3 * Copyright (c) 2002,2000 Apple Computer, Inc. All rights reserved.
5 * @APPLE_LICENSE_HEADER_START@
7 * Copyright (c) 1999-2003 Apple Computer, Inc. All Rights Reserved.
9 * This file contains Original Code and/or Modifications of Original Code
10 * as defined in and that are subject to the Apple Public Source License
11 * Version 2.0 (the 'License'). You may not use this file except in
12 * compliance with the License. Please obtain a copy of the License at
13 * http://www.opensource.apple.com/apsl/ and read it before using this
16 * The Original Code and all software distributed under the License are
17 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
18 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
19 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
21 * Please see the License for the specific language governing rights and
22 * limitations under the License.
24 * @APPLE_LICENSE_HEADER_END@
29 * File: vm/task_working_set.c
30 * Author: Chris Youngworth
33 * Working set detection and maintainence module
37 #include <mach/mach_types.h>
38 #include <mach/shared_memory_server.h>
39 #include <vm/task_working_set.h>
40 #include <vm/vm_kern.h>
41 #include <vm/vm_map.h>
42 #include <vm/vm_page.h>
43 #include <vm/vm_pageout.h>
44 #include <kern/sched.h>
46 extern unsigned sched_tick
;
47 extern zone_t lsf_zone
;
49 /* declarations for internal use only routines */
52 tws_create_startup_list(
56 tws_startup_list_lookup(
57 tws_startup_t startup
,
61 tws_internal_startup_send(
65 tws_traverse_address_hash_list (
68 vm_offset_t page_addr
,
70 vm_object_offset_t offset
,
72 tws_hash_ptr_t
*target_ele
,
73 tws_hash_ptr_t
**previous_ptr
,
74 tws_hash_ptr_t
**free_list
,
75 unsigned int exclusive_addr
);
78 tws_traverse_object_hash_list (
82 vm_object_offset_t offset
,
83 unsigned int page_mask
,
84 tws_hash_ptr_t
*target_ele
,
85 tws_hash_ptr_t
**previous_ptr
,
86 tws_hash_ptr_t
**free_list
);
100 int tws_test_for_community(
103 vm_object_offset_t offset
,
104 unsigned int threshold
,
105 unsigned int *page_mask
);
107 /* Note: all of the routines below depend on the associated map lock for */
108 /* synchronization, the map lock will be on when the routines are called */
109 /* and on when they return */
121 if ((style
!= TWS_HASH_STYLE_BASIC
) &&
122 (style
!= TWS_HASH_STYLE_BASIC
)) {
123 return((tws_hash_t
)NULL
);
127 tws
= (tws_hash_t
)(kalloc(sizeof(struct tws_hash
)));
128 if(tws
== (tws_hash_t
)NULL
)
131 if((tws
->table
[0] = (tws_hash_ptr_t
*)
132 kalloc(sizeof(tws_hash_ptr_t
) * lines
* rows
))
134 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
135 return (tws_hash_t
)NULL
;
137 if((tws
->table_ele
[0] = (tws_hash_ptr_t
)
138 kalloc(sizeof(struct tws_hash_ptr
) * lines
* rows
))
140 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ptr_t
)
142 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
143 return (tws_hash_t
)NULL
;
145 if((tws
->alt_ele
[0] = (tws_hash_ptr_t
)
146 kalloc(sizeof(struct tws_hash_ptr
) * lines
* rows
))
148 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ptr_t
)
150 kfree((vm_offset_t
)tws
->table_ele
[0],
151 sizeof(struct tws_hash_ptr
)
153 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
154 return (tws_hash_t
)NULL
;
156 if((tws
->cache
[0] = (struct tws_hash_line
