1 /*
2 * Copyright (c) 2000-2020 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * @OSF_COPYRIGHT@
30 */
31 /*
32 * Mach Operating System
33 * Copyright (c) 1991,1990,1989,1988 Carnegie Mellon University
34 * All Rights Reserved.
35 *
36 * Permission to use, copy, modify and distribute this software and its
37 * documentation is hereby granted, provided that both the copyright
38 * notice and this permission notice appear in all copies of the
39 * software, derivative works or modified versions, and any portions
40 * thereof, and that both notices appear in supporting documentation.
41 *
42 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
43 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
44 * ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
45 *
46 * Carnegie Mellon requests users of this software to return to
47 *
48 * Software Distribution Coordinator or [email protected]
49 * School of Computer Science
50 * Carnegie Mellon University
51 * Pittsburgh PA 15213-3890
52 *
53 * any improvements or extensions that they make and grant Carnegie Mellon
54 * the rights to redistribute these changes.
55 */
56 /*
57 */
58 /*
59 * File: vm/vm_page.h
60 * Author: Avadis Tevanian, Jr., Michael Wayne Young
61 * Date: 1985
62 *
63 * Resident memory system definitions.
64 */
65
66 #ifndef _VM_VM_PAGE_H_
67 #define _VM_VM_PAGE_H_
68
69 #include <debug.h>
70 #include <vm/vm_options.h>
71 #include <vm/vm_protos.h>
72 #include <mach/boolean.h>
73 #include <mach/vm_prot.h>
74 #include <mach/vm_param.h>
75 #include <mach/memory_object_types.h> /* for VMP_CS_BITS... */
76
77
78 #if defined(__LP64__)
79
80 /*
81 * in order to make the size of a vm_page_t 64 bytes (cache line size for both arm64 and x86_64)
82 * we'll keep the next_m pointer packed... as long as the kernel virtual space where we allocate
83 * vm_page_t's from doesn't span more then 256 Gbytes, we're safe. There are live tests in the
84 * vm_page_t array allocation and the zone init code to determine if we can safely pack and unpack
85 * pointers from the 2 ends of these spaces
86 */
87 typedef uint32_t vm_page_packed_t;
88
89 struct vm_page_packed_queue_entry {
90 vm_page_packed_t next; /* next element */
91 vm_page_packed_t prev; /* previous element */
92 };
93
94 typedef struct vm_page_packed_queue_entry *vm_page_queue_t;
95 typedef struct vm_page_packed_queue_entry vm_page_queue_head_t;
96 typedef struct vm_page_packed_queue_entry vm_page_queue_chain_t;
97 typedef struct vm_page_packed_queue_entry *vm_page_queue_entry_t;
98
99 typedef vm_page_packed_t vm_page_object_t;
100
101 #else // __LP64__
102
103 /*
104 * we can't do the packing trick on 32 bit architectures
105 * so just turn the macros into noops.
106 */
107 typedef struct vm_page *vm_page_packed_t;
108
109 #define vm_page_queue_t queue_t
110 #define vm_page_queue_head_t queue_head_t
111 #define vm_page_queue_chain_t queue_chain_t
112 #define vm_page_queue_entry_t queue_entry_t
113
114 #define vm_page_object_t vm_object_t
115 #endif // __LP64__
116
117
118 #include <vm/vm_object.h>
119 #include <kern/queue.h>
120 #include <kern/locks.h>
121
122 #include <kern/macro_help.h>
123 #include <libkern/OSAtomic.h>
124
125
126
127 #define VM_PAGE_COMPRESSOR_COUNT (compressor_object->resident_page_count)
128
129 /*
130 * Management of resident (logical) pages.
131 *
132 * A small structure is kept for each resident
133 * page, indexed by page number. Each structure
134 * is an element of several lists:
135 *
136 * A hash table bucket used to quickly
137 * perform object/offset lookups
138 *
139 * A list of all pages for a given object,
140 * so they can be quickly deactivated at
141 * time of deallocation.
142 *
143 * An ordered list of pages due for pageout.
144 *
145 * In addition, the structure contains the object
146 * and offset to which this page belongs (for pageout),
147 * and sundry status bits.
148 *
149 * Fields in this structure are locked either by the lock on the
150 * object that the page belongs to (O) or by the lock on the page
151 * queues (P). [Some fields require that both locks be held to
152 * change that field; holding either lock is sufficient to read.]
153 */
154
155 #define VM_PAGE_NULL ((vm_page_t) 0)
156
157 extern char vm_page_inactive_states[];
158 extern char vm_page_pageable_states[];
159 extern char vm_page_non_speculative_pageable_states[];
160 extern char vm_page_active_or_inactive_states[];
161
162
163 #define VM_PAGE_INACTIVE(m) (vm_page_inactive_states[m->vmp_q_state])
164 #define VM_PAGE_PAGEABLE(m) (vm_page_pageable_states[m->vmp_q_state])
165 #define VM_PAGE_NON_SPECULATIVE_PAGEABLE(m) (vm_page_non_speculative_pageable_states[m->vmp_q_state])
166 #define VM_PAGE_ACTIVE_OR_INACTIVE(m) (vm_page_active_or_inactive_states[m->vmp_q_state])
167
168
169 #define VM_PAGE_NOT_ON_Q 0 /* page is not present on any queue, nor is it wired... mainly a transient state */
170 #define VM_PAGE_IS_WIRED 1 /* page is currently wired */
171 #define VM_PAGE_USED_BY_COMPRESSOR 2 /* page is in use by the compressor to hold compressed data */
172 #define VM_PAGE_ON_FREE_Q 3 /* page is on the main free queue */
173 #define VM_PAGE_ON_FREE_LOCAL_Q 4 /* page is on one of the per-CPU free queues */
174 #define VM_PAGE_ON_FREE_LOPAGE_Q 5 /* page is on the lopage pool free list */
175 #define VM_PAGE_ON_THROTTLED_Q 6 /* page is on the throttled queue... we stash anonymous pages here when not paging */
176 #define VM_PAGE_ON_PAGEOUT_Q 7 /* page is on one of the pageout queues (internal/external) awaiting processing */
177 #define VM_PAGE_ON_SPECULATIVE_Q 8 /* page is on one of the speculative queues */
178 #define VM_PAGE_ON_ACTIVE_LOCAL_Q 9 /* page has recently been created and is being held in one of the per-CPU local queues */
179 #define VM_PAGE_ON_ACTIVE_Q 10 /* page is in global active queue */
180 #define VM_PAGE_ON_INACTIVE_INTERNAL_Q 11 /* page is on the inactive internal queue a.k.a. anonymous queue */
181 #define VM_PAGE_ON_INACTIVE_EXTERNAL_Q 12 /* page in on the inactive external queue a.k.a. file backed queue */
182 #define VM_PAGE_ON_INACTIVE_CLEANED_Q 13 /* page has been cleaned to a backing file and is ready to be stolen */
183 #define VM_PAGE_ON_SECLUDED_Q 14 /* page is on secluded queue */
184 #define VM_PAGE_Q_STATE_LAST_VALID_VALUE 14 /* we currently use 4 bits for the state... don't let this go beyond 15 */
185
186 #define VM_PAGE_Q_STATE_ARRAY_SIZE (VM_PAGE_Q_STATE_LAST_VALID_VALUE+1)
187
188
189 /*
190 * The structure itself. See the block comment above for what (O) and (P) mean.
191 */
192 #define vmp_pageq vmp_q_un.vmp_q_pageq
193 #define vmp_snext vmp_q_un.vmp_q_snext
194
195 struct vm_page {
196 union {
197 vm_page_queue_chain_t vmp_q_pageq; /* queue info for FIFO queue or free list (P) */
198 struct vm_page *vmp_q_snext;
199 } vmp_q_un;
200
201 vm_page_queue_chain_t vmp_listq; /* all pages in same object (O) */
202
203 vm_page_queue_chain_t vmp_specialq; /* anonymous pages in the special queues (P) */
204 vm_object_offset_t vmp_offset; /* offset into that object (O,P) */
205
206 vm_page_object_t vmp_object; /* which object am I in (O&P) */
207
208 /*
209 * The following word of flags used to be protected by the "page queues" lock.
210 * That's no longer true and what lock, if any, is needed may depend on the
211 * value of vmp_q_state.
212 *
213 * We use 'vmp_wire_count' to store the local queue id if local queues are enabled.
214 * See the comments at 'vm_page_queues_remove' as to why this is safe to do.
215 */
216 #define VM_PAGE_SPECIAL_Q_EMPTY (0)
217 #define VM_PAGE_SPECIAL_Q_BG (1)
218 #define VM_PAGE_SPECIAL_Q_DONATE (2)
219 #define VM_PAGE_SPECIAL_Q_FG (3)
220 #define vmp_local_id vmp_wire_count
221 unsigned int vmp_wire_count:16, /* how many wired down maps use me? (O&P) */
222 vmp_q_state:4, /* which q is the page on (P) */
223 vmp_on_specialq:2,
224 vmp_gobbled:1, /* page used internally (P) */
225 vmp_laundry:1, /* page is being cleaned now (P)*/
226 vmp_no_cache:1, /* page is not to be cached and should */
227 /* be reused ahead of other pages (P) */
228 vmp_private:1, /* Page should not be returned to the free list (P) */
229 vmp_reference:1, /* page has been used (P) */
230 vmp_lopage:1,
231 vmp_unused_page_bits:4;
232
233 /*
234 * MUST keep the 2 32 bit words used as bit fields
235 * separated since the compiler has a nasty habit
236 * of using 64 bit loads and stores on them as
237 * if they were a single 64 bit field... since
238 * they are protected by 2 different locks, this
239 * is a real problem
240 */
241 vm_page_packed_t vmp_next_m; /* VP bucket link (O) */
242
243 /*
244 * The following word of flags is protected by the "VM object" lock.
245 *
246 * IMPORTANT: the "vmp_pmapped", "vmp_xpmapped" and "vmp_clustered" bits can be modified while holding the
247 * VM object "shared" lock + the page lock provided through the pmap_lock_phys_page function.
