_state = VM_PAGE_ON_SPECULATIVE_Q; |
4519 | vm_page_speculative_count++; |
4520 | vm_page_pageable_external_count++; |
4521 | |
4522 | if (new == TRUE) { |
4523 | vm_object_lock_assert_exclusive(m_object); |
4524 | |
4525 | m_object->pages_created++; |
4526 | #if DEVELOPMENT || DEBUG |
4527 | vm_page_speculative_created++; |
4528 | #endif |
4529 | } |
4530 | } |
4531 | VM_PAGE_CHECK(m); |
4532 | } |
4533 | |
4534 | |
4535 | /* |
4536 | * move pages from the specified aging bin to |
4537 | * the speculative bin that pageout_scan claims from |
4538 | * |
4539 | * The page queues must be locked. |
4540 | */ |
4541 | void |
4542 | vm_page_speculate_ageit(struct vm_speculative_age_q *aq) |
4543 | { |
4544 | struct vm_speculative_age_q *sq; |
4545 | vm_page_t t; |
4546 | |
4547 | sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q]; |
4548 | |
4549 | if (vm_page_queue_empty(&sq->age_q)) { |
4550 | sq->age_q.next = aq->age_q.next; |
4551 | sq->age_q.prev = aq->age_q.prev; |
4552 | |
4553 | t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.next); |
4554 | t->vmp_pageq.prev = VM_PAGE_PACK_PTR(&sq->age_q); |
4555 | |
4556 | t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev); |
4557 | t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q); |
4558 | } else { |
4559 | t = (vm_page_t)VM_PAGE_UNPACK_PTR(sq->age_q.prev); |
4560 | t->vmp_pageq.next = aq->age_q.next; |
4561 | |
4562 | t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.next); |
4563 | t->vmp_pageq.prev = sq->age_q.prev; |
4564 | |
4565 | t = (vm_page_t)VM_PAGE_UNPACK_PTR(aq->age_q.prev); |
4566 | t->vmp_pageq.next = VM_PAGE_PACK_PTR(&sq->age_q); |
4567 | |
4568 | sq->age_q.prev = aq->age_q.prev; |
4569 | } |
4570 | vm_page_queue_init(&aq->age_q); |
4571 | } |
4572 | |
4573 | |
4574 | void |
4575 | vm_page_lru( |
4576 | vm_page_t m) |
4577 | { |
4578 | VM_PAGE_CHECK(m); |
4579 | assert(VM_PAGE_OBJECT(m) != kernel_object); |
4580 | assert(VM_PAGE_GET_PHYS_PAGE(m) != vm_page_guard_addr); |
4581 | |
4582 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
4583 | |
4584 | if (m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q) { |
4585 | /* |
4586 | * we don't need to do all the other work that |
4587 | * vm_page_queues_remove and vm_page_enqueue_inactive |
4588 | * bring along for the ride |
4589 | */ |
4590 | assert(!m->vmp_laundry); |
4591 | assert(!m->vmp_private); |
4592 | |
4593 | m->vmp_no_cache = FALSE; |
4594 | |
4595 | vm_page_queue_remove(&vm_page_queue_inactive, m, vm_page_t, vmp_pageq); |
4596 | vm_page_queue_enter(&vm_page_queue_inactive, m, vm_page_t, vmp_pageq); |
4597 | |
4598 | return; |
4599 | } |
4600 | /* |
4601 | * if this page is currently on the pageout queue, we can't do the |
4602 | * vm_page_queues_remove (which doesn't handle the pageout queue case) |
4603 | * and we can't remove it manually since we would need the object lock |
4604 | * (which is not required here) to decrement the activity_in_progress |
4605 | * reference which is held on the object while the page is in the pageout queue... |
4606 | * just let the normal laundry processing proceed |
4607 | */ |
4608 | if (m->vmp_laundry || m->vmp_private || |
4609 | (m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) || |
4610 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) || |
4611 | VM_PAGE_WIRED(m)) |
4612 | return; |
4613 | |
4614 | m->vmp_no_cache = FALSE; |
4615 | |
4616 | vm_page_queues_remove(m, FALSE); |
4617 | |
4618 | vm_page_enqueue_inactive(m, FALSE); |
4619 | } |
4620 | |
4621 | |
4622 | void |
4623 | vm_page_reactivate_all_throttled(void) |
4624 | { |
4625 | vm_page_t first_throttled, last_throttled; |
4626 | vm_page_t first_active; |
4627 | vm_page_t m; |
4628 | int ; |
4629 | int , ; |
4630 | vm_object_t m_object; |
4631 | |
4632 | if (!VM_DYNAMIC_PAGING_ENABLED()) |
4633 | return; |
4634 | |
4635 | extra_active_count = 0; |
4636 | extra_internal_count = 0; |
4637 | extra_external_count = 0; |
4638 | vm_page_lock_queues(); |
4639 | if (! vm_page_queue_empty(&vm_page_queue_throttled)) { |
4640 | /* |
4641 | * Switch "throttled" pages to "active". |
4642 | */ |
4643 | vm_page_queue_iterate(&vm_page_queue_throttled, m, vm_page_t, vmp_pageq) { |
4644 | VM_PAGE_CHECK(m); |
4645 | assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q); |
4646 | |
4647 | m_object = VM_PAGE_OBJECT(m); |
4648 | |
4649 | extra_active_count++; |
4650 | if (m_object->internal) { |
4651 | extra_internal_count++; |
4652 | } else { |
4653 | extra_external_count++; |
4654 | } |
4655 | |
4656 | m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; |
4657 | VM_PAGE_CHECK(m); |
4658 | #if CONFIG_BACKGROUND_QUEUE |
4659 | if (m->vmp_in_background) |
4660 | vm_page_add_to_backgroundq(m, FALSE); |
4661 | #endif |
4662 | } |
4663 | |
4664 | /* |
4665 | * Transfer the entire throttled queue to a regular LRU page queues. |
4666 | * We insert it at the head of the active queue, so that these pages |
4667 | * get re-evaluated by the LRU algorithm first, since they've been |
4668 | * completely out of it until now. |
4669 | */ |
4670 | first_throttled = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled); |
4671 | last_throttled = (vm_page_t) vm_page_queue_last(&vm_page_queue_throttled); |
4672 | first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); |
4673 | if (vm_page_queue_empty(&vm_page_queue_active)) { |
4674 | vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled); |
4675 | } else { |
4676 | first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_throttled); |
4677 | } |
4678 | vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_throttled); |
4679 | first_throttled->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active); |
4680 | last_throttled->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active); |
4681 | |
4682 | #if DEBUG |
4683 | printf("reactivated %d throttled pages\n" , vm_page_throttled_count); |
4684 | #endif |
4685 | vm_page_queue_init(&vm_page_queue_throttled); |
4686 | /* |
4687 | * Adjust the global page counts. |
4688 | */ |
4689 | vm_page_active_count += extra_active_count; |
4690 | vm_page_pageable_internal_count += extra_internal_count; |
4691 | vm_page_pageable_external_count += extra_external_count; |
4692 | vm_page_throttled_count = 0; |
4693 | } |
4694 | assert(vm_page_throttled_count == 0); |
4695 | assert(vm_page_queue_empty(&vm_page_queue_throttled)); |
4696 | vm_page_unlock_queues(); |
4697 | } |
4698 | |
4699 | |
4700 | /* |
4701 | * move pages from the indicated local queue to the global active queue |
4702 | * its ok to fail if we're below the hard limit and force == FALSE |
4703 | * the nolocks == TRUE case is to allow this function to be run on |
4704 | * the hibernate path |
4705 | */ |
4706 | |
4707 | void |
4708 | vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks) |
4709 | { |
4710 | struct vpl *lq; |
4711 | vm_page_t first_local, last_local; |
4712 | vm_page_t first_active; |
4713 | vm_page_t m; |
4714 | uint32_t count = 0; |
4715 | |
4716 | if (vm_page_local_q == NULL) |
4717 | return; |
4718 | |
4719 | lq = &vm_page_local_q[lid].vpl_un.vpl; |
4720 | |
4721 | if (nolocks == FALSE) { |
4722 | if (lq->vpl_count < vm_page_local_q_hard_limit && force == FALSE) { |
4723 | if ( !vm_page_trylockspin_queues()) |
4724 | return; |
4725 | } else |
4726 | vm_page_lockspin_queues(); |
4727 | |
4728 | VPL_LOCK(&lq->vpl_lock); |
4729 | } |
4730 | if (lq->vpl_count) { |
4731 | /* |
4732 | * Switch "local" pages to "active". |
4733 | */ |
4734 | assert(!vm_page_queue_empty(&lq->vpl_queue)); |
4735 | |
4736 | vm_page_queue_iterate(&lq->vpl_queue, m, vm_page_t, vmp_pageq) { |
4737 | VM_PAGE_CHECK(m); |
4738 | vm_page_check_pageable_safe(m); |
4739 | assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_LOCAL_Q); |
4740 | assert(!m->vmp_fictitious); |
4741 | |
4742 | if (m->vmp_local_id != lid) |
4743 | panic("vm_page_reactivate_local: found vm_page_t(%p) with wrong cpuid" , m); |
4744 | |
4745 | m->vmp_local_id = 0; |
4746 | m->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; |
4747 | VM_PAGE_CHECK(m); |
4748 | #if CONFIG_BACKGROUND_QUEUE |
4749 | if (m->vmp_in_background) |
4750 | vm_page_add_to_backgroundq(m, FALSE); |
4751 | #endif |
4752 | count++; |
4753 | } |
4754 | if (count != lq->vpl_count) |
4755 | panic("vm_page_reactivate_local: count = %d, vm_page_local_count = %d\n" , count, lq->vpl_count); |
4756 | |
4757 | /* |
4758 | * Transfer the entire local queue to a regular LRU page queues. |
4759 | */ |
4760 | first_local = (vm_page_t) vm_page_queue_first(&lq->vpl_queue); |
4761 | last_local = (vm_page_t) vm_page_queue_last(&lq->vpl_queue); |
4762 | first_active = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); |
4763 | |
4764 | if (vm_page_queue_empty(&vm_page_queue_active)) { |
4765 | vm_page_queue_active.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local); |
4766 | } else { |
4767 | first_active->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(last_local); |
4768 | } |
4769 | vm_page_queue_active.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_local); |
4770 | first_local->vmp_pageq.prev = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(&vm_page_queue_active); |
4771 | last_local->vmp_pageq.next = VM_PAGE_CONVERT_TO_QUEUE_ENTRY(first_active); |
4772 | |
4773 | vm_page_queue_init(&lq->vpl_queue); |
4774 | /* |
4775 | * Adjust the global page counts. |
4776 | */ |
4777 | vm_page_active_count += lq->vpl_count; |
4778 | vm_page_pageable_internal_count += lq->vpl_internal_count; |
4779 | vm_page_pageable_external_count += lq->vpl_external_count; |
4780 | lq->vpl_count = 0; |
4781 | lq->vpl_internal_count = 0; |
4782 | lq->vpl_external_count = 0; |
4783 | } |
4784 | assert(vm_page_queue_empty(&lq->vpl_queue)); |
4785 | |
4786 | if (nolocks == FALSE) { |
4787 | VPL_UNLOCK(&lq->vpl_lock); |
4788 | |
4789 | vm_page_balance_inactive(count / 4); |
4790 | vm_page_unlock_queues(); |
4791 | } |
4792 | } |
4793 | |
4794 | /* |
4795 | * vm_page_part_zero_fill: |
4796 | * |
4797 | * Zero-fill a part of the page. |
4798 | */ |
4799 | #define PMAP_ZERO_PART_PAGE_IMPLEMENTED |
4800 | void |
4801 | vm_page_part_zero_fill( |
4802 | vm_page_t m, |
4803 | vm_offset_t m_pa, |
4804 | vm_size_t len) |
4805 | { |
4806 | |
4807 | #if 0 |
4808 | /* |
4809 | * we don't hold the page queue lock |
4810 | * so this check isn't safe to make |
4811 | */ |
4812 | VM_PAGE_CHECK(m); |
4813 | #endif |
4814 | |
4815 | #ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED |
4816 | pmap_zero_part_page(VM_PAGE_GET_PHYS_PAGE(m), m_pa, len); |
4817 | #else |
4818 | vm_page_t tmp; |
4819 | while (1) { |
4820 | tmp = vm_page_grab(); |
4821 | if (tmp == VM_PAGE_NULL) { |
4822 | vm_page_wait(THREAD_UNINT); |
4823 | continue; |
4824 | } |
4825 | break; |
4826 | } |
4827 | vm_page_zero_fill(tmp); |
4828 | if(m_pa != 0) { |
4829 | vm_page_part_copy(m, 0, tmp, 0, m_pa); |
4830 | } |
4831 | if((m_pa + len) < PAGE_SIZE) { |
4832 | vm_page_part_copy(m, m_pa + len, tmp, |
4833 | m_pa + len, PAGE_SIZE - (m_pa + len)); |
4834 | } |
4835 | vm_page_copy(tmp,m); |
4836 | VM_PAGE_FREE(tmp); |
4837 | #endif |
4838 | |
4839 | } |
4840 | |
4841 | /* |
4842 | * vm_page_zero_fill: |
4843 | * |
4844 | * Zero-fill the specified page. |
4845 | */ |
4846 | void |
4847 | vm_page_zero_fill( |
4848 | vm_page_t m) |
4849 | { |
4850 | XPR(XPR_VM_PAGE, |
4851 | "vm_page_zero_fill, object 0x%X offset 0x%X page 0x%X\n" , |
4852 | VM_PAGE_OBJECT(m), m->vmp_offset, m, 0,0); |
4853 | #if 0 |
4854 | /* |
4855 | * we don't hold the page queue lock |
4856 | * so this check isn't safe to make |
4857 | */ |
4858 | VM_PAGE_CHECK(m); |
4859 | #endif |
4860 | |
4861 | // dbgTrace(0xAEAEAEAE, VM_PAGE_GET_PHYS_PAGE(m), 0); /* (BRINGUP) */ |
4862 | pmap_zero_page(VM_PAGE_GET_PHYS_PAGE(m)); |
4863 | } |
4864 | |
4865 | /* |
4866 | * vm_page_part_copy: |
4867 | * |
4868 | * copy part of one page to another |
4869 | */ |
4870 | |
4871 | void |
4872 | vm_page_part_copy( |
4873 | vm_page_t src_m, |
4874 | vm_offset_t src_pa, |
4875 | vm_page_t dst_m, |
4876 | vm_offset_t dst_pa, |
4877 | vm_size_t len) |
4878 | { |
4879 | #if 0 |
4880 | /* |
4881 | * we don't hold the page queue lock |
4882 | * so this check isn't safe to make |
4883 | */ |
4884 | VM_PAGE_CHECK(src_m); |
4885 | VM_PAGE_CHECK(dst_m); |
4886 | #endif |
4887 | pmap_copy_part_page(VM_PAGE_GET_PHYS_PAGE(src_m), src_pa, |
4888 | VM_PAGE_GET_PHYS_PAGE(dst_m), dst_pa, len); |
4889 | } |
4890 | |
4891 | /* |
4892 | * vm_page_copy: |
4893 | * |
4894 | * Copy one page to another |
4895 | */ |
4896 | |
4897 | int vm_page_copy_cs_validations = 0; |
4898 | int vm_page_copy_cs_tainted = 0; |
4899 | |
4900 | void |
4901 | vm_page_copy( |
4902 | vm_page_t src_m, |
4903 | vm_page_t dest_m) |
4904 | { |
4905 | vm_object_t src_m_object; |
4906 | |
4907 | src_m_object = VM_PAGE_OBJECT(src_m); |
4908 | |
4909 | XPR(XPR_VM_PAGE, |
4910 | "vm_page_copy, object 0x%X offset 0x%X to object 0x%X offset 0x%X\n" , |
4911 | src_m_object, src_m->vmp_offset, |
4912 | VM_PAGE_OBJECT(dest_m), dest_m->vmp_offset, |
4913 | 0); |
4914 | #if 0 |
4915 | /* |
4916 | * we don't hold the page queue lock |
4917 | * so this check isn't safe to make |
4918 | */ |
4919 | VM_PAGE_CHECK(src_m); |
4920 | VM_PAGE_CHECK(dest_m); |
4921 | #endif |
4922 | vm_object_lock_assert_held(src_m_object); |
4923 | |
4924 | if (src_m_object != VM_OBJECT_NULL && |
4925 | src_m_object->code_signed) { |
4926 | /* |
4927 | * We're copying a page from a code-signed object. |
4928 | * Whoever ends up mapping the copy page might care about |
4929 | * the original page's integrity, so let's validate the |
4930 | * source page now. |
4931 | */ |
4932 | vm_page_copy_cs_validations++; |
4933 | vm_page_validate_cs(src_m); |
4934 | #if DEVELOPMENT || DEBUG |
4935 | DTRACE_VM4(codesigned_copy, |
4936 | vm_object_t, src_m_object, |
4937 | vm_object_offset_t, src_m->vmp_offset, |
4938 | int, src_m->vmp_cs_validated, |
4939 | int, src_m->vmp_cs_tainted); |
4940 | #endif /* DEVELOPMENT || DEBUG */ |
4941 | |
4942 | } |
4943 | |
4944 | /* |
4945 | * Propagate the cs_tainted bit to the copy page. Do not propagate |
4946 | * the cs_validated bit. |
4947 | */ |
4948 | dest_m->vmp_cs_tainted = src_m->vmp_cs_tainted; |
4949 | if (dest_m->vmp_cs_tainted) { |
4950 | vm_page_copy_cs_tainted++; |
4951 | } |
4952 | dest_m->vmp_error = src_m->vmp_error; /* sliding src_m might have failed... */ |
4953 | pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(src_m), VM_PAGE_GET_PHYS_PAGE(dest_m)); |
4954 | } |
4955 | |
4956 | #if MACH_ASSERT |
4957 | static void |
4958 | _vm_page_print( |
4959 | vm_page_t p) |
4960 | { |
4961 | printf("vm_page %p: \n" , p); |
4962 | printf(" pageq: next=%p prev=%p\n" , |
4963 | (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.next), |
4964 | (vm_page_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev)); |
4965 | printf(" listq: next=%p prev=%p\n" , |
4966 | (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.next)), |
4967 | (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_listq.prev))); |
4968 | printf(" next=%p\n" , (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m))); |
4969 | printf(" object=%p offset=0x%llx\n" ,VM_PAGE_OBJECT(p), p->vmp_offset); |
4970 | printf(" wire_count=%u\n" , p->vmp_wire_count); |
4971 | printf(" q_state=%u\n" , p->vmp_q_state); |
4972 | |
4973 | printf(" %slaundry, %sref, %sgobbled, %sprivate\n" , |
4974 | (p->vmp_laundry ? "" : "!" ), |
4975 | (p->vmp_reference ? "" : "!" ), |
4976 | (p->vmp_gobbled ? "" : "!" ), |
4977 | (p->vmp_private ? "" : "!" )); |
4978 | printf(" %sbusy, %swanted, %stabled, %sfictitious, %spmapped, %swpmapped\n" , |
4979 | (p->vmp_busy ? "" : "!" ), |
4980 | (p->vmp_wanted ? "" : "!" ), |
4981 | (p->vmp_tabled ? "" : "!" ), |
4982 | (p->vmp_fictitious ? "" : "!" ), |
4983 | (p->vmp_pmapped ? "" : "!" ), |
4984 | (p->vmp_wpmapped ? "" : "!" )); |
4985 | printf(" %sfree_when_done, %sabsent, %serror, %sdirty, %scleaning, %sprecious, %sclustered\n" , |
4986 | (p->vmp_free_when_done ? "" : "!" ), |
4987 | (p->vmp_absent ? "" : "!" ), |
4988 | (p->vmp_error ? "" : "!" ), |
4989 | (p->vmp_dirty ? "" : "!" ), |
4990 | (p->vmp_cleaning ? "" : "!" ), |
4991 | (p->vmp_precious ? "" : "!" ), |
4992 | (p->vmp_clustered ? "" : "!" )); |
4993 | printf(" %soverwriting, %srestart, %sunusual\n" , |
4994 | (p->vmp_overwriting ? "" : "!" ), |
4995 | (p->vmp_restart ? "" : "!" ), |
4996 | (p->vmp_unusual ? "" : "!" )); |
4997 | printf(" %scs_validated, %scs_tainted, %scs_nx, %sno_cache\n" , |
4998 | (p->vmp_cs_validated ? "" : "!" ), |
4999 | (p->vmp_cs_tainted ? "" : "!" ), |
5000 | (p->vmp_cs_nx ? "" : "!" ), |
5001 | (p->vmp_no_cache ? "" : "!" )); |
5002 | |
5003 | printf("phys_page=0x%x\n" , VM_PAGE_GET_PHYS_PAGE(p)); |
5004 | } |
5005 | |
5006 | /* |
5007 | * Check that the list of pages is ordered by |
5008 | * ascending physical address and has no holes. |
5009 | */ |
5010 | static int |
5011 | vm_page_verify_contiguous( |
5012 | vm_page_t pages, |
5013 | unsigned int npages) |
5014 | { |
5015 | vm_page_t m; |
5016 | unsigned int page_count; |
5017 | vm_offset_t prev_addr; |
5018 | |
5019 | prev_addr = VM_PAGE_GET_PHYS_PAGE(pages); |
5020 | page_count = 1; |
5021 | for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) { |
5022 | if (VM_PAGE_GET_PHYS_PAGE(m) != prev_addr + 1) { |
5023 | printf("m %p prev_addr 0x%lx, current addr 0x%x\n" , |
5024 | m, (long)prev_addr, VM_PAGE_GET_PHYS_PAGE(m)); |
5025 | printf("pages %p page_count %d npages %d\n" , pages, page_count, npages); |
5026 | panic("vm_page_verify_contiguous: not contiguous!" ); |
5027 | } |
5028 | prev_addr = VM_PAGE_GET_PHYS_PAGE(m); |
5029 | ++page_count; |
5030 | } |
5031 | if (page_count != npages) { |
5032 | printf("pages %p actual count 0x%x but requested 0x%x\n" , |
5033 | pages, page_count, npages); |
5034 | panic("vm_page_verify_contiguous: count error" ); |
5035 | } |
5036 | return 1; |
5037 | } |
5038 | |
5039 | |
5040 | /* |
5041 | * Check the free lists for proper length etc. |
5042 | */ |
5043 | static boolean_t vm_page_verify_this_free_list_enabled = FALSE; |
5044 | static unsigned int |
5045 | vm_page_verify_free_list( |
5046 | vm_page_queue_head_t *vm_page_queue, |
5047 | unsigned int color, |
5048 | vm_page_t look_for_page, |
5049 | boolean_t expect_page) |
5050 | { |
5051 | unsigned int npages; |
5052 | vm_page_t m; |
5053 | vm_page_t prev_m; |
5054 | boolean_t found_page; |
5055 | |
5056 | if (! vm_page_verify_this_free_list_enabled) |
5057 | return 0; |
5058 | |
5059 | found_page = FALSE; |
5060 | npages = 0; |
5061 | prev_m = (vm_page_t)((uintptr_t)vm_page_queue); |
5062 | |
5063 | vm_page_queue_iterate(vm_page_queue, |
5064 | m, |
5065 | vm_page_t, |
5066 | vmp_pageq) { |
5067 | |
5068 | if (m == look_for_page) { |
5069 | found_page = TRUE; |
5070 | } |
5071 | if ((vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev) != prev_m) |
5072 | panic("vm_page_verify_free_list(color=%u, npages=%u): page %p corrupted prev ptr %p instead of %p\n" , |
5073 | color, npages, m, (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.prev), prev_m); |
5074 | if ( ! m->vmp_busy ) |
5075 | panic("vm_page_verify_free_list(color=%u, npages=%u): page %p not busy\n" , |
5076 | color, npages, m); |
5077 | if (color != (unsigned int) -1) { |
5078 | if (VM_PAGE_GET_COLOR(m) != color) |
5079 | panic("vm_page_verify_free_list(color=%u, npages=%u): page %p wrong color %u instead of %u\n" , |
5080 | color, npages, m, VM_PAGE_GET_COLOR(m), color); |
5081 | if (m->vmp_q_state != VM_PAGE_ON_FREE_Q) |
5082 | panic("vm_page_verify_free_list(color=%u, npages=%u): page %p - expecting q_state == VM_PAGE_ON_FREE_Q, found %d\n" , |
5083 | color, npages, m, m->vmp_q_state); |
5084 | } else { |
5085 | if (m->vmp_q_state != VM_PAGE_ON_FREE_LOCAL_Q) |