*)
157 kalloc(sizeof(struct tws_hash_line
) * lines
))
159 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ptr_t
)
161 kfree((vm_offset_t
)tws
->table_ele
[0],
162 sizeof(struct tws_hash_ptr
)
164 kfree((vm_offset_t
)tws
->alt_ele
[0], sizeof(struct tws_hash_ptr
)
166 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
167 return (tws_hash_t
)NULL
;
169 tws
->free_hash_ele
[0] = (tws_hash_ptr_t
)0;
170 tws
->obj_free_count
[0] = 0;
171 tws
->addr_free_count
[0] = 0;
173 /* most defaults are such that a bzero will initialize */
174 bzero((char *)tws
->table
[0],sizeof(tws_hash_ptr_t
)
176 bzero((char *)tws
->table_ele
[0],sizeof(struct tws_hash_ptr
)
178 bzero((char *)tws
->alt_ele
[0],sizeof(struct tws_hash_ptr
)
180 bzero((char *)tws
->cache
[0], sizeof(struct tws_hash_line
)
183 mutex_init(&tws
->lock
, ETAP_VM_MAP
);
185 tws
->current_line
= 0;
186 tws
->pageout_count
= 0;
188 tws
->startup_cache
= NULL
;
189 tws
->startup_name
= NULL
;
190 tws
->number_of_lines
= lines
;
191 tws
->number_of_elements
= rows
;
192 tws
->expansion_count
= 1;
193 tws
->lookup_count
= 0;
194 tws
->insert_count
= 0;
195 tws
->time_of_creation
= sched_tick
;
204 tws_hash_line_t hash_line
,
207 struct tws_hash_ele
*hash_ele
;
208 struct tws_hash_ptr
**trailer
;
209 struct tws_hash_ptr
**free_list
;
210 tws_hash_ele_t addr_ele
;
212 unsigned int i
, j
, k
;
217 if(tws
->line_count
< tws
->number_of_lines
) {
221 if(tws
->pageout_count
!= vm_pageout_scan_event_counter
) {
223 vm_pageout_scan_event_counter
;
230 hash_line
->ele_count
= 0;
232 for (i
=0; i
<tws
->number_of_elements
; i
++) {
234 hash_ele
= &(hash_line
->list
[i
]);
235 if(hash_ele
->object
!= 0) {
237 vm_object_offset_t local_off
= 0;
238 tws_hash_ptr_t cache_ele
;
240 index
= alt_tws_hash(
241 hash_ele
->page_addr
& TWS_HASH_OFF_MASK
,
242 tws
->number_of_elements
,
243 tws
->number_of_lines
);
245 tws_traverse_address_hash_list(tws
, index
,
246 hash_ele
->page_addr
, hash_ele
->object
,
247 hash_ele
->offset
, hash_ele
->map
,
248 &cache_ele
, &trailer
, &free_list
, 0);
249 if(cache_ele
!= NULL
) {
250 addr_ele
= (tws_hash_ele_t
)((unsigned int)
251 (cache_ele
->element
) & ~TWS_ADDR_HASH
);
252 if(addr_ele
!= hash_ele
)
253 panic("tws_hash_line_clear:"
255 cache_ele
->element
= 0;
256 *trailer
= cache_ele
->next
;
257 cache_ele
->next
= *free_list
;
258 *free_list
= cache_ele
;
261 index
= alt_tws_hash(
262 (hash_ele
->page_addr
- 0x1f000)
264 tws
->number_of_elements
,
265 tws
->number_of_lines
);
267 tws_traverse_address_hash_list(tws
, index
,
268 hash_ele
->page_addr
, hash_ele
->object
,
269 hash_ele
->offset
, hash_ele
->map
,
270 &cache_ele
, &trailer
, &free_list
, 0);
272 if(cache_ele
!= NULL
) {
273 addr_ele
= (tws_hash_ele_t
)((unsigned int)
274 (cache_ele
->element
) & ~TWS_ADDR_HASH
);
275 if(addr_ele
!= hash_ele
)
276 panic("tws_hash_line_clear: "
278 cache_ele
->element
= 0;
279 *trailer
= cache_ele
->next
;
280 cache_ele
->next
= *free_list
;
281 *free_list
= cache_ele
;
285 if((hash_ele
->map
!= NULL
) && (live
)) {
288 for (j
= 0x1; j
!= 0; j
= j
<<1) {
289 if(j
& hash_ele
->page_cache
) {
290 p
= vm_page_lookup(hash_ele
->object
,
291 hash_ele
->offset
+ local_off
);
292 if((p
!= NULL
) && (p
->wire_count
== 0)
293 && (dump_pmap
== 1)) {