248 * This is done in vm_fault_enter() and the CONSUME_CLUSTERED macro.
249 * It's also ok to modify them behind just the VM object "exclusive" lock.
250 */
251 unsigned int vmp_busy:1, /* page is in transit (O) */
252 vmp_wanted:1, /* someone is waiting for page (O) */
253 vmp_tabled:1, /* page is in VP table (O) */
254 vmp_hashed:1, /* page is in vm_page_buckets[] (O) + the bucket lock */
255 vmp_fictitious:1, /* Physical page doesn't exist (O) */
256 vmp_clustered:1, /* page is not the faulted page (O) or (O-shared AND pmap_page) */
257 vmp_pmapped:1, /* page has at some time been entered into a pmap (O) or */
258 /* (O-shared AND pmap_page) */
259 vmp_xpmapped:1, /* page has been entered with execute permission (O) or */
260 /* (O-shared AND pmap_page) */
261 vmp_wpmapped:1, /* page has been entered at some point into a pmap for write (O) */
262 vmp_free_when_done:1, /* page is to be freed once cleaning is completed (O) */
263 vmp_absent:1, /* Data has been requested, but is not yet available (O) */
264 vmp_error:1, /* Data manager was unable to provide data due to error (O) */
265 vmp_dirty:1, /* Page must be cleaned (O) */
266 vmp_cleaning:1, /* Page clean has begun (O) */
267 vmp_precious:1, /* Page is precious; data must be returned even if clean (O) */
268 vmp_overwriting:1, /* Request to unlock has been made without having data. (O) */
269 /* [See vm_fault_page_overwrite] */
270 vmp_restart:1, /* Page was pushed higher in shadow chain by copy_call-related pagers */
271 /* start again at top of chain */
272 vmp_unusual:1, /* Page is absent, error, restart or page locked */
273 vmp_cs_validated:VMP_CS_BITS, /* code-signing: page was checked */
274 vmp_cs_tainted:VMP_CS_BITS, /* code-signing: page is tainted */
275 vmp_cs_nx:VMP_CS_BITS, /* code-signing: page is nx */
276 vmp_reusable:1,
277 vmp_written_by_kernel:1; /* page was written by kernel (i.e. decompressed) */
278
279 #if !defined(__arm64__)
280 ppnum_t vmp_phys_page; /* Physical page number of the page */
281 #endif
282 };
283
284 typedef struct vm_page *vm_page_t;
285 extern vm_page_t vm_pages;
286 extern vm_page_t vm_page_array_beginning_addr;
287 extern vm_page_t vm_page_array_ending_addr;
288
289 static inline int
VMP_CS_FOR_OFFSET(vm_map_offset_t fault_phys_offset)290 VMP_CS_FOR_OFFSET(
291 vm_map_offset_t fault_phys_offset)
292 {
293 assertf(fault_phys_offset < PAGE_SIZE &&
294 !(fault_phys_offset & FOURK_PAGE_MASK),
295 "offset 0x%llx\n", (uint64_t)fault_phys_offset);
296 return 1 << (fault_phys_offset >> FOURK_PAGE_SHIFT);
297 }
298 static inline bool
VMP_CS_VALIDATED(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset)299 VMP_CS_VALIDATED(
300 vm_page_t p,
301 vm_map_size_t fault_page_size,
302 vm_map_offset_t fault_phys_offset)
303 {
304 assertf(fault_page_size <= PAGE_SIZE,
305 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
306 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
307 if (fault_page_size == PAGE_SIZE) {
308 return p->vmp_cs_validated == VMP_CS_ALL_TRUE;
309 }
310 return p->vmp_cs_validated & VMP_CS_FOR_OFFSET(fault_phys_offset);
311 }
312 static inline bool
VMP_CS_TAINTED(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset)313 VMP_CS_TAINTED(
314 vm_page_t p,
315 vm_map_size_t fault_page_size,
316 vm_map_offset_t fault_phys_offset)
317 {
318 assertf(fault_page_size <= PAGE_SIZE,
319 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
320 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
321 if (fault_page_size == PAGE_SIZE) {
322 return p->vmp_cs_tainted != VMP_CS_ALL_FALSE;
323 }
324 return p->vmp_cs_tainted & VMP_CS_FOR_OFFSET(fault_phys_offset);
325 }
326 static inline bool
VMP_CS_NX(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset)327 VMP_CS_NX(
328 vm_page_t p,
329 vm_map_size_t fault_page_size,
330 vm_map_offset_t fault_phys_offset)
331 {
332 assertf(fault_page_size <= PAGE_SIZE,
333 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
334 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
335 if (fault_page_size == PAGE_SIZE) {
336 return p->vmp_cs_nx != VMP_CS_ALL_FALSE;
337 }
338 return p->vmp_cs_nx & VMP_CS_FOR_OFFSET(fault_phys_offset);
339 }
340 static inline void
VMP_CS_SET_VALIDATED(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset,boolean_t value)341 VMP_CS_SET_VALIDATED(
342 vm_page_t p,
343 vm_map_size_t fault_page_size,
344 vm_map_offset_t fault_phys_offset,
345 boolean_t value)
346 {
347 assertf(fault_page_size <= PAGE_SIZE,
348 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
349 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
350 if (value) {
351 if (fault_page_size == PAGE_SIZE) {
352 p->vmp_cs_validated = VMP_CS_ALL_TRUE;
353 }
354 p->vmp_cs_validated |= VMP_CS_FOR_OFFSET(fault_phys_offset);
355 } else {
356 if (fault_page_size == PAGE_SIZE) {
357 p->vmp_cs_validated = VMP_CS_ALL_FALSE;
358 }
359 p->vmp_cs_validated &= ~VMP_CS_FOR_OFFSET(fault_phys_offset);
360 }
361 }
362 static inline void
VMP_CS_SET_TAINTED(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset,boolean_t value)363 VMP_CS_SET_TAINTED(
364 vm_page_t p,
365 vm_map_size_t fault_page_size,
366 vm_map_offset_t fault_phys_offset,
367 boolean_t value)
368 {
369 assertf(fault_page_size <= PAGE_SIZE,
370 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
371 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
372 if (value) {
373 if (fault_page_size == PAGE_SIZE) {
374 p->vmp_cs_tainted = VMP_CS_ALL_TRUE;
375 }
376 p->vmp_cs_tainted |= VMP_CS_FOR_OFFSET(fault_phys_offset);
377 } else {
378 if (fault_page_size == PAGE_SIZE) {
379 p->vmp_cs_tainted = VMP_CS_ALL_FALSE;
380 }
381 p->vmp_cs_tainted &= ~VMP_CS_FOR_OFFSET(fault_phys_offset);
382 }
383 }
384 static inline void
VMP_CS_SET_NX(vm_page_t p,vm_map_size_t fault_page_size,vm_map_offset_t fault_phys_offset,boolean_t value)385 VMP_CS_SET_NX(
386 vm_page_t p,
387 vm_map_size_t fault_page_size,
388 vm_map_offset_t fault_phys_offset,
389 boolean_t value)
390 {
391 assertf(fault_page_size <= PAGE_SIZE,
392 "fault_page_size 0x%llx fault_phys_offset 0x%llx\n",
393 (uint64_t)fault_page_size, (uint64_t)fault_phys_offset);
394 if (value) {
395 if (fault_page_size == PAGE_SIZE) {
396 p->vmp_cs_nx = VMP_CS_ALL_TRUE;
397 }
398 p->vmp_cs_nx |= VMP_CS_FOR_OFFSET(fault_phys_offset);
399 } else {
400 if (fault_page_size == PAGE_SIZE) {
401 p->vmp_cs_nx = VMP_CS_ALL_FALSE;
402 }
403 p->vmp_cs_nx &= ~VMP_CS_FOR_OFFSET(fault_phys_offset);
404 }
405 }
406
407
408 #if defined(__arm64__)
409
410 extern unsigned int vm_first_phys_ppnum;
411
412 struct vm_page_with_ppnum {
413 struct vm_page vm_page_wo_ppnum;
414
415 ppnum_t vmp_phys_page;
416 };
417 typedef struct vm_page_with_ppnum *vm_page_with_ppnum_t;
418
419
420 static inline ppnum_t
VM_PAGE_GET_PHYS_PAGE(vm_page_t m)421 VM_PAGE_GET_PHYS_PAGE(vm_page_t m)
422 {
423 if (m >= vm_page_array_beginning_addr && m < vm_page_array_ending_addr) {
424 return (ppnum_t)((uintptr_t)(m - vm_page_array_beginning_addr) + vm_first_phys_ppnum);
425 } else {
426 return ((vm_page_with_ppnum_t)m)->vmp_phys_page;
427 }
428 }
429
430 #define VM_PAGE_SET_PHYS_PAGE(m, ppnum) \
431 MACRO_BEGIN \
432 if ((m) < vm_page_array_beginning_addr || (m) >= vm_page_array_ending_addr) \
433 ((vm_page_with_ppnum_t)(m))->vmp_phys_page = ppnum; \
434 assert(ppnum == VM_PAGE_GET_PHYS_PAGE(m)); \
435 MACRO_END
436
437 #define VM_PAGE_GET_COLOR(m) (VM_PAGE_GET_PHYS_PAGE(m) & vm_color_mask)
438
439 #else /* defined(__arm64__) */
440
441
442 struct vm_page_with_ppnum {
443 struct vm_page vm_page_with_ppnum;
444 };
445 typedef struct vm_page_with_ppnum *vm_page_with_ppnum_t;
446
447
448 #define VM_PAGE_GET_PHYS_PAGE(page) (page)->vmp_phys_page
449 #define VM_PAGE_SET_PHYS_PAGE(page, ppnum) \
450 MACRO_BEGIN \
451 (page)->vmp_phys_page = ppnum; \
452 MACRO_END
453
454 #define VM_PAGE_GET_CLUMP(m) ((VM_PAGE_GET_PHYS_PAGE(m)) >> vm_clump_shift)
455 #define VM_PAGE_GET_COLOR(m) ((VM_PAGE_GET_CLUMP(m)) & vm_color_mask)
456
457 #endif /* defined(__arm64__) */
458
459
460
461 #if defined(__LP64__)
462 /*
463 * Parameters for pointer packing
464 *
465 *
466 * VM Pages pointers might point to:
467 *
468 * 1. VM_PAGE_PACKED_ALIGNED aligned kernel globals,
469 *
470 * 2. VM_PAGE_PACKED_ALIGNED aligned heap allocated vm pages
471 *
472 * 3. entries in the vm_pages array (whose entries aren't VM_PAGE_PACKED_ALIGNED
473 * aligned).