5086 | panic("vm_page_verify_free_list(npages=%u): local page %p - expecting q_state == VM_PAGE_ON_FREE_LOCAL_Q, found %d\n" , |
5087 | npages, m, m->vmp_q_state); |
5088 | } |
5089 | ++npages; |
5090 | prev_m = m; |
5091 | } |
5092 | if (look_for_page != VM_PAGE_NULL) { |
5093 | unsigned int other_color; |
5094 | |
5095 | if (expect_page && !found_page) { |
5096 | printf("vm_page_verify_free_list(color=%u, npages=%u): page %p not found phys=%u\n" , |
5097 | color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page)); |
5098 | _vm_page_print(look_for_page); |
5099 | for (other_color = 0; |
5100 | other_color < vm_colors; |
5101 | other_color++) { |
5102 | if (other_color == color) |
5103 | continue; |
5104 | vm_page_verify_free_list(&vm_page_queue_free[other_color].qhead, |
5105 | other_color, look_for_page, FALSE); |
5106 | } |
5107 | if (color == (unsigned int) -1) { |
5108 | vm_page_verify_free_list(&vm_lopage_queue_free, |
5109 | (unsigned int) -1, look_for_page, FALSE); |
5110 | } |
5111 | panic("vm_page_verify_free_list(color=%u)\n" , color); |
5112 | } |
5113 | if (!expect_page && found_page) { |
5114 | printf("vm_page_verify_free_list(color=%u, npages=%u): page %p found phys=%u\n" , |
5115 | color, npages, look_for_page, VM_PAGE_GET_PHYS_PAGE(look_for_page)); |
5116 | } |
5117 | } |
5118 | return npages; |
5119 | } |
5120 | |
5121 | static boolean_t vm_page_verify_all_free_lists_enabled = FALSE; |
5122 | static void |
5123 | vm_page_verify_free_lists( void ) |
5124 | { |
5125 | unsigned int color, npages, nlopages; |
5126 | boolean_t toggle = TRUE; |
5127 | |
5128 | if (! vm_page_verify_all_free_lists_enabled) |
5129 | return; |
5130 | |
5131 | npages = 0; |
5132 | |
5133 | lck_mtx_lock(&vm_page_queue_free_lock); |
5134 | |
5135 | if (vm_page_verify_this_free_list_enabled == TRUE) { |
5136 | /* |
5137 | * This variable has been set globally for extra checking of |
5138 | * each free list Q. Since we didn't set it, we don't own it |
5139 | * and we shouldn't toggle it. |
5140 | */ |
5141 | toggle = FALSE; |
5142 | } |
5143 | |
5144 | if (toggle == TRUE) { |
5145 | vm_page_verify_this_free_list_enabled = TRUE; |
5146 | } |
5147 | |
5148 | for( color = 0; color < vm_colors; color++ ) { |
5149 | npages += vm_page_verify_free_list(&vm_page_queue_free[color].qhead, |
5150 | color, VM_PAGE_NULL, FALSE); |
5151 | } |
5152 | nlopages = vm_page_verify_free_list(&vm_lopage_queue_free, |
5153 | (unsigned int) -1, |
5154 | VM_PAGE_NULL, FALSE); |
5155 | if (npages != vm_page_free_count || nlopages != vm_lopage_free_count) |
5156 | panic("vm_page_verify_free_lists: " |
5157 | "npages %u free_count %d nlopages %u lo_free_count %u" , |
5158 | npages, vm_page_free_count, nlopages, vm_lopage_free_count); |
5159 | |
5160 | if (toggle == TRUE) { |
5161 | vm_page_verify_this_free_list_enabled = FALSE; |
5162 | } |
5163 | |
5164 | lck_mtx_unlock(&vm_page_queue_free_lock); |
5165 | } |
5166 | |
5167 | #endif /* MACH_ASSERT */ |
5168 | |
5169 | |
5170 | |
5171 | #if __arm64__ |
5172 | /* |
5173 | * 1 or more clients (currently only SEP) ask for a large contiguous chunk of memory |
5174 | * after the system has 'aged'. To ensure that other allocation requests don't mess |
5175 | * with the chances of that request being satisfied, we pre-allocate a single contiguous |
5176 | * 10MB buffer and hand it out to the first request of >= 4MB. |
5177 | */ |
5178 | |
5179 | kern_return_t cpm_preallocate_early(void); |
5180 | |
5181 | vm_page_t cpm_preallocated_pages_list = NULL; |
5182 | boolean_t preallocated_buffer_available = FALSE; |
5183 | |
5184 | #define PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT ((10 * 1024 * 1024) / PAGE_SIZE_64) /* 10 MB */ |
5185 | #define MIN_CONTIG_PAGES_REQUEST_FOR_PREALLOCATED_BUFFER ((4 * 1024 *1024) / PAGE_SIZE_64) /* 4 MB */ |
5186 | |
5187 | kern_return_t |
5188 | cpm_preallocate_early(void) |
5189 | { |
5190 | |
5191 | kern_return_t kr = KERN_SUCCESS; |
5192 | vm_map_size_t prealloc_size = (PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT * PAGE_SIZE_64); |
5193 | |
5194 | printf("cpm_preallocate_early called to preallocate contiguous buffer of %llu pages\n" , PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT); |
5195 | |
5196 | kr = cpm_allocate(CAST_DOWN(vm_size_t, prealloc_size), &cpm_preallocated_pages_list, 0, 0, TRUE, 0); |
5197 | |
5198 | if (kr != KERN_SUCCESS) { |
5199 | printf("cpm_allocate for preallocated contig buffer failed with %d.\n" , kr); |
5200 | } else { |
5201 | preallocated_buffer_available = TRUE; |
5202 | } |
5203 | |
5204 | return kr; |
5205 | } |
5206 | #endif /* __arm64__ */ |
5207 | |
5208 | |
5209 | extern boolean_t (* volatile consider_buffer_cache_collect)(int); |
5210 | |
5211 | /* |
5212 | * CONTIGUOUS PAGE ALLOCATION |
5213 | * |
5214 | * Find a region large enough to contain at least n pages |
5215 | * of contiguous physical memory. |
5216 | * |
5217 | * This is done by traversing the vm_page_t array in a linear fashion |
5218 | * we assume that the vm_page_t array has the avaiable physical pages in an |
5219 | * ordered, ascending list... this is currently true of all our implementations |
5220 | * and must remain so... there can be 'holes' in the array... we also can |
5221 | * no longer tolerate the vm_page_t's in the list being 'freed' and reclaimed |
5222 | * which use to happen via 'vm_page_convert'... that function was no longer |
5223 | * being called and was removed... |
5224 | * |
5225 | * The basic flow consists of stabilizing some of the interesting state of |
5226 | * a vm_page_t behind the vm_page_queue and vm_page_free locks... we start our |
5227 | * sweep at the beginning of the array looking for pages that meet our criterea |
5228 | * for a 'stealable' page... currently we are pretty conservative... if the page |
5229 | * meets this criterea and is physically contiguous to the previous page in the 'run' |
5230 | * we keep developing it. If we hit a page that doesn't fit, we reset our state |
5231 | * and start to develop a new run... if at this point we've already considered |
5232 | * at least MAX_CONSIDERED_BEFORE_YIELD pages, we'll drop the 2 locks we hold, |
5233 | * and mutex_pause (which will yield the processor), to keep the latency low w/r |
5234 | * to other threads trying to acquire free pages (or move pages from q to q), |
5235 | * and then continue from the spot we left off... we only make 1 pass through the |
5236 | * array. Once we have a 'run' that is long enough, we'll go into the loop which |
5237 | * which steals the pages from the queues they're currently on... pages on the free |
5238 | * queue can be stolen directly... pages that are on any of the other queues |
5239 | * must be removed from the object they are tabled on... this requires taking the |
5240 | * object lock... we do this as a 'try' to prevent deadlocks... if the 'try' fails |
5241 | * or if the state of the page behind the vm_object lock is no longer viable, we'll |
5242 | * dump the pages we've currently stolen back to the free list, and pick up our |
5243 | * scan from the point where we aborted the 'current' run. |
5244 | * |
5245 | * |
5246 | * Requirements: |
5247 | * - neither vm_page_queue nor vm_free_list lock can be held on entry |
5248 | * |
5249 | * Returns a pointer to a list of gobbled/wired pages or VM_PAGE_NULL. |
5250 | * |
5251 | * Algorithm: |
5252 | */ |
5253 | |
5254 | #define MAX_CONSIDERED_BEFORE_YIELD 1000 |
5255 | |
5256 | |
5257 | #define RESET_STATE_OF_RUN() \ |
5258 | MACRO_BEGIN \ |
5259 | prevcontaddr = -2; \ |
5260 | start_pnum = -1; \ |
5261 | free_considered = 0; \ |
5262 | substitute_needed = 0; \ |
5263 | npages = 0; \ |
5264 | MACRO_END |
5265 | |
5266 | /* |
5267 | * Can we steal in-use (i.e. not free) pages when searching for |
5268 | * physically-contiguous pages ? |
5269 | */ |
5270 | #define VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL 1 |
5271 | |
5272 | static unsigned int vm_page_find_contiguous_last_idx = 0, vm_page_lomem_find_contiguous_last_idx = 0; |
5273 | #if DEBUG |
5274 | int vm_page_find_contig_debug = 0; |
5275 | #endif |
5276 | |
5277 | static vm_page_t |
5278 | vm_page_find_contiguous( |
5279 | unsigned int contig_pages, |
5280 | ppnum_t max_pnum, |
5281 | ppnum_t pnum_mask, |
5282 | boolean_t wire, |
5283 | int flags) |
5284 | { |
5285 | vm_page_t m = NULL; |
5286 | ppnum_t prevcontaddr = 0; |
5287 | ppnum_t start_pnum = 0; |
5288 | unsigned int npages = 0, considered = 0, scanned = 0; |
5289 | unsigned int page_idx = 0, start_idx = 0, last_idx = 0, orig_last_idx = 0; |
5290 | unsigned int idx_last_contig_page_found = 0; |
5291 | int free_considered = 0, free_available = 0; |
5292 | int substitute_needed = 0; |
5293 | boolean_t wrapped, zone_gc_called = FALSE; |
5294 | kern_return_t kr; |
5295 | #if DEBUG |
5296 | clock_sec_t tv_start_sec = 0, tv_end_sec = 0; |
5297 | clock_usec_t tv_start_usec = 0, tv_end_usec = 0; |
5298 | #endif |
5299 | |
5300 | int yielded = 0; |
5301 | int dumped_run = 0; |
5302 | int stolen_pages = 0; |
5303 | int compressed_pages = 0; |
5304 | |
5305 | |
5306 | if (contig_pages == 0) |
5307 | return VM_PAGE_NULL; |
5308 | |
5309 | full_scan_again: |
5310 | |
5311 | #if MACH_ASSERT |
5312 | vm_page_verify_free_lists(); |
5313 | #endif |
5314 | #if DEBUG |
5315 | clock_get_system_microtime(&tv_start_sec, &tv_start_usec); |
5316 | #endif |
5317 | PAGE_REPLACEMENT_ALLOWED(TRUE); |
5318 | |
5319 | vm_page_lock_queues(); |
5320 | |
5321 | #if __arm64__ |
5322 | if (preallocated_buffer_available) { |
5323 | |
5324 | if ((contig_pages >= MIN_CONTIG_PAGES_REQUEST_FOR_PREALLOCATED_BUFFER) && (contig_pages <= PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT)) { |
5325 | |
5326 | m = cpm_preallocated_pages_list; |
5327 | |
5328 | start_idx = (unsigned int) (m - &vm_pages[0]); |
5329 | |
5330 | if (wire == FALSE) { |
5331 | |
5332 | last_idx = start_idx; |
5333 | |
5334 | for(npages = 0; npages < contig_pages; npages++, last_idx++) { |
5335 | |
5336 | assert(vm_pages[last_idx].vmp_gobbled == FALSE); |
5337 | |
5338 | vm_pages[last_idx].vmp_gobbled = TRUE; |
5339 | vm_page_gobble_count++; |
5340 | |
5341 | assert(1 == vm_pages[last_idx].vmp_wire_count); |
5342 | /* |
5343 | * Gobbled pages are counted as wired pages. So no need to drop |
5344 | * the global wired page count. Just the page's wire count is fine. |
5345 | */ |
5346 | vm_pages[last_idx].vmp_wire_count--; |
5347 | vm_pages[last_idx].vmp_q_state = VM_PAGE_NOT_ON_Q; |
5348 | } |
5349 | |
5350 | } |
5351 | |
5352 | last_idx = start_idx + contig_pages - 1; |
5353 | |
5354 | vm_pages[last_idx].vmp_snext = NULL; |
5355 | |
5356 | printf("Using preallocated buffer: Requested size (pages):%d... index range: %d-%d...freeing %llu pages\n" , contig_pages, start_idx, last_idx, PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT - contig_pages); |
5357 | |
5358 | last_idx += 1; |
5359 | for(npages = contig_pages; npages < PREALLOCATED_CONTIG_BUFFER_PAGES_COUNT; npages++, last_idx++) { |
5360 | |
5361 | VM_PAGE_ZERO_PAGEQ_ENTRY(&vm_pages[last_idx]); |
5362 | vm_page_free(&vm_pages[last_idx]); |
5363 | } |
5364 | |
5365 | cpm_preallocated_pages_list = NULL; |
5366 | preallocated_buffer_available = FALSE; |
5367 | |
5368 | goto done_scanning; |
5369 | } |
5370 | } |
5371 | #endif /* __arm64__ */ |
5372 | |
5373 | lck_mtx_lock(&vm_page_queue_free_lock); |
5374 | |
5375 | RESET_STATE_OF_RUN(); |
5376 | |
5377 | scanned = 0; |
5378 | considered = 0; |
5379 | free_available = vm_page_free_count - vm_page_free_reserved; |
5380 | |
5381 | wrapped = FALSE; |
5382 | |
5383 | if(flags & KMA_LOMEM) |
5384 | idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx; |
5385 | else |
5386 | idx_last_contig_page_found = vm_page_find_contiguous_last_idx; |
5387 | |
5388 | orig_last_idx = idx_last_contig_page_found; |
5389 | last_idx = orig_last_idx; |
5390 | |
5391 | for (page_idx = last_idx, start_idx = last_idx; |
5392 | npages < contig_pages && page_idx < vm_pages_count; |
5393 | page_idx++) { |
5394 | retry: |
5395 | if (wrapped && |
5396 | npages == 0 && |
5397 | page_idx >= orig_last_idx) { |
5398 | /* |
5399 | * We're back where we started and we haven't |
5400 | * found any suitable contiguous range. Let's |
5401 | * give up. |
5402 | */ |
5403 | break; |
5404 | } |
5405 | scanned++; |
5406 | m = &vm_pages[page_idx]; |
5407 | |
5408 | assert(!m->vmp_fictitious); |
5409 | assert(!m->vmp_private); |
5410 | |
5411 | if (max_pnum && VM_PAGE_GET_PHYS_PAGE(m) > max_pnum) { |
5412 | /* no more low pages... */ |
5413 | break; |
5414 | } |
5415 | if (!npages & ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0)) { |
5416 | /* |
5417 | * not aligned |
5418 | */ |
5419 | RESET_STATE_OF_RUN(); |
5420 | |
5421 | } else if (VM_PAGE_WIRED(m) || m->vmp_gobbled || |
5422 | m->vmp_laundry || m->vmp_wanted || |
5423 | m->vmp_cleaning || m->vmp_overwriting || m->vmp_free_when_done) { |
5424 | /* |
5425 | * page is in a transient state |
5426 | * or a state we don't want to deal |
5427 | * with, so don't consider it which |
5428 | * means starting a new run |
5429 | */ |
5430 | RESET_STATE_OF_RUN(); |
5431 | |
5432 | } else if ((m->vmp_q_state == VM_PAGE_NOT_ON_Q) || |
5433 | (m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q) || |
5434 | (m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q) || |
5435 | (m->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { |
5436 | /* |
5437 | * page needs to be on one of our queues (other then the pageout or special free queues) |
5438 | * or it needs to belong to the compressor pool (which is now indicated |
5439 | * by vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR and falls out |
5440 | * from the check for VM_PAGE_NOT_ON_Q) |
5441 | * in order for it to be stable behind the |
5442 | * locks we hold at this point... |
5443 | * if not, don't consider it which |
5444 | * means starting a new run |
5445 | */ |
5446 | RESET_STATE_OF_RUN(); |
5447 | |
5448 | } else if ((m->vmp_q_state != VM_PAGE_ON_FREE_Q) && (!m->vmp_tabled || m->vmp_busy)) { |
5449 | /* |
5450 | * pages on the free list are always 'busy' |
5451 | * so we couldn't test for 'busy' in the check |
5452 | * for the transient states... pages that are |
5453 | * 'free' are never 'tabled', so we also couldn't |
5454 | * test for 'tabled'. So we check here to make |
5455 | * sure that a non-free page is not busy and is |
5456 | * tabled on an object... |
5457 | * if not, don't consider it which |
5458 | * means starting a new run |
5459 | */ |
5460 | RESET_STATE_OF_RUN(); |
5461 | |
5462 | } else { |
5463 | if (VM_PAGE_GET_PHYS_PAGE(m) != prevcontaddr + 1) { |
5464 | if ((VM_PAGE_GET_PHYS_PAGE(m) & pnum_mask) != 0) { |
5465 | RESET_STATE_OF_RUN(); |
5466 | goto did_consider; |
5467 | } else { |
5468 | npages = 1; |
5469 | start_idx = page_idx; |
5470 | start_pnum = VM_PAGE_GET_PHYS_PAGE(m); |
5471 | } |
5472 | } else { |
5473 | npages++; |
5474 | } |
5475 | prevcontaddr = VM_PAGE_GET_PHYS_PAGE(m); |
5476 | |
5477 | VM_PAGE_CHECK(m); |
5478 | if (m->vmp_q_state == VM_PAGE_ON_FREE_Q) { |
5479 | free_considered++; |
5480 | } else { |
5481 | /* |
5482 | * This page is not free. |
5483 | * If we can't steal used pages, |
5484 | * we have to give up this run |
5485 | * and keep looking. |
5486 | * Otherwise, we might need to |
5487 | * move the contents of this page |
5488 | * into a substitute page. |
5489 | */ |
5490 | #if VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL |
5491 | if (m->vmp_pmapped || m->vmp_dirty || m->vmp_precious) { |
5492 | substitute_needed++; |
5493 | } |
5494 | #else |
5495 | RESET_STATE_OF_RUN(); |
5496 | #endif |
5497 | } |
5498 | |
5499 | if ((free_considered + substitute_needed) > free_available) { |
5500 | /* |
5501 | * if we let this run continue |
5502 | * we will end up dropping the vm_page_free_count |
5503 | * below the reserve limit... we need to abort |
5504 | * this run, but we can at least re-consider this |
5505 | * page... thus the jump back to 'retry' |
5506 | */ |
5507 | RESET_STATE_OF_RUN(); |
5508 | |
5509 | if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) { |
5510 | considered++; |
5511 | goto retry; |
5512 | } |
5513 | /* |
5514 | * free_available == 0 |
5515 | * so can't consider any free pages... if |
5516 | * we went to retry in this case, we'd |
5517 | * get stuck looking at the same page |
5518 | * w/o making any forward progress |
5519 | * we also want to take this path if we've already |
5520 | * reached our limit that controls the lock latency |
5521 | */ |
5522 | } |
5523 | } |
5524 | did_consider: |
5525 | if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) { |
5526 | |
5527 | PAGE_REPLACEMENT_ALLOWED(FALSE); |
5528 | |
5529 | lck_mtx_unlock(&vm_page_queue_free_lock); |
5530 | vm_page_unlock_queues(); |
5531 | |
5532 | mutex_pause(0); |
5533 | |
5534 | PAGE_REPLACEMENT_ALLOWED(TRUE); |
5535 | |
5536 | vm_page_lock_queues(); |
5537 | lck_mtx_lock(&vm_page_queue_free_lock); |
5538 | |
5539 | RESET_STATE_OF_RUN(); |
5540 | /* |
5541 | * reset our free page limit since we |
5542 | * dropped the lock protecting the vm_page_free_queue |
5543 | */ |
5544 | free_available = vm_page_free_count - vm_page_free_reserved; |
5545 | considered = 0; |
5546 | |
5547 | yielded++; |
5548 | |
5549 | goto retry; |
5550 | } |
5551 | considered++; |
5552 | } |
5553 | m = VM_PAGE_NULL; |
5554 | |
5555 | if (npages != contig_pages) { |
5556 | if (!wrapped) { |
5557 | /* |
5558 | * We didn't find a contiguous range but we didn't |
5559 | * start from the very first page. |
5560 | * Start again from the very first page. |
5561 | */ |
5562 | RESET_STATE_OF_RUN(); |
5563 | if( flags & KMA_LOMEM) |
5564 | idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = 0; |
5565 | else |
5566 | idx_last_contig_page_found = vm_page_find_contiguous_last_idx = 0; |
5567 | last_idx = 0; |
5568 | page_idx = last_idx; |
5569 | wrapped = TRUE; |
5570 | goto retry; |
5571 | } |
5572 | lck_mtx_unlock(&vm_page_queue_free_lock); |
5573 | } else { |
5574 | vm_page_t m1; |
5575 | vm_page_t m2; |
5576 | unsigned int cur_idx; |
5577 | unsigned int tmp_start_idx; |
5578 | vm_object_t locked_object = VM_OBJECT_NULL; |
5579 | boolean_t abort_run = FALSE; |
5580 | |
5581 | assert(page_idx - start_idx == contig_pages); |
5582 | |
5583 | tmp_start_idx = start_idx; |
5584 | |
5585 | /* |
5586 | * first pass through to pull the free pages |
5587 | * off of the free queue so that in case we |
5588 | * need substitute pages, we won't grab any |
5589 | * of the free pages in the run... we'll clear |
5590 | * the 'free' bit in the 2nd pass, and even in |
5591 | * an abort_run case, we'll collect all of the |
5592 | * free pages in this run and return them to the free list |
5593 | */ |
5594 | while (start_idx < page_idx) { |
5595 | |
5596 | m1 = &vm_pages[start_idx++]; |
5597 | |
5598 | #if !VM_PAGE_FIND_CONTIGUOUS_CAN_STEAL |
5599 | assert(m1->vmp_q_state == VM_PAGE_ON_FREE_Q); |
5600 | #endif |
5601 | |
5602 | if (m1->vmp_q_state == VM_PAGE_ON_FREE_Q) { |
5603 | unsigned int color; |
5604 | |
5605 | color = VM_PAGE_GET_COLOR(m1); |
5606 | #if MACH_ASSERT |
5607 | vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, m1, TRUE); |
5608 | #endif |
5609 | vm_page_queue_remove(&vm_page_queue_free[color].qhead, |
5610 | m1, |
5611 | vm_page_t, |
5612 | vmp_pageq); |
5613 | |
5614 | VM_PAGE_ZERO_PAGEQ_ENTRY(m1); |
5615 | #if MACH_ASSERT |
5616 | vm_page_verify_free_list(&vm_page_queue_free[color].qhead, color, VM_PAGE_NULL, FALSE); |
5617 | #endif |
5618 | /* |
5619 | * Clear the "free" bit so that this page |
5620 | * does not get considered for another |
5621 | * concurrent physically-contiguous allocation. |
5622 | */ |
5623 | m1->vmp_q_state = VM_PAGE_NOT_ON_Q; |
5624 | assert(m1->vmp_busy); |
5625 | |
5626 | vm_page_free_count--; |
5627 | } |