294 pmap_remove_some_phys((pmap_t
)
300 local_off
+= PAGE_SIZE_64
;
304 if(tws
->style
== TWS_HASH_STYLE_SIGNAL
) {
305 vm_object_deallocate(hash_ele
->object
);
306 vm_map_deallocate(hash_ele
->map
);
309 index
= do_tws_hash(hash_ele
->object
, hash_ele
->offset
,
310 tws
->number_of_elements
,
311 tws
->number_of_lines
);
313 tws_traverse_object_hash_list(tws
,
314 index
, hash_ele
->object
, hash_ele
->offset
,
315 0xFFFFFFFF, &cache_ele
, &trailer
, &free_list
);
316 if((cache_ele
!= NULL
) && (cache_ele
->element
== hash_ele
)) {
317 cache_ele
->element
= 0;
318 *trailer
= cache_ele
->next
;
319 cache_ele
->next
= *free_list
;
320 *free_list
= cache_ele
;
322 hash_ele
->object
= 0;
330 vm_object_offset_t offset
,
332 tws_hash_line_t
*line
)
339 tws_hash_ptr_t cache_ele
;
340 tws_hash_ptr_t
*trailer
;
341 tws_hash_ptr_t
*free_list
;
343 /* don't cache private objects */
347 index
= do_tws_hash(object
, offset
,
348 tws
->number_of_elements
, tws
->number_of_lines
);
352 if(tws
->lookup_count
== 0)
353 tws
->insert_count
= 0;
354 if(tws
->startup_name
!= NULL
) {
356 age_of_cache
= ((sched_tick
357 - tws
->time_of_creation
) >> SCHED_TICK_SHIFT
);
358 if (age_of_cache
> 35) {
359 return KERN_OPERATION_TIMED_OUT
;
363 if(tws
->lookup_count
> (4 * tws
->expansion_count
364 * tws
->number_of_elements
* tws
->number_of_lines
) &&
365 (tws
->lookup_count
> (2 * tws
->insert_count
))) {
366 if(tws
->startup_cache
) {
368 age_of_cache
= ((sched_tick
369 - tws
->time_of_creation
) >> SCHED_TICK_SHIFT
);
370 if (age_of_cache
> 60) {
371 return KERN_OPERATION_TIMED_OUT
;
376 pagenum
= (vm_offset_t
)(offset
& TWS_INDEX_MASK
);
377 pagenum
= pagenum
>> 12;
378 pagenum
= 1 << pagenum
; /* get the appropriate page in 32 page block */
379 tws_traverse_object_hash_list(tws
, index
, object
, offset
, pagenum
,
380 &cache_ele
, &trailer
, &free_list
);
381 if(cache_ele
!= NULL
) {
382 set
= cache_ele
->element
->line
/tws
->number_of_lines
;
383 ele_line
= cache_ele
->element
->line
- set
;
384 *line
= &tws
->cache
[set
][ele_line
];
396 vm_object_offset_t offset
,
398 tws_hash_line_t
*line
)
402 if(!tws_lock_try(tws
)) {
405 kr
= tws_internal_lookup(tws
,
406 offset
, object
, line
);
412 tws_expand_working_set(
420 struct tws_hash temp
;
422 old_tws
= (tws_hash_t
)tws
;
424 /* Note we do an elaborate dance to preserve the header that */
425 /* task is pointing to. In this way we can avoid taking a task */
426 /* lock every time we want to access the tws */
428 if (old_tws
->number_of_lines
>= line_count
) {
431 if((new_tws
= tws_hash_create(line_count
,
432 old_tws
->number_of_elements
, old_tws
->style
)) == 0) {
433 return(KERN_NO_SPACE
);
438 for(i
= 0; i
<old_tws
->number_of_lines
; i
++) {
439 for(j
= 0; j
<old_tws
->number_of_elements
; j
++) {
440 for(k
= 0; k
<old_tws
->expansion_count
; k
++) {
441 tws_hash_ele_t entry
;
442 vm_object_offset_t paddr
;
443 unsigned int page_index
;
444 entry
= &old_tws
->cache
[k
][i
].list
[j
];
445 if(entry
->object
!= 0) {
447 for(page_index
= 1; page_index
!= 0;
448 page_index
= page_index
<< 1); {
449 if (entry
->page_cache
& page_index
) {
453 entry
->page_addr
+paddr
,
465 temp
.style
= new_tws
->style
;
466 temp
.current_line
= new_tws
->current_line
;
467 temp
.pageout_count
= new_tws