474 *
475 *
476 * The current scheme uses 31 bits of storage and 6 bits of shift using the
477 * VM_PACK_POINTER() scheme for (1-2), and packs (3) as an index within the
478 * vm_pages array, setting the top bit (VM_PAGE_PACKED_FROM_ARRAY).
479 *
480 * This scheme gives us a reach of 128G from VM_MIN_KERNEL_AND_KEXT_ADDRESS.
481 */
482 #define VM_VPLQ_ALIGNMENT 128
483 #define VM_PAGE_PACKED_PTR_ALIGNMENT 64 /* must be a power of 2 */
484 #define VM_PAGE_PACKED_ALIGNED __attribute__((aligned(VM_PAGE_PACKED_PTR_ALIGNMENT)))
485 #define VM_PAGE_PACKED_PTR_BITS 31
486 #define VM_PAGE_PACKED_PTR_SHIFT 6
487 #define VM_PAGE_PACKED_PTR_BASE ((uintptr_t)VM_MIN_KERNEL_AND_KEXT_ADDRESS)
488
489 #define VM_PAGE_PACKED_FROM_ARRAY 0x80000000
490
491 static inline vm_page_packed_t
vm_page_pack_ptr(uintptr_t p)492 vm_page_pack_ptr(uintptr_t p)
493 {
494 if (p >= (uintptr_t)vm_page_array_beginning_addr &&
495 p < (uintptr_t)vm_page_array_ending_addr) {
496 ptrdiff_t diff = (vm_page_t)p - vm_page_array_beginning_addr;
497 assert((vm_page_t)p == &vm_pages[diff]);
498 return (vm_page_packed_t)(diff | VM_PAGE_PACKED_FROM_ARRAY);
499 }
500
501 VM_ASSERT_POINTER_PACKABLE(p, VM_PAGE_PACKED_PTR);
502 vm_offset_t packed = VM_PACK_POINTER(p, VM_PAGE_PACKED_PTR);
503 return CAST_DOWN_EXPLICIT(vm_page_packed_t, packed);
504 }
505
506
507 static inline uintptr_t
vm_page_unpack_ptr(uintptr_t p)508 vm_page_unpack_ptr(uintptr_t p)
509 {
510 extern unsigned int vm_pages_count;
511
512 if (p >= VM_PAGE_PACKED_FROM_ARRAY) {
513 p &= ~VM_PAGE_PACKED_FROM_ARRAY;
514 assert(p < (uintptr_t)vm_pages_count);
515 return (uintptr_t)&vm_pages[p];
516 }
517
518 return VM_UNPACK_POINTER(p, VM_PAGE_PACKED_PTR);
519 }
520
521
522 #define VM_PAGE_PACK_PTR(p) vm_page_pack_ptr((uintptr_t)(p))
523 #define VM_PAGE_UNPACK_PTR(p) vm_page_unpack_ptr((uintptr_t)(p))
524
525 #define VM_OBJECT_PACK(o) ((vm_page_object_t)VM_PACK_POINTER((uintptr_t)(o), VM_PAGE_PACKED_PTR))
526 #define VM_OBJECT_UNPACK(p) ((vm_object_t)VM_UNPACK_POINTER(p, VM_PAGE_PACKED_PTR))
527
528 #define VM_PAGE_OBJECT(p) VM_OBJECT_UNPACK((p)->vmp_object)
529 #define VM_PAGE_PACK_OBJECT(o) VM_OBJECT_PACK(o)
530
531
532 #define VM_PAGE_ZERO_PAGEQ_ENTRY(p) \
533 MACRO_BEGIN \
534 (p)->vmp_snext = 0; \
535 MACRO_END
536
537
538 #define VM_PAGE_CONVERT_TO_QUEUE_ENTRY(p) VM_PAGE_PACK_PTR(p)
539
540
541 static __inline__ void
vm_page_enqueue_tail(vm_page_queue_t que,vm_page_queue_entry_t elt)542 vm_page_enqueue_tail(
543 vm_page_queue_t que,
544 vm_page_queue_entry_t elt)
545 {
546 vm_page_queue_entry_t old_tail;
547
548 old_tail = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(que->prev);
549 elt->next = VM_PAGE_PACK_PTR(que);
550 elt->prev = que->prev;
551 que->prev = old_tail->next = VM_PAGE_PACK_PTR(elt);
552 }
553
554
555 static __inline__ void
vm_page_remque(vm_page_queue_entry_t elt)556 vm_page_remque(
557 vm_page_queue_entry_t elt)
558 {
559 vm_page_queue_entry_t next;
560 vm_page_queue_entry_t prev;
561 vm_page_packed_t next_pck = elt->next;
562 vm_page_packed_t prev_pck = elt->prev;
563
564 next = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(next_pck);
565
566 /* next may equal prev (and the queue head) if elt was the only element */
567 prev = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(prev_pck);
568
569 next->prev = prev_pck;
570 prev->next = next_pck;
571
572 elt->next = 0;
573 elt->prev = 0;
574 }
575
576
577 /*
578 * Macro: vm_page_queue_init
579 * Function:
580 * Initialize the given queue.
581 * Header:
582 * void vm_page_queue_init(q)
583 * vm_page_queue_t q; \* MODIFIED *\
584 */
585 #define vm_page_queue_init(q) \
586 MACRO_BEGIN \
587 VM_ASSERT_POINTER_PACKABLE((vm_offset_t)(q), VM_PAGE_PACKED_PTR); \
588 (q)->next = VM_PAGE_PACK_PTR(q); \
589 (q)->prev = VM_PAGE_PACK_PTR(q); \
590 MACRO_END
591
592
593 /*
594 * Macro: vm_page_queue_enter
595 * Function:
596 * Insert a new element at the tail of the vm_page queue.
597 * Header:
598 * void vm_page_queue_enter(q, elt, field)
599 * queue_t q;
600 * vm_page_t elt;
601 * <field> is the list field in vm_page_t
602 *
603 * This macro's arguments have to match the generic "queue_enter()" macro which is
604 * what is used for this on 32 bit kernels.
605 */
606 #define vm_page_queue_enter(head, elt, field) \
607 MACRO_BEGIN \
608 vm_page_packed_t __pck_elt = VM_PAGE_PACK_PTR(elt); \
609 vm_page_packed_t __pck_head = VM_PAGE_PACK_PTR(head); \
610 vm_page_packed_t __pck_prev = (head)->prev; \
611 \
612 if (__pck_head == __pck_prev) { \
613 (head)->next = __pck_elt; \
614 } else { \
615 vm_page_t __prev; \
616 __prev = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_prev); \
617 __prev->field.next = __pck_elt; \
618 } \
619 (elt)->field.prev = __pck_prev; \
620 (elt)->field.next = __pck_head; \
621 (head)->prev = __pck_elt; \
622 MACRO_END
623
624
625 #if defined(__x86_64__)
626 /*
627 * These are helper macros for vm_page_queue_enter_clump to assist
628 * with conditional compilation (release / debug / development)
629 */
630 #if DEVELOPMENT || DEBUG
631
632 #define __DEBUG_CHECK_BUDDIES(__prev, __p, field) \
633 MACRO_BEGIN \
634 if (__prev != NULL) { \
635 assert(__p == (vm_page_t)VM_PAGE_UNPACK_PTR(__prev->next)); \
636 assert(__prev == (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(__p->field.prev)); \
637 } \
638 MACRO_END
639
640 #define __DEBUG_VERIFY_LINKS(__first, __n_free, __last_next) \
641 MACRO_BEGIN \
642 unsigned int __i; \
643 vm_page_queue_entry_t __tmp; \
644 for (__i = 0, __tmp = __first; __i < __n_free; __i++) { \
645 __tmp = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(__tmp->next); \
646 } \
647 assert(__tmp == __last_next); \
648 MACRO_END
649
650 #define __DEBUG_STAT_INCREMENT_INRANGE vm_clump_inrange++
651 #define __DEBUG_STAT_INCREMENT_INSERTS vm_clump_inserts++
652 #define __DEBUG_STAT_INCREMENT_PROMOTES(__n_free) vm_clump_promotes+=__n_free
653
654 #else
655
656 #define __DEBUG_CHECK_BUDDIES(__prev, __p, field)
657 #define __DEBUG_VERIFY_LINKS(__first, __n_free, __last_next)
658 #define __DEBUG_STAT_INCREMENT_INRANGE
659 #define __DEBUG_STAT_INCREMENT_INSERTS
660 #define __DEBUG_STAT_INCREMENT_PROMOTES(__n_free)
661
662 #endif /* if DEVELOPMENT || DEBUG */
663
664 /*
665 * Insert a new page into a free queue and clump pages within the same 16K boundary together
666 */
667 static inline void
vm_page_queue_enter_clump(vm_page_queue_t head,vm_page_t elt)668 vm_page_queue_enter_clump(
669 vm_page_queue_t head,
670 vm_page_t elt)
671 {
672 vm_page_queue_entry_t first = NULL; /* first page in the clump */
673 vm_page_queue_entry_t last = NULL; /* last page in the clump */
674 vm_page_queue_entry_t prev = NULL;
675 vm_page_queue_entry_t next;
676 uint_t n_free = 1;
677 extern unsigned int vm_pages_count;
678 extern unsigned int vm_clump_size, vm_clump_mask, vm_clump_shift, vm_clump_promote_threshold;
679 extern unsigned long vm_clump_allocs, vm_clump_inserts, vm_clump_inrange, vm_clump_promotes;
680
681 /*
682 * If elt is part of the vm_pages[] array, find its neighboring buddies in the array.