5628 | } |
5629 | if( flags & KMA_LOMEM) |
5630 | vm_page_lomem_find_contiguous_last_idx = page_idx; |
5631 | else |
5632 | vm_page_find_contiguous_last_idx = page_idx; |
5633 | |
5634 | /* |
5635 | * we can drop the free queue lock at this point since |
5636 | * we've pulled any 'free' candidates off of the list |
5637 | * we need it dropped so that we can do a vm_page_grab |
5638 | * when substituing for pmapped/dirty pages |
5639 | */ |
5640 | lck_mtx_unlock(&vm_page_queue_free_lock); |
5641 | |
5642 | start_idx = tmp_start_idx; |
5643 | cur_idx = page_idx - 1; |
5644 | |
5645 | while (start_idx++ < page_idx) { |
5646 | /* |
5647 | * must go through the list from back to front |
5648 | * so that the page list is created in the |
5649 | * correct order - low -> high phys addresses |
5650 | */ |
5651 | m1 = &vm_pages[cur_idx--]; |
5652 | |
5653 | if (m1->vmp_object == 0) { |
5654 | /* |
5655 | * page has already been removed from |
5656 | * the free list in the 1st pass |
5657 | */ |
5658 | assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q); |
5659 | assert(m1->vmp_offset == (vm_object_offset_t) -1); |
5660 | assert(m1->vmp_busy); |
5661 | assert(!m1->vmp_wanted); |
5662 | assert(!m1->vmp_laundry); |
5663 | } else { |
5664 | vm_object_t object; |
5665 | int refmod; |
5666 | boolean_t disconnected, reusable; |
5667 | |
5668 | if (abort_run == TRUE) |
5669 | continue; |
5670 | |
5671 | assert(m1->vmp_q_state != VM_PAGE_NOT_ON_Q); |
5672 | |
5673 | object = VM_PAGE_OBJECT(m1); |
5674 | |
5675 | if (object != locked_object) { |
5676 | if (locked_object) { |
5677 | vm_object_unlock(locked_object); |
5678 | locked_object = VM_OBJECT_NULL; |
5679 | } |
5680 | if (vm_object_lock_try(object)) |
5681 | locked_object = object; |
5682 | } |
5683 | if (locked_object == VM_OBJECT_NULL || |
5684 | (VM_PAGE_WIRED(m1) || m1->vmp_gobbled || |
5685 | m1->vmp_laundry || m1->vmp_wanted || |
5686 | m1->vmp_cleaning || m1->vmp_overwriting || m1->vmp_free_when_done || m1->vmp_busy) || |
5687 | (m1->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q)) { |
5688 | |
5689 | if (locked_object) { |
5690 | vm_object_unlock(locked_object); |
5691 | locked_object = VM_OBJECT_NULL; |
5692 | } |
5693 | tmp_start_idx = cur_idx; |
5694 | abort_run = TRUE; |
5695 | continue; |
5696 | } |
5697 | |
5698 | disconnected = FALSE; |
5699 | reusable = FALSE; |
5700 | |
5701 | if ((m1->vmp_reusable || |
5702 | object->all_reusable) && |
5703 | (m1->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q) && |
5704 | !m1->vmp_dirty && |
5705 | !m1->vmp_reference) { |
5706 | /* reusable page... */ |
5707 | refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1)); |
5708 | disconnected = TRUE; |
5709 | if (refmod == 0) { |
5710 | /* |
5711 | * ... not reused: can steal |
5712 | * without relocating contents. |
5713 | */ |
5714 | reusable = TRUE; |
5715 | } |
5716 | } |
5717 | |
5718 | if ((m1->vmp_pmapped && |
5719 | ! reusable) || |
5720 | m1->vmp_dirty || |
5721 | m1->vmp_precious) { |
5722 | vm_object_offset_t offset; |
5723 | |
5724 | m2 = vm_page_grab(); |
5725 | |
5726 | if (m2 == VM_PAGE_NULL) { |
5727 | if (locked_object) { |
5728 | vm_object_unlock(locked_object); |
5729 | locked_object = VM_OBJECT_NULL; |
5730 | } |
5731 | tmp_start_idx = cur_idx; |
5732 | abort_run = TRUE; |
5733 | continue; |
5734 | } |
5735 | if (! disconnected) { |
5736 | if (m1->vmp_pmapped) |
5737 | refmod = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m1)); |
5738 | else |
5739 | refmod = 0; |
5740 | } |
5741 | |
5742 | /* copy the page's contents */ |
5743 | pmap_copy_page(VM_PAGE_GET_PHYS_PAGE(m1), VM_PAGE_GET_PHYS_PAGE(m2)); |
5744 | /* copy the page's state */ |
5745 | assert(!VM_PAGE_WIRED(m1)); |
5746 | assert(m1->vmp_q_state != VM_PAGE_ON_FREE_Q); |
5747 | assert(m1->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q); |
5748 | assert(!m1->vmp_laundry); |
5749 | m2->vmp_reference = m1->vmp_reference; |
5750 | assert(!m1->vmp_gobbled); |
5751 | assert(!m1->vmp_private); |
5752 | m2->vmp_no_cache = m1->vmp_no_cache; |
5753 | m2->vmp_xpmapped = 0; |
5754 | assert(!m1->vmp_busy); |
5755 | assert(!m1->vmp_wanted); |
5756 | assert(!m1->vmp_fictitious); |
5757 | m2->vmp_pmapped = m1->vmp_pmapped; /* should flush cache ? */ |
5758 | m2->vmp_wpmapped = m1->vmp_wpmapped; |
5759 | assert(!m1->vmp_free_when_done); |
5760 | m2->vmp_absent = m1->vmp_absent; |
5761 | m2->vmp_error = m1->vmp_error; |
5762 | m2->vmp_dirty = m1->vmp_dirty; |
5763 | assert(!m1->vmp_cleaning); |
5764 | m2->vmp_precious = m1->vmp_precious; |
5765 | m2->vmp_clustered = m1->vmp_clustered; |
5766 | assert(!m1->vmp_overwriting); |
5767 | m2->vmp_restart = m1->vmp_restart; |
5768 | m2->vmp_unusual = m1->vmp_unusual; |
5769 | m2->vmp_cs_validated = m1->vmp_cs_validated; |
5770 | m2->vmp_cs_tainted = m1->vmp_cs_tainted; |
5771 | m2->vmp_cs_nx = m1->vmp_cs_nx; |
5772 | |
5773 | /* |
5774 | * If m1 had really been reusable, |
5775 | * we would have just stolen it, so |
5776 | * let's not propagate it's "reusable" |
5777 | * bit and assert that m2 is not |
5778 | * marked as "reusable". |
5779 | */ |
5780 | // m2->vmp_reusable = m1->vmp_reusable; |
5781 | assert(!m2->vmp_reusable); |
5782 | |
5783 | // assert(!m1->vmp_lopage); |
5784 | |
5785 | if (m1->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) |
5786 | m2->vmp_q_state = VM_PAGE_USED_BY_COMPRESSOR; |
5787 | |
5788 | /* |
5789 | * page may need to be flushed if |
5790 | * it is marshalled into a UPL |
5791 | * that is going to be used by a device |
5792 | * that doesn't support coherency |
5793 | */ |
5794 | m2->vmp_written_by_kernel = TRUE; |
5795 | |
5796 | /* |
5797 | * make sure we clear the ref/mod state |
5798 | * from the pmap layer... else we risk |
5799 | * inheriting state from the last time |
5800 | * this page was used... |
5801 | */ |
5802 | pmap_clear_refmod(VM_PAGE_GET_PHYS_PAGE(m2), VM_MEM_MODIFIED | VM_MEM_REFERENCED); |
5803 | |
5804 | if (refmod & VM_MEM_REFERENCED) |
5805 | m2->vmp_reference = TRUE; |
5806 | if (refmod & VM_MEM_MODIFIED) { |
5807 | SET_PAGE_DIRTY(m2, TRUE); |
5808 | } |
5809 | offset = m1->vmp_offset; |
5810 | |
5811 | /* |
5812 | * completely cleans up the state |
5813 | * of the page so that it is ready |
5814 | * to be put onto the free list, or |
5815 | * for this purpose it looks like it |
5816 | * just came off of the free list |
5817 | */ |
5818 | vm_page_free_prepare(m1); |
5819 | |
5820 | /* |
5821 | * now put the substitute page |
5822 | * on the object |
5823 | */ |
5824 | vm_page_insert_internal(m2, locked_object, offset, VM_KERN_MEMORY_NONE, TRUE, TRUE, FALSE, FALSE, NULL); |
5825 | |
5826 | if (m2->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) { |
5827 | m2->vmp_pmapped = TRUE; |
5828 | m2->vmp_wpmapped = TRUE; |
5829 | |
5830 | PMAP_ENTER(kernel_pmap, m2->vmp_offset, m2, |
5831 | VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE, kr); |
5832 | |
5833 | assert(kr == KERN_SUCCESS); |
5834 | |
5835 | compressed_pages++; |
5836 | |
5837 | } else { |
5838 | if (m2->vmp_reference) |
5839 | vm_page_activate(m2); |
5840 | else |
5841 | vm_page_deactivate(m2); |
5842 | } |
5843 | PAGE_WAKEUP_DONE(m2); |
5844 | |
5845 | } else { |
5846 | assert(m1->vmp_q_state != VM_PAGE_USED_BY_COMPRESSOR); |
5847 | |
5848 | /* |
5849 | * completely cleans up the state |
5850 | * of the page so that it is ready |
5851 | * to be put onto the free list, or |
5852 | * for this purpose it looks like it |
5853 | * just came off of the free list |
5854 | */ |
5855 | vm_page_free_prepare(m1); |
5856 | } |
5857 | |
5858 | stolen_pages++; |
5859 | |
5860 | } |
5861 | #if CONFIG_BACKGROUND_QUEUE |
5862 | vm_page_assign_background_state(m1); |
5863 | #endif |
5864 | VM_PAGE_ZERO_PAGEQ_ENTRY(m1); |
5865 | m1->vmp_snext = m; |
5866 | m = m1; |
5867 | } |
5868 | if (locked_object) { |
5869 | vm_object_unlock(locked_object); |
5870 | locked_object = VM_OBJECT_NULL; |
5871 | } |
5872 | |
5873 | if (abort_run == TRUE) { |
5874 | /* |
5875 | * want the index of the last |
5876 | * page in this run that was |
5877 | * successfully 'stolen', so back |
5878 | * it up 1 for the auto-decrement on use |
5879 | * and 1 more to bump back over this page |
5880 | */ |
5881 | page_idx = tmp_start_idx + 2; |
5882 | if (page_idx >= vm_pages_count) { |
5883 | if (wrapped) { |
5884 | if (m != VM_PAGE_NULL) { |
5885 | vm_page_unlock_queues(); |
5886 | vm_page_free_list(m, FALSE); |
5887 | vm_page_lock_queues(); |
5888 | m = VM_PAGE_NULL; |
5889 | } |
5890 | dumped_run++; |
5891 | goto done_scanning; |
5892 | } |
5893 | page_idx = last_idx = 0; |
5894 | wrapped = TRUE; |
5895 | } |
5896 | abort_run = FALSE; |
5897 | |
5898 | /* |
5899 | * We didn't find a contiguous range but we didn't |
5900 | * start from the very first page. |
5901 | * Start again from the very first page. |
5902 | */ |
5903 | RESET_STATE_OF_RUN(); |
5904 | |
5905 | if( flags & KMA_LOMEM) |
5906 | idx_last_contig_page_found = vm_page_lomem_find_contiguous_last_idx = page_idx; |
5907 | else |
5908 | idx_last_contig_page_found = vm_page_find_contiguous_last_idx = page_idx; |
5909 | |
5910 | last_idx = page_idx; |
5911 | |
5912 | if (m != VM_PAGE_NULL) { |
5913 | vm_page_unlock_queues(); |
5914 | vm_page_free_list(m, FALSE); |
5915 | vm_page_lock_queues(); |
5916 | m = VM_PAGE_NULL; |
5917 | } |
5918 | dumped_run++; |
5919 | |
5920 | lck_mtx_lock(&vm_page_queue_free_lock); |
5921 | /* |
5922 | * reset our free page limit since we |
5923 | * dropped the lock protecting the vm_page_free_queue |
5924 | */ |
5925 | free_available = vm_page_free_count - vm_page_free_reserved; |
5926 | goto retry; |
5927 | } |
5928 | |
5929 | for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) { |
5930 | |
5931 | assert(m1->vmp_q_state == VM_PAGE_NOT_ON_Q); |
5932 | assert(m1->vmp_wire_count == 0); |
5933 | |
5934 | if (wire == TRUE) { |
5935 | m1->vmp_wire_count++; |
5936 | m1->vmp_q_state = VM_PAGE_IS_WIRED; |
5937 | } else |
5938 | m1->vmp_gobbled = TRUE; |
5939 | } |
5940 | if (wire == FALSE) |
5941 | vm_page_gobble_count += npages; |
5942 | |
5943 | /* |
5944 | * gobbled pages are also counted as wired pages |
5945 | */ |
5946 | vm_page_wire_count += npages; |
5947 | |
5948 | assert(vm_page_verify_contiguous(m, npages)); |
5949 | } |
5950 | done_scanning: |
5951 | PAGE_REPLACEMENT_ALLOWED(FALSE); |
5952 | |
5953 | vm_page_unlock_queues(); |
5954 | |
5955 | #if DEBUG |
5956 | clock_get_system_microtime(&tv_end_sec, &tv_end_usec); |
5957 | |
5958 | tv_end_sec -= tv_start_sec; |
5959 | if (tv_end_usec < tv_start_usec) { |
5960 | tv_end_sec--; |
5961 | tv_end_usec += 1000000; |
5962 | } |
5963 | tv_end_usec -= tv_start_usec; |
5964 | if (tv_end_usec >= 1000000) { |
5965 | tv_end_sec++; |
5966 | tv_end_sec -= 1000000; |
5967 | } |
5968 | if (vm_page_find_contig_debug) { |
5969 | 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" , |
5970 | __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT, |
5971 | (long)tv_end_sec, tv_end_usec, orig_last_idx, |
5972 | scanned, yielded, dumped_run, stolen_pages, compressed_pages); |
5973 | } |
5974 | |
5975 | #endif |
5976 | #if MACH_ASSERT |
5977 | vm_page_verify_free_lists(); |
5978 | #endif |
5979 | if (m == NULL && zone_gc_called == FALSE) { |
5980 | 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" , |
5981 | __func__, contig_pages, max_pnum, npages, (vm_object_offset_t)start_pnum << PAGE_SHIFT, |
5982 | scanned, yielded, dumped_run, stolen_pages, compressed_pages, vm_page_wire_count); |
5983 | |
5984 | if (consider_buffer_cache_collect != NULL) { |
5985 | (void)(*consider_buffer_cache_collect)(1); |
5986 | } |
5987 | |
5988 | consider_zone_gc(FALSE); |
5989 | |
5990 | zone_gc_called = TRUE; |
5991 | |
5992 | printf("vm_page_find_contiguous: zone_gc called... wired count is %d\n" , vm_page_wire_count); |
5993 | goto full_scan_again; |
5994 | } |
5995 | |
5996 | return m; |
5997 | } |
5998 | |
5999 | /* |
6000 | * Allocate a list of contiguous, wired pages. |
6001 | */ |
6002 | kern_return_t |
6003 | cpm_allocate( |
6004 | vm_size_t size, |
6005 | vm_page_t *list, |
6006 | ppnum_t max_pnum, |
6007 | ppnum_t pnum_mask, |
6008 | boolean_t wire, |
6009 | int flags) |
6010 | { |
6011 | vm_page_t pages; |
6012 | unsigned int npages; |
6013 | |
6014 | if (size % PAGE_SIZE != 0) |
6015 | return KERN_INVALID_ARGUMENT; |
6016 | |
6017 | npages = (unsigned int) (size / PAGE_SIZE); |
6018 | if (npages != size / PAGE_SIZE) { |
6019 | /* 32-bit overflow */ |
6020 | return KERN_INVALID_ARGUMENT; |
6021 | } |
6022 | |
6023 | /* |
6024 | * Obtain a pointer to a subset of the free |
6025 | * list large enough to satisfy the request; |
6026 | * the region will be physically contiguous. |
6027 | */ |
6028 | pages = vm_page_find_contiguous(npages, max_pnum, pnum_mask, wire, flags); |
6029 | |
6030 | if (pages == VM_PAGE_NULL) |
6031 | return KERN_NO_SPACE; |
6032 | /* |
6033 | * determine need for wakeups |
6034 | */ |
6035 | if (vm_page_free_count < vm_page_free_min) |
6036 | thread_wakeup((event_t) &vm_page_free_wanted); |
6037 | |
6038 | VM_CHECK_MEMORYSTATUS; |
6039 | |
6040 | /* |
6041 | * The CPM pages should now be available and |
6042 | * ordered by ascending physical address. |
6043 | */ |
6044 | assert(vm_page_verify_contiguous(pages, npages)); |
6045 | |
6046 | *list = pages; |
6047 | return KERN_SUCCESS; |
6048 | } |
6049 | |
6050 | |
6051 | unsigned int vm_max_delayed_work_limit = DEFAULT_DELAYED_WORK_LIMIT; |
6052 | |
6053 | /* |
6054 | * when working on a 'run' of pages, it is necessary to hold |
6055 | * the vm_page_queue_lock (a hot global lock) for certain operations |
6056 | * on the page... however, the majority of the work can be done |
6057 | * while merely holding the object lock... in fact there are certain |
6058 | * collections of pages that don't require any work brokered by the |
6059 | * vm_page_queue_lock... to mitigate the time spent behind the global |
6060 | * lock, go to a 2 pass algorithm... collect pages up to DELAYED_WORK_LIMIT |
6061 | * while doing all of the work that doesn't require the vm_page_queue_lock... |
6062 | * then call vm_page_do_delayed_work to acquire the vm_page_queue_lock and do the |
6063 | * necessary work for each page... we will grab the busy bit on the page |
6064 | * if it's not already held so that vm_page_do_delayed_work can drop the object lock |
6065 | * if it can't immediately take the vm_page_queue_lock in order to compete |
6066 | * for the locks in the same order that vm_pageout_scan takes them. |
6067 | * the operation names are modeled after the names of the routines that |
6068 | * need to be called in order to make the changes very obvious in the |
6069 | * original loop |
6070 | */ |
6071 | |
6072 | void |
6073 | vm_page_do_delayed_work( |
6074 | vm_object_t object, |
6075 | vm_tag_t tag, |
6076 | struct vm_page_delayed_work *dwp, |
6077 | int dw_count) |
6078 | { |
6079 | int j; |
6080 | vm_page_t m; |
6081 | vm_page_t local_free_q = VM_PAGE_NULL; |
6082 | |
6083 | /* |
6084 | * pageout_scan takes the vm_page_lock_queues first |
6085 | * then tries for the object lock... to avoid what |
6086 | * is effectively a lock inversion, we'll go to the |
6087 | * trouble of taking them in that same order... otherwise |
6088 | * if this object contains the majority of the pages resident |
6089 | * in the UBC (or a small set of large objects actively being |
6090 | * worked on contain the majority of the pages), we could |
6091 | * cause the pageout_scan thread to 'starve' in its attempt |
6092 | * to find pages to move to the free queue, since it has to |
6093 | * successfully acquire the object lock of any candidate page |
6094 | * before it can steal/clean it. |
6095 | */ |
6096 | if (!vm_page_trylockspin_queues()) { |
6097 | vm_object_unlock(object); |
6098 | |
6099 | vm_page_lockspin_queues(); |
6100 | |
6101 | for (j = 0; ; j++) { |
6102 | if (!vm_object_lock_avoid(object) && |
6103 | _vm_object_lock_try(object)) |
6104 | break; |
6105 | vm_page_unlock_queues(); |
6106 | mutex_pause(j); |
6107 | vm_page_lockspin_queues(); |
6108 | } |
6109 | } |
6110 | for (j = 0; j < dw_count; j++, dwp++) { |
6111 | |
6112 | m = dwp->dw_m; |
6113 | |
6114 | if (dwp->dw_mask & DW_vm_pageout_throttle_up) |
6115 | vm_pageout_throttle_up(m); |
6116 | #if CONFIG_PHANTOM_CACHE |
6117 | if (dwp->dw_mask & DW_vm_phantom_cache_update) |
6118 | vm_phantom_cache_update(m); |
6119 | #endif |
6120 | if (dwp->dw_mask & DW_vm_page_wire) |
6121 | vm_page_wire(m, tag, FALSE); |
6122 | else if (dwp->dw_mask & DW_vm_page_unwire) { |
6123 | boolean_t queueit; |
6124 | |
6125 | queueit = (dwp->dw_mask & (DW_vm_page_free | DW_vm_page_deactivate_internal)) ? FALSE : TRUE; |
6126 | |
6127 | vm_page_unwire(m, queueit); |
6128 | } |
6129 | if (dwp->dw_mask & DW_vm_page_free) { |
6130 | vm_page_free_prepare_queues(m); |
6131 | |
6132 | assert(m->vmp_pageq.next == 0 && m->vmp_pageq.prev == 0); |
6133 | /* |
6134 | * Add this page to our list of reclaimed pages, |
6135 | * to be freed later. |
6136 | */ |
6137 | m->vmp_snext = local_free_q; |
6138 | local_free_q = m; |
6139 | } else { |
6140 | if (dwp->dw_mask & DW_vm_page_deactivate_internal) |
6141 | vm_page_deactivate_internal(m, FALSE); |
6142 | else if (dwp->dw_mask & DW_vm_page_activate) { |
6143 | if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) { |
6144 | vm_page_activate(m); |
6145 | } |
6146 | } |
6147 | else if (dwp->dw_mask & DW_vm_page_speculate) |
6148 | vm_page_speculate(m, TRUE); |
6149 | else if (dwp->dw_mask & DW_enqueue_cleaned) { |
6150 | /* |
6151 | * if we didn't hold the object lock and did this, |
6152 | * we might disconnect the page, then someone might |
6153 | * soft fault it back in, then we would put it on the |
6154 | * cleaned queue, and so we would have a referenced (maybe even dirty) |
6155 | * page on that queue, which we don't want |
6156 | */ |
6157 | int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); |
6158 | |
6159 | if ((refmod_state & VM_MEM_REFERENCED)) { |
6160 | /* |
6161 | * this page has been touched since it got cleaned; let's activate it |
6162 | * if it hasn't already been |
6163 | */ |
6164 | VM_PAGEOUT_DEBUG(vm_pageout_enqueued_cleaned, 1); |
6165 | VM_PAGEOUT_DEBUG(vm_pageout_cleaned_reactivated, 1); |
6166 | |
6167 | if (m->vmp_q_state != VM_PAGE_ON_ACTIVE_Q) |
6168 | vm_page_activate(m); |
6169 | } else { |
6170 | m->vmp_reference = FALSE; |
6171 | vm_page_enqueue_cleaned(m); |
6172 | } |
6173 | } |
6174 | else if (dwp->dw_mask & DW_vm_page_lru) |
6175 | vm_page_lru(m); |
6176 | else if (dwp->dw_mask & DW_VM_PAGE_QUEUES_REMOVE) { |
6177 | if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) |
6178 | vm_page_queues_remove(m, TRUE); |
6179 | } |
6180 | if (dwp->dw_mask & DW_set_reference) |
6181 | m->vmp_reference = TRUE; |
6182 | else if (dwp->dw_mask & DW_clear_reference) |
6183 | m->vmp_reference = FALSE; |
6184 | |
6185 | if (dwp->dw_mask & DW_move_page) { |
6186 | if (m->vmp_q_state != VM_PAGE_ON_PAGEOUT_Q) { |
6187 | vm_page_queues_remove(m, FALSE); |
6188 | |
6189 | assert(VM_PAGE_OBJECT(m) != kernel_object); |
6190 | |
6191 | vm_page_enqueue_inactive(m, FALSE); |
6192 | } |
6193 | } |
6194 | if (dwp->dw_mask & DW_clear_busy) |
6195 | m->vmp_busy = FALSE; |
6196 | |
6197 | if (dwp->dw_mask & DW_PAGE_WAKEUP) |
6198 | PAGE_WAKEUP(m); |
6199 | } |
6200 | } |
6201 | vm_page_unlock_queues(); |
6202 | |
6203 | if (local_free_q) |
6204 | vm_page_free_list(local_free_q, TRUE); |
6205 | |
6206 | VM_CHECK_MEMORYSTATUS; |
6207 | |
6208 | } |
6209 | |
6210 | kern_return_t |
6211 | vm_page_alloc_list( |
6212 | int page_count, |
6213 | int flags, |
6214 | vm_page_t *list) |
6215 | { |
6216 | vm_page_t lo_page_list = VM_PAGE_NULL; |
6217 | vm_page_t mem; |
6218 | int i; |
6219 | |
6220 | if ( !(flags & KMA_LOMEM)) |
