]> git.saurik.com Git - apple/xnu.git/blame - osfmk/vm/vm_compressor_backing_store.c
projects / apple / xnu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xnu-2422.100.13.tar.gz
[apple/xnu.git] / osfmk / vm / vm_compressor_backing_store.c
CommitLineData
39236c6e
A		 1/*
2 * Copyright (c) 2000-2013 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#include "vm_compressor_backing_store.h"
30#include <vm/vm_protos.h>
31
32#include <IOKit/IOHibernatePrivate.h>
33
34
35boolean_t compressor_store_stop_compaction = FALSE;
36boolean_t vm_swap_up = FALSE;
37boolean_t vm_swapfile_mgmt_needed = FALSE;
38
39int swapper_throttle = -1;
40boolean_t swapper_throttle_inited = FALSE;
41uint64_t vm_swapout_thread_id;
42
43uint64_t vm_swap_put_failures = 0;
44uint64_t vm_swap_get_failures = 0;
45int vm_num_swap_files = 0;
46int vm_swapout_thread_processed_segments = 0;
47int vm_swapout_thread_awakened = 0;
48int vm_swapfile_mgmt_thread_awakened = 0;
49int vm_swapfile_mgmt_thread_running = 0;
50
51unsigned int vm_swapfile_total_segs_alloced = 0;
52unsigned int vm_swapfile_total_segs_used = 0;
53
54
55#define SWAP_READY 0x1 /* Swap file is ready to be used */
56#define SWAP_RECLAIM 0x2 /* Swap file is marked to be reclaimed */
57#define SWAP_WANTED 0x4 /* Swap file has waiters */
58#define SWAP_REUSE 0x8 /* Swap file is on the Q and has a name. Reuse after init-ing.*/
59
60struct swapfile{
61 queue_head_t swp_queue; /* list of swap files */
62 char *swp_path; /* saved pathname of swap file */
63 struct vnode *swp_vp; /* backing vnode */
64 uint64_t swp_size; /* size of this swap file */
65 uint8_t *swp_bitmap; /* bitmap showing the alloced/freed slots in the swap file */
66 unsigned int swp_pathlen; /* length of pathname */
67 unsigned int swp_nsegs; /* #segments we can use */
68 unsigned int swp_nseginuse; /* #segments in use */
69 unsigned int swp_index; /* index of this swap file */
70 unsigned int swp_flags; /* state of swap file */
71 unsigned int swp_free_hint; /* offset of 1st free chunk */
72 unsigned int swp_io_count; /* count of outstanding I/Os */
73 c_segment_t *swp_csegs; /* back pointers to the c_segments. Used during swap reclaim. */
74
75 struct trim_list *swp_delayed_trim_list_head;
76 unsigned int swp_delayed_trim_count;
77 boolean_t swp_trim_supported;
78};
79
80queue_head_t swf_global_queue;
81
82#define VM_SWAPFILE_DELAYED_TRIM_MAX 128
83
84extern clock_sec_t dont_trim_until_ts;
85clock_sec_t vm_swapfile_last_failed_to_create_ts = 0;
86
87static void vm_swapout_thread_throttle_adjust(void);
88static void vm_swap_free_now(struct swapfile *swf, uint64_t f_offset);
89static void vm_swapout_thread(void);
90static void vm_swapfile_mgmt_thread(void);
91static void vm_swap_defragment();
92static void vm_swap_handle_delayed_trims(boolean_t);
93static void vm_swap_do_delayed_trim();
94
95
96#define VM_SWAPFILE_DELAYED_CREATE 30
97#define VM_SWAP_SHOULD_DEFRAGMENT() (c_swappedout_sparse_count > (vm_swapfile_total_segs_used / 4) ? 1 : 0)
98#define VM_SWAP_SHOULD_RECLAIM() (((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) >= SWAPFILE_RECLAIM_THRESHOLD_SEGS) ? 1 : 0)
99#define VM_SWAP_SHOULD_CREATE(cur_ts) (((vm_swapfile_total_segs_alloced - vm_swapfile_total_segs_used) < (unsigned int)VM_SWAPFILE_HIWATER_SEGS) && \
100 ((cur_ts - vm_swapfile_last_failed_to_create_ts) > VM_SWAPFILE_DELAYED_CREATE) ? 1 : 0)
101#define VM_SWAP_SHOULD_TRIM(swf) ((swf->swp_delayed_trim_count >= VM_SWAPFILE_DELAYED_TRIM_MAX) ? 1 : 0)
102
103
104#define VM_SWAP_BUSY() ((c_swapout_count && (swapper_throttle == THROTTLE_LEVEL_COMPRESSOR_TIER1 || swapper_throttle == THROTTLE_LEVEL_COMPRESSOR_TIER0)) ? 1 : 0)
105
106
107#if CHECKSUM_THE_SWAP
108extern unsigned int hash_string(char *cp, int len);
109#endif
110
111#if CRYPTO
112extern boolean_t swap_crypt_ctx_initialized;
113extern void swap_crypt_ctx_initialize(void);
114extern const unsigned char swap_crypt_null_iv[AES_BLOCK_SIZE];
115extern aes_ctx swap_crypt_ctx;
116extern unsigned long vm_page_encrypt_counter;
117extern unsigned long vm_page_decrypt_counter;
118#endif /* CRYPTO */
119
120extern void vm_pageout_io_throttle(void);
121
122struct swapfile *vm_swapfile_for_handle(uint64_t);
123
124/*
125 * Called with the vm_swap_data_lock held.
126 */
127
128struct swapfile *
129vm_swapfile_for_handle(uint64_t f_offset)
130{
131
132 uint64_t file_offset = 0;
133 unsigned int swapfile_index = 0;
134 struct swapfile* swf = NULL;
135
136 file_offset = (f_offset & SWAP_SLOT_MASK);
137 swapfile_index = (f_offset >> SWAP_DEVICE_SHIFT);
138
139 swf = (struct swapfile*) queue_first(&swf_global_queue);
140
141 while(queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
142
143 if (swapfile_index == swf->swp_index) {
144 break;
145 }
146
147 swf = (struct swapfile*) queue_next(&swf->swp_queue);
148 }
149
150 if (queue_end(&swf_global_queue, (queue_entry_t) swf)) {
151 swf = NULL;
152 }
153
154 return swf;
155}
156
157void
158vm_swap_init()
159{
160 static boolean_t vm_swap_try_init = FALSE;
161 thread_t thread = NULL;
162
163 if (vm_swap_try_init == TRUE) {
164 return;
165 }
166
167 vm_swap_try_init = TRUE;
168
169 lck_grp_attr_setdefault(&vm_swap_data_lock_grp_attr);
170 lck_grp_init(&vm_swap_data_lock_grp,
171 "vm_swap_data",
172 &vm_swap_data_lock_grp_attr);
173 lck_attr_setdefault(&vm_swap_data_lock_attr);
174 lck_mtx_init_ext(&vm_swap_data_lock,
175 &vm_swap_data_lock_ext,
176 &vm_swap_data_lock_grp,
177 &vm_swap_data_lock_attr);
178
179 queue_init(&swf_global_queue);
180
181 if (vm_swap_create_file()) {
182
183 if (kernel_thread_start_priority((thread_continue_t)vm_swapout_thread, NULL,
184 BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
185 panic("vm_swapout_thread: create failed");
186 }
187 thread->options |= TH_OPT_VMPRIV;
188 vm_swapout_thread_id = thread->thread_id;
189
190 thread_deallocate(thread);
191
192 if (kernel_thread_start_priority((thread_continue_t)vm_swapfile_mgmt_thread, NULL,
193 BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
194 panic("vm_swapfile_mgmt_thread: create failed");
195 }
196 thread->options |= TH_OPT_VMPRIV;
197
198 thread_deallocate(thread);
199
200#if CRYPTO
201 if (swap_crypt_ctx_initialized == FALSE) {
202 swap_crypt_ctx_initialize();
203 }
204#endif /* CRYPTO */
205
206 vm_swap_up = TRUE;
207
208#if SANITY_CHECK_SWAP_ROUTINES
209extern lck_attr_t *vm_compressor_lck_attr;
210extern lck_grp_t *vm_compressor_lck_grp;
211
212 /*
213 * Changes COMPRESSED_SWAP_CHUNK_SIZE to make it (4*KB).