683 */
684 if (vm_page_array_beginning_addr <= elt && elt < &vm_pages[vm_pages_count]) {
685 vm_page_t p;
686 uint_t i;
687 uint_t n;
688 ppnum_t clump_num;
689
690 first = last = (vm_page_queue_entry_t)elt;
691 clump_num = VM_PAGE_GET_CLUMP(elt);
692 n = VM_PAGE_GET_PHYS_PAGE(elt) & vm_clump_mask;
693
694 /*
695 * Check for preceeding vm_pages[] entries in the same chunk
696 */
697 for (i = 0, p = elt - 1; i < n && vm_page_array_beginning_addr <= p; i++, p--) {
698 if (p->vmp_q_state == VM_PAGE_ON_FREE_Q && clump_num == VM_PAGE_GET_CLUMP(p)) {
699 if (prev == NULL) {
700 prev = (vm_page_queue_entry_t)p;
701 }
702 first = (vm_page_queue_entry_t)p;
703 n_free++;
704 }
705 }
706
707 /*
708 * Check the following vm_pages[] entries in the same chunk
709 */
710 for (i = n + 1, p = elt + 1; i < vm_clump_size && p < &vm_pages[vm_pages_count]; i++, p++) {
711 if (p->vmp_q_state == VM_PAGE_ON_FREE_Q && clump_num == VM_PAGE_GET_CLUMP(p)) {
712 if (last == (vm_page_queue_entry_t)elt) { /* first one only */
713 __DEBUG_CHECK_BUDDIES(prev, p, vmp_pageq);
714 }
715
716 if (prev == NULL) {
717 prev = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(p->vmp_pageq.prev);
718 }
719 last = (vm_page_queue_entry_t)p;
720 n_free++;
721 }
722 }
723 __DEBUG_STAT_INCREMENT_INRANGE;
724 }
725
726 /* if elt is not part of vm_pages or if 1st page in clump, insert at tail */
727 if (prev == NULL) {
728 prev = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(head->prev);
729 }
730
731 /* insert the element */
732 next = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(prev->next);
733 elt->vmp_pageq.next = prev->next;
734 elt->vmp_pageq.prev = next->prev;
735 prev->next = next->prev = VM_PAGE_PACK_PTR(elt);
736 __DEBUG_STAT_INCREMENT_INSERTS;
737
738 /*
739 * Check if clump needs to be promoted to head.
740 */
741 if (n_free >= vm_clump_promote_threshold && n_free > 1) {
742 vm_page_queue_entry_t first_prev;
743
744 first_prev = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(first->prev);
745
746 /* If not at head already */
747 if (first_prev != head) {
748 vm_page_queue_entry_t last_next;
749 vm_page_queue_entry_t head_next;
750
751 last_next = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(last->next);
752
753 /* verify that the links within the clump are consistent */
754 __DEBUG_VERIFY_LINKS(first, n_free, last_next);
755
756 /* promote clump to head */
757 first_prev->next = last->next;
758 last_next->prev = first->prev;
759 first->prev = VM_PAGE_PACK_PTR(head);
760 last->next = head->next;
761
762 head_next = (vm_page_queue_entry_t)VM_PAGE_UNPACK_PTR(head->next);
763 head_next->prev = VM_PAGE_PACK_PTR(last);
764 head->next = VM_PAGE_PACK_PTR(first);
765 __DEBUG_STAT_INCREMENT_PROMOTES(n_free);
766 }
767 }
768 }
769 #endif
770
771 /*
772 * Macro: vm_page_queue_enter_first
773 * Function:
774 * Insert a new element at the head of the vm_page queue.
775 * Header:
776 * void queue_enter_first(q, elt, , field)
777 * queue_t q;
778 * vm_page_t elt;
779 * <field> is the linkage field in vm_page
780 *
781 * This macro's arguments have to match the generic "queue_enter_first()" macro which is
782 * what is used for this on 32 bit kernels.
783 */
784 #define vm_page_queue_enter_first(head, elt, field) \
785 MACRO_BEGIN \
786 vm_page_packed_t __pck_next = (head)->next; \
787 vm_page_packed_t __pck_head = VM_PAGE_PACK_PTR(head); \
788 vm_page_packed_t __pck_elt = VM_PAGE_PACK_PTR(elt); \
789 \
790 if (__pck_head == __pck_next) { \
791 (head)->prev = __pck_elt; \
792 } else { \
793 vm_page_t __next; \
794 __next = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_next); \
795 __next->field.prev = __pck_elt; \
796 } \
797 \
798 (elt)->field.next = __pck_next; \
799 (elt)->field.prev = __pck_head; \
800 (head)->next = __pck_elt; \
801 MACRO_END
802
803
804 /*
805 * Macro: vm_page_queue_remove
806 * Function:
807 * Remove an arbitrary page from a vm_page queue.
808 * Header:
809 * void vm_page_queue_remove(q, qe, field)
810 * arguments as in vm_page_queue_enter
811 *
812 * This macro's arguments have to match the generic "queue_enter()" macro which is
813 * what is used for this on 32 bit kernels.
814 */
815 #define vm_page_queue_remove(head, elt, field) \
816 MACRO_BEGIN \
817 vm_page_packed_t __pck_next = (elt)->field.next; \
818 vm_page_packed_t __pck_prev = (elt)->field.prev; \
819 vm_page_t __next = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_next); \
820 vm_page_t __prev = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_prev); \
821 \
822 if ((void *)(head) == (void *)__next) { \
823 (head)->prev = __pck_prev; \
824 } else { \
825 __next->field.prev = __pck_prev; \
826 } \
827 \
828 if ((void *)(head) == (void *)__prev) { \
829 (head)->next = __pck_next; \
830 } else { \
831 __prev->field.next = __pck_next; \
832 } \
833 \
834 (elt)->field.next = 0; \
835 (elt)->field.prev = 0; \
836 MACRO_END
837
838
839 /*
840 * Macro: vm_page_queue_remove_first
841 *
842 * Function:
843 * Remove and return the entry at the head of a vm_page queue.
844 *
845 * Header:
846 * vm_page_queue_remove_first(head, entry, field)
847 * N.B. entry is returned by reference
848 *
849 * This macro's arguments have to match the generic "queue_remove_first()" macro which is
850 * what is used for this on 32 bit kernels.
851 */
852 #define vm_page_queue_remove_first(head, entry, field) \
853 MACRO_BEGIN \
854 vm_page_packed_t __pck_head = VM_PAGE_PACK_PTR(head); \
855 vm_page_packed_t __pck_next; \
856 vm_page_t __next; \
857 \
858 (entry) = (vm_page_t)VM_PAGE_UNPACK_PTR((head)->next); \
859 __pck_next = (entry)->field.next; \
860 __next = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_next); \
861 \
862 if (__pck_head == __pck_next) { \
863 (head)->prev = __pck_head; \
864 } else { \
865 __next->field.prev = __pck_head; \
866 } \
867 \
868 (head)->next = __pck_next; \
869 (entry)->field.next = 0; \
870 (entry)->field.prev = 0; \
871 MACRO_END
872
873
874 #if defined(__x86_64__)
875 /*
876 * Macro: vm_page_queue_remove_first_with_clump
877 * Function:
878 * Remove and return the entry at the head of the free queue
879 * end is set to 1 to indicate that we just returned the last page in a clump
880 *
881 * Header:
882 * vm_page_queue_remove_first_with_clump(head, entry, end)
883 * entry is returned by reference
884 * end is returned by reference
885 */
886 #define vm_page_queue_remove_first_with_clump(head, entry, end) \
887 MACRO_BEGIN \
888 vm_page_packed_t __pck_head = VM_PAGE_PACK_PTR(head); \
889 vm_page_packed_t __pck_next; \
890 vm_page_t __next; \
891 \
892 (entry) = (vm_page_t)VM_PAGE_UNPACK_PTR((head)->next); \
893 __pck_next = (entry)->vmp_pageq.next; \
894 __next = (vm_page_t)VM_PAGE_UNPACK_PTR(__pck_next); \
895 \
896 (end) = 0; \
897 if (__pck_head == __pck_next) { \
898 (head)->prev = __pck_head; \
899 (end) = 1; \
900 } else { \
901 __next->vmp_pageq.prev = __pck_head; \
902 if (VM_PAGE_GET_CLUMP(entry) != VM_PAGE_GET_CLUMP(__next)) { \
903 (end) = 1; \
904 } \
905 } \
906 \
907 (head)->next = __pck_next; \
908 (entry)->vmp_pageq.next = 0; \
909 (entry)->vmp_pageq.prev = 0; \
910 MACRO_END
911 #endif
912
913 /*
914 * Macro: vm_page_queue_end
915 * Function:
916 * Tests whether a new entry is really the end of
917 * the queue.
918 * Header:
919 * boolean_t vm_page_queue_end(q, qe)
920 * vm_page_queue_t q;
921 * vm_page_queue_entry_t qe;
922 */
923 #define vm_page_queue_end(q, qe) ((q) == (qe))
924
925
926 /*
927 * Macro: vm_page_queue_empty
928 * Function:
929 * Tests whether a queue is empty.
930 * Header:
931 * boolean_t vm_page_queue_empty(q)
932 * vm_page_queue_t q;
933 */
934 #define vm_page_queue_empty(q) vm_page_queue_end((q), ((vm_page_queue_entry_t)vm_page_queue_first(q)))
935
936
937
938 /*
939 * Macro: vm_page_queue_first
940 * Function:
941 * Returns the first entry in the queue,
942 * Header:
943 * uintpr_t vm_page_queue_first(q)
944 * vm_page_queue_t q; \* IN *\
945 */
946 #define vm_page_queue_first(q) (VM_PAGE_UNPACK_PTR((q)->next))
947
948
949
950 /*
951 * Macro: vm_page_queue_last
952 * Function:
953 * Returns the last entry in the queue.