6221 | panic("vm_page_alloc_list: called w/o KMA_LOMEM" ); |
6222 | |
6223 | for (i = 0; i < page_count; i++) { |
6224 | |
6225 | mem = vm_page_grablo(); |
6226 | |
6227 | if (mem == VM_PAGE_NULL) { |
6228 | if (lo_page_list) |
6229 | vm_page_free_list(lo_page_list, FALSE); |
6230 | |
6231 | *list = VM_PAGE_NULL; |
6232 | |
6233 | return (KERN_RESOURCE_SHORTAGE); |
6234 | } |
6235 | mem->vmp_snext = lo_page_list; |
6236 | lo_page_list = mem; |
6237 | } |
6238 | *list = lo_page_list; |
6239 | |
6240 | return (KERN_SUCCESS); |
6241 | } |
6242 | |
6243 | void |
6244 | vm_page_set_offset(vm_page_t page, vm_object_offset_t offset) |
6245 | { |
6246 | page->vmp_offset = offset; |
6247 | } |
6248 | |
6249 | vm_page_t |
6250 | vm_page_get_next(vm_page_t page) |
6251 | { |
6252 | return (page->vmp_snext); |
6253 | } |
6254 | |
6255 | vm_object_offset_t |
6256 | vm_page_get_offset(vm_page_t page) |
6257 | { |
6258 | return (page->vmp_offset); |
6259 | } |
6260 | |
6261 | ppnum_t |
6262 | vm_page_get_phys_page(vm_page_t page) |
6263 | { |
6264 | return (VM_PAGE_GET_PHYS_PAGE(page)); |
6265 | } |
6266 | |
6267 | |
6268 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ |
6269 | |
6270 | #if HIBERNATION |
6271 | |
6272 | static vm_page_t hibernate_gobble_queue; |
6273 | |
6274 | static int hibernate_drain_pageout_queue(struct vm_pageout_queue *); |
6275 | static int hibernate_flush_dirty_pages(int); |
6276 | static int hibernate_flush_queue(vm_page_queue_head_t *, int); |
6277 | |
6278 | void hibernate_flush_wait(void); |
6279 | void hibernate_mark_in_progress(void); |
6280 | void hibernate_clear_in_progress(void); |
6281 | |
6282 | void hibernate_free_range(int, int); |
6283 | void hibernate_hash_insert_page(vm_page_t); |
6284 | uint32_t hibernate_mark_as_unneeded(addr64_t, addr64_t, hibernate_page_list_t *, hibernate_page_list_t *); |
6285 | void hibernate_rebuild_vm_structs(void); |
6286 | uint32_t hibernate_teardown_vm_structs(hibernate_page_list_t *, hibernate_page_list_t *); |
6287 | ppnum_t hibernate_lookup_paddr(unsigned int); |
6288 | |
6289 | struct hibernate_statistics { |
6290 | int hibernate_considered; |
6291 | int hibernate_reentered_on_q; |
6292 | int hibernate_found_dirty; |
6293 | int hibernate_skipped_cleaning; |
6294 | int hibernate_skipped_transient; |
6295 | int hibernate_skipped_precious; |
6296 | int hibernate_skipped_external; |
6297 | int hibernate_queue_nolock; |
6298 | int hibernate_queue_paused; |
6299 | int hibernate_throttled; |
6300 | int hibernate_throttle_timeout; |
6301 | int hibernate_drained; |
6302 | int hibernate_drain_timeout; |
6303 | int cd_lock_failed; |
6304 | int cd_found_precious; |
6305 | int cd_found_wired; |
6306 | int cd_found_busy; |
6307 | int cd_found_unusual; |
6308 | int cd_found_cleaning; |
6309 | int cd_found_laundry; |
6310 | int cd_found_dirty; |
6311 | int cd_found_xpmapped; |
6312 | int cd_skipped_xpmapped; |
6313 | int cd_local_free; |
6314 | int cd_total_free; |
6315 | int cd_vm_page_wire_count; |
6316 | int cd_vm_struct_pages_unneeded; |
6317 | int cd_pages; |
6318 | int cd_discarded; |
6319 | int cd_count_wire; |
6320 | } hibernate_stats; |
6321 | |
6322 | |
6323 | /* |
6324 | * clamp the number of 'xpmapped' pages we'll sweep into the hibernation image |
6325 | * so that we don't overrun the estimated image size, which would |
6326 | * result in a hibernation failure. |
6327 | */ |
6328 | #define HIBERNATE_XPMAPPED_LIMIT 40000 |
6329 | |
6330 | |
6331 | static int |
6332 | hibernate_drain_pageout_queue(struct vm_pageout_queue *q) |
6333 | { |
6334 | wait_result_t wait_result; |
6335 | |
6336 | vm_page_lock_queues(); |
6337 | |
6338 | while ( !vm_page_queue_empty(&q->pgo_pending) ) { |
6339 | |
6340 | q->pgo_draining = TRUE; |
6341 | |
6342 | assert_wait_timeout((event_t) (&q->pgo_laundry+1), THREAD_INTERRUPTIBLE, 5000, 1000*NSEC_PER_USEC); |
6343 | |
6344 | vm_page_unlock_queues(); |
6345 | |
6346 | wait_result = thread_block(THREAD_CONTINUE_NULL); |
6347 | |
6348 | if (wait_result == THREAD_TIMED_OUT && !vm_page_queue_empty(&q->pgo_pending)) { |
6349 | hibernate_stats.hibernate_drain_timeout++; |
6350 | |
6351 | if (q == &vm_pageout_queue_external) |
6352 | return (0); |
6353 | |
6354 | return (1); |
6355 | } |
6356 | vm_page_lock_queues(); |
6357 | |
6358 | hibernate_stats.hibernate_drained++; |
6359 | } |
6360 | vm_page_unlock_queues(); |
6361 | |
6362 | return (0); |
6363 | } |
6364 | |
6365 | |
6366 | boolean_t hibernate_skip_external = FALSE; |
6367 | |
6368 | static int |
6369 | hibernate_flush_queue(vm_page_queue_head_t *q, int qcount) |
6370 | { |
6371 | vm_page_t m; |
6372 | vm_object_t l_object = NULL; |
6373 | vm_object_t m_object = NULL; |
6374 | int refmod_state = 0; |
6375 | int try_failed_count = 0; |
6376 | int retval = 0; |
6377 | int current_run = 0; |
6378 | struct vm_pageout_queue *iq; |
6379 | struct vm_pageout_queue *eq; |
6380 | struct vm_pageout_queue *tq; |
6381 | |
6382 | KDBG(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_START, |
6383 | VM_KERNEL_UNSLIDE_OR_PERM(q), qcount); |
6384 | |
6385 | iq = &vm_pageout_queue_internal; |
6386 | eq = &vm_pageout_queue_external; |
6387 | |
6388 | vm_page_lock_queues(); |
6389 | |
6390 | while (qcount && !vm_page_queue_empty(q)) { |
6391 | |
6392 | if (current_run++ == 1000) { |
6393 | if (hibernate_should_abort()) { |
6394 | retval = 1; |
6395 | break; |
6396 | } |
6397 | current_run = 0; |
6398 | } |
6399 | |
6400 | m = (vm_page_t) vm_page_queue_first(q); |
6401 | m_object = VM_PAGE_OBJECT(m); |
6402 | |
6403 | /* |
6404 | * check to see if we currently are working |
6405 | * with the same object... if so, we've |
6406 | * already got the lock |
6407 | */ |
6408 | if (m_object != l_object) { |
6409 | /* |
6410 | * the object associated with candidate page is |
6411 | * different from the one we were just working |
6412 | * with... dump the lock if we still own it |
6413 | */ |
6414 | if (l_object != NULL) { |
6415 | vm_object_unlock(l_object); |
6416 | l_object = NULL; |
6417 | } |
6418 | /* |
6419 | * Try to lock object; since we've alread got the |
6420 | * page queues lock, we can only 'try' for this one. |
6421 | * if the 'try' fails, we need to do a mutex_pause |
6422 | * to allow the owner of the object lock a chance to |
6423 | * run... |
6424 | */ |
6425 | if ( !vm_object_lock_try_scan(m_object)) { |
6426 | |
6427 | if (try_failed_count > 20) { |
6428 | hibernate_stats.hibernate_queue_nolock++; |
6429 | |
6430 | goto reenter_pg_on_q; |
6431 | } |
6432 | |
6433 | vm_page_unlock_queues(); |
6434 | mutex_pause(try_failed_count++); |
6435 | vm_page_lock_queues(); |
6436 | |
6437 | hibernate_stats.hibernate_queue_paused++; |
6438 | continue; |
6439 | } else { |
6440 | l_object = m_object; |
6441 | } |
6442 | } |
6443 | if ( !m_object->alive || m->vmp_cleaning || m->vmp_laundry || m->vmp_busy || m->vmp_absent || m->vmp_error) { |
6444 | /* |
6445 | * page is not to be cleaned |
6446 | * put it back on the head of its queue |
6447 | */ |
6448 | if (m->vmp_cleaning) |
6449 | hibernate_stats.hibernate_skipped_cleaning++; |
6450 | else |
6451 | hibernate_stats.hibernate_skipped_transient++; |
6452 | |
6453 | goto reenter_pg_on_q; |
6454 | } |
6455 | if (m_object->copy == VM_OBJECT_NULL) { |
6456 | if (m_object->purgable == VM_PURGABLE_VOLATILE || m_object->purgable == VM_PURGABLE_EMPTY) { |
6457 | /* |
6458 | * let the normal hibernate image path |
6459 | * deal with these |
6460 | */ |
6461 | goto reenter_pg_on_q; |
6462 | } |
6463 | } |
6464 | if ( !m->vmp_dirty && m->vmp_pmapped) { |
6465 | refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m)); |
6466 | |
6467 | if ((refmod_state & VM_MEM_MODIFIED)) { |
6468 | SET_PAGE_DIRTY(m, FALSE); |
6469 | } |
6470 | } else |
6471 | refmod_state = 0; |
6472 | |
6473 | if ( !m->vmp_dirty) { |
6474 | /* |
6475 | * page is not to be cleaned |
6476 | * put it back on the head of its queue |
6477 | */ |
6478 | if (m->vmp_precious) |
6479 | hibernate_stats.hibernate_skipped_precious++; |
6480 | |
6481 | goto reenter_pg_on_q; |
6482 | } |
6483 | |
6484 | if (hibernate_skip_external == TRUE && !m_object->internal) { |
6485 | |
6486 | hibernate_stats.hibernate_skipped_external++; |
6487 | |
6488 | goto reenter_pg_on_q; |
6489 | } |
6490 | tq = NULL; |
6491 | |
6492 | if (m_object->internal) { |
6493 | if (VM_PAGE_Q_THROTTLED(iq)) |
6494 | tq = iq; |
6495 | } else if (VM_PAGE_Q_THROTTLED(eq)) |
6496 | tq = eq; |
6497 | |
6498 | if (tq != NULL) { |
6499 | wait_result_t wait_result; |
6500 | int wait_count = 5; |
6501 | |
6502 | if (l_object != NULL) { |
6503 | vm_object_unlock(l_object); |
6504 | l_object = NULL; |
6505 | } |
6506 | |
6507 | while (retval == 0) { |
6508 | |
6509 | tq->pgo_throttled = TRUE; |
6510 | |
6511 | assert_wait_timeout((event_t) &tq->pgo_laundry, THREAD_INTERRUPTIBLE, 1000, 1000*NSEC_PER_USEC); |
6512 | |
6513 | vm_page_unlock_queues(); |
6514 | |
6515 | wait_result = thread_block(THREAD_CONTINUE_NULL); |
6516 | |
6517 | vm_page_lock_queues(); |
6518 | |
6519 | if (wait_result != THREAD_TIMED_OUT) |
6520 | break; |
6521 | if (!VM_PAGE_Q_THROTTLED(tq)) |
6522 | break; |
6523 | |
6524 | if (hibernate_should_abort()) |
6525 | retval = 1; |
6526 | |
6527 | if (--wait_count == 0) { |
6528 | |
6529 | hibernate_stats.hibernate_throttle_timeout++; |
6530 | |
6531 | if (tq == eq) { |
6532 | hibernate_skip_external = TRUE; |
6533 | break; |
6534 | } |
6535 | retval = 1; |
6536 | } |
6537 | } |
6538 | if (retval) |
6539 | break; |
6540 | |
6541 | hibernate_stats.hibernate_throttled++; |
6542 | |
6543 | continue; |
6544 | } |
6545 | /* |
6546 | * we've already factored out pages in the laundry which |
6547 | * means this page can't be on the pageout queue so it's |
6548 | * safe to do the vm_page_queues_remove |
6549 | */ |
6550 | vm_page_queues_remove(m, TRUE); |
6551 | |
6552 | if (m_object->internal == TRUE) |
6553 | pmap_disconnect_options(VM_PAGE_GET_PHYS_PAGE(m), PMAP_OPTIONS_COMPRESSOR, NULL); |
6554 | |
6555 | vm_pageout_cluster(m); |
6556 | |
6557 | hibernate_stats.hibernate_found_dirty++; |
6558 | |
6559 | goto next_pg; |
6560 | |
6561 | reenter_pg_on_q: |
6562 | vm_page_queue_remove(q, m, vm_page_t, vmp_pageq); |
6563 | vm_page_queue_enter(q, m, vm_page_t, vmp_pageq); |
6564 | |
6565 | hibernate_stats.hibernate_reentered_on_q++; |
6566 | next_pg: |
6567 | hibernate_stats.hibernate_considered++; |
6568 | |
6569 | qcount--; |
6570 | try_failed_count = 0; |
6571 | } |
6572 | if (l_object != NULL) { |
6573 | vm_object_unlock(l_object); |
6574 | l_object = NULL; |
6575 | } |
6576 | |
6577 | vm_page_unlock_queues(); |
6578 | |
6579 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 4) | DBG_FUNC_END, hibernate_stats.hibernate_found_dirty, retval, 0, 0, 0); |
6580 | |
6581 | return (retval); |
6582 | } |
6583 | |
6584 | |
6585 | static int |
6586 | hibernate_flush_dirty_pages(int pass) |
6587 | { |
6588 | struct vm_speculative_age_q *aq; |
6589 | uint32_t i; |
6590 | |
6591 | if (vm_page_local_q) { |
6592 | for (i = 0; i < vm_page_local_q_count; i++) |
6593 | vm_page_reactivate_local(i, TRUE, FALSE); |
6594 | } |
6595 | |
6596 | for (i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++) { |
6597 | int qcount; |
6598 | vm_page_t m; |
6599 | |
6600 | aq = &vm_page_queue_speculative[i]; |
6601 | |
6602 | if (vm_page_queue_empty(&aq->age_q)) |
6603 | continue; |
6604 | qcount = 0; |
6605 | |
6606 | vm_page_lockspin_queues(); |
6607 | |
6608 | vm_page_queue_iterate(&aq->age_q, |
6609 | m, |
6610 | vm_page_t, |
6611 | vmp_pageq) |
6612 | { |
6613 | qcount++; |
6614 | } |
6615 | vm_page_unlock_queues(); |
6616 | |
6617 | if (qcount) { |
6618 | if (hibernate_flush_queue(&aq->age_q, qcount)) |
6619 | return (1); |
6620 | } |
6621 | } |
6622 | if (hibernate_flush_queue(&vm_page_queue_inactive, vm_page_inactive_count - vm_page_anonymous_count - vm_page_cleaned_count)) |
6623 | return (1); |
6624 | /* XXX FBDP TODO: flush secluded queue */ |
6625 | if (hibernate_flush_queue(&vm_page_queue_anonymous, vm_page_anonymous_count)) |
6626 | return (1); |
6627 | if (hibernate_flush_queue(&vm_page_queue_cleaned, vm_page_cleaned_count)) |
6628 | return (1); |
6629 | if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) |
6630 | return (1); |
6631 | |
6632 | if (pass == 1) |
6633 | vm_compressor_record_warmup_start(); |
6634 | |
6635 | if (hibernate_flush_queue(&vm_page_queue_active, vm_page_active_count)) { |
6636 | if (pass == 1) |
6637 | vm_compressor_record_warmup_end(); |
6638 | return (1); |
6639 | } |
6640 | if (hibernate_drain_pageout_queue(&vm_pageout_queue_internal)) { |
6641 | if (pass == 1) |
6642 | vm_compressor_record_warmup_end(); |
6643 | return (1); |
6644 | } |
6645 | if (pass == 1) |
6646 | vm_compressor_record_warmup_end(); |
6647 | |
6648 | if (hibernate_skip_external == FALSE && hibernate_drain_pageout_queue(&vm_pageout_queue_external)) |
6649 | return (1); |
6650 | |
6651 | return (0); |
6652 | } |
6653 | |
6654 | |
6655 | void |
6656 | hibernate_reset_stats() |
6657 | { |
6658 | bzero(&hibernate_stats, sizeof(struct hibernate_statistics)); |
6659 | } |
6660 | |
6661 | |
6662 | int |
6663 | hibernate_flush_memory() |
6664 | { |
6665 | int retval; |
6666 | |
6667 | assert(VM_CONFIG_COMPRESSOR_IS_PRESENT); |
6668 | |
6669 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_START, vm_page_free_count, 0, 0, 0, 0); |
6670 | |
6671 | hibernate_cleaning_in_progress = TRUE; |
6672 | hibernate_skip_external = FALSE; |
6673 | |
6674 | if ((retval = hibernate_flush_dirty_pages(1)) == 0) { |
6675 | |
6676 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_START, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0); |
6677 | |
6678 | vm_compressor_flush(); |
6679 | |
6680 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 10) | DBG_FUNC_END, VM_PAGE_COMPRESSOR_COUNT, 0, 0, 0, 0); |
6681 | |
6682 | if (consider_buffer_cache_collect != NULL) { |
6683 | unsigned int orig_wire_count; |
6684 | |
6685 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_START, 0, 0, 0, 0, 0); |
6686 | orig_wire_count = vm_page_wire_count; |
6687 | |
6688 | (void)(*consider_buffer_cache_collect)(1); |
6689 | consider_zone_gc(FALSE); |
6690 | |
6691 | HIBLOG("hibernate_flush_memory: buffer_cache_gc freed up %d wired pages\n" , orig_wire_count - vm_page_wire_count); |
6692 | |
6693 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 7) | DBG_FUNC_END, orig_wire_count - vm_page_wire_count, 0, 0, 0, 0); |
6694 | } |
6695 | } |
6696 | hibernate_cleaning_in_progress = FALSE; |
6697 | |
6698 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 3) | DBG_FUNC_END, vm_page_free_count, hibernate_stats.hibernate_found_dirty, retval, 0, 0); |
6699 | |
6700 | if (retval) |
6701 | HIBLOG("hibernate_flush_memory() failed to finish - vm_page_compressor_count(%d)\n" , VM_PAGE_COMPRESSOR_COUNT); |
6702 | |
6703 | |
6704 | HIBPRINT("hibernate_flush_memory() considered(%d) reentered_on_q(%d) found_dirty(%d)\n" , |
6705 | hibernate_stats.hibernate_considered, |
6706 | hibernate_stats.hibernate_reentered_on_q, |
6707 | hibernate_stats.hibernate_found_dirty); |
6708 | HIBPRINT(" skipped_cleaning(%d) skipped_transient(%d) skipped_precious(%d) skipped_external(%d) queue_nolock(%d)\n" , |
6709 | hibernate_stats.hibernate_skipped_cleaning, |
6710 | hibernate_stats.hibernate_skipped_transient, |
6711 | hibernate_stats.hibernate_skipped_precious, |
6712 | hibernate_stats.hibernate_skipped_external, |
6713 | hibernate_stats.hibernate_queue_nolock); |
6714 | HIBPRINT(" queue_paused(%d) throttled(%d) throttle_timeout(%d) drained(%d) drain_timeout(%d)\n" , |
6715 | hibernate_stats.hibernate_queue_paused, |
6716 | hibernate_stats.hibernate_throttled, |
6717 | hibernate_stats.hibernate_throttle_timeout, |
6718 | hibernate_stats.hibernate_drained, |
6719 | hibernate_stats.hibernate_drain_timeout); |
6720 | |
6721 | return (retval); |
6722 | } |
6723 | |
6724 | |
6725 | static void |
6726 | hibernate_page_list_zero(hibernate_page_list_t *list) |
6727 | { |
6728 | uint32_t bank; |
6729 | hibernate_bitmap_t * bitmap; |
6730 | |
6731 | bitmap = &list->bank_bitmap[0]; |
6732 | for (bank = 0; bank < list->bank_count; bank++) |
6733 | { |
6734 | uint32_t last_bit; |
6735 | |
6736 | bzero((void *) &bitmap->bitmap[0], bitmap->bitmapwords << 2); |
6737 | // set out-of-bound bits at end of bitmap. |
6738 | last_bit = ((bitmap->last_page - bitmap->first_page + 1) & 31); |
6739 | if (last_bit) |
6740 | bitmap->bitmap[bitmap->bitmapwords - 1] = (0xFFFFFFFF >> last_bit); |
6741 | |
6742 | bitmap = (hibernate_bitmap_t *) &bitmap->bitmap[bitmap->bitmapwords]; |
6743 | } |
6744 | } |
6745 | |
6746 | void |
6747 | hibernate_free_gobble_pages(void) |
6748 | { |
6749 | vm_page_t m, next; |
6750 | uint32_t count = 0; |
6751 | |
6752 | m = (vm_page_t) hibernate_gobble_queue; |
6753 | while(m) |
6754 | { |
6755 | next = m->vmp_snext; |
6756 | vm_page_free(m); |
6757 | count++; |
6758 | m = next; |
6759 | } |
6760 | hibernate_gobble_queue = VM_PAGE_NULL; |
6761 | |
6762 | if (count) |
6763 | HIBLOG("Freed %d pages\n" , count); |
6764 | } |
6765 | |
6766 | static boolean_t |
6767 | hibernate_consider_discard(vm_page_t m, boolean_t preflight) |
6768 | { |
6769 | vm_object_t object = NULL; |
6770 | int refmod_state; |
6771 | boolean_t discard = FALSE; |
6772 | |
6773 | do |
6774 | { |
6775 | if (m->vmp_private) |
6776 | panic("hibernate_consider_discard: private" ); |
6777 | |
6778 | object = VM_PAGE_OBJECT(m); |
6779 | |
6780 | if (!vm_object_lock_try(object)) { |
6781 | object = NULL; |
6782 | if (!preflight) hibernate_stats.cd_lock_failed++; |
6783 | break; |
6784 | } |
6785 | if (VM_PAGE_WIRED(m)) { |
6786 | if (!preflight) hibernate_stats.cd_found_wired++; |
6787 | break; |
6788 | } |
6789 | if (m->vmp_precious) { |
6790 | if (!preflight) hibernate_stats.cd_found_precious++; |
6791 | break; |
6792 | } |
6793 | if (m->vmp_busy || !object->alive) { |
6794 | /* |
6795 | * Somebody is playing with this page. |
6796 | */ |
6797 | if (!preflight) hibernate_stats.cd_found_busy++; |
6798 | break; |
6799 | } |
6800 | if (m->vmp_absent || m->vmp_unusual || m->vmp_error) { |
6801 | /* |
6802 | * If it's unusual in anyway, ignore it |
6803 | */ |
6804 | if (!preflight) hibernate_stats.cd_found_unusual++; |
6805 | break; |
6806 | } |
6807 | if (m->vmp_cleaning) { |
6808 | if (!preflight) hibernate_stats.cd_found_cleaning++; |
6809 | break; |
6810 | } |
6811 | if (m->vmp_laundry) { |
6812 | if (!preflight) hibernate_stats.cd_found_laundry++; |
6813 | break; |
6814 | } |
6815 | if (!m->vmp_dirty) |
6816 | { |
6817 | refmod_state = pmap_get_refmod(VM_PAGE_GET_PHYS_PAGE(m)); |
6818 | |
6819 | if (refmod_state & VM_MEM_REFERENCED) |
6820 | m->vmp_reference = TRUE; |
6821 | if (refmod_state & VM_MEM_MODIFIED) { |
6822 | SET_PAGE_DIRTY(m, FALSE); |
6823 | } |
6824 | } |
6825 | |
6826 | /* |
6827 | * If it's clean or purgeable we can discard the page on wakeup. |
6828 | */ |
6829 | discard = (!m->vmp_dirty) |
6830 | || (VM_PURGABLE_VOLATILE == object->purgable) |
6831 | || (VM_PURGABLE_EMPTY == object->purgable); |
6832 | |
6833 | |
6834 | if (discard == FALSE) { |
6835 | if (!preflight) |
6836 | hibernate_stats.cd_found_dirty++; |
6837 | } else if (m->vmp_xpmapped && m->vmp_reference && !object->internal) { |
6838 | if (hibernate_stats.cd_found_xpmapped < HIBERNATE_XPMAPPED_LIMIT) { |
6839 | if (!preflight) |
6840 | hibernate_stats.cd_found_xpmapped++; |
6841 | discard = FALSE; |
6842 | } else { |
6843 | if (!preflight) |
6844 | hibernate_stats.cd_skipped_xpmapped++; |
6845 | } |
6846 | } |
6847 | } |
6848 | while (FALSE); |
6849 | |
6850 | if (object) |
6851 | vm_object_unlock(object); |
6852 | |
6853 | return (discard); |
6854 | } |
6855 | |
6856 | |
6857 | static void |
6858 | hibernate_discard_page(vm_page_t m) |
6859 | { |
6860 | vm_object_t m_object; |
6861 | |