214 * Changes MIN_SWAP_FILE_SIZE to (4*KB).
215 * Changes MAX_SWAP_FILE_SIZE to (4*KB).
216 * That will then cause the below allocations to create
217 * 4 new swap files and put/get/free from them.
218 */
219 {
220 c_segment_t c_seg = NULL, c_seg1 = NULL, c_seg2 = NULL, c_seg3 = NULL;
221 vm_offset_t addr = 0;
222 vm_offset_t dup_addr = 0;
223 kern_return_t kr = KERN_SUCCESS;
224 uint64_t f_offset = 0;
225 uint64_t f_offset1 = 0;
226 uint64_t f_offset2 = 0;
227 uint64_t f_offset3 = 0;
228
229 if ((kr = kernel_memory_allocate(kernel_map,
230 &addr,
231 4 * COMPRESSED_SWAP_CHUNK_SIZE,
232 0,
233 KMA_KOBJECT))) {
234 printf("kernel_memory_allocate failed with %d\n", kr);
235 goto done;
236 }
237
238 if ((kr = kernel_memory_allocate(kernel_map,
239 &dup_addr,
240 4 * COMPRESSED_SWAP_CHUNK_SIZE,
241 0,
242 KMA_KOBJECT))) {
243 printf("kernel_memory_allocate failed with %d\n", kr);
244 goto done;
245 }
246
247 c_seg = (c_segment_t) kalloc(sizeof(*c_seg));
248 memset(c_seg, 0, sizeof(*c_seg));
249#if __i386__ || __x86_64__
250 lck_mtx_init(&c_seg->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
251#else /* __i386__ || __x86_64__ */
252 lck_spin_init(&c_seg->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
253#endif /* __i386__ || __x86_64__ */
254
255
256 c_seg1 = (c_segment_t) kalloc(sizeof(*c_seg));
257 memset(c_seg1, 0, sizeof(*c_seg));
258#if __i386__ || __x86_64__
259 lck_mtx_init(&c_seg1->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
260#else /* __i386__ || __x86_64__ */
261 lck_spin_init(&c_seg1->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
262#endif /* __i386__ || __x86_64__ */
263
264
265 c_seg2 = (c_segment_t) kalloc(sizeof(*c_seg));
266 memset(c_seg2, 0, sizeof(*c_seg));
267#if __i386__ || __x86_64__
268 lck_mtx_init(&c_seg2->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
269#else /* __i386__ || __x86_64__ */
270 lck_spin_init(&c_seg2->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
271#endif /* __i386__ || __x86_64__ */
272
273
274 c_seg3 = (c_segment_t) kalloc(sizeof(*c_seg));
275 memset(c_seg3, 0, sizeof(*c_seg));
276#if __i386__ || __x86_64__
277 lck_mtx_init(&c_seg3->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
278#else /* __i386__ || __x86_64__ */
279 lck_spin_init(&c_seg3->c_lock, vm_compressor_lck_grp, vm_compressor_lck_attr);
280#endif /* __i386__ || __x86_64__ */
281
282
283 memset((void*)addr, (int) 'a', PAGE_SIZE_64);
284 memset((void*)(addr + PAGE_SIZE_64), (int) 'b', PAGE_SIZE_64);
285 memset((void*)(addr + (2 * PAGE_SIZE_64)), (int) 'c', PAGE_SIZE_64);
286 memset((void*)(addr + (3 * PAGE_SIZE_64)), (int) 'd', PAGE_SIZE_64);
287
288 vm_swap_put(addr, &f_offset, PAGE_SIZE_64, c_seg);
289 c_seg->c_store.c_swap_handle = f_offset;
290
291 vm_swap_put(addr + PAGE_SIZE_64, &f_offset1, PAGE_SIZE_64, c_seg1);
292 c_seg1->c_store.c_swap_handle = f_offset1;
293
294 vm_swap_put(addr + (2 * PAGE_SIZE_64), &f_offset2, PAGE_SIZE_64, c_seg2);
295 c_seg2->c_store.c_swap_handle = f_offset2;
296
297 vm_swap_put(addr + (3 * PAGE_SIZE_64), &f_offset3, PAGE_SIZE_64, c_seg3);
298 c_seg3->c_store.c_swap_handle = f_offset3;
299
300 //vm_swap_free(f_offset);
301 vm_swap_get(dup_addr, f_offset, PAGE_SIZE_64);
302
303 //vm_swap_free(f_offset1);
304 vm_swap_reclaim();
305 vm_swap_get(dup_addr + PAGE_SIZE_64, c_seg1->c_store.c_swap_handle, PAGE_SIZE_64);
306
307 //vm_swap_free(f_offset2);
308 vm_swap_reclaim();
309 vm_swap_get(dup_addr + (2 * PAGE_SIZE_64), c_seg2->c_store.c_swap_handle, PAGE_SIZE_64);
310
311 //vm_swap_free(f_offset3);
312 vm_swap_reclaim();
313 vm_swap_get(dup_addr + (3 * PAGE_SIZE_64), c_seg3->c_store.c_swap_handle, PAGE_SIZE_64);
314
315 if (memcmp((void*)addr, (void*)dup_addr, PAGE_SIZE_64)) {
316 panic("First page data mismatch\n");
317 kr = KERN_FAILURE;
318 goto done;
319 }
320
321 if (memcmp((void*)(addr + PAGE_SIZE_64), (void*)(dup_addr + PAGE_SIZE_64), PAGE_SIZE_64)) {
322 panic("Second page data mismatch 0x%lx, 0x%lxn", addr, dup_addr);
323 kr = KERN_FAILURE;
324 goto done;
325 }
326
327 if (memcmp((void*)(addr + (2 * PAGE_SIZE_64)), (void*)(dup_addr + (2 * PAGE_SIZE_64)), PAGE_SIZE_64)) {
328 panic("Third page data mismatch\n");
329 kr = KERN_FAILURE;
330 goto done;
331 }
332
333 if (memcmp((void*)(addr + (3 * PAGE_SIZE_64)), (void*)(dup_addr + (3 * PAGE_SIZE_64)), PAGE_SIZE_64)) {
334 panic("Fourth page data mismatch 0x%lx, 0x%lxn", addr, dup_addr);
335 kr = KERN_FAILURE;
336 goto done;
337 }
338
339done:
340 printf("Sanity check %s\n", ((kr != KERN_SUCCESS) ? "FAILED" : "SUCCEEDED"));
341 kfree((void*)addr, 4 * COMPRESSED_SWAP_CHUNK_SIZE);
342 addr = 0;
343 kfree((void*)dup_addr, 4 * COMPRESSED_SWAP_CHUNK_SIZE);
344 dup_addr = 0;
345 }
346#endif /* SANITY_CHECK_SWAP_ROUTINES */
347 }
348
349 printf("VM Swap Subsystem is %s\n", (vm_swap_up == TRUE) ? "ON" : "OFF");
350}
351
352#if CRYPTO
353void
354vm_swap_encrypt(c_segment_t c_seg)
355{
356 vm_offset_t kernel_vaddr = 0;
357 uint64_t size = 0;
358
359 union {
360 unsigned char aes_iv[AES_BLOCK_SIZE];
361 void *c_seg;
362 } encrypt_iv;
363
364 assert(swap_crypt_ctx_initialized);
365
366 bzero(&encrypt_iv.aes_iv[0], sizeof (encrypt_iv.aes_iv));
367
368 encrypt_iv.c_seg = (void*)c_seg;
369
370 /* encrypt the "initial vector" */
371 aes_encrypt_cbc((const unsigned char *) &encrypt_iv.aes_iv[0],
372 swap_crypt_null_iv,
373 1,
374 &encrypt_iv.aes_iv[0],
375 &swap_crypt_ctx.encrypt);
376
377 kernel_vaddr = (vm_offset_t) c_seg->c_store.c_buffer;
378 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
379
380 /*
381 * Encrypt the c_segment.