954 * Header:
955 * vm_page_queue_entry_t queue_last(q)
956 * queue_t q; \* IN *\
957 */
958 #define vm_page_queue_last(q) (VM_PAGE_UNPACK_PTR((q)->prev))
959
960
961
962 /*
963 * Macro: vm_page_queue_next
964 * Function:
965 * Returns the entry after an item in the queue.
966 * Header:
967 * uintpr_t vm_page_queue_next(qc)
968 * vm_page_queue_t qc;
969 */
970 #define vm_page_queue_next(qc) (VM_PAGE_UNPACK_PTR((qc)->next))
971
972
973
974 /*
975 * Macro: vm_page_queue_prev
976 * Function:
977 * Returns the entry before an item in the queue.
978 * Header:
979 * uinptr_t vm_page_queue_prev(qc)
980 * vm_page_queue_t qc;
981 */
982 #define vm_page_queue_prev(qc) (VM_PAGE_UNPACK_PTR((qc)->prev))
983
984
985
986 /*
987 * Macro: vm_page_queue_iterate
988 * Function:
989 * iterate over each item in a vm_page queue.
990 * Generates a 'for' loop, setting elt to
991 * each item in turn (by reference).
992 * Header:
993 * vm_page_queue_iterate(q, elt, field)
994 * queue_t q;
995 * vm_page_t elt;
996 * <field> is the chain field in vm_page_t
997 */
998 #define vm_page_queue_iterate(head, elt, field) \
999 for ((elt) = (vm_page_t)vm_page_queue_first(head); \
1000 !vm_page_queue_end((head), (vm_page_queue_entry_t)(elt)); \
1001 (elt) = (vm_page_t)vm_page_queue_next(&(elt)->field)) \
1002
1003 #else // LP64
1004
1005 #define VM_VPLQ_ALIGNMENT 128
1006 #define VM_PAGE_PACKED_PTR_ALIGNMENT sizeof(vm_offset_t)
1007 #define VM_PAGE_PACKED_ALIGNED
1008 #define VM_PAGE_PACKED_PTR_BITS 32
1009 #define VM_PAGE_PACKED_PTR_SHIFT 0
1010 #define VM_PAGE_PACKED_PTR_BASE 0
1011
1012 #define VM_PAGE_PACKED_FROM_ARRAY 0
1013
1014 #define VM_PAGE_PACK_PTR(p) (p)
1015 #define VM_PAGE_UNPACK_PTR(p) ((uintptr_t)(p))
1016
1017 #define VM_OBJECT_PACK(o) ((vm_page_object_t)(o))
1018 #define VM_OBJECT_UNPACK(p) ((vm_object_t)(p))
1019
1020 #define VM_PAGE_PACK_OBJECT(o) VM_OBJECT_PACK(o)
1021 #define VM_PAGE_OBJECT(p) VM_OBJECT_UNPACK((p)->vmp_object)
1022
1023
1024 #define VM_PAGE_ZERO_PAGEQ_ENTRY(p) \
1025 MACRO_BEGIN \
1026 (p)->vmp_pageq.next = 0; \
1027 (p)->vmp_pageq.prev = 0; \
1028 MACRO_END
1029
1030 #define VM_PAGE_CONVERT_TO_QUEUE_ENTRY(p) ((queue_entry_t)(p))
1031
1032 #define vm_page_remque remque
1033 #define vm_page_enqueue_tail enqueue_tail
1034 #define vm_page_queue_init queue_init
1035 #define vm_page_queue_enter(h, e, f) queue_enter(h, e, vm_page_t, f)
1036 #define vm_page_queue_enter_first(h, e, f) queue_enter_first(h, e, vm_page_t, f)
1037 #define vm_page_queue_remove(h, e, f) queue_remove(h, e, vm_page_t, f)
1038 #define vm_page_queue_remove_first(h, e, f) queue_remove_first(h, e, vm_page_t, f)
1039 #define vm_page_queue_end queue_end
1040 #define vm_page_queue_empty queue_empty
1041 #define vm_page_queue_first queue_first
1042 #define vm_page_queue_last queue_last
1043 #define vm_page_queue_next queue_next
1044 #define vm_page_queue_prev queue_prev
1045 #define vm_page_queue_iterate(h, e, f) queue_iterate(h, e, vm_page_t, f)
1046
1047 #endif // __LP64__
1048
1049
1050
1051 /*
1052 * VM_PAGE_MIN_SPECULATIVE_AGE_Q through VM_PAGE_MAX_SPECULATIVE_AGE_Q
1053 * represents a set of aging bins that are 'protected'...
1054 *
1055 * VM_PAGE_SPECULATIVE_AGED_Q is a list of the speculative pages that have
1056 * not yet been 'claimed' but have been aged out of the protective bins
1057 * this occurs in vm_page_speculate when it advances to the next bin
1058 * and discovers that it is still occupied... at that point, all of the
1059 * pages in that bin are moved to the VM_PAGE_SPECULATIVE_AGED_Q. the pages
1060 * in that bin are all guaranteed to have reached at least the maximum age
1061 * we allow for a protected page... they can be older if there is no
1062 * memory pressure to pull them from the bin, or there are no new speculative pages
1063 * being generated to push them out.
1064 * this list is the one that vm_pageout_scan will prefer when looking
1065 * for pages to move to the underweight free list
1066 *
1067 * VM_PAGE_MAX_SPECULATIVE_AGE_Q * VM_PAGE_SPECULATIVE_Q_AGE_MS
1068 * defines the amount of time a speculative page is normally
1069 * allowed to live in the 'protected' state (i.e. not available
1070 * to be stolen if vm_pageout_scan is running and looking for
1071 * pages)... however, if the total number of speculative pages
1072 * in the protected state exceeds our limit (defined in vm_pageout.c)
1073 * and there are none available in VM_PAGE_SPECULATIVE_AGED_Q, then
1074 * vm_pageout_scan is allowed to steal pages from the protected
1075 * bucket even if they are underage.
1076 *
1077 * vm_pageout_scan is also allowed to pull pages from a protected
1078 * bin if the bin has reached the "age of consent" we've set
1079 */
1080 #define VM_PAGE_MAX_SPECULATIVE_AGE_Q 10
1081 #define VM_PAGE_MIN_SPECULATIVE_AGE_Q 1
1082 #define VM_PAGE_SPECULATIVE_AGED_Q 0
1083
1084 #define VM_PAGE_SPECULATIVE_Q_AGE_MS 500
1085
1086 struct vm_speculative_age_q {
1087 /*
1088 * memory queue for speculative pages via clustered pageins
1089 */
1090 vm_page_queue_head_t age_q;
1091 mach_timespec_t age_ts;
1092 } VM_PAGE_PACKED_ALIGNED;
1093
1094
1095
1096 extern
1097 struct vm_speculative_age_q vm_page_queue_speculative[];
1098
1099 extern int speculative_steal_index;
1100 extern int speculative_age_index;
1101 extern unsigned int vm_page_speculative_q_age_ms;
1102
1103
1104 typedef struct vm_locks_array {
1105 char pad __attribute__ ((aligned(64)));
1106 lck_mtx_t vm_page_queue_lock2 __attribute__ ((aligned(64)));
1107 lck_mtx_t vm_page_queue_free_lock2 __attribute__ ((aligned(64)));
1108 char pad2 __attribute__ ((aligned(64)));
1109 } vm_locks_array_t;
1110
1111
1112 extern void vm_page_assign_special_state(vm_page_t mem, int mode);
1113 extern void vm_page_update_special_state(vm_page_t mem);
1114 extern void vm_page_add_to_specialq(vm_page_t mem, boolean_t first);
1115 extern void vm_page_remove_from_specialq(vm_page_t mem);
1116
1117 #define VM_PAGE_WIRED(m) ((m)->vmp_q_state == VM_PAGE_IS_WIRED)
1118 #define NEXT_PAGE(m) ((m)->vmp_snext)
1119 #define NEXT_PAGE_PTR(m) (&(m)->vmp_snext)
1120
1121 /*
1122 * XXX The unusual bit should not be necessary. Most of the bit
1123 * XXX fields above really want to be masks.
1124 */
1125
1126 /*
1127 * For debugging, this macro can be defined to perform
1128 * some useful check on a page structure.
1129 * INTENTIONALLY left as a no-op so that the
1130 * current call-sites can be left intact for future uses.
1131 */
1132
1133 #define VM_PAGE_CHECK(mem) \
1134 MACRO_BEGIN \
1135 MACRO_END
1136
1137 /* Page coloring:
1138 *
1139 * The free page list is actually n lists, one per color,
1140 * where the number of colors is a function of the machine's
1141 * cache geometry set at system initialization. To disable
1142 * coloring, set vm_colors to 1 and vm_color_mask to 0.
1143 * The boot-arg "colors" may be used to override vm_colors.
1144 * Note that there is little harm in having more colors than needed.
1145 */
1146
1147 #define MAX_COLORS 128
1148 #define DEFAULT_COLORS 32
1149
1150 extern
1151 unsigned int vm_colors; /* must be in range 1..MAX_COLORS */
1152 extern
1153 unsigned int vm_color_mask; /* must be (vm_colors-1) */
1154 extern
1155 unsigned int vm_cache_geometry_colors; /* optimal #colors based on cache geometry */
1156
1157 /*
1158 * Wired memory is a very limited resource and we can't let users exhaust it
1159 * and deadlock the entire system. We enforce the following limits:
1160 *
1161 * vm_per_task_user_wire_limit
1162 * how much memory can be user-wired in one user task
1163 *
1164 * vm_global_user_wire_limit (default: same as vm_per_task_user_wire_limit)
1165 * how much memory can be user-wired in all user tasks
1166 *
1167 * These values are set to defaults based on the number of pages managed
1168 * by the VM system. They can be overriden via sysctls.
1169 * See kmem_set_user_wire_limits for details on the default values.
1170 *
1171 * Regardless of the amount of memory in the system, we never reserve
1172 * more than VM_NOT_USER_WIREABLE_MAX bytes as unlockable.