6862 | if (m->vmp_absent || m->vmp_unusual || m->vmp_error) |
6863 | /* |
6864 | * If it's unusual in anyway, ignore |
6865 | */ |
6866 | return; |
6867 | |
6868 | m_object = VM_PAGE_OBJECT(m); |
6869 | |
6870 | #if MACH_ASSERT || DEBUG |
6871 | if (!vm_object_lock_try(m_object)) |
6872 | panic("hibernate_discard_page(%p) !vm_object_lock_try" , m); |
6873 | #else |
6874 | /* No need to lock page queue for token delete, hibernate_vm_unlock() |
6875 | makes sure these locks are uncontended before sleep */ |
6876 | #endif /* MACH_ASSERT || DEBUG */ |
6877 | |
6878 | if (m->vmp_pmapped == TRUE) |
6879 | { |
6880 | __unused int refmod_state = pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(m)); |
6881 | } |
6882 | |
6883 | if (m->vmp_laundry) |
6884 | panic("hibernate_discard_page(%p) laundry" , m); |
6885 | if (m->vmp_private) |
6886 | panic("hibernate_discard_page(%p) private" , m); |
6887 | if (m->vmp_fictitious) |
6888 | panic("hibernate_discard_page(%p) fictitious" , m); |
6889 | |
6890 | if (VM_PURGABLE_VOLATILE == m_object->purgable) |
6891 | { |
6892 | /* object should be on a queue */ |
6893 | assert((m_object->objq.next != NULL) && (m_object->objq.prev != NULL)); |
6894 | purgeable_q_t old_queue = vm_purgeable_object_remove(m_object); |
6895 | assert(old_queue); |
6896 | if (m_object->purgeable_when_ripe) { |
6897 | vm_purgeable_token_delete_first(old_queue); |
6898 | } |
6899 | vm_object_lock_assert_exclusive(m_object); |
6900 | m_object->purgable = VM_PURGABLE_EMPTY; |
6901 | |
6902 | /* |
6903 | * Purgeable ledgers: pages of VOLATILE and EMPTY objects are |
6904 | * accounted in the "volatile" ledger, so no change here. |
6905 | * We have to update vm_page_purgeable_count, though, since we're |
6906 | * effectively purging this object. |
6907 | */ |
6908 | unsigned int delta; |
6909 | assert(m_object->resident_page_count >= m_object->wired_page_count); |
6910 | delta = (m_object->resident_page_count - m_object->wired_page_count); |
6911 | assert(vm_page_purgeable_count >= delta); |
6912 | assert(delta > 0); |
6913 | OSAddAtomic(-delta, (SInt32 *)&vm_page_purgeable_count); |
6914 | } |
6915 | |
6916 | vm_page_free(m); |
6917 | |
6918 | #if MACH_ASSERT || DEBUG |
6919 | vm_object_unlock(m_object); |
6920 | #endif /* MACH_ASSERT || DEBUG */ |
6921 | } |
6922 | |
6923 | /* |
6924 | Grab locks for hibernate_page_list_setall() |
6925 | */ |
6926 | void |
6927 | hibernate_vm_lock_queues(void) |
6928 | { |
6929 | vm_object_lock(compressor_object); |
6930 | vm_page_lock_queues(); |
6931 | lck_mtx_lock(&vm_page_queue_free_lock); |
6932 | lck_mtx_lock(&vm_purgeable_queue_lock); |
6933 | |
6934 | if (vm_page_local_q) { |
6935 | uint32_t i; |
6936 | for (i = 0; i < vm_page_local_q_count; i++) { |
6937 | struct vpl *lq; |
6938 | lq = &vm_page_local_q[i].vpl_un.vpl; |
6939 | VPL_LOCK(&lq->vpl_lock); |
6940 | } |
6941 | } |
6942 | } |
6943 | |
6944 | void |
6945 | hibernate_vm_unlock_queues(void) |
6946 | { |
6947 | if (vm_page_local_q) { |
6948 | uint32_t i; |
6949 | for (i = 0; i < vm_page_local_q_count; i++) { |
6950 | struct vpl *lq; |
6951 | lq = &vm_page_local_q[i].vpl_un.vpl; |
6952 | VPL_UNLOCK(&lq->vpl_lock); |
6953 | } |
6954 | } |
6955 | lck_mtx_unlock(&vm_purgeable_queue_lock); |
6956 | lck_mtx_unlock(&vm_page_queue_free_lock); |
6957 | vm_page_unlock_queues(); |
6958 | vm_object_unlock(compressor_object); |
6959 | } |
6960 | |
6961 | /* |
6962 | Bits zero in the bitmaps => page needs to be saved. All pages default to be saved, |
6963 | pages known to VM to not need saving are subtracted. |
6964 | Wired pages to be saved are present in page_list_wired, pageable in page_list. |
6965 | */ |
6966 | |
6967 | void |
6968 | hibernate_page_list_setall(hibernate_page_list_t * page_list, |
6969 | hibernate_page_list_t * page_list_wired, |
6970 | hibernate_page_list_t * page_list_pal, |
6971 | boolean_t preflight, |
6972 | boolean_t will_discard, |
6973 | uint32_t * pagesOut) |
6974 | { |
6975 | uint64_t start, end, nsec; |
6976 | vm_page_t m; |
6977 | vm_page_t next; |
6978 | uint32_t pages = page_list->page_count; |
6979 | uint32_t count_anonymous = 0, count_throttled = 0, count_compressor = 0; |
6980 | uint32_t count_inactive = 0, count_active = 0, count_speculative = 0, count_cleaned = 0; |
6981 | uint32_t count_wire = pages; |
6982 | uint32_t count_discard_active = 0; |
6983 | uint32_t count_discard_inactive = 0; |
6984 | uint32_t count_discard_cleaned = 0; |
6985 | uint32_t count_discard_purgeable = 0; |
6986 | uint32_t count_discard_speculative = 0; |
6987 | uint32_t count_discard_vm_struct_pages = 0; |
6988 | uint32_t i; |
6989 | uint32_t bank; |
6990 | hibernate_bitmap_t * bitmap; |
6991 | hibernate_bitmap_t * bitmap_wired; |
6992 | boolean_t discard_all; |
6993 | boolean_t discard; |
6994 | |
6995 | HIBLOG("hibernate_page_list_setall(preflight %d) start\n" , preflight); |
6996 | |
6997 | if (preflight) { |
6998 | page_list = NULL; |
6999 | page_list_wired = NULL; |
7000 | page_list_pal = NULL; |
7001 | discard_all = FALSE; |
7002 | } else { |
7003 | discard_all = will_discard; |
7004 | } |
7005 | |
7006 | #if MACH_ASSERT || DEBUG |
7007 | if (!preflight) |
7008 | { |
7009 | assert(hibernate_vm_locks_are_safe()); |
7010 | vm_page_lock_queues(); |
7011 | if (vm_page_local_q) { |
7012 | for (i = 0; i < vm_page_local_q_count; i++) { |
7013 | struct vpl *lq; |
7014 | lq = &vm_page_local_q[i].vpl_un.vpl; |
7015 | VPL_LOCK(&lq->vpl_lock); |
7016 | } |
7017 | } |
7018 | } |
7019 | #endif /* MACH_ASSERT || DEBUG */ |
7020 | |
7021 | |
7022 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_START, count_wire, 0, 0, 0, 0); |
7023 | |
7024 | clock_get_uptime(&start); |
7025 | |
7026 | if (!preflight) { |
7027 | hibernate_page_list_zero(page_list); |
7028 | hibernate_page_list_zero(page_list_wired); |
7029 | hibernate_page_list_zero(page_list_pal); |
7030 | |
7031 | hibernate_stats.cd_vm_page_wire_count = vm_page_wire_count; |
7032 | hibernate_stats.cd_pages = pages; |
7033 | } |
7034 | |
7035 | if (vm_page_local_q) { |
7036 | for (i = 0; i < vm_page_local_q_count; i++) |
7037 | vm_page_reactivate_local(i, TRUE, !preflight); |
7038 | } |
7039 | |
7040 | if (preflight) { |
7041 | vm_object_lock(compressor_object); |
7042 | vm_page_lock_queues(); |
7043 | lck_mtx_lock(&vm_page_queue_free_lock); |
7044 | } |
7045 | |
7046 | m = (vm_page_t) hibernate_gobble_queue; |
7047 | while (m) |
7048 | { |
7049 | pages--; |
7050 | count_wire--; |
7051 | if (!preflight) { |
7052 | hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7053 | hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7054 | } |
7055 | m = m->vmp_snext; |
7056 | } |
7057 | |
7058 | if (!preflight) for( i = 0; i < real_ncpus; i++ ) |
7059 | { |
7060 | if (cpu_data_ptr[i] && cpu_data_ptr[i]->cpu_processor) |
7061 | { |
7062 | for (m = PROCESSOR_DATA(cpu_data_ptr[i]->cpu_processor, free_pages); m; m = m->vmp_snext) |
7063 | { |
7064 | assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOCAL_Q); |
7065 | |
7066 | pages--; |
7067 | count_wire--; |
7068 | hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7069 | hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7070 | |
7071 | hibernate_stats.cd_local_free++; |
7072 | hibernate_stats.cd_total_free++; |
7073 | } |
7074 | } |
7075 | } |
7076 | |
7077 | for( i = 0; i < vm_colors; i++ ) |
7078 | { |
7079 | vm_page_queue_iterate(&vm_page_queue_free[i].qhead, |
7080 | m, |
7081 | vm_page_t, |
7082 | vmp_pageq) |
7083 | { |
7084 | assert(m->vmp_q_state == VM_PAGE_ON_FREE_Q); |
7085 | |
7086 | pages--; |
7087 | count_wire--; |
7088 | if (!preflight) { |
7089 | hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7090 | hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7091 | |
7092 | hibernate_stats.cd_total_free++; |
7093 | } |
7094 | } |
7095 | } |
7096 | |
7097 | vm_page_queue_iterate(&vm_lopage_queue_free, |
7098 | m, |
7099 | vm_page_t, |
7100 | vmp_pageq) |
7101 | { |
7102 | assert(m->vmp_q_state == VM_PAGE_ON_FREE_LOPAGE_Q); |
7103 | |
7104 | pages--; |
7105 | count_wire--; |
7106 | if (!preflight) { |
7107 | hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7108 | hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7109 | |
7110 | hibernate_stats.cd_total_free++; |
7111 | } |
7112 | } |
7113 | |
7114 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_throttled); |
7115 | while (m && !vm_page_queue_end(&vm_page_queue_throttled, (vm_page_queue_entry_t)m)) |
7116 | { |
7117 | assert(m->vmp_q_state == VM_PAGE_ON_THROTTLED_Q); |
7118 | |
7119 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7120 | discard = FALSE; |
7121 | if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) |
7122 | && hibernate_consider_discard(m, preflight)) |
7123 | { |
7124 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7125 | count_discard_inactive++; |
7126 | discard = discard_all; |
7127 | } |
7128 | else |
7129 | count_throttled++; |
7130 | count_wire--; |
7131 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7132 | |
7133 | if (discard) hibernate_discard_page(m); |
7134 | m = next; |
7135 | } |
7136 | |
7137 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_anonymous); |
7138 | while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) |
7139 | { |
7140 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q); |
7141 | |
7142 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7143 | discard = FALSE; |
7144 | if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) |
7145 | && hibernate_consider_discard(m, preflight)) |
7146 | { |
7147 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7148 | if (m->vmp_dirty) |
7149 | count_discard_purgeable++; |
7150 | else |
7151 | count_discard_inactive++; |
7152 | discard = discard_all; |
7153 | } |
7154 | else |
7155 | count_anonymous++; |
7156 | count_wire--; |
7157 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7158 | if (discard) hibernate_discard_page(m); |
7159 | m = next; |
7160 | } |
7161 | |
7162 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned); |
7163 | while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) |
7164 | { |
7165 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q); |
7166 | |
7167 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7168 | discard = FALSE; |
7169 | if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) |
7170 | && hibernate_consider_discard(m, preflight)) |
7171 | { |
7172 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7173 | if (m->vmp_dirty) |
7174 | count_discard_purgeable++; |
7175 | else |
7176 | count_discard_cleaned++; |
7177 | discard = discard_all; |
7178 | } |
7179 | else |
7180 | count_cleaned++; |
7181 | count_wire--; |
7182 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7183 | if (discard) hibernate_discard_page(m); |
7184 | m = next; |
7185 | } |
7186 | |
7187 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); |
7188 | while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) |
7189 | { |
7190 | assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q); |
7191 | |
7192 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7193 | discard = FALSE; |
7194 | if ((kIOHibernateModeDiscardCleanActive & gIOHibernateMode) |
7195 | && hibernate_consider_discard(m, preflight)) |
7196 | { |
7197 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7198 | if (m->vmp_dirty) |
7199 | count_discard_purgeable++; |
7200 | else |
7201 | count_discard_active++; |
7202 | discard = discard_all; |
7203 | } |
7204 | else |
7205 | count_active++; |
7206 | count_wire--; |
7207 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7208 | if (discard) hibernate_discard_page(m); |
7209 | m = next; |
7210 | } |
7211 | |
7212 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive); |
7213 | while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) |
7214 | { |
7215 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q); |
7216 | |
7217 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7218 | discard = FALSE; |
7219 | if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) |
7220 | && hibernate_consider_discard(m, preflight)) |
7221 | { |
7222 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7223 | if (m->vmp_dirty) |
7224 | count_discard_purgeable++; |
7225 | else |
7226 | count_discard_inactive++; |
7227 | discard = discard_all; |
7228 | } |
7229 | else |
7230 | count_inactive++; |
7231 | count_wire--; |
7232 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7233 | if (discard) hibernate_discard_page(m); |
7234 | m = next; |
7235 | } |
7236 | /* XXX FBDP TODO: secluded queue */ |
7237 | |
7238 | for( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) |
7239 | { |
7240 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q); |
7241 | while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) |
7242 | { |
7243 | assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q); |
7244 | assertf(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q, |
7245 | "Bad page: %p (0x%x:0x%x) on queue %d has state: %d (Discard: %d, Preflight: %d)" , |
7246 | m, m->vmp_pageq.next, m->vmp_pageq.prev, i, m->vmp_q_state, discard, preflight); |
7247 | |
7248 | next = (vm_page_t)VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7249 | discard = FALSE; |
7250 | if ((kIOHibernateModeDiscardCleanInactive & gIOHibernateMode) |
7251 | && hibernate_consider_discard(m, preflight)) |
7252 | { |
7253 | if (!preflight) hibernate_page_bitset(page_list, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7254 | count_discard_speculative++; |
7255 | discard = discard_all; |
7256 | } |
7257 | else |
7258 | count_speculative++; |
7259 | count_wire--; |
7260 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7261 | if (discard) hibernate_discard_page(m); |
7262 | m = next; |
7263 | } |
7264 | } |
7265 | |
7266 | vm_page_queue_iterate(&compressor_object->memq, m, vm_page_t, vmp_listq) |
7267 | { |
7268 | assert(m->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR); |
7269 | |
7270 | count_compressor++; |
7271 | count_wire--; |
7272 | if (!preflight) hibernate_page_bitset(page_list_wired, TRUE, VM_PAGE_GET_PHYS_PAGE(m)); |
7273 | } |
7274 | |
7275 | if (preflight == FALSE && discard_all == TRUE) { |
7276 | KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_START); |
7277 | |
7278 | HIBLOG("hibernate_teardown started\n" ); |
7279 | count_discard_vm_struct_pages = hibernate_teardown_vm_structs(page_list, page_list_wired); |
7280 | HIBLOG("hibernate_teardown completed - discarded %d\n" , count_discard_vm_struct_pages); |
7281 | |
7282 | pages -= count_discard_vm_struct_pages; |
7283 | count_wire -= count_discard_vm_struct_pages; |
7284 | |
7285 | hibernate_stats.cd_vm_struct_pages_unneeded = count_discard_vm_struct_pages; |
7286 | |
7287 | KDBG(IOKDBG_CODE(DBG_HIBERNATE, 12) | DBG_FUNC_END); |
7288 | } |
7289 | |
7290 | if (!preflight) { |
7291 | // pull wired from hibernate_bitmap |
7292 | bitmap = &page_list->bank_bitmap[0]; |
7293 | bitmap_wired = &page_list_wired->bank_bitmap[0]; |
7294 | for (bank = 0; bank < page_list->bank_count; bank++) |
7295 | { |
7296 | for (i = 0; i < bitmap->bitmapwords; i++) |
7297 | bitmap->bitmap[i] = bitmap->bitmap[i] | ~bitmap_wired->bitmap[i]; |
7298 | bitmap = (hibernate_bitmap_t *) &bitmap->bitmap [bitmap->bitmapwords]; |
7299 | bitmap_wired = (hibernate_bitmap_t *) &bitmap_wired->bitmap[bitmap_wired->bitmapwords]; |
7300 | } |
7301 | } |
7302 | |
7303 | // machine dependent adjustments |
7304 | hibernate_page_list_setall_machine(page_list, page_list_wired, preflight, &pages); |
7305 | |
7306 | if (!preflight) { |
7307 | hibernate_stats.cd_count_wire = count_wire; |
7308 | hibernate_stats.cd_discarded = count_discard_active + count_discard_inactive + count_discard_purgeable + |
7309 | count_discard_speculative + count_discard_cleaned + count_discard_vm_struct_pages; |
7310 | } |
7311 | |
7312 | clock_get_uptime(&end); |
7313 | absolutetime_to_nanoseconds(end - start, &nsec); |
7314 | HIBLOG("hibernate_page_list_setall time: %qd ms\n" , nsec / 1000000ULL); |
7315 | |
7316 | 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\n" , |
7317 | pages, count_wire, count_active, count_inactive, count_cleaned, count_speculative, count_anonymous, count_throttled, count_compressor, hibernate_stats.cd_found_xpmapped, |
7318 | discard_all ? "did" : "could" , |
7319 | count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned); |
7320 | |
7321 | if (hibernate_stats.cd_skipped_xpmapped) |
7322 | HIBLOG("WARNING: hibernate_page_list_setall skipped %d xpmapped pages\n" , hibernate_stats.cd_skipped_xpmapped); |
7323 | |
7324 | *pagesOut = pages - count_discard_active - count_discard_inactive - count_discard_purgeable - count_discard_speculative - count_discard_cleaned; |
7325 | |
7326 | if (preflight && will_discard) *pagesOut -= count_compressor + count_throttled + count_anonymous + count_inactive + count_cleaned + count_speculative + count_active; |
7327 | |
7328 | #if MACH_ASSERT || DEBUG |
7329 | if (!preflight) |
7330 | { |
7331 | if (vm_page_local_q) { |
7332 | for (i = 0; i < vm_page_local_q_count; i++) { |
7333 | struct vpl *lq; |
7334 | lq = &vm_page_local_q[i].vpl_un.vpl; |
7335 | VPL_UNLOCK(&lq->vpl_lock); |
7336 | } |
7337 | } |
7338 | vm_page_unlock_queues(); |
7339 | } |
7340 | #endif /* MACH_ASSERT || DEBUG */ |
7341 | |
7342 | if (preflight) { |
7343 | lck_mtx_unlock(&vm_page_queue_free_lock); |
7344 | vm_page_unlock_queues(); |
7345 | vm_object_unlock(compressor_object); |
7346 | } |
7347 | |
7348 | KERNEL_DEBUG_CONSTANT(IOKDBG_CODE(DBG_HIBERNATE, 8) | DBG_FUNC_END, count_wire, *pagesOut, 0, 0, 0); |
7349 | } |
7350 | |
7351 | void |
7352 | hibernate_page_list_discard(hibernate_page_list_t * page_list) |
7353 | { |
7354 | uint64_t start, end, nsec; |
7355 | vm_page_t m; |
7356 | vm_page_t next; |
7357 | uint32_t i; |
7358 | uint32_t count_discard_active = 0; |
7359 | uint32_t count_discard_inactive = 0; |
7360 | uint32_t count_discard_purgeable = 0; |
7361 | uint32_t count_discard_cleaned = 0; |
7362 | uint32_t count_discard_speculative = 0; |
7363 | |
7364 | |
7365 | #if MACH_ASSERT || DEBUG |
7366 | vm_page_lock_queues(); |
7367 | if (vm_page_local_q) { |
7368 | for (i = 0; i < vm_page_local_q_count; i++) { |
7369 | struct vpl *lq; |
7370 | lq = &vm_page_local_q[i].vpl_un.vpl; |
7371 | VPL_LOCK(&lq->vpl_lock); |
7372 | } |
7373 | } |
7374 | #endif /* MACH_ASSERT || DEBUG */ |
7375 | |
7376 | clock_get_uptime(&start); |
7377 | |
7378 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_anonymous); |
7379 | while (m && !vm_page_queue_end(&vm_page_queue_anonymous, (vm_page_queue_entry_t)m)) |
7380 | { |
7381 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_INTERNAL_Q); |
7382 | |
7383 | next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7384 | if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) |
7385 | { |
7386 | if (m->vmp_dirty) |
7387 | count_discard_purgeable++; |
7388 | else |
7389 | count_discard_inactive++; |
7390 | hibernate_discard_page(m); |
7391 | } |
7392 | m = next; |
7393 | } |
7394 | |
7395 | for( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) |
7396 | { |
7397 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_speculative[i].age_q); |
7398 | while (m && !vm_page_queue_end(&vm_page_queue_speculative[i].age_q, (vm_page_queue_entry_t)m)) |
7399 | { |
7400 | assert(m->vmp_q_state == VM_PAGE_ON_SPECULATIVE_Q); |
7401 | |
7402 | next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7403 | if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) |
7404 | { |
7405 | count_discard_speculative++; |
7406 | hibernate_discard_page(m); |
7407 | } |
7408 | m = next; |
7409 | } |
7410 | } |
7411 | |
7412 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_inactive); |
7413 | while (m && !vm_page_queue_end(&vm_page_queue_inactive, (vm_page_queue_entry_t)m)) |
7414 | { |