382 */
383 aes_encrypt_cbc((const unsigned char *) kernel_vaddr,
384 &encrypt_iv.aes_iv[0],
385 (unsigned int)(size / AES_BLOCK_SIZE),
386 (unsigned char *) kernel_vaddr,
387 &swap_crypt_ctx.encrypt);
388
389 vm_page_encrypt_counter += (size/PAGE_SIZE_64);
390}
391
392void
393vm_swap_decrypt(c_segment_t c_seg)
394{
395
396 vm_offset_t kernel_vaddr = 0;
397 uint64_t size = 0;
398
399 union {
400 unsigned char aes_iv[AES_BLOCK_SIZE];
401 void *c_seg;
402 } decrypt_iv;
403
404
405 assert(swap_crypt_ctx_initialized);
406
407 /*
408 * Prepare an "initial vector" for the decryption.
409 * It has to be the same as the "initial vector" we
410 * used to encrypt that page.
411 */
412 bzero(&decrypt_iv.aes_iv[0], sizeof (decrypt_iv.aes_iv));
413
414 decrypt_iv.c_seg = (void*)c_seg;
415
416 /* encrypt the "initial vector" */
417 aes_encrypt_cbc((const unsigned char *) &decrypt_iv.aes_iv[0],
418 swap_crypt_null_iv,
419 1,
420 &decrypt_iv.aes_iv[0],
421 &swap_crypt_ctx.encrypt);
422
423 kernel_vaddr = (vm_offset_t) c_seg->c_store.c_buffer;
424 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
425
426 /*
427 * Decrypt the c_segment.
428 */
429 aes_decrypt_cbc((const unsigned char *) kernel_vaddr,
430 &decrypt_iv.aes_iv[0],
431 (unsigned int) (size / AES_BLOCK_SIZE),
432 (unsigned char *) kernel_vaddr,
433 &swap_crypt_ctx.decrypt);
434
435 vm_page_decrypt_counter += (size/PAGE_SIZE_64);
436}
437#endif /* CRYPTO */
438
439
440void
441vm_swap_consider_defragmenting()
442{
443 if (compressor_store_stop_compaction == FALSE && !VM_SWAP_BUSY() && (VM_SWAP_SHOULD_DEFRAGMENT() || VM_SWAP_SHOULD_RECLAIM())) {
444
445 if (!vm_swapfile_mgmt_thread_running) {
446 lck_mtx_lock(&vm_swap_data_lock);
447
448 if (!vm_swapfile_mgmt_thread_running)
449 thread_wakeup((event_t) &vm_swapfile_mgmt_needed);
450
451 lck_mtx_unlock(&vm_swap_data_lock);
452 }
453 }
454}
455
456
457int vm_swap_defragment_yielded = 0;
458int vm_swap_defragment_swapin = 0;
459int vm_swap_defragment_free = 0;
460int vm_swap_defragment_busy = 0;
461
462
463static void
464vm_swap_defragment()
465{
466 c_segment_t c_seg;
467
468 /*
469 * have to grab the master lock w/o holding
470 * any locks in spin mode
471 */
472 PAGE_REPLACEMENT_DISALLOWED(TRUE);
473
474 lck_mtx_lock_spin_always(c_list_lock);
475
476 while (!queue_empty(&c_swappedout_sparse_list_head)) {
477
478 if (compressor_store_stop_compaction == TRUE || VM_SWAP_BUSY()) {
479 vm_swap_defragment_yielded++;
480 break;
481 }
482 c_seg = (c_segment_t)queue_first(&c_swappedout_sparse_list_head);
483
484 lck_mtx_lock_spin_always(&c_seg->c_lock);
485
486 assert(c_seg->c_on_swappedout_sparse_q);
487
488 if (c_seg->c_busy) {
489 lck_mtx_unlock_always(c_list_lock);
490
491 PAGE_REPLACEMENT_DISALLOWED(FALSE);
492 /*
493 * c_seg_wait_on_busy consumes c_seg->c_lock
494 */
495 c_seg_wait_on_busy(c_seg);
496
497 PAGE_REPLACEMENT_DISALLOWED(TRUE);
498
499 lck_mtx_lock_spin_always(c_list_lock);
500
501 vm_swap_defragment_busy++;
502 continue;
503 }
504 if (c_seg->c_bytes_used == 0) {
505 /*
506 * c_seg_free_locked consumes the c_list_lock
507 * and c_seg->c_lock
508 */
509 c_seg_free_locked(c_seg);
510
511 vm_swap_defragment_free++;
512 } else {
513 lck_mtx_unlock_always(c_list_lock);
514
515 c_seg_swapin(c_seg, TRUE);
516 lck_mtx_unlock_always(&c_seg->c_lock);
517
518 vm_swap_defragment_swapin++;
519 }
520 PAGE_REPLACEMENT_DISALLOWED(FALSE);
521
522 vm_pageout_io_throttle();
523
524 /*
525 * because write waiters have privilege over readers,
526 * dropping and immediately retaking the master lock will
527 * still allow any thread waiting to acquire the
528 * master lock exclusively an opportunity to take it
529 */
530 PAGE_REPLACEMENT_DISALLOWED(TRUE);
531
532 lck_mtx_lock_spin_always(c_list_lock);
533 }
534 lck_mtx_unlock_always(c_list_lock);
535
536 PAGE_REPLACEMENT_DISALLOWED(FALSE);
537}
538
539
540
541static void
542vm_swapfile_mgmt_thread(void)
543{
544
545 boolean_t did_work = FALSE;
546 clock_sec_t sec;
547 clock_nsec_t nsec;
548
549 vm_swapfile_mgmt_thread_awakened++;
550 vm_swapfile_mgmt_thread_running = 1;
551
552try_again:
553
554 do {
555 if (vm_swap_up == FALSE)
556 break;
557 did_work = FALSE;
558 clock_get_system_nanotime(&sec, &nsec);
559
560 /*
561 * walk through the list of swap files
562 * and do the delayed frees/trims for
563 * any swap file whose count of delayed
564 * frees is above the batch limit
565 */
566 vm_swap_handle_delayed_trims(FALSE);
567
568 if (VM_SWAP_SHOULD_CREATE(sec)) {
569 if (vm_swap_create_file() == TRUE)
570 did_work = TRUE;
571 else {
572 vm_swapfile_last_failed_to_create_ts = sec;
573 HIBLOG("vm_swap_create_file failed @ %lu secs\n", sec);
574 }
575 }
576 if (VM_SWAP_SHOULD_DEFRAGMENT()) {
577 proc_set_task_policy_thread(kernel_task, current_thread()->thread_id,
578 TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
579
580 vm_swap_defragment();
581
582 if (!VM_SWAP_BUSY())
583 did_work = TRUE;
584
585 proc_set_task_policy_thread(kernel_task, current_thread()->thread_id,
586 TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER1);
587 }
588 if (VM_SWAP_SHOULD_RECLAIM()) {
589 proc_set_task_policy_thread(kernel_task, current_thread()->thread_id,
590 TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
591
592 vm_swap_defragment();
593 vm_swap_reclaim();
594