1173 */
1174 #if defined(__LP64__)
1175 #define VM_NOT_USER_WIREABLE_MAX (32ULL*1024*1024*1024) /* 32GB */
1176 #else
1177 #define VM_NOT_USER_WIREABLE_MAX (1UL*1024*1024*1024) /* 1GB */
1178 #endif /* __LP64__ */
1179 extern
1180 vm_map_size_t vm_per_task_user_wire_limit;
1181 extern
1182 vm_map_size_t vm_global_user_wire_limit;
1183 extern
1184 uint64_t vm_add_wire_count_over_global_limit;
1185 extern
1186 uint64_t vm_add_wire_count_over_user_limit;
1187
1188 /*
1189 * Each pageable resident page falls into one of three lists:
1190 *
1191 * free
1192 * Available for allocation now. The free list is
1193 * actually an array of lists, one per color.
1194 * inactive
1195 * Not referenced in any map, but still has an
1196 * object/offset-page mapping, and may be dirty.
1197 * This is the list of pages that should be
1198 * paged out next. There are actually two
1199 * inactive lists, one for pages brought in from
1200 * disk or other backing store, and another
1201 * for "zero-filled" pages. See vm_pageout_scan()
1202 * for the distinction and usage.
1203 * active
1204 * A list of pages which have been placed in
1205 * at least one physical map. This list is
1206 * ordered, in LRU-like fashion.
1207 */
1208
1209
1210 #define VPL_LOCK_SPIN 1
1211
1212 struct vpl {
1213 vm_page_queue_head_t vpl_queue;
1214 unsigned int vpl_count;
1215 unsigned int vpl_internal_count;
1216 unsigned int vpl_external_count;
1217 lck_spin_t vpl_lock;
1218 };
1219
1220 extern
1221 struct vpl * /* __zpercpu */ vm_page_local_q;
1222 extern
1223 unsigned int vm_page_local_q_soft_limit;
1224 extern
1225 unsigned int vm_page_local_q_hard_limit;
1226 extern
1227 vm_locks_array_t vm_page_locks;
1228
1229 extern
1230 vm_page_queue_head_t vm_lopage_queue_free; /* low memory free queue */
1231 extern
1232 vm_page_queue_head_t vm_page_queue_active; /* active memory queue */
1233 extern
1234 vm_page_queue_head_t vm_page_queue_inactive; /* inactive memory queue for normal pages */
1235 #if CONFIG_SECLUDED_MEMORY
1236 extern
1237 vm_page_queue_head_t vm_page_queue_secluded; /* reclaimable pages secluded for Camera */
1238 #endif /* CONFIG_SECLUDED_MEMORY */
1239 extern
1240 vm_page_queue_head_t vm_page_queue_cleaned; /* clean-queue inactive memory */
1241 extern
1242 vm_page_queue_head_t vm_page_queue_anonymous; /* inactive memory queue for anonymous pages */
1243 extern
1244 vm_page_queue_head_t vm_page_queue_throttled; /* memory queue for throttled pageout pages */
1245
1246 extern
1247 queue_head_t vm_objects_wired;
1248 extern
1249 lck_spin_t vm_objects_wired_lock;
1250
1251 #define VM_PAGE_DONATE_DISABLED 0
1252 #define VM_PAGE_DONATE_ENABLED 1
1253 extern
1254 uint32_t vm_page_donate_mode;
1255 extern
1256 bool vm_page_donate_queue_ripe;
1257
1258 #define VM_PAGE_BACKGROUND_TARGET_MAX 50000
1259 #define VM_PAGE_BG_DISABLED 0
1260 #define VM_PAGE_BG_ENABLED 1
1261
1262 extern
1263 vm_page_queue_head_t vm_page_queue_background;
1264 extern
1265 uint64_t vm_page_background_promoted_count;
1266 extern
1267 uint32_t vm_page_background_count;
1268 extern
1269 uint32_t vm_page_background_target;
1270 extern
1271 uint32_t vm_page_background_internal_count;
1272 extern
1273 uint32_t vm_page_background_external_count;
1274 extern
1275 uint32_t vm_page_background_mode;
1276 extern
1277 uint32_t vm_page_background_exclude_external;
1278
1279 extern
1280 vm_page_queue_head_t vm_page_queue_donate;
1281 extern
1282 uint32_t vm_page_donate_count;
1283 extern
1284 uint32_t vm_page_donate_target_low;
1285 extern
1286 uint32_t vm_page_donate_target_high;
1287 #define VM_PAGE_DONATE_TARGET_LOWWATER (100)
1288 #define VM_PAGE_DONATE_TARGET_HIGHWATER ((unsigned int)(atop_64(max_mem) / 8))
1289
1290 extern
1291 vm_offset_t first_phys_addr; /* physical address for first_page */
1292 extern
1293 vm_offset_t last_phys_addr; /* physical address for last_page */
1294
1295 extern
1296 unsigned int vm_page_free_count; /* How many pages are free? (sum of all colors) */
1297 extern
1298 unsigned int vm_page_active_count; /* How many pages are active? */
1299 extern
1300 unsigned int vm_page_inactive_count; /* How many pages are inactive? */
1301 extern
1302 unsigned int vm_page_kernelcache_count; /* How many pages are used for the kernelcache? */
1303 #if CONFIG_SECLUDED_MEMORY
1304 extern
1305 unsigned int vm_page_secluded_count; /* How many pages are secluded? */
1306 extern
1307 unsigned int vm_page_secluded_count_free; /* how many of them are free? */
1308 extern
1309 unsigned int vm_page_secluded_count_inuse; /* how many of them are in use? */
1310 /*
1311 * We keep filling the secluded pool with new eligible pages and
1312 * we can overshoot our target by a lot.
1313 * When there's memory pressure, vm_pageout_scan() will re-balance the queues,
1314 * pushing the extra secluded pages to the active or free queue.
1315 * Since these "over target" secluded pages are actually "available", jetsam
1316 * should consider them as such, so make them visible to jetsam via the
1317 * "vm_page_secluded_count_over_target" counter and update it whenever we
1318 * update vm_page_secluded_count or vm_page_secluded_target.
1319 */
1320 extern
1321 unsigned int vm_page_secluded_count_over_target;
1322 #define VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE() \
1323 MACRO_BEGIN \
1324 if (vm_page_secluded_count > vm_page_secluded_target) { \
1325 vm_page_secluded_count_over_target = \
1326 (vm_page_secluded_count - vm_page_secluded_target); \
1327 } else { \
1328 vm_page_secluded_count_over_target = 0; \
1329 } \
1330 MACRO_END
1331 #define VM_PAGE_SECLUDED_COUNT_OVER_TARGET() vm_page_secluded_count_over_target
1332 #else /* CONFIG_SECLUDED_MEMORY */
1333 #define VM_PAGE_SECLUDED_COUNT_OVER_TARGET_UPDATE() \
1334 MACRO_BEGIN \
1335 MACRO_END
1336 #define VM_PAGE_SECLUDED_COUNT_OVER_TARGET() 0
1337 #endif /* CONFIG_SECLUDED_MEMORY */
1338 extern
1339 unsigned int vm_page_cleaned_count; /* How many pages are in the clean queue? */
1340 extern
1341 unsigned int vm_page_throttled_count;/* How many inactives are throttled */
1342 extern
1343 unsigned int vm_page_speculative_count; /* How many speculative pages are unclaimed? */
1344 extern unsigned int vm_page_pageable_internal_count;
1345 extern unsigned int vm_page_pageable_external_count;
1346 extern
1347 unsigned int vm_page_xpmapped_external_count; /* How many pages are mapped executable? */
1348 extern
1349 unsigned int vm_page_external_count; /* How many pages are file-backed? */
1350 extern
1351 unsigned int vm_page_internal_count; /* How many pages are anonymous? */
1352 extern
1353 unsigned int vm_page_wire_count; /* How many pages are wired? */
1354 extern
1355 unsigned int vm_page_wire_count_initial; /* How many pages wired at startup */
1356 extern
1357 unsigned int vm_page_wire_count_on_boot; /* even earlier than _initial */
1358 extern
1359 unsigned int vm_page_free_target; /* How many do we want free? */
1360 extern
1361 unsigned int vm_page_free_min; /* When to wakeup pageout */
1362 extern
1363 unsigned int vm_page_throttle_limit; /* When to throttle new page creation */
1364 extern
1365 unsigned int vm_page_inactive_target;/* How many do we want inactive? */
1366 #if CONFIG_SECLUDED_MEMORY
1367 extern
1368 unsigned int vm_page_secluded_target;/* How many do we want secluded? */
1369 #endif /* CONFIG_SECLUDED_MEMORY */
1370 extern
1371 unsigned int vm_page_anonymous_min; /* When it's ok to pre-clean */
1372 extern
1373 unsigned int vm_page_free_reserved; /* How many pages reserved to do pageout */
1374 extern
1375 unsigned int vm_page_gobble_count;
1376 extern
1377 unsigned int vm_page_stolen_count; /* Count of stolen pages not acccounted in zones */
1378 extern
1379 unsigned int vm_page_kern_lpage_count; /* Count of large pages used in early boot */
1380
1381
1382 #if DEVELOPMENT || DEBUG
1383 extern
1384 unsigned int vm_page_speculative_used;
1385 #endif
1386
1387 extern
1388 unsigned int vm_page_purgeable_count;/* How many pages are purgeable now ? */
1389 extern
1390 unsigned int vm_page_purgeable_wired_count;/* How many purgeable pages are wired now ? */
1391 extern
1392 uint64_t vm_page_purged_count; /* How many pages got purged so far ? */
1393
1394 extern unsigned int vm_page_free_wanted;
1395 /* how many threads are waiting for memory */
1396
1397 extern unsigned int vm_page_free_wanted_privileged;
1398 /* how many VM privileged threads are waiting for memory */
1399 #if CONFIG_SECLUDED_MEMORY
1400 extern unsigned int vm_page_free_wanted_secluded;
1401 /* how many threads are waiting for secluded memory */
1402 #endif /* CONFIG_SECLUDED_MEMORY */
1403
1404 extern const ppnum_t vm_page_fictitious_addr;
1405 /* (fake) phys_addr of fictitious pages */
1406
1407 extern const ppnum_t vm_page_guard_addr;
1408 /* (fake) phys_addr of guard pages */
1409
1410
1411 extern boolean_t vm_page_deactivate_hint;
1412
1413 extern int vm_compressor_mode;
1414
1415 /*
1416 * Defaults to true, so highest memory is used first.