->pageout_count
;
468 temp
.line_count
= new_tws
->line_count
;
469 temp
.number_of_lines
= new_tws
->number_of_lines
;
470 temp
.number_of_elements
= new_tws
->number_of_elements
;
471 temp
.expansion_count
= new_tws
->expansion_count
;
472 temp
.lookup_count
= new_tws
->lookup_count
;
473 temp
.insert_count
= new_tws
->insert_count
;
474 for(i
= 0; i
<new_tws
->expansion_count
; i
++) {
475 temp
.obj_free_count
[i
] = new_tws
->obj_free_count
[i
];
476 temp
.addr_free_count
[i
] = new_tws
->addr_free_count
[i
];
477 temp
.free_hash_ele
[i
] = new_tws
->free_hash_ele
[i
];
478 temp
.table
[i
] = new_tws
->table
[i
];
479 temp
.table_ele
[i
] = new_tws
->table_ele
[i
];
480 temp
.alt_ele
[i
] = new_tws
->alt_ele
[i
];
481 temp
.cache
[i
] = new_tws
->cache
[i
];
484 new_tws
->style
= old_tws
->style
;
485 new_tws
->current_line
= old_tws
->current_line
;
486 new_tws
->pageout_count
= old_tws
->pageout_count
;
487 new_tws
->line_count
= old_tws
->line_count
;
488 new_tws
->number_of_lines
= old_tws
->number_of_lines
;
489 new_tws
->number_of_elements
= old_tws
->number_of_elements
;
490 new_tws
->expansion_count
= old_tws
->expansion_count
;
491 new_tws
->lookup_count
= old_tws
->lookup_count
;
492 new_tws
->insert_count
= old_tws
->insert_count
;
493 for(i
= 0; i
<old_tws
->expansion_count
; i
++) {
494 new_tws
->obj_free_count
[i
] = old_tws
->obj_free_count
[i
];
495 new_tws
->addr_free_count
[i
] = old_tws
->addr_free_count
[i
];
496 new_tws
->free_hash_ele
[i
] = old_tws
->free_hash_ele
[i
];
497 new_tws
->table
[i
] = old_tws
->table
[i
];
498 new_tws
->table_ele
[i
] = old_tws
->table_ele
[i
];
499 new_tws
->alt_ele
[i
] = old_tws
->alt_ele
[i
];
500 new_tws
->cache
[i
] = old_tws
->cache
[i
];
503 old_tws
->style
= temp
.style
;
504 old_tws
->current_line
= temp
.current_line
;
505 old_tws
->pageout_count
= temp
.pageout_count
;
506 old_tws
->line_count
= temp
.line_count
;
507 old_tws
->number_of_lines
= temp
.number_of_lines
;
508 old_tws
->number_of_elements
= temp
.number_of_elements
;
509 old_tws
->expansion_count
= temp
.expansion_count
;
510 old_tws
->lookup_count
= temp
.lookup_count
;
511 old_tws
->insert_count
= temp
.insert_count
;
512 for(i
= 0; i
<temp
.expansion_count
; i
++) {
513 old_tws
->obj_free_count
[i
] = temp
.obj_free_count
[i
];;
514 old_tws
->addr_free_count
[i
] = temp
.addr_free_count
[i
];;
515 old_tws
->free_hash_ele
[i
] = NULL
;
516 old_tws
->table
[i
] = temp
.table
[i
];
517 old_tws
->table_ele
[i
] = temp
.table_ele
[i
];
518 old_tws
->alt_ele
[i
] = temp
.alt_ele
[i
];
519 old_tws
->cache
[i
] = temp
.cache
[i
];
522 tws_hash_destroy(new_tws
);
527 tws_hash_t test_tws
= 0;
532 vm_object_offset_t offset
,
534 vm_offset_t page_addr
,
539 unsigned int alt_index
;
540 unsigned int index_enum
[2];
541 unsigned int ele_index
;
542 tws_hash_ptr_t cache_ele
;
543 tws_hash_ptr_t obj_ele
= NULL
;
544 tws_hash_ptr_t addr_ele
= NULL
;
545 tws_hash_ptr_t
*trailer
;
546 tws_hash_ptr_t
*free_list
;
547 tws_hash_ele_t target_element
= NULL
;
552 unsigned int startup_cache_line
;
553 vm_offset_t startup_page_addr
;
555 int ask_for_startup_cache_release
= 0;
558 if(!tws_lock_try(tws
)) {
562 current_line
= 0xFFFFFFFF;
564 startup_cache_line
= 0;
566 page_addr
- (offset
- (offset
& TWS_HASH_OFF_MASK
));