595 if (!VM_SWAP_BUSY())
596 did_work = TRUE;
597
598 proc_set_task_policy_thread(kernel_task, current_thread()->thread_id,
599 TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER1);
600 }
601
602 } while (did_work == TRUE);
603
604 lck_mtx_lock(&vm_swap_data_lock);
605
606 clock_get_system_nanotime(&sec, &nsec);
607
608 if (vm_swap_up == TRUE && (VM_SWAP_SHOULD_CREATE(sec) || ((!VM_SWAP_BUSY() && compressor_store_stop_compaction == FALSE) &&
609 (VM_SWAP_SHOULD_DEFRAGMENT() || VM_SWAP_SHOULD_RECLAIM())))) {
610 lck_mtx_unlock(&vm_swap_data_lock);
611 goto try_again;
612 }
613
614 vm_swapfile_mgmt_thread_running = 0;
615
616 assert_wait((event_t)&vm_swapfile_mgmt_needed, THREAD_UNINT);
617
618 lck_mtx_unlock(&vm_swap_data_lock);
619
620 thread_block((thread_continue_t)vm_swapfile_mgmt_thread);
621
622 /* NOTREACHED */
623}
624
625
626
627int swapper_entered_T0 = 0;
628int swapper_entered_T1 = 0;
629int swapper_entered_T2 = 0;
630
631static void
632vm_swapout_thread_throttle_adjust(void)
633{
634 int swapper_throttle_new;
635
636 if (swapper_throttle_inited == FALSE) {
637 /*
638 * force this thread to be set to the correct
639 * throttling tier
640 */
641 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
642 swapper_throttle = THROTTLE_LEVEL_COMPRESSOR_TIER1;
643 swapper_throttle_inited = TRUE;
644 swapper_entered_T2++;
645 goto done;
646 }
647 swapper_throttle_new = swapper_throttle;
648
649
650 switch(swapper_throttle) {
651
652 case THROTTLE_LEVEL_COMPRESSOR_TIER2:
653
654 if (SWAPPER_NEEDS_TO_UNTHROTTLE() || swapout_target_age || hibernate_flushing == TRUE) {
655 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER1;
656 swapper_entered_T1++;
657 break;
658 }
659 break;
660
661 case THROTTLE_LEVEL_COMPRESSOR_TIER1:
662
663 if (VM_PAGEOUT_SCAN_NEEDS_TO_THROTTLE()) {
664 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER0;
665 swapper_entered_T0++;
666 break;
667 }
668 if (COMPRESSOR_NEEDS_TO_SWAP() == 0 && swapout_target_age == 0 && hibernate_flushing == FALSE) {
669 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
670 swapper_entered_T2++;
671 break;
672 }
673 break;
674
675 case THROTTLE_LEVEL_COMPRESSOR_TIER0:
676
677 if (COMPRESSOR_NEEDS_TO_SWAP() == 0) {
678 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER2;
679 swapper_entered_T2++;
680 break;
681 }
682 if (SWAPPER_NEEDS_TO_UNTHROTTLE() == 0) {
683 swapper_throttle_new = THROTTLE_LEVEL_COMPRESSOR_TIER1;
684 swapper_entered_T1++;
685 break;
686 }
687 break;
688 }
689done:
690 if (swapper_throttle != swapper_throttle_new) {
691 proc_set_task_policy_thread(kernel_task, vm_swapout_thread_id,
692 TASK_POLICY_INTERNAL, TASK_POLICY_IO, swapper_throttle_new);
693 proc_set_task_policy_thread(kernel_task, vm_swapout_thread_id,
694 TASK_POLICY_INTERNAL, TASK_POLICY_PASSIVE_IO, TASK_POLICY_ENABLE);
695
696 swapper_throttle = swapper_throttle_new;
697 }
698}
699
700
701static void
702vm_swapout_thread(void)
703{
704 uint64_t f_offset = 0;
705 uint32_t size = 0;
706 c_segment_t c_seg = NULL;
707 kern_return_t kr = KERN_SUCCESS;
708 vm_offset_t addr = 0;
709
710 vm_swapout_thread_awakened++;
711
712 lck_mtx_lock_spin_always(c_list_lock);
713
714 while (!queue_empty(&c_swapout_list_head)) {
715
716 c_seg = (c_segment_t)queue_first(&c_swapout_list_head);
717
718 lck_mtx_lock_spin_always(&c_seg->c_lock);
719
720 assert(c_seg->c_on_swapout_q);
721
722 if (c_seg->c_busy) {
723 lck_mtx_unlock_always(&c_seg->c_lock);
724 lck_mtx_unlock_always(c_list_lock);
725
726 mutex_pause(2);
727
728 lck_mtx_lock_spin_always(c_list_lock);
729
730 continue;
731 }
732 queue_remove(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
733 c_seg->c_on_swapout_q = 0;
734 c_swapout_count--;
735
736 c_seg->c_busy = 1;
737 c_seg->c_busy_swapping = 1;
738
739 vm_swapout_thread_processed_segments++;
740
741 thread_wakeup((event_t)&compaction_swapper_running);
742
743 lck_mtx_unlock_always(c_list_lock);
744
745 addr = (vm_offset_t) c_seg->c_store.c_buffer;
746
747 size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
748
749 lck_mtx_unlock_always(&c_seg->c_lock);
750
751#if CHECKSUM_THE_SWAP
752 c_seg->cseg_hash = hash_string((char*)addr, (int)size);
753 c_seg->cseg_swap_size = size;
754#endif /* CHECKSUM_THE_SWAP */
755
756#if CRYPTO
757 vm_swap_encrypt(c_seg);
758#endif /* CRYPTO */
759
760 vm_swapout_thread_throttle_adjust();
761
762 kr = vm_swap_put((vm_offset_t) addr, &f_offset, size, c_seg);
763
764 PAGE_REPLACEMENT_DISALLOWED(TRUE);
765
766 lck_mtx_lock_spin_always(c_list_lock);
767 lck_mtx_lock_spin_always(&c_seg->c_lock);
768
769 if (kr == KERN_SUCCESS) {
770
771 if (C_SEG_ONDISK_IS_SPARSE(c_seg) && hibernate_flushing == FALSE) {
772
773 c_seg_insert_into_q(&c_swappedout_sparse_list_head, c_seg);
774 c_seg->c_on_swappedout_sparse_q = 1;
775 c_swappedout_sparse_count++;
776
777 } else {
778 if (hibernate_flushing == TRUE && (c_seg->c_generation_id >= first_c_segment_to_warm_generation_id &&
779 c_seg->c_generation_id <= last_c_segment_to_warm_generation_id))
780 queue_enter_first(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
781 else
782 queue_enter(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
783 c_seg->c_on_swappedout_q = 1;
784 c_swappedout_count++;
785 }
786 c_seg->c_store.c_swap_handle = f_offset;
787 c_seg->c_ondisk = 1;