1417 */
1418 extern boolean_t vm_himemory_mode;
1419
1420 extern boolean_t vm_lopage_needed;
1421 extern uint32_t vm_lopage_free_count;
1422 extern uint32_t vm_lopage_free_limit;
1423 extern uint32_t vm_lopage_lowater;
1424 extern boolean_t vm_lopage_refill;
1425 extern uint64_t max_valid_dma_address;
1426 extern ppnum_t max_valid_low_ppnum;
1427
1428 /*
1429 * Prototypes for functions exported by this module.
1430 */
1431 extern void vm_page_bootstrap(
1432 vm_offset_t *startp,
1433 vm_offset_t *endp);
1434
1435 extern void vm_page_init_local_q(unsigned int num_cpus);
1436
1437 extern void vm_page_create(
1438 ppnum_t start,
1439 ppnum_t end);
1440
1441 extern void vm_page_create_retired(
1442 ppnum_t pn);
1443
1444 extern vm_page_t kdp_vm_page_lookup(
1445 vm_object_t object,
1446 vm_object_offset_t offset);
1447
1448 extern vm_page_t vm_page_lookup(
1449 vm_object_t object,
1450 vm_object_offset_t offset);
1451
1452 extern vm_page_t vm_page_grab_fictitious(boolean_t canwait);
1453
1454 extern vm_page_t vm_page_grab_guard(boolean_t canwait);
1455
1456 extern void vm_page_release_fictitious(
1457 vm_page_t page);
1458
1459 extern void vm_free_delayed_pages(void);
1460
1461 extern bool vm_pool_low(void);
1462
1463 extern vm_page_t vm_page_grab(void);
1464 extern vm_page_t vm_page_grab_options(int flags);
1465
1466 #define VM_PAGE_GRAB_OPTIONS_NONE 0x00000000
1467 #if CONFIG_SECLUDED_MEMORY
1468 #define VM_PAGE_GRAB_SECLUDED 0x00000001
1469 #endif /* CONFIG_SECLUDED_MEMORY */
1470 #define VM_PAGE_GRAB_Q_LOCK_HELD 0x00000002
1471
1472 extern vm_page_t vm_page_grablo(void);
1473
1474 extern void vm_page_release(
1475 vm_page_t page,
1476 boolean_t page_queues_locked);
1477
1478 extern boolean_t vm_page_wait(
1479 int interruptible );
1480
1481 extern vm_page_t vm_page_alloc(
1482 vm_object_t object,
1483 vm_object_offset_t offset);
1484
1485 extern void vm_page_init(
1486 vm_page_t page,
1487 ppnum_t phys_page,
1488 boolean_t lopage);
1489
1490 extern void vm_page_free(
1491 vm_page_t page);
1492
1493 extern void vm_page_free_unlocked(
1494 vm_page_t page,
1495 boolean_t remove_from_hash);
1496
1497 extern void vm_page_balance_inactive(
1498 int max_to_move);
1499
1500 extern void vm_page_activate(
1501 vm_page_t page);
1502
1503 extern void vm_page_deactivate(
1504 vm_page_t page);
1505
1506 extern void vm_page_deactivate_internal(
1507 vm_page_t page,
1508 boolean_t clear_hw_reference);
1509
1510 extern void vm_page_enqueue_cleaned(vm_page_t page);
1511
1512 extern void vm_page_lru(
1513 vm_page_t page);
1514
1515 extern void vm_page_speculate(
1516 vm_page_t page,
1517 boolean_t new);
1518
1519 extern void vm_page_speculate_ageit(
1520 struct vm_speculative_age_q *aq);
1521
1522 extern void vm_page_reactivate_all_throttled(void);
1523
1524 extern void vm_page_reactivate_local(uint32_t lid, boolean_t force, boolean_t nolocks);
1525
1526 extern void vm_page_rename(
1527 vm_page_t page,
1528 vm_object_t new_object,
1529 vm_object_offset_t new_offset);
1530
1531 extern void vm_page_insert(
1532 vm_page_t page,
1533 vm_object_t object,
1534 vm_object_offset_t offset);
1535
1536 extern void vm_page_insert_wired(
1537 vm_page_t page,
1538 vm_object_t object,
1539 vm_object_offset_t offset,
1540 vm_tag_t tag);
1541
1542 extern void vm_page_insert_internal(
1543 vm_page_t page,
1544 vm_object_t object,
1545 vm_object_offset_t offset,
1546 vm_tag_t tag,
1547 boolean_t queues_lock_held,
1548 boolean_t insert_in_hash,
1549 boolean_t batch_pmap_op,
1550 boolean_t delayed_accounting,
1551 uint64_t *delayed_ledger_update);
1552
1553 extern void vm_page_replace(
1554 vm_page_t mem,
1555 vm_object_t object,
1556 vm_object_offset_t offset);
1557
1558 extern void vm_page_remove(
1559 vm_page_t page,
1560 boolean_t remove_from_hash);
1561
1562 extern void vm_page_zero_fill(
1563 vm_page_t page);
1564
1565 extern void vm_page_part_zero_fill(
1566 vm_page_t m,
1567 vm_offset_t m_pa,
1568 vm_size_t len);
1569
1570 extern void vm_page_copy(
1571 vm_page_t src_page,
1572 vm_page_t dest_page);
1573
1574 extern void vm_page_part_copy(
1575 vm_page_t src_m,
1576 vm_offset_t src_pa,
1577 vm_page_t dst_m,
1578 vm_offset_t dst_pa,
1579 vm_size_t len);
1580
1581 extern void vm_page_wire(
1582 vm_page_t page,
1583 vm_tag_t tag,
1584 boolean_t check_memorystatus);
1585
1586 extern void vm_page_unwire(
1587 vm_page_t page,
1588 boolean_t queueit);
1589
1590 extern void vm_set_page_size(void);
1591
1592 extern void vm_page_gobble(
1593 vm_page_t page);
1594
1595 extern void vm_page_validate_cs(
1596 vm_page_t page,
1597 vm_map_size_t fault_page_size,
1598 vm_map_offset_t fault_phys_offset);
1599 extern void vm_page_validate_cs_mapped(
1600 vm_page_t page,
1601 vm_map_size_t fault_page_size,
1602 vm_map_offset_t fault_phys_offset,
1603 const void *kaddr);
1604 extern void vm_page_validate_cs_mapped_slow(
1605 vm_page_t page,
1606 const void *kaddr);
1607 extern void vm_page_validate_cs_mapped_chunk(
1608 vm_page_t page,
1609 const void *kaddr,
1610 vm_offset_t chunk_offset,
1611 vm_size_t chunk_size,
1612 boolean_t *validated,
1613 unsigned *tainted);
1614
1615 extern void vm_page_free_prepare_queues(
1616 vm_page_t page);
1617
1618 extern void vm_page_free_prepare_object(
1619 vm_page_t page,
1620 boolean_t remove_from_hash);
1621
1622 #if CONFIG_IOSCHED
1623 extern wait_result_t vm_page_sleep(
1624 vm_object_t object,
1625 vm_page_t m,
1626 int interruptible);
1627 #endif
1628
1629 extern void vm_pressure_response(void);
1630
1631 #if CONFIG_JETSAM
1632 extern void memorystatus_pages_update(unsigned int pages_avail);
1633
1634 #define VM_CHECK_MEMORYSTATUS do { \
1635 memorystatus_pages_update( \
1636 vm_page_pageable_external_count + \
1637 vm_page_free_count + \
1638 VM_PAGE_SECLUDED_COUNT_OVER_TARGET() + \
1639 (VM_DYNAMIC_PAGING_ENABLED() ? 0 : vm_page_purgeable_count) \
1640 ); \
1641 } while(0)
1642
1643 #else /* CONFIG_JETSAM */
1644
1645 #if !XNU_TARGET_OS_OSX
1646
1647 #define VM_CHECK_MEMORYSTATUS do {} while(0)
1648
1649 #else /* !XNU_TARGET_OS_OSX */
1650
1651 #define VM_CHECK_MEMORYSTATUS vm_pressure_response()
1652
1653 #endif /* !XNU_TARGET_OS_OSX */
1654
1655 #endif /* CONFIG_JETSAM */
1656
1657 /*
1658 * Functions implemented as macros. m->vmp_wanted and m->vmp_busy are
1659 * protected by the object lock.