567 if(tws
->startup_cache
) {
569 age_of_cache
= ((sched_tick
- tws
->time_of_creation
)
570 >> SCHED_TICK_SHIFT
);
571 startup_cache_line
= tws_startup_list_lookup(
572 tws
->startup_cache
, startup_page_addr
);
573 if(tws
== test_tws
) {
574 printf("cache_lookup, result = 0x%x, addr = 0x%x, object 0x%x, offset 0x%x%x\n", startup_cache_line
, startup_page_addr
, object
, offset
);
576 if(age_of_cache
> 60) {
577 ask_for_startup_cache_release
= 1;
580 /* This next bit of code, the and alternate hash */
581 /* are all made necessary because of IPC COW */
583 /* Note: the use of page_addr modified by delta from offset */
584 /* frame base means we may miss some previous entries. However */
585 /* we will not miss the present entry. This is most important */
586 /* in avoiding duplication of entries against long lived non-cow */
588 index_enum
[0] = alt_tws_hash(
589 page_addr
& TWS_HASH_OFF_MASK
,
590 tws
->number_of_elements
, tws
->number_of_lines
);
592 index_enum
[1] = alt_tws_hash(
593 (page_addr
- 0x1f000) & TWS_HASH_OFF_MASK
,
594 tws
->number_of_elements
, tws
->number_of_lines
);
596 for(ctr
= 0; ctr
< 2;) {
597 tws_hash_ele_t resident
;
598 tws_traverse_address_hash_list(tws
,
599 index_enum
[ctr
], page_addr
, NULL
,
601 &cache_ele
, &trailer
, &free_list
, 1);
602 if(cache_ele
!= NULL
) {
604 resident
= (tws_hash_ele_t
)((unsigned int)
605 cache_ele
->element
& ~TWS_ADDR_HASH
);
606 if((object
== resident
->object
) &&
608 (offset
& TWS_HASH_OFF_MASK
)) {
609 /* This is our object/offset */
611 |= startup_cache_line
;
612 resident
->page_cache
|=
614 (offset
& TWS_INDEX_MASK
))>>12));
616 if(ask_for_startup_cache_release
)
617 return KERN_OPERATION_TIMED_OUT
;
620 if((object
->shadow
==
623 + object
->shadow_offset
)
624 == (offset
& TWS_HASH_OFF_MASK
))) {
625 /* if we just shadowed, inherit */
626 /* access pattern from parent */
627 startup_cache_line
|=
628 resident
->page_cache
;
629 /* thow out old entry */
630 resident
->page_cache
= 0;
633 resident
->page_cache
&=
634 ~(1<<(((vm_offset_t
)(page_addr
635 - resident
->page_addr
))
638 /* Throw out old entry if there are no */
639 /* more pages in cache */
640 if(resident
->page_cache
== 0) {
641 /* delete addr hash entry */
642 cache_ele
->element
= 0;
643 *trailer
= cache_ele
->next
;
644 cache_ele
->next
= *free_list
;
645 *free_list
= cache_ele
;
646 /* go after object hash */
650 tws
->number_of_elements
,
651 tws
->number_of_lines
);
652 tws_traverse_object_hash_list(tws
,
653 index
, resident
->object
,
655 0xFFFFFFFF, &cache_ele
,
656 &trailer
, &free_list
);
657 if(cache_ele
!= NULL
) {
659 TWS_HASH_STYLE_SIGNAL
) {
660 vm_object_deallocate(
661 cache_ele
->element
->object
);
663 cache_ele
->element
->map
);
666 cache_ele
->element
->line
;
668 /tws
->number_of_lines
;
669 current_line
-= set
*
670 tws
->number_of_lines
;
671 if(cache_ele
->element
->object
!= 0) {
672 cache_ele
->element
->object
= 0;
674 [current_line
].ele_count
--;
676 cache_ele
->element
= 0;
677 *trailer
= cache_ele
->next
;
678 cache_ele
->next
= *free_list
;
679 *free_list
= cache_ele
;
688 * We may or may not have a current line setting coming out of
689 * the code above. If we have a current line it means we can
690 * choose to back-fill the spot vacated by a previous entry.