7415 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_EXTERNAL_Q); |
7416 | |
7417 | next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7418 | if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) |
7419 | { |
7420 | if (m->vmp_dirty) |
7421 | count_discard_purgeable++; |
7422 | else |
7423 | count_discard_inactive++; |
7424 | hibernate_discard_page(m); |
7425 | } |
7426 | m = next; |
7427 | } |
7428 | /* XXX FBDP TODO: secluded queue */ |
7429 | |
7430 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_active); |
7431 | while (m && !vm_page_queue_end(&vm_page_queue_active, (vm_page_queue_entry_t)m)) |
7432 | { |
7433 | assert(m->vmp_q_state == VM_PAGE_ON_ACTIVE_Q); |
7434 | |
7435 | next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7436 | if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) |
7437 | { |
7438 | if (m->vmp_dirty) |
7439 | count_discard_purgeable++; |
7440 | else |
7441 | count_discard_active++; |
7442 | hibernate_discard_page(m); |
7443 | } |
7444 | m = next; |
7445 | } |
7446 | |
7447 | m = (vm_page_t) vm_page_queue_first(&vm_page_queue_cleaned); |
7448 | while (m && !vm_page_queue_end(&vm_page_queue_cleaned, (vm_page_queue_entry_t)m)) |
7449 | { |
7450 | assert(m->vmp_q_state == VM_PAGE_ON_INACTIVE_CLEANED_Q); |
7451 | |
7452 | next = (vm_page_t) VM_PAGE_UNPACK_PTR(m->vmp_pageq.next); |
7453 | if (hibernate_page_bittst(page_list, VM_PAGE_GET_PHYS_PAGE(m))) |
7454 | { |
7455 | if (m->vmp_dirty) |
7456 | count_discard_purgeable++; |
7457 | else |
7458 | count_discard_cleaned++; |
7459 | hibernate_discard_page(m); |
7460 | } |
7461 | m = next; |
7462 | } |
7463 | |
7464 | #if MACH_ASSERT || DEBUG |
7465 | if (vm_page_local_q) { |
7466 | for (i = 0; i < vm_page_local_q_count; i++) { |
7467 | struct vpl *lq; |
7468 | lq = &vm_page_local_q[i].vpl_un.vpl; |
7469 | VPL_UNLOCK(&lq->vpl_lock); |
7470 | } |
7471 | } |
7472 | vm_page_unlock_queues(); |
7473 | #endif /* MACH_ASSERT || DEBUG */ |
7474 | |
7475 | clock_get_uptime(&end); |
7476 | absolutetime_to_nanoseconds(end - start, &nsec); |
7477 | HIBLOG("hibernate_page_list_discard time: %qd ms, discarded act %d inact %d purgeable %d spec %d cleaned %d\n" , |
7478 | nsec / 1000000ULL, |
7479 | count_discard_active, count_discard_inactive, count_discard_purgeable, count_discard_speculative, count_discard_cleaned); |
7480 | } |
7481 | |
7482 | boolean_t hibernate_paddr_map_inited = FALSE; |
7483 | unsigned int hibernate_teardown_last_valid_compact_indx = -1; |
7484 | vm_page_t hibernate_rebuild_hash_list = NULL; |
7485 | |
7486 | unsigned int hibernate_teardown_found_tabled_pages = 0; |
7487 | unsigned int hibernate_teardown_found_created_pages = 0; |
7488 | unsigned int hibernate_teardown_found_free_pages = 0; |
7489 | unsigned int hibernate_teardown_vm_page_free_count; |
7490 | |
7491 | |
7492 | struct ppnum_mapping { |
7493 | struct ppnum_mapping *ppnm_next; |
7494 | ppnum_t ppnm_base_paddr; |
7495 | unsigned int ppnm_sindx; |
7496 | unsigned int ppnm_eindx; |
7497 | }; |
7498 | |
7499 | struct ppnum_mapping *ppnm_head; |
7500 | struct ppnum_mapping *ppnm_last_found = NULL; |
7501 | |
7502 | |
7503 | void |
7504 | hibernate_create_paddr_map() |
7505 | { |
7506 | unsigned int i; |
7507 | ppnum_t next_ppnum_in_run = 0; |
7508 | struct ppnum_mapping *ppnm = NULL; |
7509 | |
7510 | if (hibernate_paddr_map_inited == FALSE) { |
7511 | |
7512 | for (i = 0; i < vm_pages_count; i++) { |
7513 | |
7514 | if (ppnm) |
7515 | ppnm->ppnm_eindx = i; |
7516 | |
7517 | if (ppnm == NULL || VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) != next_ppnum_in_run) { |
7518 | |
7519 | ppnm = kalloc(sizeof(struct ppnum_mapping)); |
7520 | |
7521 | ppnm->ppnm_next = ppnm_head; |
7522 | ppnm_head = ppnm; |
7523 | |
7524 | ppnm->ppnm_sindx = i; |
7525 | ppnm->ppnm_base_paddr = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]); |
7526 | } |
7527 | next_ppnum_in_run = VM_PAGE_GET_PHYS_PAGE(&vm_pages[i]) + 1; |
7528 | } |
7529 | ppnm->ppnm_eindx++; |
7530 | |
7531 | hibernate_paddr_map_inited = TRUE; |
7532 | } |
7533 | } |
7534 | |
7535 | ppnum_t |
7536 | hibernate_lookup_paddr(unsigned int indx) |
7537 | { |
7538 | struct ppnum_mapping *ppnm = NULL; |
7539 | |
7540 | ppnm = ppnm_last_found; |
7541 | |
7542 | if (ppnm) { |
7543 | if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) |
7544 | goto done; |
7545 | } |
7546 | for (ppnm = ppnm_head; ppnm; ppnm = ppnm->ppnm_next) { |
7547 | |
7548 | if (indx >= ppnm->ppnm_sindx && indx < ppnm->ppnm_eindx) { |
7549 | ppnm_last_found = ppnm; |
7550 | break; |
7551 | } |
7552 | } |
7553 | if (ppnm == NULL) |
7554 | panic("hibernate_lookup_paddr of %d failed\n" , indx); |
7555 | done: |
7556 | return (ppnm->ppnm_base_paddr + (indx - ppnm->ppnm_sindx)); |
7557 | } |
7558 | |
7559 | |
7560 | uint32_t |
7561 | hibernate_mark_as_unneeded(addr64_t saddr, addr64_t eaddr, hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired) |
7562 | { |
7563 | addr64_t saddr_aligned; |
7564 | addr64_t eaddr_aligned; |
7565 | addr64_t addr; |
7566 | ppnum_t paddr; |
7567 | unsigned int mark_as_unneeded_pages = 0; |
7568 | |
7569 | saddr_aligned = (saddr + PAGE_MASK_64) & ~PAGE_MASK_64; |
7570 | eaddr_aligned = eaddr & ~PAGE_MASK_64; |
7571 | |
7572 | for (addr = saddr_aligned; addr < eaddr_aligned; addr += PAGE_SIZE_64) { |
7573 | |
7574 | paddr = pmap_find_phys(kernel_pmap, addr); |
7575 | |
7576 | assert(paddr); |
7577 | |
7578 | hibernate_page_bitset(page_list, TRUE, paddr); |
7579 | hibernate_page_bitset(page_list_wired, TRUE, paddr); |
7580 | |
7581 | mark_as_unneeded_pages++; |
7582 | } |
7583 | return (mark_as_unneeded_pages); |
7584 | } |
7585 | |
7586 | |
7587 | void |
7588 | hibernate_hash_insert_page(vm_page_t mem) |
7589 | { |
7590 | vm_page_bucket_t *bucket; |
7591 | int hash_id; |
7592 | vm_object_t m_object; |
7593 | |
7594 | m_object = VM_PAGE_OBJECT(mem); |
7595 | |
7596 | assert(mem->vmp_hashed); |
7597 | assert(m_object); |
7598 | assert(mem->vmp_offset != (vm_object_offset_t) -1); |
7599 | |
7600 | /* |
7601 | * Insert it into the object_object/offset hash table |
7602 | */ |
7603 | hash_id = vm_page_hash(m_object, mem->vmp_offset); |
7604 | bucket = &vm_page_buckets[hash_id]; |
7605 | |
7606 | mem->vmp_next_m = bucket->page_list; |
7607 | bucket->page_list = VM_PAGE_PACK_PTR(mem); |
7608 | } |
7609 | |
7610 | |
7611 | void |
7612 | hibernate_free_range(int sindx, int eindx) |
7613 | { |
7614 | vm_page_t mem; |
7615 | unsigned int color; |
7616 | |
7617 | while (sindx < eindx) { |
7618 | mem = &vm_pages[sindx]; |
7619 | |
7620 | vm_page_init(mem, hibernate_lookup_paddr(sindx), FALSE); |
7621 | |
7622 | mem->vmp_lopage = FALSE; |
7623 | mem->vmp_q_state = VM_PAGE_ON_FREE_Q; |
7624 | |
7625 | color = VM_PAGE_GET_COLOR(mem); |
7626 | #if defined(__x86_64__) |
7627 | vm_page_queue_enter_clump(&vm_page_queue_free[color].qhead, |
7628 | mem, |
7629 | vm_page_t, |
7630 | vmp_pageq); |
7631 | #else |
7632 | vm_page_queue_enter(&vm_page_queue_free[color].qhead, |
7633 | mem, |
7634 | vm_page_t, |
7635 | vmp_pageq); |
7636 | #endif |
7637 | vm_page_free_count++; |
7638 | |
7639 | sindx++; |
7640 | } |
7641 | } |
7642 | |
7643 | |
7644 | extern void hibernate_rebuild_pmap_structs(void); |
7645 | |
7646 | void |
7647 | hibernate_rebuild_vm_structs(void) |
7648 | { |
7649 | int i, cindx, sindx, eindx; |
7650 | vm_page_t mem, tmem, mem_next; |
7651 | AbsoluteTime startTime, endTime; |
7652 | uint64_t nsec; |
7653 | |
7654 | if (hibernate_rebuild_needed == FALSE) |
7655 | return; |
7656 | |
7657 | KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_START); |
7658 | HIBLOG("hibernate_rebuild started\n" ); |
7659 | |
7660 | clock_get_uptime(&startTime); |
7661 | |
7662 | hibernate_rebuild_pmap_structs(); |
7663 | |
7664 | bzero(&vm_page_buckets[0], vm_page_bucket_count * sizeof(vm_page_bucket_t)); |
7665 | eindx = vm_pages_count; |
7666 | |
7667 | /* |
7668 | * Mark all the vm_pages[] that have not been initialized yet as being |
7669 | * transient. This is needed to ensure that buddy page search is corrrect. |
7670 | * Without this random data in these vm_pages[] can trip the buddy search |
7671 | */ |
7672 | for (i = hibernate_teardown_last_valid_compact_indx+1; i < eindx; ++i) |
7673 | vm_pages[i].vmp_q_state = VM_PAGE_NOT_ON_Q; |
7674 | |
7675 | for (cindx = hibernate_teardown_last_valid_compact_indx; cindx >= 0; cindx--) { |
7676 | |
7677 | mem = &vm_pages[cindx]; |
7678 | assert(mem->vmp_q_state != VM_PAGE_ON_FREE_Q); |
7679 | /* |
7680 | * hibernate_teardown_vm_structs leaves the location where |
7681 | * this vm_page_t must be located in "next". |
7682 | */ |
7683 | tmem = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); |
7684 | mem->vmp_next_m = VM_PAGE_PACK_PTR(NULL); |
7685 | |
7686 | sindx = (int)(tmem - &vm_pages[0]); |
7687 | |
7688 | if (mem != tmem) { |
7689 | /* |
7690 | * this vm_page_t was moved by hibernate_teardown_vm_structs, |
7691 | * so move it back to its real location |
7692 | */ |
7693 | *tmem = *mem; |
7694 | mem = tmem; |
7695 | } |
7696 | if (mem->vmp_hashed) |
7697 | hibernate_hash_insert_page(mem); |
7698 | /* |
7699 | * the 'hole' between this vm_page_t and the previous |
7700 | * vm_page_t we moved needs to be initialized as |
7701 | * a range of free vm_page_t's |
7702 | */ |
7703 | hibernate_free_range(sindx + 1, eindx); |
7704 | |
7705 | eindx = sindx; |
7706 | } |
7707 | if (sindx) |
7708 | hibernate_free_range(0, sindx); |
7709 | |
7710 | assert(vm_page_free_count == hibernate_teardown_vm_page_free_count); |
7711 | |
7712 | /* |
7713 | * process the list of vm_page_t's that were entered in the hash, |
7714 | * but were not located in the vm_pages arrary... these are |
7715 | * vm_page_t's that were created on the fly (i.e. fictitious) |
7716 | */ |
7717 | for (mem = hibernate_rebuild_hash_list; mem; mem = mem_next) { |
7718 | mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); |
7719 | |
7720 | mem->vmp_next_m = 0; |
7721 | hibernate_hash_insert_page(mem); |
7722 | } |
7723 | hibernate_rebuild_hash_list = NULL; |
7724 | |
7725 | clock_get_uptime(&endTime); |
7726 | SUB_ABSOLUTETIME(&endTime, &startTime); |
7727 | absolutetime_to_nanoseconds(endTime, &nsec); |
7728 | |
7729 | HIBLOG("hibernate_rebuild completed - took %qd msecs\n" , nsec / 1000000ULL); |
7730 | |
7731 | hibernate_rebuild_needed = FALSE; |
7732 | |
7733 | KDBG(IOKDBG_CODE(DBG_HIBERNATE, 13) | DBG_FUNC_END); |
7734 | } |
7735 | |
7736 | |
7737 | extern void hibernate_teardown_pmap_structs(addr64_t *, addr64_t *); |
7738 | |
7739 | uint32_t |
7740 | hibernate_teardown_vm_structs(hibernate_page_list_t *page_list, hibernate_page_list_t *page_list_wired) |
7741 | { |
7742 | unsigned int i; |
7743 | unsigned int compact_target_indx; |
7744 | vm_page_t mem, mem_next; |
7745 | vm_page_bucket_t *bucket; |
7746 | unsigned int mark_as_unneeded_pages = 0; |
7747 | unsigned int unneeded_vm_page_bucket_pages = 0; |
7748 | unsigned int unneeded_vm_pages_pages = 0; |
7749 | unsigned int unneeded_pmap_pages = 0; |
7750 | addr64_t start_of_unneeded = 0; |
7751 | addr64_t end_of_unneeded = 0; |
7752 | |
7753 | |
7754 | if (hibernate_should_abort()) |
7755 | return (0); |
7756 | |
7757 | hibernate_rebuild_needed = TRUE; |
7758 | |
7759 | 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" , |
7760 | vm_page_wire_count, vm_page_free_count, vm_page_active_count, vm_page_inactive_count, vm_page_speculative_count, |
7761 | vm_page_cleaned_count, compressor_object->resident_page_count); |
7762 | |
7763 | for (i = 0; i < vm_page_bucket_count; i++) { |
7764 | |
7765 | bucket = &vm_page_buckets[i]; |
7766 | |
7767 | for (mem = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); mem != VM_PAGE_NULL; mem = mem_next) { |
7768 | assert(mem->vmp_hashed); |
7769 | |
7770 | mem_next = (vm_page_t)(VM_PAGE_UNPACK_PTR(mem->vmp_next_m)); |
7771 | |
7772 | if (mem < &vm_pages[0] || mem >= &vm_pages[vm_pages_count]) { |
7773 | mem->vmp_next_m = VM_PAGE_PACK_PTR(hibernate_rebuild_hash_list); |
7774 | hibernate_rebuild_hash_list = mem; |
7775 | } |
7776 | } |
7777 | } |
7778 | 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); |
7779 | mark_as_unneeded_pages += unneeded_vm_page_bucket_pages; |
7780 | |
7781 | hibernate_teardown_vm_page_free_count = vm_page_free_count; |
7782 | |
7783 | compact_target_indx = 0; |
7784 | |
7785 | for (i = 0; i < vm_pages_count; i++) { |
7786 | |
7787 | mem = &vm_pages[i]; |
7788 | |
7789 | if (mem->vmp_q_state == VM_PAGE_ON_FREE_Q) { |
7790 | unsigned int color; |
7791 | |
7792 | assert(mem->vmp_busy); |
7793 | assert(!mem->vmp_lopage); |
7794 | |
7795 | color = VM_PAGE_GET_COLOR(mem); |
7796 | |
7797 | vm_page_queue_remove(&vm_page_queue_free[color].qhead, |
7798 | mem, |
7799 | vm_page_t, |
7800 | vmp_pageq); |
7801 | |
7802 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
7803 | |
7804 | vm_page_free_count--; |
7805 | |
7806 | hibernate_teardown_found_free_pages++; |
7807 | |
7808 | if (vm_pages[compact_target_indx].vmp_q_state != VM_PAGE_ON_FREE_Q) |
7809 | compact_target_indx = i; |
7810 | } else { |
7811 | /* |
7812 | * record this vm_page_t's original location |
7813 | * we need this even if it doesn't get moved |
7814 | * as an indicator to the rebuild function that |
7815 | * we don't have to move it |
7816 | */ |
7817 | mem->vmp_next_m = VM_PAGE_PACK_PTR(mem); |
7818 | |
7819 | if (vm_pages[compact_target_indx].vmp_q_state == VM_PAGE_ON_FREE_Q) { |
7820 | /* |
7821 | * we've got a hole to fill, so |
7822 | * move this vm_page_t to it's new home |
7823 | */ |
7824 | vm_pages[compact_target_indx] = *mem; |
7825 | mem->vmp_q_state = VM_PAGE_ON_FREE_Q; |
7826 | |
7827 | hibernate_teardown_last_valid_compact_indx = compact_target_indx; |
7828 | compact_target_indx++; |
7829 | } else |
7830 | hibernate_teardown_last_valid_compact_indx = i; |
7831 | } |
7832 | } |
7833 | unneeded_vm_pages_pages = hibernate_mark_as_unneeded((addr64_t)&vm_pages[hibernate_teardown_last_valid_compact_indx+1], |
7834 | (addr64_t)&vm_pages[vm_pages_count-1], page_list, page_list_wired); |
7835 | mark_as_unneeded_pages += unneeded_vm_pages_pages; |
7836 | |
7837 | hibernate_teardown_pmap_structs(&start_of_unneeded, &end_of_unneeded); |
7838 | |
7839 | if (start_of_unneeded) { |
7840 | unneeded_pmap_pages = hibernate_mark_as_unneeded(start_of_unneeded, end_of_unneeded, page_list, page_list_wired); |
7841 | mark_as_unneeded_pages += unneeded_pmap_pages; |
7842 | } |
7843 | HIBLOG("hibernate_teardown: mark_as_unneeded_pages %d, %d, %d\n" , unneeded_vm_page_bucket_pages, unneeded_vm_pages_pages, unneeded_pmap_pages); |
7844 | |
7845 | return (mark_as_unneeded_pages); |
7846 | } |
7847 | |
7848 | |
7849 | #endif /* HIBERNATION */ |
7850 | |
7851 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ |
7852 | |
7853 | #include <mach_vm_debug.h> |
7854 | #if MACH_VM_DEBUG |
7855 | |
7856 | #include <mach_debug/hash_info.h> |
7857 | #include <vm/vm_debug.h> |
7858 | |
7859 | /* |
7860 | * Routine: vm_page_info |
7861 | * Purpose: |
7862 | * Return information about the global VP table. |
7863 | * Fills the buffer with as much information as possible |
7864 | * and returns the desired size of the buffer. |
7865 | * Conditions: |
7866 | * Nothing locked. The caller should provide |
7867 | * possibly-pageable memory. |
7868 | */ |
7869 | |
7870 | unsigned int |
7871 | vm_page_info( |
7872 | hash_info_bucket_t *info, |
7873 | unsigned int count) |
7874 | { |
7875 | unsigned int i; |
7876 | lck_spin_t *bucket_lock; |
7877 | |
7878 | if (vm_page_bucket_count < count) |
7879 | count = vm_page_bucket_count; |
7880 | |
7881 | for (i = 0; i < count; i++) { |
7882 | vm_page_bucket_t *bucket = &vm_page_buckets[i]; |
7883 | unsigned int bucket_count = 0; |
7884 | vm_page_t m; |
7885 | |
7886 | bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK]; |
7887 | lck_spin_lock(bucket_lock); |
7888 | |
7889 | for (m = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); |
7890 | m != VM_PAGE_NULL; |
7891 | m = (vm_page_t)(VM_PAGE_UNPACK_PTR(m->vmp_next_m))) |
7892 | bucket_count++; |
7893 | |
7894 | lck_spin_unlock(bucket_lock); |
7895 | |
7896 | /* don't touch pageable memory while holding locks */ |
7897 | info[i].hib_count = bucket_count; |
7898 | } |
7899 | |
7900 | return vm_page_bucket_count; |
7901 | } |
7902 | #endif /* MACH_VM_DEBUG */ |
7903 | |
7904 | #if VM_PAGE_BUCKETS_CHECK |
7905 | void |
7906 | vm_page_buckets_check(void) |
7907 | { |
7908 | unsigned int i; |
7909 | vm_page_t p; |
7910 | unsigned int p_hash; |
7911 | vm_page_bucket_t *bucket; |
7912 | lck_spin_t *bucket_lock; |
7913 | |
7914 | if (!vm_page_buckets_check_ready) { |
7915 | return; |
7916 | } |
7917 | |
7918 | #if HIBERNATION |
7919 | if (hibernate_rebuild_needed || |
7920 | hibernate_rebuild_hash_list) { |
7921 | panic("BUCKET_CHECK: hibernation in progress: " |
7922 | "rebuild_needed=%d rebuild_hash_list=%p\n" , |
7923 | hibernate_rebuild_needed, |
7924 | hibernate_rebuild_hash_list); |
7925 | } |
7926 | #endif /* HIBERNATION */ |
7927 | |
7928 | #if VM_PAGE_FAKE_BUCKETS |
7929 | char *cp; |
7930 | for (cp = (char *) vm_page_fake_buckets_start; |
7931 | cp < (char *) vm_page_fake_buckets_end; |
7932 | cp++) { |
7933 | if (*cp != 0x5a) { |
7934 | panic("BUCKET_CHECK: corruption at %p in fake buckets " |
7935 | "[0x%llx:0x%llx]\n" , |
7936 | cp, |
7937 | (uint64_t) vm_page_fake_buckets_start, |
7938 | (uint64_t) vm_page_fake_buckets_end); |
7939 | } |
7940 | } |
7941 | #endif /* VM_PAGE_FAKE_BUCKETS */ |
7942 | |
7943 | for (i = 0; i < vm_page_bucket_count; i++) { |
7944 | vm_object_t p_object; |
7945 | |
7946 | bucket = &vm_page_buckets[i]; |
7947 | if (!bucket->page_list) { |
7948 | continue; |
7949 | } |
7950 | |
7951 | bucket_lock = &vm_page_bucket_locks[i / BUCKETS_PER_LOCK]; |
7952 | lck_spin_lock(bucket_lock); |
7953 | p = (vm_page_t)(VM_PAGE_UNPACK_PTR(bucket->page_list)); |
7954 | |
7955 | while (p != VM_PAGE_NULL) { |
7956 | p_object = VM_PAGE_OBJECT(p); |
7957 | |
7958 | if (!p->vmp_hashed) { |
7959 | panic("BUCKET_CHECK: page %p (%p,0x%llx) " |
7960 | "hash %d in bucket %d at %p " |
7961 | "is not hashed\n" , |
7962 | p, p_object, p->vmp_offset, |
7963 | p_hash, i, bucket); |
7964 | } |
7965 | p_hash = vm_page_hash(p_object, p->vmp_offset); |
7966 | if (p_hash != i) { |
7967 | panic("BUCKET_CHECK: corruption in bucket %d " |
7968 | "at %p: page %p object %p offset 0x%llx " |
7969 | "hash %d\n" , |
7970 | i, bucket, p, p_object, p->vmp_offset, |
7971 | p_hash); |
7972 | } |
7973 | p = (vm_page_t)(VM_PAGE_UNPACK_PTR(p->vmp_next_m)); |
7974 | } |
7975 | lck_spin_unlock(bucket_lock); |
7976 | } |
7977 | |
7978 | // printf("BUCKET_CHECK: checked buckets\n"); |
7979 | } |
7980 | #endif /* VM_PAGE_BUCKETS_CHECK */ |
7981 | |
7982 | /* |
7983 | * 'vm_fault_enter' will place newly created pages (zero-fill and COW) onto the |
7984 | * local queues if they exist... its the only spot in the system where we add pages |
7985 | * to those queues... once on those queues, those pages can only move to one of the |
7986 | * global page queues or the free queues... they NEVER move from local q to local q. |
7987 | * the 'local' state is stable when vm_page_queues_remove is called since we're behind |
7988 | * the global vm_page_queue_lock at this point... we still need to take the local lock |
7989 | * in case this operation is being run on a different CPU then the local queue's identity, |
7990 | * but we don't have to worry about the page moving to a global queue or becoming wired |
7991 | * while we're grabbing the local lock since those operations would require the global |
7992 | * vm_page_queue_lock to be held, and we already own it. |
7993 | * |
7994 | * this is why its safe to utilze the wire_count field in the vm_page_t as the local_id... |
7995 | * 'wired' and local are ALWAYS mutually exclusive conditions. |