1660 */
1661
1662 #if !XNU_TARGET_OS_OSX
1663 #define SET_PAGE_DIRTY(m, set_pmap_modified) \
1664 MACRO_BEGIN \
1665 vm_page_t __page__ = (m); \
1666 if (__page__->vmp_pmapped == TRUE && \
1667 __page__->vmp_wpmapped == TRUE && \
1668 __page__->vmp_dirty == FALSE && \
1669 (set_pmap_modified)) { \
1670 pmap_set_modify(VM_PAGE_GET_PHYS_PAGE(__page__)); \
1671 } \
1672 __page__->vmp_dirty = TRUE; \
1673 MACRO_END
1674 #else /* !XNU_TARGET_OS_OSX */
1675 #define SET_PAGE_DIRTY(m, set_pmap_modified) \
1676 MACRO_BEGIN \
1677 vm_page_t __page__ = (m); \
1678 __page__->vmp_dirty = TRUE; \
1679 MACRO_END
1680 #endif /* !XNU_TARGET_OS_OSX */
1681
1682 #define PAGE_ASSERT_WAIT(m, interruptible) \
1683 (((m)->vmp_wanted = TRUE), \
1684 assert_wait((event_t) (m), (interruptible)))
1685
1686 #if CONFIG_IOSCHED
1687 #define PAGE_SLEEP(o, m, interruptible) \
1688 vm_page_sleep(o, m, interruptible)
1689 #else
1690 #define PAGE_SLEEP(o, m, interruptible) \
1691 (((m)->vmp_wanted = TRUE), \
1692 thread_sleep_vm_object((o), (m), (interruptible)))
1693 #endif
1694
1695 #define PAGE_WAKEUP_DONE(m) \
1696 MACRO_BEGIN \
1697 (m)->vmp_busy = FALSE; \
1698 if ((m)->vmp_wanted) { \
1699 (m)->vmp_wanted = FALSE; \
1700 thread_wakeup((event_t) (m)); \
1701 } \
1702 MACRO_END
1703
1704 #define PAGE_WAKEUP(m) \
1705 MACRO_BEGIN \
1706 if ((m)->vmp_wanted) { \
1707 (m)->vmp_wanted = FALSE; \
1708 thread_wakeup((event_t) (m)); \
1709 } \
1710 MACRO_END
1711
1712 #define VM_PAGE_FREE(p) \
1713 MACRO_BEGIN \
1714 vm_page_free_unlocked(p, TRUE); \
1715 MACRO_END
1716
1717 #define VM_PAGE_WAIT() ((void)vm_page_wait(THREAD_UNINT))
1718
1719 #define vm_page_queue_lock (vm_page_locks.vm_page_queue_lock2)
1720 #define vm_page_queue_free_lock (vm_page_locks.vm_page_queue_free_lock2)
1721
1722 static inline void
vm_free_page_lock(void)1723 vm_free_page_lock(void)
1724 {
1725 lck_mtx_lock(&vm_page_queue_free_lock);
1726 }
1727
1728 static inline void
vm_free_page_lock_spin(void)1729 vm_free_page_lock_spin(void)
1730 {
1731 lck_mtx_lock_spin(&vm_page_queue_free_lock);
1732 }
1733
1734 static inline void
vm_free_page_unlock(void)1735 vm_free_page_unlock(void)
1736 {
1737 lck_mtx_unlock(&vm_page_queue_free_lock);
1738 }
1739
1740
1741 static inline void
vm_page_lock_queues(void)1742 vm_page_lock_queues(void)
1743 {
1744 lck_mtx_lock(&vm_page_queue_lock);
1745 }
1746
1747 static inline boolean_t
vm_page_trylock_queues(void)1748 vm_page_trylock_queues(void)
1749 {
1750 boolean_t ret;
1751 ret = lck_mtx_try_lock(&vm_page_queue_lock);
1752 return ret;
1753 }
1754
1755 static inline void
vm_page_unlock_queues(void)1756 vm_page_unlock_queues(void)
1757 {
1758 lck_mtx_unlock(&vm_page_queue_lock);
1759 }
1760
1761 static inline void
vm_page_lockspin_queues(void)1762 vm_page_lockspin_queues(void)
1763 {
1764 lck_mtx_lock_spin(&vm_page_queue_lock);
1765 }
1766
1767 static inline boolean_t
vm_page_trylockspin_queues(void)1768 vm_page_trylockspin_queues(void)
1769 {
1770 boolean_t ret;
1771 ret = lck_mtx_try_lock_spin(&vm_page_queue_lock);
1772 return ret;
1773 }
1774 #define vm_page_lockconvert_queues() lck_mtx_convert_spin(&vm_page_queue_lock)
1775
1776 #ifdef VPL_LOCK_SPIN
1777 extern lck_grp_t vm_page_lck_grp_local;
1778
1779 #define VPL_LOCK_INIT(vlq, vpl_grp, vpl_attr) lck_spin_init(&vlq->vpl_lock, vpl_grp, vpl_attr)
1780 #define VPL_LOCK(vpl) lck_spin_lock_grp(vpl, &vm_page_lck_grp_local)
1781 #define VPL_UNLOCK(vpl) lck_spin_unlock(vpl)
1782 #else
1783 #define VPL_LOCK_INIT(vlq, vpl_grp, vpl_attr) lck_mtx_init(&vlq->vpl_lock, vpl_grp, vpl_attr)
1784 #define VPL_LOCK(vpl) lck_mtx_lock_spin(vpl)
1785 #define VPL_UNLOCK(vpl) lck_mtx_unlock(vpl)
1786 #endif
1787
1788
1789 #if DEVELOPMENT || DEBUG
1790 #define VM_PAGE_SPECULATIVE_USED_ADD() \
1791 MACRO_BEGIN \
1792 OSAddAtomic(1, &vm_page_speculative_used); \
1793 MACRO_END
1794 #else
1795 #define VM_PAGE_SPECULATIVE_USED_ADD()
1796 #endif
1797
1798
1799 #define VM_PAGE_CONSUME_CLUSTERED(mem) \
1800 MACRO_BEGIN \
1801 ppnum_t __phys_page; \
1802 __phys_page = VM_PAGE_GET_PHYS_PAGE(mem); \
1803 pmap_lock_phys_page(__phys_page); \
1804 if (mem->vmp_clustered) { \
1805 vm_object_t o; \
1806 o = VM_PAGE_OBJECT(mem); \
1807 assert(o); \
1808 o->pages_used++; \
1809 mem->vmp_clustered = FALSE; \
1810 VM_PAGE_SPECULATIVE_USED_ADD(); \
1811 } \
1812 pmap_unlock_phys_page(__phys_page); \
1813 MACRO_END
1814
1815
1816 #define VM_PAGE_COUNT_AS_PAGEIN(mem) \
1817 MACRO_BEGIN \
1818 { \
1819 vm_object_t o; \
1820 o = VM_PAGE_OBJECT(mem); \
1821 DTRACE_VM2(pgin, int, 1, (uint64_t *), NULL); \
1822 counter_inc(¤t_task()->pageins); \
1823 if (o->internal) { \
1824 DTRACE_VM2(anonpgin, int, 1, (uint64_t *), NULL); \
1825 } else { \
1826 DTRACE_VM2(fspgin, int, 1, (uint64_t *), NULL); \
1827 } \
1828 } \
1829 MACRO_END
1830
1831 /* adjust for stolen pages accounted elsewhere */
1832 #define VM_PAGE_MOVE_STOLEN(page_count) \
1833 MACRO_BEGIN \
1834 vm_page_stolen_count -= (page_count); \
1835 vm_page_wire_count_initial -= (page_count); \
1836 MACRO_END
1837
1838 #define DW_vm_page_unwire 0x01
1839 #define DW_vm_page_wire 0x02
1840 #define DW_vm_page_free 0x04
1841 #define DW_vm_page_activate 0x08
1842 #define DW_vm_page_deactivate_internal 0x10
1843 #define DW_vm_page_speculate 0x20
1844 #define DW_vm_page_lru 0x40
1845 #define DW_vm_pageout_throttle_up 0x80
1846 #define DW_PAGE_WAKEUP 0x100
1847 #define DW_clear_busy 0x200
1848 #define DW_clear_reference 0x400
1849 #define DW_set_reference 0x800
1850 #define DW_move_page 0x1000
1851 #define DW_VM_PAGE_QUEUES_REMOVE 0x2000
1852 #define DW_enqueue_cleaned 0x4000
1853 #define DW_vm_phantom_cache_update 0x8000
1854
1855 struct vm_page_delayed_work {
1856 vm_page_t dw_m;
1857 int dw_mask;
1858 };
1859
1860 #define DEFAULT_DELAYED_WORK_LIMIT 32
1861
1862 struct vm_page_delayed_work_ctx {
1863 struct vm_page_delayed_work dwp[DEFAULT_DELAYED_WORK_LIMIT];
1864 thread_t delayed_owner;
1865 };
1866
1867 void vm_page_do_delayed_work(vm_object_t object, vm_tag_t tag, struct vm_page_delayed_work *dwp, int dw_count);
1868
1869 extern unsigned int vm_max_delayed_work_limit;
1870
1871 #define DELAYED_WORK_LIMIT(max) ((vm_max_delayed_work_limit >= max ? max : vm_max_delayed_work_limit))
1872
1873 /*
1874 * vm_page_do_delayed_work may need to drop the object lock...
1875 * if it does, we need the pages it's looking at to
1876 * be held stable via the busy bit, so if busy isn't already
1877 * set, we need to set it and ask vm_page_do_delayed_work
1878 * to clear it and wakeup anyone that might have blocked on
1879 * it once we're done processing the page.
1880 */
1881
1882 #define VM_PAGE_ADD_DELAYED_WORK(dwp, mem, dw_cnt) \
1883 MACRO_BEGIN \
1884 if (mem->vmp_busy == FALSE) { \
1885 mem->vmp_busy = TRUE; \
1886 if ( !(dwp->dw_mask & DW_vm_page_free)) \
1887 dwp->dw_mask |= (DW_clear_busy | DW_PAGE_WAKEUP); \
1888 } \
1889 dwp->dw_m = mem; \
1890 dwp++; \
1891 dw_cnt++; \
1892 MACRO_END
1893
1894 extern vm_page_t vm_object_page_grab(vm_object_t);
1895
1896 #if VM_PAGE_BUCKETS_CHECK
1897 extern void vm_page_buckets_check(void);
1898 #endif /* VM_PAGE_BUCKETS_CHECK */
1899
1900 extern void vm_page_queues_remove(vm_page_t mem, boolean_t remove_from_specialq);
1901 extern void vm_page_remove_internal(vm_page_t page);
1902 extern void vm_page_enqueue_inactive(vm_page_t mem, boolean_t first);
1903 extern void vm_page_enqueue_active(vm_page_t mem, boolean_t first);
1904 extern void vm_page_check_pageable_safe(vm_page_t page);
1905
1906 #if CONFIG_SECLUDED_MEMORY
1907 extern uint64_t secluded_shutoff_trigger;
1908 extern uint64_t secluded_shutoff_headroom;
1909 extern void start_secluded_suppression(task_t);
1910 extern void stop_secluded_suppression(task_t);
1911 #endif /* CONFIG_SECLUDED_MEMORY */
1912
1913 extern void vm_retire_boot_pages(void);
1914
1915
1916 #define VMP_ERROR_GET(p) ((p)->vmp_error)
1917
1918
1919 #endif /* _VM_VM_PAGE_H_ */
1920