691 * We have yet to do a definitive check using the original obj/off
692 * We will do that now and override the current line if we
696 index
= do_tws_hash(object
, offset
,
697 tws
->number_of_elements
, tws
->number_of_lines
);
699 alt_index
= index_enum
[0];
701 tws_traverse_object_hash_list(tws
, index
, object
, offset
,
702 0xFFFFFFFF, &cache_ele
, &trailer
, &free_list
);
703 if(cache_ele
!= NULL
) {
705 current_line
= cache_ele
->element
->line
;
706 set
= current_line
/tws
->number_of_lines
;
707 current_line
-= set
* tws
->number_of_lines
;
708 target_element
= cache_ele
->element
;
710 /* Now check to see if we have a hash addr for it */
711 tws_traverse_address_hash_list(tws
,
712 alt_index
, obj_ele
->element
->page_addr
,
713 obj_ele
->element
->object
,
714 obj_ele
->element
->offset
,
715 obj_ele
->element
->map
,
716 &cache_ele
, &trailer
, &free_list
, 0);
717 if(cache_ele
!= NULL
) {
718 addr_ele
= cache_ele
;
720 addr_ele
= new_addr_hash(tws
, set
, alt_index
);
721 /* if cannot allocate just do without */
722 /* we'll get it next time around */
728 if(tws
->style
== TWS_HASH_STYLE_SIGNAL
) {
729 vm_object_reference(object
);
730 vm_map_reference(map
);
733 if(current_line
== 0xFFFFFFFF) {
734 current_line
= tws
->current_line
;
735 set
= current_line
/tws
->number_of_lines
;
736 current_line
= current_line
- (set
* tws
->number_of_lines
);
740 tws
->current_line
= tws
->number_of_lines
- 1;
743 if(tws
->cache
[set
][current_line
].ele_count
744 >= tws
->number_of_elements
) {
747 if(current_line
== tws
->number_of_lines
) {
750 if (set
== tws
->expansion_count
) {
751 if((tws
->lookup_count
<
752 (2 * tws
->insert_count
)) &&
753 (set
<TWS_HASH_EXPANSION_MAX
)) {
754 tws
->lookup_count
= 0;
755 tws
->insert_count
= 0;
756 if(tws
->number_of_lines
757 < TWS_HASH_LINE_COUNT
) {
760 return KERN_NO_SPACE
;
762 /* object persistence is guaranteed by */
763 /* an elevated paging or object */
764 /* reference count in the caller. */
765 vm_object_unlock(object
);
766 if((tws
->table
[set
] = (tws_hash_ptr_t
*)
767 kalloc(sizeof(tws_hash_ptr_t
)
768 * tws
->number_of_lines
769 * tws
->number_of_elements
))
772 } else if((tws
->table_ele
[set
] =
774 kalloc(sizeof(struct tws_hash_ptr
)
775 * tws
->number_of_lines
776 * tws
->number_of_elements
))
778 kfree((vm_offset_t
)tws
->table
[set
],
779 sizeof(tws_hash_ptr_t
)
780 * tws
->number_of_lines
781 * tws
->number_of_elements
);
783 } else if((tws
->alt_ele
[set
] =
785 kalloc(sizeof(struct tws_hash_ptr
)
786 * tws
->number_of_lines
787 * tws
->number_of_elements
))
789 kfree((vm_offset_t
)tws
->table_ele
[set
],
790 sizeof(struct tws_hash_ptr
)
791 * tws
->number_of_lines
792 * tws
->number_of_elements
);
793 kfree((vm_offset_t
)tws
->table
[set
],
794 sizeof(tws_hash_ptr_t
)
795 * tws
->number_of_lines
796 * tws
->number_of_elements
);
797 tws
->table
[set
] = NULL
;
800 } else if((tws
->cache
[set
] =
801 (struct tws_hash_line
*)
803 (struct tws_hash_line
)
804 * tws
->number_of_lines
))
806 kfree((vm_offset_t
)tws
->alt_ele
[set
],
807 sizeof(struct tws_hash_ptr
)
808 * tws
->number_of_lines
809 * tws
->number_of_elements
);