788
789 VM_STAT_INCR_BY(swapouts, size >> PAGE_SHIFT);
790
791 if (c_seg->c_bytes_used)
792 OSAddAtomic64(-c_seg->c_bytes_used, &compressor_bytes_used);
793 } else {
794#if CRYPTO
795 vm_swap_decrypt(c_seg);
796#endif /* CRYPTO */
797 c_seg_insert_into_q(&c_age_list_head, c_seg);
798 c_seg->c_on_age_q = 1;
799 c_age_count++;
800
801 vm_swap_put_failures++;
802 }
803 lck_mtx_unlock_always(c_list_lock);
804
805 c_seg->c_busy_swapping = 0;
806
807 C_SEG_WAKEUP_DONE(c_seg);
808
809 if (c_seg->c_must_free)
810 c_seg_free(c_seg);
811 else
812 lck_mtx_unlock_always(&c_seg->c_lock);
813
814 if (kr == KERN_SUCCESS)
815 kernel_memory_depopulate(kernel_map, (vm_offset_t) addr, size, KMA_COMPRESSOR);
816
817 PAGE_REPLACEMENT_DISALLOWED(FALSE);
818
819 if (kr == KERN_SUCCESS)
820 kmem_free(kernel_map, (vm_offset_t) addr, C_SEG_ALLOCSIZE);
821
822 vm_pageout_io_throttle();
823
824 if (c_swapout_count == 0)
825 vm_swap_consider_defragmenting();
826
827 lck_mtx_lock_spin_always(c_list_lock);
828 }
829
830 assert_wait((event_t)&c_swapout_list_head, THREAD_UNINT);
831
832 lck_mtx_unlock_always(c_list_lock);
833
834 thread_block((thread_continue_t)vm_swapout_thread);
835
836 /* NOTREACHED */
837}
838
839boolean_t
840vm_swap_create_file()
841{
842 uint64_t size = 0;
843 int namelen = 0;
844 boolean_t swap_file_created = FALSE;
845 boolean_t swap_file_reuse = FALSE;
846 struct swapfile *swf = NULL;
847
848
849 if (DEFAULT_PAGER_IS_ACTIVE || DEFAULT_FREEZER_IS_ACTIVE) {
850 }
851
852 /*
853 * Any swapfile structure ready for re-use?
854 */
855
856 lck_mtx_lock(&vm_swap_data_lock);
857
858 swf = (struct swapfile*) queue_first(&swf_global_queue);
859
860 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
861 if (swf->swp_flags == SWAP_REUSE) {
862 swap_file_reuse = TRUE;
863 break;
864 }
865 swf = (struct swapfile*) queue_next(&swf->swp_queue);
866 }
867
868 lck_mtx_unlock(&vm_swap_data_lock);
869
870 if (swap_file_reuse == FALSE) {
871
872 namelen = SWAPFILENAME_LEN + SWAPFILENAME_INDEX_LEN + 1;
873
874 swf = (struct swapfile*) kalloc(sizeof *swf);
875 memset(swf, 0, sizeof(*swf));
876
877 swf->swp_index = vm_num_swap_files + 1;
878 swf->swp_pathlen = namelen;
879 swf->swp_path = (char*)kalloc(swf->swp_pathlen);
880
881 memset(swf->swp_path, 0, namelen);
882
883 snprintf(swf->swp_path, namelen, "%s%d", SWAP_FILE_NAME, vm_num_swap_files + 1);
884 }
885
886 vm_swapfile_open(swf->swp_path, &swf->swp_vp);
887
888 if (swf->swp_vp == NULL) {
889 if (swap_file_reuse == FALSE) {
890 kfree(swf->swp_path, swf->swp_pathlen);
891 kfree(swf, sizeof *swf);
892 }
893 return FALSE;
894 }
895 size = MAX_SWAP_FILE_SIZE;
896
897 while (size >= MIN_SWAP_FILE_SIZE) {
898
899 if (vm_swapfile_preallocate(swf->swp_vp, &size) == 0) {
900
901 int num_bytes_for_bitmap = 0;
902
903 swap_file_created = TRUE;
904
905 swf->swp_size = size;
906 swf->swp_nsegs = (unsigned int) (size / COMPRESSED_SWAP_CHUNK_SIZE);
907 swf->swp_nseginuse = 0;
908 swf->swp_free_hint = 0;
909
910 num_bytes_for_bitmap = MAX((swf->swp_nsegs >> 3) , 1);
911 /*
912 * Allocate a bitmap that describes the
913 * number of segments held by this swapfile.
914 */
915 swf->swp_bitmap = (uint8_t*)kalloc(num_bytes_for_bitmap);
916 memset(swf->swp_bitmap, 0, num_bytes_for_bitmap);
917
918 swf->swp_csegs = (c_segment_t *) kalloc(swf->swp_nsegs * sizeof(c_segment_t));
919 memset(swf->swp_csegs, 0, (swf->swp_nsegs * sizeof(c_segment_t)));
920
921 /*
922 * passing a NULL trim_list into vnode_trim_list
923 * will return ENOTSUP if trim isn't supported
924 * and 0 if it is
925 */
926 if (vnode_trim_list(swf->swp_vp, NULL))
927 swf->swp_trim_supported = FALSE;
928 else
929 swf->swp_trim_supported = TRUE;
930
931 lck_mtx_lock(&vm_swap_data_lock);
932
933 swf->swp_flags = SWAP_READY;
934
935 if (swap_file_reuse == FALSE) {
936 queue_enter(&swf_global_queue, swf, struct swapfile*, swp_queue);
937 }
938
939 vm_num_swap_files++;
940
941 vm_swapfile_total_segs_alloced += swf->swp_nsegs;
942
943 lck_mtx_unlock(&vm_swap_data_lock);
944
945 thread_wakeup((event_t) &vm_num_swap_files);
946
947 break;
948 } else {
949
950 size = size / 2;
951 }
952 }
953 if (swap_file_created == FALSE) {
954
955 vm_swapfile_close((uint64_t)(swf->swp_path), swf->swp_vp);
956
957 swf->swp_vp = NULL;
958
959 if (swap_file_reuse == FALSE) {
960 kfree(swf->swp_path, swf->swp_pathlen);
961 kfree(swf, sizeof *swf);
962 }
963 }
964 return swap_file_created;
965}
966
967
968kern_return_t
969vm_swap_get(vm_offset_t addr, uint64_t f_offset, uint64_t size)
970{
971 struct swapfile *swf = NULL;
972 uint64_t file_offset = 0;
973 int retval;
974
975 if (addr == 0) {
976 return KERN_FAILURE;
977 }
978
979 lck_mtx_lock(&vm_swap_data_lock);
980
981 swf = vm_swapfile_for_handle(f_offset);
982
983 if (swf) {
984 if ((swf->swp_flags & SWAP_READY) || (swf->swp_flags & SWAP_RECLAIM)) {
985
986 swf->swp_io_count++;
987 file_offset = (f_offset & SWAP_SLOT_MASK);
988
989 lck_mtx_unlock(&vm_swap_data_lock);
990
991 } else {
992
993 lck_mtx_unlock(&vm_swap_data_lock);
994 return KERN_FAILURE;
995 }
996 } else {
997
998 lck_mtx_unlock(&vm_swap_data_lock);
999 return KERN_FAILURE;
1000 }
1001
1002 retval = vm_swapfile_io(swf->swp_vp, file_offset, addr, (int)(size / PAGE_SIZE_64), SWAP_READ);
1003
1004 /*
1005 * Free this slot in the swap structure.