7996 | */ |
7997 | |
7998 | #if CONFIG_BACKGROUND_QUEUE |
7999 | void |
8000 | vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_backgroundq) |
8001 | #else |
8002 | void |
8003 | vm_page_queues_remove(vm_page_t mem, boolean_t __unused remove_from_backgroundq) |
8004 | #endif |
8005 | { |
8006 | boolean_t was_pageable = TRUE; |
8007 | vm_object_t m_object; |
8008 | |
8009 | m_object = VM_PAGE_OBJECT(mem); |
8010 | |
8011 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
8012 | |
8013 | if (mem->vmp_q_state == VM_PAGE_NOT_ON_Q) |
8014 | { |
8015 | assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); |
8016 | #if CONFIG_BACKGROUND_QUEUE |
8017 | if (remove_from_backgroundq == TRUE) { |
8018 | vm_page_remove_from_backgroundq(mem); |
8019 | } |
8020 | if (mem->vmp_on_backgroundq) { |
8021 | assert(mem->vmp_backgroundq.next != 0); |
8022 | assert(mem->vmp_backgroundq.prev != 0); |
8023 | } else { |
8024 | assert(mem->vmp_backgroundq.next == 0); |
8025 | assert(mem->vmp_backgroundq.prev == 0); |
8026 | } |
8027 | #endif /* CONFIG_BACKGROUND_QUEUE */ |
8028 | return; |
8029 | } |
8030 | |
8031 | if (mem->vmp_q_state == VM_PAGE_USED_BY_COMPRESSOR) |
8032 | { |
8033 | assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); |
8034 | #if CONFIG_BACKGROUND_QUEUE |
8035 | assert(mem->vmp_backgroundq.next == 0 && |
8036 | mem->vmp_backgroundq.prev == 0 && |
8037 | mem->vmp_on_backgroundq == FALSE); |
8038 | #endif |
8039 | return; |
8040 | } |
8041 | if (mem->vmp_q_state == VM_PAGE_IS_WIRED) { |
8042 | /* |
8043 | * might put these guys on a list for debugging purposes |
8044 | * if we do, we'll need to remove this assert |
8045 | */ |
8046 | assert(mem->vmp_pageq.next == 0 && mem->vmp_pageq.prev == 0); |
8047 | #if CONFIG_BACKGROUND_QUEUE |
8048 | assert(mem->vmp_backgroundq.next == 0 && |
8049 | mem->vmp_backgroundq.prev == 0 && |
8050 | mem->vmp_on_backgroundq == FALSE); |
8051 | #endif |
8052 | return; |
8053 | } |
8054 | |
8055 | assert(m_object != compressor_object); |
8056 | assert(m_object != kernel_object); |
8057 | assert(m_object != vm_submap_object); |
8058 | assert(!mem->vmp_fictitious); |
8059 | |
8060 | switch(mem->vmp_q_state) { |
8061 | |
8062 | case VM_PAGE_ON_ACTIVE_LOCAL_Q: |
8063 | { |
8064 | struct vpl *lq; |
8065 | |
8066 | lq = &vm_page_local_q[mem->vmp_local_id].vpl_un.vpl; |
8067 | VPL_LOCK(&lq->vpl_lock); |
8068 | vm_page_queue_remove(&lq->vpl_queue, |
8069 | mem, vm_page_t, vmp_pageq); |
8070 | mem->vmp_local_id = 0; |
8071 | lq->vpl_count--; |
8072 | if (m_object->internal) { |
8073 | lq->vpl_internal_count--; |
8074 | } else { |
8075 | lq->vpl_external_count--; |
8076 | } |
8077 | VPL_UNLOCK(&lq->vpl_lock); |
8078 | was_pageable = FALSE; |
8079 | break; |
8080 | } |
8081 | case VM_PAGE_ON_ACTIVE_Q: |
8082 | { |
8083 | vm_page_queue_remove(&vm_page_queue_active, |
8084 | mem, vm_page_t, vmp_pageq); |
8085 | vm_page_active_count--; |
8086 | break; |
8087 | } |
8088 | |
8089 | case VM_PAGE_ON_INACTIVE_INTERNAL_Q: |
8090 | { |
8091 | assert(m_object->internal == TRUE); |
8092 | |
8093 | vm_page_inactive_count--; |
8094 | vm_page_queue_remove(&vm_page_queue_anonymous, |
8095 | mem, vm_page_t, vmp_pageq); |
8096 | vm_page_anonymous_count--; |
8097 | |
8098 | vm_purgeable_q_advance_all(); |
8099 | vm_page_balance_inactive(3); |
8100 | break; |
8101 | } |
8102 | |
8103 | case VM_PAGE_ON_INACTIVE_EXTERNAL_Q: |
8104 | { |
8105 | assert(m_object->internal == FALSE); |
8106 | |
8107 | vm_page_inactive_count--; |
8108 | vm_page_queue_remove(&vm_page_queue_inactive, |
8109 | mem, vm_page_t, vmp_pageq); |
8110 | vm_purgeable_q_advance_all(); |
8111 | vm_page_balance_inactive(3); |
8112 | break; |
8113 | } |
8114 | |
8115 | case VM_PAGE_ON_INACTIVE_CLEANED_Q: |
8116 | { |
8117 | assert(m_object->internal == FALSE); |
8118 | |
8119 | vm_page_inactive_count--; |
8120 | vm_page_queue_remove(&vm_page_queue_cleaned, |
8121 | mem, vm_page_t, vmp_pageq); |
8122 | vm_page_cleaned_count--; |
8123 | vm_page_balance_inactive(3); |
8124 | break; |
8125 | } |
8126 | |
8127 | case VM_PAGE_ON_THROTTLED_Q: |
8128 | { |
8129 | assert(m_object->internal == TRUE); |
8130 | |
8131 | vm_page_queue_remove(&vm_page_queue_throttled, |
8132 | mem, vm_page_t, vmp_pageq); |
8133 | vm_page_throttled_count--; |
8134 | was_pageable = FALSE; |
8135 | break; |
8136 | } |
8137 | |
8138 | case VM_PAGE_ON_SPECULATIVE_Q: |
8139 | { |
8140 | assert(m_object->internal == FALSE); |
8141 | |
8142 | vm_page_remque(&mem->vmp_pageq); |
8143 | vm_page_speculative_count--; |
8144 | vm_page_balance_inactive(3); |
8145 | break; |
8146 | } |
8147 | |
8148 | #if CONFIG_SECLUDED_MEMORY |
8149 | case VM_PAGE_ON_SECLUDED_Q: |
8150 | { |
8151 | vm_page_queue_remove(&vm_page_queue_secluded, |
8152 | mem, vm_page_t, vmp_pageq); |
8153 | vm_page_secluded_count--; |
8154 | if (m_object == VM_OBJECT_NULL) { |
8155 | vm_page_secluded_count_free--; |
8156 | was_pageable = FALSE; |
8157 | } else { |
8158 | assert(!m_object->internal); |
8159 | vm_page_secluded_count_inuse--; |
8160 | was_pageable = FALSE; |
8161 | // was_pageable = TRUE; |
8162 | } |
8163 | break; |
8164 | } |
8165 | #endif /* CONFIG_SECLUDED_MEMORY */ |
8166 | |
8167 | default: |
8168 | { |
8169 | /* |
8170 | * if (mem->vmp_q_state == VM_PAGE_ON_PAGEOUT_Q) |
8171 | * NOTE: vm_page_queues_remove does not deal with removing pages from the pageout queue... |
8172 | * the caller is responsible for determing if the page is on that queue, and if so, must |
8173 | * either first remove it (it needs both the page queues lock and the object lock to do |
8174 | * this via vm_pageout_steal_laundry), or avoid the call to vm_page_queues_remove |
8175 | * |
8176 | * we also don't expect to encounter VM_PAGE_ON_FREE_Q, VM_PAGE_ON_FREE_LOCAL_Q, VM_PAGE_ON_FREE_LOPAGE_Q |
8177 | * or any of the undefined states |
8178 | */ |
8179 | panic("vm_page_queues_remove - bad page q_state (%p, %d)\n" , mem, mem->vmp_q_state); |
8180 | break; |
8181 | } |
8182 | |
8183 | } |
8184 | VM_PAGE_ZERO_PAGEQ_ENTRY(mem); |
8185 | mem->vmp_q_state = VM_PAGE_NOT_ON_Q; |
8186 | |
8187 | #if CONFIG_BACKGROUND_QUEUE |
8188 | if (remove_from_backgroundq == TRUE) |
8189 | vm_page_remove_from_backgroundq(mem); |
8190 | #endif |
8191 | if (was_pageable) { |
8192 | if (m_object->internal) { |
8193 | vm_page_pageable_internal_count--; |
8194 | } else { |
8195 | vm_page_pageable_external_count--; |
8196 | } |
8197 | } |
8198 | } |
8199 | |
8200 | void |
8201 | vm_page_remove_internal(vm_page_t page) |
8202 | { |
8203 | vm_object_t __object = VM_PAGE_OBJECT(page); |
8204 | if (page == __object->memq_hint) { |
8205 | vm_page_t __new_hint; |
8206 | vm_page_queue_entry_t __qe; |
8207 | __qe = (vm_page_queue_entry_t)vm_page_queue_next(&page->vmp_listq); |
8208 | if (vm_page_queue_end(&__object->memq, __qe)) { |
8209 | __qe = (vm_page_queue_entry_t)vm_page_queue_prev(&page->vmp_listq); |
8210 | if (vm_page_queue_end(&__object->memq, __qe)) { |
8211 | __qe = NULL; |
8212 | } |
8213 | } |
8214 | __new_hint = (vm_page_t)((uintptr_t) __qe); |
8215 | __object->memq_hint = __new_hint; |
8216 | } |
8217 | vm_page_queue_remove(&__object->memq, page, vm_page_t, vmp_listq); |
8218 | #if CONFIG_SECLUDED_MEMORY |
8219 | if (__object->eligible_for_secluded) { |
8220 | vm_page_secluded.eligible_for_secluded--; |
8221 | } |
8222 | #endif /* CONFIG_SECLUDED_MEMORY */ |
8223 | } |
8224 | |
8225 | void |
8226 | vm_page_enqueue_inactive(vm_page_t mem, boolean_t first) |
8227 | { |
8228 | vm_object_t m_object; |
8229 | |
8230 | m_object = VM_PAGE_OBJECT(mem); |
8231 | |
8232 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
8233 | assert(!mem->vmp_fictitious); |
8234 | assert(!mem->vmp_laundry); |
8235 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
8236 | vm_page_check_pageable_safe(mem); |
8237 | |
8238 | if (m_object->internal) { |
8239 | mem->vmp_q_state = VM_PAGE_ON_INACTIVE_INTERNAL_Q; |
8240 | |
8241 | if (first == TRUE) |
8242 | vm_page_queue_enter_first(&vm_page_queue_anonymous, mem, vm_page_t, vmp_pageq); |
8243 | else |
8244 | vm_page_queue_enter(&vm_page_queue_anonymous, mem, vm_page_t, vmp_pageq); |
8245 | |
8246 | vm_page_anonymous_count++; |
8247 | vm_page_pageable_internal_count++; |
8248 | } else { |
8249 | mem->vmp_q_state = VM_PAGE_ON_INACTIVE_EXTERNAL_Q; |
8250 | |
8251 | if (first == TRUE) |
8252 | vm_page_queue_enter_first(&vm_page_queue_inactive, mem, vm_page_t, vmp_pageq); |
8253 | else |
8254 | vm_page_queue_enter(&vm_page_queue_inactive, mem, vm_page_t, vmp_pageq); |
8255 | |
8256 | vm_page_pageable_external_count++; |
8257 | } |
8258 | vm_page_inactive_count++; |
8259 | token_new_pagecount++; |
8260 | |
8261 | #if CONFIG_BACKGROUND_QUEUE |
8262 | if (mem->vmp_in_background) |
8263 | vm_page_add_to_backgroundq(mem, FALSE); |
8264 | #endif |
8265 | } |
8266 | |
8267 | void |
8268 | vm_page_enqueue_active(vm_page_t mem, boolean_t first) |
8269 | { |
8270 | vm_object_t m_object; |
8271 | |
8272 | m_object = VM_PAGE_OBJECT(mem); |
8273 | |
8274 | LCK_MTX_ASSERT(&vm_page_queue_lock, LCK_MTX_ASSERT_OWNED); |
8275 | assert(!mem->vmp_fictitious); |
8276 | assert(!mem->vmp_laundry); |
8277 | assert(mem->vmp_q_state == VM_PAGE_NOT_ON_Q); |
8278 | vm_page_check_pageable_safe(mem); |
8279 | |
8280 | mem->vmp_q_state = VM_PAGE_ON_ACTIVE_Q; |
8281 | if (first == TRUE) |
8282 | vm_page_queue_enter_first(&vm_page_queue_active, mem, vm_page_t, vmp_pageq); |
8283 | else |
8284 | vm_page_queue_enter(&vm_page_queue_active, mem, vm_page_t, vmp_pageq); |
8285 | vm_page_active_count++; |
8286 | |
8287 | if (m_object->internal) { |
8288 | vm_page_pageable_internal_count++; |
8289 | } else { |
8290 | vm_page_pageable_external_count++; |
8291 | } |
8292 | |
8293 | #if CONFIG_BACKGROUND_QUEUE |
8294 | if (mem->vmp_in_background) |
8295 | vm_page_add_to_backgroundq(mem, FALSE); |
8296 | #endif |
8297 | vm_page_balance_inactive(3); |
8298 | } |
8299 | |
8300 | /* |
8301 | * Pages from special kernel objects shouldn't |
8302 | * be placed on pageable queues. |
8303 | */ |
8304 | void |
8305 | vm_page_check_pageable_safe(vm_page_t page) |
8306 | { |
8307 | vm_object_t page_object; |
8308 | |
8309 | page_object = VM_PAGE_OBJECT(page); |
8310 | |
8311 | if (page_object == kernel_object) { |
8312 | panic("vm_page_check_pageable_safe: trying to add page" \ |
8313 | "from kernel object (%p) to pageable queue" , kernel_object); |
8314 | } |
8315 | |
8316 | if (page_object == compressor_object) { |
8317 | panic("vm_page_check_pageable_safe: trying to add page" \ |
8318 | "from compressor object (%p) to pageable queue" , compressor_object); |
8319 | } |
8320 | |
8321 | if (page_object == vm_submap_object) { |
8322 | panic("vm_page_check_pageable_safe: trying to add page" \ |
8323 | "from submap object (%p) to pageable queue" , vm_submap_object); |
8324 | } |
8325 | } |
8326 | |
8327 | /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * |
8328 | * wired page diagnose |
8329 | * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ |
8330 | |
8331 | #include <libkern/OSKextLibPrivate.h> |
8332 | |
8333 | #define KA_SIZE(namelen, subtotalscount) \ |
8334 | (sizeof(struct vm_allocation_site) + (namelen) + 1 + ((subtotalscount) * sizeof(struct vm_allocation_total))) |
8335 | |
8336 | #define KA_NAME(alloc) \ |
8337 | ((char *)(&(alloc)->subtotals[(alloc->subtotalscount)])) |
8338 | |
8339 | #define KA_NAME_LEN(alloc) \ |
8340 | (VM_TAG_NAME_LEN_MAX & (alloc->flags >> VM_TAG_NAME_LEN_SHIFT)) |
8341 | |
8342 | vm_tag_t |
8343 | vm_tag_bt(void) |
8344 | { |
8345 | uintptr_t* frameptr; |
8346 | uintptr_t* frameptr_next; |
8347 | uintptr_t retaddr; |
8348 | uintptr_t kstackb, kstackt; |
8349 | const vm_allocation_site_t * site; |
8350 | thread_t cthread; |
8351 | kern_allocation_name_t name; |
8352 | |
8353 | cthread = current_thread(); |
8354 | if (__improbable(cthread == NULL)) return VM_KERN_MEMORY_OSFMK; |
8355 | |
8356 | if ((name = thread_get_kernel_state(cthread)->allocation_name)) |
8357 | { |
8358 | if (!name->tag) vm_tag_alloc(name); |
8359 | return name->tag; |
8360 | } |
8361 | |
8362 | kstackb = cthread->kernel_stack; |
8363 | kstackt = kstackb + kernel_stack_size; |
8364 | |
8365 | /* Load stack frame pointer (EBP on x86) into frameptr */ |
8366 | frameptr = __builtin_frame_address(0); |
8367 | site = NULL; |
8368 | while (frameptr != NULL) |
8369 | { |
8370 | /* Verify thread stack bounds */ |
8371 | if (((uintptr_t)(frameptr + 2) > kstackt) || ((uintptr_t)frameptr < kstackb)) break; |
8372 | |
8373 | /* Next frame pointer is pointed to by the previous one */ |
8374 | frameptr_next = (uintptr_t*) *frameptr; |
8375 | |
8376 | /* Pull return address from one spot above the frame pointer */ |
8377 | retaddr = *(frameptr + 1); |
8378 | |
8379 | |
8380 | if (((retaddr < vm_kernel_builtinkmod_text_end) && (retaddr >= vm_kernel_builtinkmod_text)) |
8381 | || (retaddr < vm_kernel_stext) || (retaddr > vm_kernel_top)) |
8382 | { |
8383 | site = OSKextGetAllocationSiteForCaller(retaddr); |
8384 | break; |
8385 | } |
8386 | frameptr = frameptr_next; |
8387 | } |
8388 | |
8389 | return (site ? site->tag : VM_KERN_MEMORY_NONE); |
8390 | } |
8391 | |
8392 | static uint64_t free_tag_bits[VM_MAX_TAG_VALUE/64]; |
8393 | |
8394 | void |
8395 | vm_tag_alloc_locked(vm_allocation_site_t * site, vm_allocation_site_t ** releasesiteP) |
8396 | { |
8397 | vm_tag_t tag; |
8398 | uint64_t avail; |
8399 | uint32_t idx; |
8400 | vm_allocation_site_t * prev; |
8401 | |
8402 | if (site->tag) return; |
8403 | |
8404 | idx = 0; |
8405 | while (TRUE) |
8406 | { |
8407 | avail = free_tag_bits[idx]; |
8408 | if (avail) |
8409 | { |
8410 | tag = __builtin_clzll(avail); |
8411 | avail &= ~(1ULL << (63 - tag)); |
8412 | free_tag_bits[idx] = avail; |
8413 | tag += (idx << 6); |
8414 | break; |
8415 | } |
8416 | idx++; |
8417 | if (idx >= ARRAY_COUNT(free_tag_bits)) |
8418 | { |
8419 | for (idx = 0; idx < ARRAY_COUNT(vm_allocation_sites); idx++) |
8420 | { |
8421 | prev = vm_allocation_sites[idx]; |
8422 | if (!prev) continue; |
8423 | if (!KA_NAME_LEN(prev)) continue; |
8424 | if (!prev->tag) continue; |
8425 | if (prev->total) continue; |
8426 | if (1 != prev->refcount) continue; |
8427 | |
8428 | assert(idx == prev->tag); |
8429 | tag = idx; |
8430 | prev->tag = VM_KERN_MEMORY_NONE; |
8431 | *releasesiteP = prev; |
8432 | break; |
8433 | } |
8434 | if (idx >= ARRAY_COUNT(vm_allocation_sites)) |
8435 | { |
8436 | tag = VM_KERN_MEMORY_ANY; |
8437 | } |
8438 | break; |
8439 | } |
8440 | } |
8441 | site->tag = tag; |
8442 | |
8443 | OSAddAtomic16(1, &site->refcount); |
8444 | |
8445 | if (VM_KERN_MEMORY_ANY != tag) vm_allocation_sites[tag] = site; |
8446 | |
8447 | if (tag > vm_allocation_tag_highest) vm_allocation_tag_highest = tag; |
8448 | } |
8449 | |
8450 | static void |
8451 | vm_tag_free_locked(vm_tag_t tag) |
8452 | { |
8453 | uint64_t avail; |
8454 | uint32_t idx; |
8455 | uint64_t bit; |
8456 | |
8457 | if (VM_KERN_MEMORY_ANY == tag) return; |
8458 | |
8459 | idx = (tag >> 6); |
8460 | avail = free_tag_bits[idx]; |
8461 | tag &= 63; |
8462 | bit = (1ULL << (63 - tag)); |
8463 | assert(!(avail & bit)); |
8464 | free_tag_bits[idx] = (avail | bit); |
8465 | } |
8466 | |
8467 | static void |
8468 | vm_tag_init(void) |
8469 | { |
8470 | vm_tag_t tag; |
8471 | for (tag = VM_KERN_MEMORY_FIRST_DYNAMIC; tag < VM_KERN_MEMORY_ANY; tag++) |
8472 | { |
8473 | vm_tag_free_locked(tag); |
8474 | } |
8475 | |
8476 | for (tag = VM_KERN_MEMORY_ANY + 1; tag < VM_MAX_TAG_VALUE; tag++) |
8477 | { |
8478 | vm_tag_free_locked(tag); |
8479 | } |
8480 | } |
8481 | |
8482 | vm_tag_t |
8483 | vm_tag_alloc(vm_allocation_site_t * site) |
8484 | { |
8485 | vm_tag_t tag; |
8486 | vm_allocation_site_t * releasesite; |
8487 | |
8488 | if (VM_TAG_BT & site->flags) |
8489 | { |
8490 | tag = vm_tag_bt(); |
8491 | if (VM_KERN_MEMORY_NONE != tag) return (tag); |
8492 | } |
8493 | |
8494 | if (!site->tag) |
8495 | { |
8496 | releasesite = NULL; |
8497 | lck_spin_lock(&vm_allocation_sites_lock); |
8498 | vm_tag_alloc_locked(site, &releasesite); |
8499 | lck_spin_unlock(&vm_allocation_sites_lock); |
8500 | if (releasesite) kern_allocation_name_release(releasesite); |
8501 | } |
8502 | |
8503 | return (site->tag); |
8504 | } |
8505 | |
8506 | void |
8507 | vm_tag_update_size(vm_tag_t tag, int64_t delta) |
8508 | { |
8509 | vm_allocation_site_t * allocation; |
8510 | uint64_t prior; |
8511 | |
8512 | assert(VM_KERN_MEMORY_NONE != tag); |
8513 | assert(tag < VM_MAX_TAG_VALUE); |
8514 | |
8515 | allocation = vm_allocation_sites[tag]; |
8516 | assert(allocation); |
8517 | |
8518 | if (delta < 0) { |
8519 | assertf(allocation->total >= ((uint64_t)-delta), "tag %d, site %p" , tag, allocation); |
8520 | } |
8521 | prior = OSAddAtomic64(delta, &allocation->total); |
8522 | |
8523 | #if DEBUG || DEVELOPMENT |
8524 | |
8525 | uint64_t new, peak; |
8526 | new = prior + delta; |
8527 | do |
8528 | { |
8529 | peak = allocation->peak; |
8530 | if (new <= peak) break; |
8531 | } |
8532 | while (!OSCompareAndSwap64(peak, new, &allocation->peak)); |
8533 | |
8534 | #endif /* DEBUG || DEVELOPMENT */ |
8535 | |
8536 | if (tag < VM_KERN_MEMORY_FIRST_DYNAMIC) return; |
8537 | |
8538 | if (!prior && !allocation->tag) vm_tag_alloc(allocation); |
8539 | } |
8540 | |
8541 | void |
8542 | kern_allocation_update_size(kern_allocation_name_t allocation, int64_t delta) |
8543 | { |
8544 | uint64_t prior; |
8545 | |
8546 | if (delta < 0) { |
8547 | assertf(allocation->total >= ((uint64_t)-delta), "name %p" , allocation); |
8548 | } |
8549 | prior = OSAddAtomic64(delta, &allocation->total); |
8550 | |
8551 | #if DEBUG || DEVELOPMENT |
8552 | |
8553 | uint64_t new, peak; |
8554 | new = prior + delta; |
8555 | do |
8556 | { |
8557 | peak = allocation->peak; |
8558 | if (new <= peak) break; |
8559 | } |
8560 | while (!OSCompareAndSwap64(peak, new, &allocation->peak)); |
8561 | |
8562 | #endif /* DEBUG || DEVELOPMENT */ |
8563 | |
8564 | if (!prior && !allocation->tag) vm_tag_alloc(allocation); |
8565 | } |
8566 | |
8567 | #if VM_MAX_TAG_ZONES |
8568 | |
8569 | void |
8570 | vm_allocation_zones_init(void) |
8571 | { |
8572 | kern_return_t ret; |
8573 | vm_offset_t addr; |
8574 | vm_size_t size; |
8575 | |
8576 | size = VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t **) |
8577 | + 2 * VM_MAX_TAG_ZONES * sizeof(vm_allocation_zone_total_t); |
8578 | |
8579 | ret = kernel_memory_allocate(kernel_map, |
8580 | &addr, round_page(size), 0, |
8581 | KMA_ZERO, VM_KERN_MEMORY_DIAG); |
8582 | assert(KERN_SUCCESS == ret); |
8583 | |
8584 | vm_allocation_zone_totals = (vm_allocation_zone_total_t **) addr; |
8585 | addr += VM_MAX_TAG_VALUE * sizeof(vm_allocation_zone_total_t **); |
8586 | |
8587 | // prepopulate VM_KERN_MEMORY_DIAG & VM_KERN_MEMORY_KALLOC so allocations |
8588 | // in vm_tag_update_zone_size() won't recurse |
8589 | vm_allocation_zone_totals[VM_KERN_MEMORY_DIAG] = (vm_allocation_zone_total_t *) addr; |
8590 | addr += VM_MAX_TAG_ZONES * sizeof(vm_allocation_zone_total_t); |
8591 | vm_allocation_zone_totals[VM_KERN_MEMORY_KALLOC] = (vm_allocation_zone_total_t *) addr; |
8592 | } |
8593 | |
8594 | void |
8595 | vm_tag_will_update_zone(vm_tag_t tag, uint32_t zidx) |
8596 | { |
8597 | vm_allocation_zone_total_t * zone; |
8598 | |
8599 | assert(VM_KERN_MEMORY_NONE != tag); |
8600 | assert(tag < VM_MAX_TAG_VALUE); |
8601 | |
8602 | if (zidx >= VM_MAX_TAG_ZONES) return; |
8603 | |
8604 | zone = vm_allocation_zone_totals[tag]; |
8605 | if (!zone) |
8606 | { |
8607 | zone = kalloc_tag(VM_MAX_TAG_ZONES * sizeof(*zone), VM_KERN_MEMORY_DIAG); |
8608 | if (!zone) return; |
8609 | bzero(zone, VM_MAX_TAG_ZONES * sizeof(*zone)); |
8610 | if (!OSCompareAndSwapPtr(NULL, zone, &vm_allocation_zone_totals[tag])) |
8611 | { |
8612 | kfree(zone, VM_MAX_TAG_ZONES * sizeof(*zone)); |