1006 */
1007 vm_swap_free(f_offset);
1008
1009 lck_mtx_lock(&vm_swap_data_lock);
1010 swf->swp_io_count--;
1011
1012 if ((swf->swp_flags & SWAP_WANTED) && swf->swp_io_count == 0) {
1013
1014 swf->swp_flags &= ~SWAP_WANTED;
1015 thread_wakeup((event_t) &swf->swp_flags);
1016 }
1017 if (retval == 0)
1018 VM_STAT_INCR_BY(swapins, size >> PAGE_SHIFT);
1019 lck_mtx_unlock(&vm_swap_data_lock);
1020
1021 if (retval == 0)
1022 return KERN_SUCCESS;
1023 else {
1024 vm_swap_get_failures++;
1025 return KERN_FAILURE;
1026 }
1027}
1028
1029kern_return_t
1030vm_swap_put(vm_offset_t addr, uint64_t *f_offset, uint64_t size, c_segment_t c_seg)
1031{
1032 unsigned int segidx = 0;
1033 struct swapfile *swf = NULL;
1034 uint64_t file_offset = 0;
1035 uint64_t swapfile_index = 0;
1036 unsigned int byte_for_segidx = 0;
1037 unsigned int offset_within_byte = 0;
1038 boolean_t swf_eligible = FALSE;
1039 boolean_t waiting = FALSE;
1040 int error = 0;
1041 clock_sec_t sec;
1042 clock_nsec_t nsec;
1043
1044 if (addr == 0 || f_offset == NULL) {
1045 return KERN_FAILURE;
1046 }
1047
1048 lck_mtx_lock(&vm_swap_data_lock);
1049
1050 swf = (struct swapfile*) queue_first(&swf_global_queue);
1051
1052 while(queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1053
1054 segidx = swf->swp_free_hint;
1055
1056 swf_eligible = (swf->swp_flags & SWAP_READY) && (swf->swp_nseginuse < swf->swp_nsegs);
1057
1058 if (swf_eligible) {
1059
1060 while(segidx < swf->swp_nsegs) {
1061
1062 byte_for_segidx = segidx >> 3;
1063 offset_within_byte = segidx % 8;
1064
1065 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1066 segidx++;
1067 continue;
1068 }
1069
1070 (swf->swp_bitmap)[byte_for_segidx] |= (1 << offset_within_byte);
1071
1072 file_offset = segidx * COMPRESSED_SWAP_CHUNK_SIZE;
1073 swf->swp_nseginuse++;
1074 swf->swp_io_count++;
1075 swapfile_index = swf->swp_index;
1076
1077 vm_swapfile_total_segs_used++;
1078
1079 clock_get_system_nanotime(&sec, &nsec);
1080
1081 if (VM_SWAP_SHOULD_CREATE(sec) && !vm_swapfile_mgmt_thread_running)
1082 thread_wakeup((event_t) &vm_swapfile_mgmt_needed);
1083
1084 lck_mtx_unlock(&vm_swap_data_lock);
1085
1086 goto done;
1087 }
1088 }
1089 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1090 }
1091 assert(queue_end(&swf_global_queue, (queue_entry_t) swf));
1092
1093 /*
1094 * we've run out of swap segments, but may not
1095 * be in a position to immediately create a new swap
1096 * file if we've recently failed to create due to a lack
1097 * of free space in the root filesystem... we'll try
1098 * to kick that create off, but in any event we're going
1099 * to take a breather (up to 1 second) so that we're not caught in a tight
1100 * loop back in "vm_compressor_compact_and_swap" trying to stuff
1101 * segments into swap files only to have them immediately put back
1102 * on the c_age queue due to vm_swap_put failing.
1103 *
1104 * if we're doing these puts due to a hibernation flush,
1105 * no need to block... setting hibernate_no_swapspace to TRUE,
1106 * will cause "vm_compressor_compact_and_swap" to immediately abort
1107 */
1108 clock_get_system_nanotime(&sec, &nsec);
1109
1110 if (VM_SWAP_SHOULD_CREATE(sec) && !vm_swapfile_mgmt_thread_running)
1111 thread_wakeup((event_t) &vm_swapfile_mgmt_needed);
1112
1113 if (hibernate_flushing == FALSE || VM_SWAP_SHOULD_CREATE(sec)) {
1114 waiting = TRUE;
1115 assert_wait_timeout((event_t) &vm_num_swap_files, THREAD_INTERRUPTIBLE, 1000, 1000*NSEC_PER_USEC);
1116 } else
1117 hibernate_no_swapspace = TRUE;
1118
1119 lck_mtx_unlock(&vm_swap_data_lock);
1120
1121 if (waiting == TRUE)
1122 thread_block(THREAD_CONTINUE_NULL);
1123
1124 return KERN_FAILURE;
1125
1126done:
1127 error = vm_swapfile_io(swf->swp_vp, file_offset, addr, (int) (size / PAGE_SIZE_64), SWAP_WRITE);
1128
1129 lck_mtx_lock(&vm_swap_data_lock);
1130
1131 swf->swp_csegs[segidx] = c_seg;
1132
1133 swf->swp_io_count--;
1134
1135 *f_offset = (swapfile_index << SWAP_DEVICE_SHIFT) | file_offset;
1136
1137 if ((swf->swp_flags & SWAP_WANTED) && swf->swp_io_count == 0) {
1138
1139 swf->swp_flags &= ~SWAP_WANTED;
1140 thread_wakeup((event_t) &swf->swp_flags);
1141 }
1142
1143 lck_mtx_unlock(&vm_swap_data_lock);
1144
1145#if SANITY_CHECK_SWAP_ROUTINES
1146 printf("Returned 0x%llx as offset\n", *f_offset);
1147#endif /* SANITY_CHECK_SWAP_ROUTINES */
1148
1149 if (error) {
1150 vm_swap_free(*f_offset);
1151
1152 return KERN_FAILURE;
1153 }
1154 return KERN_SUCCESS;
1155}
1156
1157
1158
1159static void
1160vm_swap_free_now(struct swapfile *swf, uint64_t f_offset)
1161{
1162 uint64_t file_offset = 0;
1163 unsigned int segidx = 0;
1164
1165
1166 if ((swf->swp_flags & SWAP_READY) || (swf->swp_flags & SWAP_RECLAIM)) {
1167
1168 unsigned int byte_for_segidx = 0;
1169 unsigned int offset_within_byte = 0;
1170
1171 file_offset = (f_offset & SWAP_SLOT_MASK);
1172 segidx = (unsigned int) (file_offset / COMPRESSED_SWAP_CHUNK_SIZE);
1173
1174 byte_for_segidx = segidx >> 3;
1175 offset_within_byte = segidx % 8;
1176
1177 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1178
1179 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1180
1181 swf->swp_csegs[segidx] = NULL;
1182
1183 swf->swp_nseginuse--;
1184 vm_swapfile_total_segs_used--;
1185
1186 if (segidx < swf->swp_free_hint) {
1187 swf->swp_free_hint = segidx;
1188 }
1189 }
1190 if (VM_SWAP_SHOULD_RECLAIM() && !vm_swapfile_mgmt_thread_running)
1191 thread_wakeup((event_t) &vm_swapfile_mgmt_needed);
1192 }
1193 lck_mtx_unlock(&vm_swap_data_lock);
1194}
1195
1196
1197uint32_t vm_swap_free_now_count = 0;
1198uint32_t vm_swap_free_delayed_count = 0;
1199
1200
1201void
1202vm_swap_free(uint64_t f_offset)
1203{
1204 struct swapfile *swf = NULL;
1205 struct trim_list *tl;
1206 clock_sec_t sec;
1207 clock_nsec_t nsec;
1208
1209 lck_mtx_lock(&vm_swap_data_lock);
1210
1211 swf = vm_swapfile_for_handle(f_offset);
1212
1213 if (swf && (swf->swp_flags & (SWAP_READY | SWAP_RECLAIM))) {
1214
1215 if (swf->swp_trim_supported == FALSE || (swf->swp_flags & SWAP_RECLAIM)) {
1216 /*
1217 * don't delay the free if the underlying disk doesn't support
1218 * trim, or we're in the midst of reclaiming this swap file since
1219 * we don't want to move segments that are technically free
1220 * but not yet handled by the delayed free mechanism
1221 */
1222 vm_swap_free_now(swf, f_offset);
1223
1224 vm_swap_free_now_count++;
1225 return;
1226 }
1227 tl = kalloc(sizeof(struct trim_list));
1228
1229 tl->tl_offset = f_offset & SWAP_SLOT_MASK;
1230 tl->tl_length = COMPRESSED_SWAP_CHUNK_SIZE;
1231
1232 tl->tl_next = swf->swp_delayed_trim_list_head;
1233 swf->swp_delayed_trim_list_head = tl;
1234 swf->swp_delayed_trim_count++;
1235
1236 if (VM_SWAP_SHOULD_TRIM(swf) && !vm_swapfile_mgmt_thread_running) {
1237 clock_get_system_nanotime(&sec, &nsec);
1238
1239 if (sec > dont_trim_until_ts)
1240 thread_wakeup((event_t) &vm_swapfile_mgmt_needed);
1241 }
1242 vm_swap_free_delayed_count++;
1243 }
1244 lck_mtx_unlock(&vm_swap_data_lock);
1245}
1246
1247
1248static void
1249vm_swap_handle_delayed_trims(boolean_t force_now)
1250{
1251 struct swapfile *swf = NULL;
1252
1253 /*
1254 * because swap files are created or reclaimed on the
1255 * same thread that calls this function, it's safe
1256 * to iterate "swf_global_queue" w/o holding
1257 * the lock since those are the only 2 cases that can
1258 * change the items on the "swf_global_queue"
1259 */
1260 swf = (struct swapfile*) queue_first(&swf_global_queue);
1261
1262 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1263
1264 assert(!(swf->swp_flags & SWAP_RECLAIM));
1265
1266 if ((swf->swp_flags & SWAP_READY) && (force_now == TRUE || VM_SWAP_SHOULD_TRIM(swf)))
1267 vm_swap_do_delayed_trim(swf);
1268
1269 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1270 }
1271}
1272
1273
1274static void
1275vm_swap_do_delayed_trim(struct swapfile *swf)
1276{
1277 struct trim_list *tl, *tl_head;
1278
1279 lck_mtx_lock(&vm_swap_data_lock);
1280
1281 tl_head = swf->swp_delayed_trim_list_head;
1282 swf->swp_delayed_trim_list_head = NULL;
1283 swf->swp_delayed_trim_count = 0;
1284
1285 lck_mtx_unlock(&vm_swap_data_lock);
1286
1287 vnode_trim_list(swf->swp_vp, tl_head);
1288
1289 while ((tl = tl_head) != NULL) {
1290 unsigned int segidx = 0;
1291 unsigned int byte_for_segidx = 0;
1292 unsigned int offset_within_byte = 0;
1293
1294 lck_mtx_lock(&vm_swap_data_lock);
1295
1296 segidx = (unsigned int) (tl->tl_offset / COMPRESSED_SWAP_CHUNK_SIZE);
1297
1298 byte_for_segidx = segidx >> 3;
1299 offset_within_byte = segidx % 8;
1300
1301 if ((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) {
1302
1303 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1304
1305 swf->swp_csegs[segidx] = NULL;
1306
1307 swf->swp_nseginuse--;
1308 vm_swapfile_total_segs_used--;
1309
1310 if (segidx < swf->swp_free_hint) {
1311 swf->swp_free_hint = segidx;
1312 }
1313 }
1314 lck_mtx_unlock(&vm_swap_data_lock);
1315
1316 tl_head = tl->tl_next;
1317
1318 kfree(tl, sizeof(struct trim_list));
1319 }
1320}
1321
1322
1323void
1324vm_swap_flush()
1325{
1326 return;
1327}
1328
1329int vm_swap_reclaim_yielded = 0;
1330
1331void
1332vm_swap_reclaim(void)
1333{
1334 vm_offset_t addr = 0;
1335 unsigned int segidx = 0;
1336 uint64_t f_offset = 0;
1337 struct swapfile *swf = NULL;
1338 struct swapfile *smallest_swf = NULL;
1339 unsigned int min_nsegs = 0;
1340 unsigned int byte_for_segidx = 0;
1341 unsigned int offset_within_byte = 0;
1342 uint32_t c_size = 0;
1343
1344 c_segment_t c_seg = NULL;
1345
1346 if (kernel_memory_allocate(kernel_map, (vm_offset_t *)(&addr), C_SEG_BUFSIZE, 0, KMA_KOBJECT) != KERN_SUCCESS) {
1347 panic("vm_swap_reclaim: kernel_memory_allocate failed\n");
1348 }
1349
1350 lck_mtx_lock(&vm_swap_data_lock);
1351
1352 swf = (struct swapfile*) queue_first(&swf_global_queue);
1353 min_nsegs = MAX_SWAP_FILE_SIZE / COMPRESSED_SWAP_CHUNK_SIZE;
1354 smallest_swf = NULL;
1355
1356 while (queue_end(&swf_global_queue, (queue_entry_t)swf) == FALSE) {
1357
1358 if ((swf->swp_flags & SWAP_READY) && (swf->swp_nseginuse <= min_nsegs)) {
1359
1360 smallest_swf = swf;
1361 min_nsegs = swf->swp_nseginuse;
1362 }
1363 swf = (struct swapfile*) queue_next(&swf->swp_queue);
1364 }
1365
1366 if (smallest_swf == NULL)
1367 goto done;
1368
1369 swf = smallest_swf;
1370
1371
1372 swf->swp_flags &= ~SWAP_READY;
1373 swf->swp_flags |= SWAP_RECLAIM;
1374
1375 if (swf->swp_delayed_trim_count) {
1376
1377 lck_mtx_unlock(&vm_swap_data_lock);
1378
1379 vm_swap_do_delayed_trim(swf);
1380
1381 lck_mtx_lock(&vm_swap_data_lock);
1382 }
1383 segidx = 0;
1384
1385 while (segidx < swf->swp_nsegs) {
1386
1387ReTry_for_cseg:
1388 if (compressor_store_stop_compaction == TRUE || (swf->swp_trim_supported == FALSE && VM_SWAP_BUSY())) {
1389 vm_swap_reclaim_yielded++;
1390 break;
1391 }
1392 /*
1393 * Wait for outgoing I/Os.