8613 | } |
8614 | } |
8615 | } |
8616 | |
8617 | void |
8618 | vm_tag_update_zone_size(vm_tag_t tag, uint32_t zidx, int64_t delta, int64_t dwaste) |
8619 | { |
8620 | vm_allocation_zone_total_t * zone; |
8621 | uint32_t new; |
8622 | |
8623 | assert(VM_KERN_MEMORY_NONE != tag); |
8624 | assert(tag < VM_MAX_TAG_VALUE); |
8625 | |
8626 | if (zidx >= VM_MAX_TAG_ZONES) return; |
8627 | |
8628 | zone = vm_allocation_zone_totals[tag]; |
8629 | assert(zone); |
8630 | zone += zidx; |
8631 | |
8632 | /* the zone is locked */ |
8633 | if (delta < 0) |
8634 | { |
8635 | assertf(zone->total >= ((uint64_t)-delta), "zidx %d, tag %d, %p" , zidx, tag, zone); |
8636 | zone->total += delta; |
8637 | } |
8638 | else |
8639 | { |
8640 | zone->total += delta; |
8641 | if (zone->total > zone->peak) zone->peak = zone->total; |
8642 | if (dwaste) |
8643 | { |
8644 | new = zone->waste; |
8645 | if (zone->wastediv < 65536) zone->wastediv++; |
8646 | else new -= (new >> 16); |
8647 | __assert_only bool ov = os_add_overflow(new, dwaste, &new); |
8648 | assert(!ov); |
8649 | zone->waste = new; |
8650 | } |
8651 | } |
8652 | } |
8653 | |
8654 | #endif /* VM_MAX_TAG_ZONES */ |
8655 | |
8656 | void |
8657 | kern_allocation_update_subtotal(kern_allocation_name_t allocation, uint32_t subtag, int64_t delta) |
8658 | { |
8659 | kern_allocation_name_t other; |
8660 | struct vm_allocation_total * total; |
8661 | uint32_t subidx; |
8662 | |
8663 | subidx = 0; |
8664 | assert(VM_KERN_MEMORY_NONE != subtag); |
8665 | for (; subidx < allocation->subtotalscount; subidx++) |
8666 | { |
8667 | if (VM_KERN_MEMORY_NONE == allocation->subtotals[subidx].tag) |
8668 | { |
8669 | allocation->subtotals[subidx].tag = subtag; |
8670 | break; |
8671 | } |
8672 | if (subtag == allocation->subtotals[subidx].tag) break; |
8673 | } |
8674 | assert(subidx < allocation->subtotalscount); |
8675 | if (subidx >= allocation->subtotalscount) return; |
8676 | |
8677 | total = &allocation->subtotals[subidx]; |
8678 | other = vm_allocation_sites[subtag]; |
8679 | assert(other); |
8680 | |
8681 | if (delta < 0) |
8682 | { |
8683 | assertf(total->total >= ((uint64_t)-delta), "name %p" , allocation); |
8684 | OSAddAtomic64(delta, &total->total); |
8685 | assertf(other->mapped >= ((uint64_t)-delta), "other %p" , other); |
8686 | OSAddAtomic64(delta, &other->mapped); |
8687 | } |
8688 | else |
8689 | { |
8690 | OSAddAtomic64(delta, &other->mapped); |
8691 | OSAddAtomic64(delta, &total->total); |
8692 | } |
8693 | } |
8694 | |
8695 | const char * |
8696 | kern_allocation_get_name(kern_allocation_name_t allocation) |
8697 | { |
8698 | return (KA_NAME(allocation)); |
8699 | } |
8700 | |
8701 | kern_allocation_name_t |
8702 | kern_allocation_name_allocate(const char * name, uint32_t subtotalscount) |
8703 | { |
8704 | uint32_t namelen; |
8705 | |
8706 | namelen = (uint32_t) strnlen(name, MACH_MEMORY_INFO_NAME_MAX_LEN - 1); |
8707 | |
8708 | kern_allocation_name_t allocation; |
8709 | allocation = kalloc(KA_SIZE(namelen, subtotalscount)); |
8710 | bzero(allocation, KA_SIZE(namelen, subtotalscount)); |
8711 | |
8712 | allocation->refcount = 1; |
8713 | allocation->subtotalscount = subtotalscount; |
8714 | allocation->flags = (namelen << VM_TAG_NAME_LEN_SHIFT); |
8715 | strlcpy(KA_NAME(allocation), name, namelen + 1); |
8716 | |
8717 | return (allocation); |
8718 | } |
8719 | |
8720 | void |
8721 | kern_allocation_name_release(kern_allocation_name_t allocation) |
8722 | { |
8723 | assert(allocation->refcount > 0); |
8724 | if (1 == OSAddAtomic16(-1, &allocation->refcount)) |
8725 | { |
8726 | kfree(allocation, KA_SIZE(KA_NAME_LEN(allocation), allocation->subtotalscount)); |
8727 | } |
8728 | } |
8729 | |
8730 | vm_tag_t |
8731 | kern_allocation_name_get_vm_tag(kern_allocation_name_t allocation) |
8732 | { |
8733 | return (vm_tag_alloc(allocation)); |
8734 | } |
8735 | |
8736 | #if ! VM_TAG_ACTIVE_UPDATE |
8737 | static void |
8738 | vm_page_count_object(mach_memory_info_t * info, unsigned int __unused num_info, vm_object_t object) |
8739 | { |
8740 | if (!object->wired_page_count) return; |
8741 | if (object != kernel_object) |
8742 | { |
8743 | assert(object->wire_tag < num_info); |
8744 | info[object->wire_tag].size += ptoa_64(object->wired_page_count); |
8745 | } |
8746 | } |
8747 | |
8748 | typedef void (*vm_page_iterate_proc)(mach_memory_info_t * info, |
8749 | unsigned int num_info, vm_object_t object); |
8750 | |
8751 | static void |
8752 | vm_page_iterate_purgeable_objects(mach_memory_info_t * info, unsigned int num_info, |
8753 | vm_page_iterate_proc proc, purgeable_q_t queue, |
8754 | int group) |
8755 | { |
8756 | vm_object_t object; |
8757 | |
8758 | for (object = (vm_object_t) queue_first(&queue->objq[group]); |
8759 | !queue_end(&queue->objq[group], (queue_entry_t) object); |
8760 | object = (vm_object_t) queue_next(&object->objq)) |
8761 | { |
8762 | proc(info, num_info, object); |
8763 | } |
8764 | } |
8765 | |
8766 | static void |
8767 | vm_page_iterate_objects(mach_memory_info_t * info, unsigned int num_info, |
8768 | vm_page_iterate_proc proc) |
8769 | { |
8770 | vm_object_t object; |
8771 | |
8772 | lck_spin_lock(&vm_objects_wired_lock); |
8773 | queue_iterate(&vm_objects_wired, |
8774 | object, |
8775 | vm_object_t, |
8776 | wired_objq) |
8777 | { |
8778 | proc(info, num_info, object); |
8779 | } |
8780 | lck_spin_unlock(&vm_objects_wired_lock); |
8781 | } |
8782 | #endif /* ! VM_TAG_ACTIVE_UPDATE */ |
8783 | |
8784 | static uint64_t |
8785 | process_account(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes, boolean_t iterated) |
8786 | { |
8787 | size_t namelen; |
8788 | unsigned int idx, count, nextinfo; |
8789 | vm_allocation_site_t * site; |
8790 | lck_spin_lock(&vm_allocation_sites_lock); |
8791 | |
8792 | for (idx = 0; idx <= vm_allocation_tag_highest; idx++) |
8793 | { |
8794 | site = vm_allocation_sites[idx]; |
8795 | if (!site) continue; |
8796 | info[idx].mapped = site->mapped; |
8797 | info[idx].tag = site->tag; |
8798 | if (!iterated) |
8799 | { |
8800 | info[idx].size = site->total; |
8801 | #if DEBUG || DEVELOPMENT |
8802 | info[idx].peak = site->peak; |
8803 | #endif /* DEBUG || DEVELOPMENT */ |
8804 | } |
8805 | else |
8806 | { |
8807 | if (!site->subtotalscount && (site->total != info[idx].size)) |
8808 | { |
8809 | printf("tag mismatch[%d] 0x%qx, iter 0x%qx\n" , idx, site->total, info[idx].size); |
8810 | info[idx].size = site->total; |
8811 | } |
8812 | } |
8813 | } |
8814 | |
8815 | nextinfo = (vm_allocation_tag_highest + 1); |
8816 | count = nextinfo; |
8817 | if (count >= num_info) count = num_info; |
8818 | |
8819 | for (idx = 0; idx < count; idx++) |
8820 | { |
8821 | site = vm_allocation_sites[idx]; |
8822 | if (!site) continue; |
8823 | info[idx].flags |= VM_KERN_SITE_WIRED; |
8824 | if (idx < VM_KERN_MEMORY_FIRST_DYNAMIC) |
8825 | { |
8826 | info[idx].site = idx; |
8827 | info[idx].flags |= VM_KERN_SITE_TAG; |
8828 | if (VM_KERN_MEMORY_ZONE == idx) |
8829 | { |
8830 | info[idx].flags |= VM_KERN_SITE_HIDE; |
8831 | info[idx].flags &= ~VM_KERN_SITE_WIRED; |
8832 | info[idx].collectable_bytes = zones_collectable_bytes; |
8833 | } |
8834 | } |
8835 | else if ((namelen = (VM_TAG_NAME_LEN_MAX & (site->flags >> VM_TAG_NAME_LEN_SHIFT)))) |
8836 | { |
8837 | info[idx].site = 0; |
8838 | info[idx].flags |= VM_KERN_SITE_NAMED; |
8839 | if (namelen > sizeof(info[idx].name)) namelen = sizeof(info[idx].name); |
8840 | strncpy(&info[idx].name[0], KA_NAME(site), namelen); |
8841 | } |
8842 | else if (VM_TAG_KMOD & site->flags) |
8843 | { |
8844 | info[idx].site = OSKextGetKmodIDForSite(site, NULL, 0); |
8845 | info[idx].flags |= VM_KERN_SITE_KMOD; |
8846 | } |
8847 | else |
8848 | { |
8849 | info[idx].site = VM_KERNEL_UNSLIDE(site); |
8850 | info[idx].flags |= VM_KERN_SITE_KERNEL; |
8851 | } |
8852 | #if VM_MAX_TAG_ZONES |
8853 | vm_allocation_zone_total_t * zone; |
8854 | unsigned int zidx; |
8855 | vm_size_t elem_size; |
8856 | |
8857 | if (vm_allocation_zone_totals |
8858 | && (zone = vm_allocation_zone_totals[idx]) |
8859 | && (nextinfo < num_info)) |
8860 | { |
8861 | for (zidx = 0; zidx < VM_MAX_TAG_ZONES; zidx++) |
8862 | { |
8863 | if (!zone[zidx].peak) continue; |
8864 | info[nextinfo] = info[idx]; |
8865 | info[nextinfo].zone = zone_index_from_tag_index(zidx, &elem_size); |
8866 | info[nextinfo].flags &= ~VM_KERN_SITE_WIRED; |
8867 | info[nextinfo].flags |= VM_KERN_SITE_ZONE; |
8868 | info[nextinfo].size = zone[zidx].total; |
8869 | info[nextinfo].peak = zone[zidx].peak; |
8870 | info[nextinfo].mapped = 0; |
8871 | if (zone[zidx].wastediv) |
8872 | { |
8873 | info[nextinfo].collectable_bytes = ((zone[zidx].waste * zone[zidx].total / elem_size) / zone[zidx].wastediv); |
8874 | } |
8875 | nextinfo++; |
8876 | } |
8877 | } |
8878 | #endif /* VM_MAX_TAG_ZONES */ |
8879 | if (site->subtotalscount) |
8880 | { |
8881 | uint64_t mapped, mapcost, take; |
8882 | uint32_t sub; |
8883 | vm_tag_t alloctag; |
8884 | |
8885 | info[idx].size = site->total; |
8886 | mapped = info[idx].size; |
8887 | info[idx].mapped = mapped; |
8888 | mapcost = 0; |
8889 | for (sub = 0; sub < site->subtotalscount; sub++) |
8890 | { |
8891 | alloctag = site->subtotals[sub].tag; |
8892 | assert(alloctag < num_info); |
8893 | if (info[alloctag].name[0]) continue; |
8894 | take = info[alloctag].mapped; |
8895 | if (take > info[alloctag].size) take = info[alloctag].size; |
8896 | if (take > mapped) take = mapped; |
8897 | info[alloctag].mapped -= take; |
8898 | info[alloctag].size -= take; |
8899 | mapped -= take; |
8900 | mapcost += take; |
8901 | } |
8902 | info[idx].size = mapcost; |
8903 | } |
8904 | } |
8905 | lck_spin_unlock(&vm_allocation_sites_lock); |
8906 | |
8907 | return (0); |
8908 | } |
8909 | |
8910 | uint32_t |
8911 | vm_page_diagnose_estimate(void) |
8912 | { |
8913 | vm_allocation_site_t * site; |
8914 | uint32_t count; |
8915 | uint32_t idx; |
8916 | |
8917 | lck_spin_lock(&vm_allocation_sites_lock); |
8918 | for (count = idx = 0; idx < VM_MAX_TAG_VALUE; idx++) |
8919 | { |
8920 | site = vm_allocation_sites[idx]; |
8921 | if (!site) continue; |
8922 | count++; |
8923 | #if VM_MAX_TAG_ZONES |
8924 | if (vm_allocation_zone_totals) |
8925 | { |
8926 | vm_allocation_zone_total_t * zone; |
8927 | zone = vm_allocation_zone_totals[idx]; |
8928 | if (!zone) continue; |
8929 | for (uint32_t zidx = 0; zidx < VM_MAX_TAG_ZONES; zidx++) if (zone[zidx].peak) count++; |
8930 | } |
8931 | #endif |
8932 | } |
8933 | lck_spin_unlock(&vm_allocation_sites_lock); |
8934 | |
8935 | /* some slop for new tags created */ |
8936 | count += 8; |
8937 | count += VM_KERN_COUNTER_COUNT; |
8938 | |
8939 | return (count); |
8940 | } |
8941 | |
8942 | |
8943 | kern_return_t |
8944 | vm_page_diagnose(mach_memory_info_t * info, unsigned int num_info, uint64_t zones_collectable_bytes) |
8945 | { |
8946 | uint64_t wired_size; |
8947 | uint64_t wired_managed_size; |
8948 | uint64_t wired_reserved_size; |
8949 | uint64_t booter_size; |
8950 | boolean_t iterate; |
8951 | mach_memory_info_t * counts; |
8952 | |
8953 | bzero(info, num_info * sizeof(mach_memory_info_t)); |
8954 | |
8955 | if (!vm_page_wire_count_initial) return (KERN_ABORTED); |
8956 | |
8957 | #if CONFIG_EMBEDDED |
8958 | wired_size = ptoa_64(vm_page_wire_count); |
8959 | wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count); |
8960 | #else |
8961 | wired_size = ptoa_64(vm_page_wire_count + vm_lopage_free_count + vm_page_throttled_count); |
8962 | wired_reserved_size = ptoa_64(vm_page_wire_count_initial - vm_page_stolen_count + vm_page_throttled_count); |
8963 | #endif |
8964 | wired_managed_size = ptoa_64(vm_page_wire_count - vm_page_wire_count_initial); |
8965 | |
8966 | booter_size = ml_get_booter_memory_size(); |
8967 | wired_size += booter_size; |
8968 | |
8969 | assert(num_info >= VM_KERN_COUNTER_COUNT); |
8970 | num_info -= VM_KERN_COUNTER_COUNT; |
8971 | counts = &info[num_info]; |
8972 | |
8973 | #define SET_COUNT(xcount, xsize, xflags) \ |
8974 | counts[xcount].tag = VM_MAX_TAG_VALUE + xcount; \ |
8975 | counts[xcount].site = (xcount); \ |
8976 | counts[xcount].size = (xsize); \ |
8977 | counts[xcount].mapped = (xsize); \ |
8978 | counts[xcount].flags = VM_KERN_SITE_COUNTER | xflags; |
8979 | |
8980 | SET_COUNT(VM_KERN_COUNT_MANAGED, ptoa_64(vm_page_pages), 0); |
8981 | SET_COUNT(VM_KERN_COUNT_WIRED, wired_size, 0); |
8982 | SET_COUNT(VM_KERN_COUNT_WIRED_MANAGED, wired_managed_size, 0); |
8983 | SET_COUNT(VM_KERN_COUNT_RESERVED, wired_reserved_size, VM_KERN_SITE_WIRED); |
8984 | SET_COUNT(VM_KERN_COUNT_STOLEN, ptoa_64(vm_page_stolen_count), VM_KERN_SITE_WIRED); |
8985 | SET_COUNT(VM_KERN_COUNT_LOPAGE, ptoa_64(vm_lopage_free_count), VM_KERN_SITE_WIRED); |
8986 | SET_COUNT(VM_KERN_COUNT_WIRED_BOOT, ptoa_64(vm_page_wire_count_on_boot), 0); |
8987 | SET_COUNT(VM_KERN_COUNT_BOOT_STOLEN, booter_size, VM_KERN_SITE_WIRED); |
8988 | |
8989 | #define SET_MAP(xcount, xsize, xfree, xlargest) \ |
8990 | counts[xcount].site = (xcount); \ |
8991 | counts[xcount].size = (xsize); \ |
8992 | counts[xcount].mapped = (xsize); \ |
8993 | counts[xcount].free = (xfree); \ |
8994 | counts[xcount].largest = (xlargest); \ |
8995 | counts[xcount].flags = VM_KERN_SITE_COUNTER; |
8996 | |
8997 | vm_map_size_t map_size, map_free, map_largest; |
8998 | |
8999 | vm_map_sizes(kernel_map, &map_size, &map_free, &map_largest); |
9000 | SET_MAP(VM_KERN_COUNT_MAP_KERNEL, map_size, map_free, map_largest); |
9001 | |
9002 | vm_map_sizes(zone_map, &map_size, &map_free, &map_largest); |
9003 | SET_MAP(VM_KERN_COUNT_MAP_ZONE, map_size, map_free, map_largest); |
9004 | |
9005 | vm_map_sizes(kalloc_map, &map_size, &map_free, &map_largest); |
9006 | SET_MAP(VM_KERN_COUNT_MAP_KALLOC, map_size, map_free, map_largest); |
9007 | |
9008 | iterate = !VM_TAG_ACTIVE_UPDATE; |
9009 | if (iterate) |
9010 | { |
9011 | enum { kMaxKernelDepth = 1 }; |
9012 | vm_map_t maps [kMaxKernelDepth]; |
9013 | vm_map_entry_t entries[kMaxKernelDepth]; |
9014 | vm_map_t map; |
9015 | vm_map_entry_t entry; |
9016 | vm_object_offset_t offset; |
9017 | vm_page_t page; |
9018 | int stackIdx, count; |
9019 | |
9020 | #if ! VM_TAG_ACTIVE_UPDATE |
9021 | vm_page_iterate_objects(info, num_info, &vm_page_count_object); |
9022 | #endif /* ! VM_TAG_ACTIVE_UPDATE */ |
9023 | |
9024 | map = kernel_map; |
9025 | stackIdx = 0; |
9026 | while (map) |
9027 | { |
9028 | vm_map_lock(map); |
9029 | for (entry = map->hdr.links.next; map; entry = entry->links.next) |
9030 | { |
9031 | if (entry->is_sub_map) |
9032 | { |
9033 | assert(stackIdx < kMaxKernelDepth); |
9034 | maps[stackIdx] = map; |
9035 | entries[stackIdx] = entry; |
9036 | stackIdx++; |
9037 | map = VME_SUBMAP(entry); |
9038 | entry = NULL; |
9039 | break; |
9040 | } |
9041 | if (VME_OBJECT(entry) == kernel_object) |
9042 | { |
9043 | count = 0; |
9044 | vm_object_lock(VME_OBJECT(entry)); |
9045 | for (offset = entry->links.start; offset < entry->links.end; offset += page_size) |
9046 | { |
9047 | page = vm_page_lookup(VME_OBJECT(entry), offset); |
9048 | if (page && VM_PAGE_WIRED(page)) count++; |
9049 | } |
9050 | vm_object_unlock(VME_OBJECT(entry)); |
9051 | |
9052 | if (count) |
9053 | { |
9054 | assert(VME_ALIAS(entry) != VM_KERN_MEMORY_NONE); |
9055 | assert(VME_ALIAS(entry) < num_info); |
9056 | info[VME_ALIAS(entry)].size += ptoa_64(count); |
9057 | } |
9058 | } |
9059 | while (map && (entry == vm_map_last_entry(map))) |
9060 | { |
9061 | vm_map_unlock(map); |
9062 | if (!stackIdx) map = NULL; |
9063 | else |
9064 | { |
9065 | --stackIdx; |
9066 | map = maps[stackIdx]; |
9067 | entry = entries[stackIdx]; |
9068 | } |
9069 | } |
9070 | } |
9071 | } |
9072 | } |
9073 | |
9074 | process_account(info, num_info, zones_collectable_bytes, iterate); |
9075 | |
9076 | return (KERN_SUCCESS); |
9077 | } |
9078 | |
9079 | #if DEBUG || DEVELOPMENT |
9080 | |
9081 | kern_return_t |
9082 | vm_kern_allocation_info(uintptr_t addr, vm_size_t * size, vm_tag_t * tag, vm_size_t * zone_size) |
9083 | { |
9084 | kern_return_t ret; |
9085 | vm_size_t zsize; |
9086 | vm_map_t map; |
9087 | vm_map_entry_t entry; |
9088 | |
9089 | zsize = zone_element_info((void *) addr, tag); |
9090 | if (zsize) |
9091 | { |
9092 | *zone_size = *size = zsize; |
9093 | return (KERN_SUCCESS); |
9094 | } |
9095 | |
9096 | *zone_size = 0; |
9097 | ret = KERN_INVALID_ADDRESS; |
9098 | for (map = kernel_map; map; ) |
9099 | { |
9100 | vm_map_lock(map); |
9101 | if (!vm_map_lookup_entry(map, addr, &entry)) break; |
9102 | if (entry->is_sub_map) |
9103 | { |
9104 | if (map != kernel_map) break; |
9105 | map = VME_SUBMAP(entry); |
9106 | continue; |
9107 | } |
9108 | if (entry->vme_start != addr) break; |
9109 | *tag = VME_ALIAS(entry); |
9110 | *size = (entry->vme_end - addr); |
9111 | ret = KERN_SUCCESS; |
9112 | break; |
9113 | } |
9114 | if (map != kernel_map) vm_map_unlock(map); |
9115 | vm_map_unlock(kernel_map); |
9116 | |
9117 | return (ret); |
9118 | } |
9119 | |
9120 | #endif /* DEBUG || DEVELOPMENT */ |
9121 | |
9122 | uint32_t |
9123 | vm_tag_get_kext(vm_tag_t tag, char * name, vm_size_t namelen) |
9124 | { |
9125 | vm_allocation_site_t * site; |
9126 | uint32_t kmodId; |
9127 | |
9128 | kmodId = 0; |
9129 | lck_spin_lock(&vm_allocation_sites_lock); |
9130 | if ((site = vm_allocation_sites[tag])) |
9131 | { |
9132 | if (VM_TAG_KMOD & site->flags) |
9133 | { |
9134 | kmodId = OSKextGetKmodIDForSite(site, name, namelen); |
9135 | } |
9136 | } |
9137 | lck_spin_unlock(&vm_allocation_sites_lock); |
9138 | |
9139 | return (kmodId); |
9140 | } |
9141 | |
9142 | |
9143 | #if CONFIG_SECLUDED_MEMORY |
9144 | /* |
9145 | * Note that there's no locking around other accesses to vm_page_secluded_target. |
9146 | * That should be OK, since these are the only place where it can be changed after |
9147 | * initialization. Other users (like vm_pageout) may see the wrong value briefly, |
9148 | * but will eventually get the correct value. This brief mismatch is OK as pageout |
9149 | * and page freeing will auto-adjust the vm_page_secluded_count to match the target |
9150 | * over time. |
9151 | */ |
9152 | unsigned int vm_page_secluded_suppress_cnt = 0; |
9153 | unsigned int vm_page_secluded_save_target; |
9154 | |
9155 | |
9156 | lck_grp_attr_t secluded_suppress_slock_grp_attr; |
9157 | lck_grp_t secluded_suppress_slock_grp; |
9158 | lck_attr_t secluded_suppress_slock_attr; |
9159 | lck_spin_t secluded_suppress_slock; |
9160 | |
9161 | void |
9162 | secluded_suppression_init(void) |
9163 | { |
9164 | lck_grp_attr_setdefault(&secluded_suppress_slock_grp_attr); |
9165 | lck_grp_init(&secluded_suppress_slock_grp, |
9166 | "secluded_suppress_slock" , &secluded_suppress_slock_grp_attr); |
9167 | lck_attr_setdefault(&secluded_suppress_slock_attr); |
9168 | lck_spin_init(&secluded_suppress_slock, |
9169 | &secluded_suppress_slock_grp, &secluded_suppress_slock_attr); |
9170 | } |
9171 | |
9172 | void |
9173 | start_secluded_suppression(task_t task) |
9174 | { |
9175 | if (task->task_suppressed_secluded) |
9176 | return; |
9177 | lck_spin_lock(&secluded_suppress_slock); |
9178 | if (!task->task_suppressed_secluded && vm_page_secluded_suppress_cnt++ == 0) { |
9179 | task->task_suppressed_secluded = TRUE; |
9180 | vm_page_secluded_save_target = vm_page_secluded_target; |
9181 | vm_page_secluded_target = 0; |
9182 | } |
9183 | lck_spin_unlock(&secluded_suppress_slock); |
9184 | } |
9185 | |
9186 | void |
9187 | stop_secluded_suppression(task_t task) |
9188 | { |
9189 | lck_spin_lock(&secluded_suppress_slock); |
9190 | if (task->task_suppressed_secluded && --vm_page_secluded_suppress_cnt == 0) { |
9191 | task->task_suppressed_secluded = FALSE; |
9192 | vm_page_secluded_target = vm_page_secluded_save_target; |
9193 | } |
9194 | lck_spin_unlock(&secluded_suppress_slock); |
9195 | } |
9196 | |
9197 | #endif /* CONFIG_SECLUDED_MEMORY */ |
9198 | |