1394 */
1395 while (swf->swp_io_count) {
1396
1397 swf->swp_flags |= SWAP_WANTED;
1398
1399 assert_wait((event_t) &swf->swp_flags, THREAD_UNINT);
1400 lck_mtx_unlock(&vm_swap_data_lock);
1401
1402 thread_block(THREAD_CONTINUE_NULL);
1403
1404 lck_mtx_lock(&vm_swap_data_lock);
1405 }
1406
1407 byte_for_segidx = segidx >> 3;
1408 offset_within_byte = segidx % 8;
1409
1410 if (((swf->swp_bitmap)[byte_for_segidx] & (1 << offset_within_byte)) == 0) {
1411
1412 segidx++;
1413 continue;
1414 }
1415
1416 c_seg = swf->swp_csegs[segidx];
1417
1418 lck_mtx_lock_spin_always(&c_seg->c_lock);
1419
1420 assert(c_seg->c_ondisk);
1421
1422 if (c_seg->c_busy) {
1423
1424 c_seg->c_wanted = 1;
1425
1426 assert_wait((event_t) (c_seg), THREAD_UNINT);
1427 lck_mtx_unlock_always(&c_seg->c_lock);
1428
1429 lck_mtx_unlock(&vm_swap_data_lock);
1430
1431 thread_block(THREAD_CONTINUE_NULL);
1432
1433 lck_mtx_lock(&vm_swap_data_lock);
1434
1435 goto ReTry_for_cseg;
1436 }
1437 (swf->swp_bitmap)[byte_for_segidx] &= ~(1 << offset_within_byte);
1438
1439 f_offset = segidx * COMPRESSED_SWAP_CHUNK_SIZE;
1440
1441 swf->swp_csegs[segidx] = NULL;
1442 swf->swp_nseginuse--;
1443
1444 vm_swapfile_total_segs_used--;
1445
1446 lck_mtx_unlock(&vm_swap_data_lock);
1447
1448 if (c_seg->c_must_free) {
1449
1450 c_seg_free(c_seg);
1451 } else {
1452
1453 c_seg->c_busy = 1;
1454 c_seg->c_busy_swapping = 1;
1455#if !CHECKSUM_THE_SWAP
1456 c_seg_trim_tail(c_seg);
1457#endif
1458
1459#if SANITY_CHECK_SWAP_ROUTINES
1460
1461 c_size = COMPRESSED_SWAP_CHUNK_SIZE;
1462
1463#else /* SANITY_CHECK_SWAP_ROUTINES */
1464
1465 c_size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
1466
1467 assert(c_size <= C_SEG_BUFSIZE);
1468
1469#endif /* SANITY_CHECK_SWAP_ROUTINES */
1470
1471 lck_mtx_unlock_always(&c_seg->c_lock);
1472
1473 if (vm_swapfile_io(swf->swp_vp, f_offset, addr, (int)(c_size / PAGE_SIZE_64), SWAP_READ)) {
1474
1475 /*
1476 * reading the data back in failed, so convert c_seg
1477 * to a swapped in c_segment that contains no data
1478 */
1479 c_seg->c_store.c_buffer = (int32_t *)NULL;
1480 c_seg_swapin_requeue(c_seg);
1481
1482 goto swap_io_failed;
1483 }
1484 VM_STAT_INCR_BY(swapins, c_size >> PAGE_SHIFT);
1485
1486 if (vm_swap_put(addr, &f_offset, c_size, c_seg)) {
1487 vm_offset_t c_buffer;
1488
1489 /*
1490 * the put failed, so convert c_seg to a fully swapped in c_segment
1491 * with valid data
1492 */
1493 if (kernel_memory_allocate(kernel_map, &c_buffer, C_SEG_ALLOCSIZE, 0, KMA_COMPRESSOR | KMA_VAONLY) != KERN_SUCCESS)
1494 panic("vm_swap_reclaim: kernel_memory_allocate failed\n");
1495 kernel_memory_populate(kernel_map, c_buffer, c_size, KMA_COMPRESSOR);
1496
1497 memcpy((char *)c_buffer, (char *)addr, c_size);
1498
1499 c_seg->c_store.c_buffer = (int32_t *)c_buffer;
1500#if CRYPTO
1501 vm_swap_decrypt(c_seg);
1502#endif /* CRYPTO */
1503 c_seg_swapin_requeue(c_seg);
1504
1505 OSAddAtomic64(c_seg->c_bytes_used, &compressor_bytes_used);
1506
1507 goto swap_io_failed;
1508 }
1509 VM_STAT_INCR_BY(swapouts, c_size >> PAGE_SHIFT);
1510
1511 lck_mtx_lock_spin_always(&c_seg->c_lock);
1512
1513 assert(c_seg->c_ondisk);
1514 /*
1515 * The c_seg will now know about the new location on disk.
1516 */
1517 c_seg->c_store.c_swap_handle = f_offset;
1518swap_io_failed:
1519 c_seg->c_busy_swapping = 0;
1520
1521 if (c_seg->c_must_free)
1522 c_seg_free(c_seg);
1523 else {
1524 C_SEG_WAKEUP_DONE(c_seg);
1525
1526 lck_mtx_unlock_always(&c_seg->c_lock);
1527 }
1528 }
1529 lck_mtx_lock(&vm_swap_data_lock);
1530 }
1531
1532 if (swf->swp_nseginuse) {
1533
1534 swf->swp_flags &= ~SWAP_RECLAIM;
1535 swf->swp_flags |= SWAP_READY;
1536
1537 goto done;
1538 }
1539 /*
1540 * We don't remove this inactive swf from the queue.
1541 * That way, we can re-use it when needed again and
1542 * preserve the namespace.
1543 */
1544 //queue_remove(&swf_global_queue, swf, struct swapfile*, swp_queue);
1545
1546 vm_num_swap_files--;
1547
1548 vm_swapfile_total_segs_alloced -= swf->swp_nsegs;
1549
1550 lck_mtx_unlock(&vm_swap_data_lock);
1551
1552 vm_swapfile_close((uint64_t)(swf->swp_path), swf->swp_vp);
1553
1554 kfree(swf->swp_csegs, swf->swp_nsegs * sizeof(c_segment_t));
1555 kfree(swf->swp_bitmap, MAX((swf->swp_nsegs >> 3), 1));
1556
1557 lck_mtx_lock(&vm_swap_data_lock);
1558
1559 swf->swp_vp = NULL;
1560 swf->swp_size = 0;
1561 swf->swp_free_hint = 0;
1562 swf->swp_nsegs = 0;
1563 swf->swp_flags = SWAP_REUSE;
1564
1565 thread_wakeup((event_t) &swf->swp_flags);
1566done:
1567 lck_mtx_unlock(&vm_swap_data_lock);
1568
1569 kmem_free(kernel_map, (vm_offset_t) addr, C_SEG_BUFSIZE);
1570}
1571
1572
1573uint64_t
1574vm_swap_get_total_space(void)
1575{
1576 uint64_t total_space = 0;
1577
1578 total_space = (uint64_t)vm_swapfile_total_segs_alloced * COMPRESSED_SWAP_CHUNK_SIZE;
1579
1580 return total_space;
1581}
1582
1583uint64_t
1584vm_swap_get_used_space(void)
1585{
1586 uint64_t used_space = 0;
1587
1588 used_space = (uint64_t)vm_swapfile_total_segs_used * COMPRESSED_SWAP_CHUNK_SIZE;
1589
1590 return used_space;
1591}
1592
1593uint64_t
1594vm_swap_get_free_space(void)
1595{
1596 return (vm_swap_get_total_space() - vm_swap_get_used_space());
1597}
mirror of XNU