810 kfree((vm_offset_t
)tws
->table_ele
[set
],
811 sizeof(struct tws_hash_ptr
)
812 * tws
->number_of_lines
813 * tws
->number_of_elements
);
814 kfree((vm_offset_t
)tws
->table
[set
],
815 sizeof(tws_hash_ptr_t
)
816 * tws
->number_of_lines
817 * tws
->number_of_elements
);
818 tws
->table
[set
] = NULL
;
822 tws
->free_hash_ele
[set
] =
824 tws
->obj_free_count
[set
] = 0;
825 tws
->addr_free_count
[set
] = 0;
826 bzero((char *)tws
->table
[set
],
827 sizeof(tws_hash_ptr_t
)
828 * tws
->number_of_lines
829 * tws
->number_of_elements
);
830 bzero((char *)tws
->table_ele
[set
],
831 sizeof(struct tws_hash_ptr
)
832 * tws
->number_of_lines
833 * tws
->number_of_elements
);
834 bzero((char *)tws
->alt_ele
[set
],
835 sizeof(struct tws_hash_ptr
)
836 * tws
->number_of_lines
837 * tws
->number_of_elements
);
838 bzero((char *)tws
->cache
[set
],
839 sizeof(struct tws_hash_line
)
840 * tws
->number_of_lines
);
842 vm_object_lock(object
);
847 tws
->time_of_creation
)
848 >> SCHED_TICK_SHIFT
);
850 if((tws
->startup_cache
) &&
851 (age_of_cache
> 60)) {
852 ask_for_startup_cache_release
= 1;
854 if((tws
->startup_name
!= NULL
) &&
855 (age_of_cache
> 15)) {
858 return KERN_OPERATION_TIMED_OUT
;
860 if((tws
->startup_name
!= NULL
) &&
861 (age_of_cache
< 15)) {
862 /* If we are creating a */
863 /* cache, don't lose the */
869 tws
->lookup_count
= 0;
870 tws
->insert_count
= 0;
874 tws
->current_line
= set
* tws
->number_of_lines
;
876 if(set
< tws
->expansion_count
) {
877 tws_hash_line_clear(tws
,
878 &(tws
->cache
[set
][current_line
]), TRUE
);
879 if(tws
->cache
[set
][current_line
].ele_count
880 >= tws
->number_of_elements
) {
881 if(tws
->style
== TWS_HASH_STYLE_SIGNAL
) {
882 vm_object_deallocate(object
);
883 vm_map_deallocate(map
);
889 tws
->expansion_count
++;
895 /* set object hash element */
896 if(obj_ele
== NULL
) {
897 obj_ele
= new_obj_hash(tws
, set
, index
);
898 if(obj_ele
== NULL
) {
899 tws
->cache
[set
][current_line
].ele_count
900 = tws
->number_of_elements
;
906 /* set address hash element */
907 if(addr_ele
== NULL
) {
908 addr_ele
= new_addr_hash(tws
, set
, alt_index
);
911 if(target_element
== NULL
) {
913 for(i
= 0; i
<tws
->number_of_elements
; i
++) {
914 if(tws
->cache
[set
][current_line
].
915 list
[ele_index
].object
== 0) {
919 if(ele_index
>= tws
->number_of_elements
)
924 if(i
== tws
->number_of_elements
)
925 panic("tws_insert: no free elements");
928 &(tws
->cache
[set
][current_line
].list
[ele_index
]);
930 tws
->cache
[set
][current_line
].ele_count
++;
933 obj_ele
->element
= target_element
;
935 addr_ele
->element
= (tws_hash_ele_t
)
936 (((unsigned int)target_element
) | TWS_ADDR_HASH
);
938 target_element
->object
= object
;
939 target_element
->offset
= offset
& TWS_HASH_OFF_MASK
;
940 target_element
->page_addr
=
941 page_addr
- (offset
- (offset
& TWS_HASH_OFF_MASK
));
942 target_element
->map
= map
;
943 target_element
->line
=
944 current_line
+ (set
* tws
->number_of_lines
);
945 if(startup_cache_line
) {
946 target_element
->page_cache
= startup_cache_line
;
948 target_element
->page_cache
|=
949 1<<(((vm_offset_t
)(offset
& TWS_INDEX_MASK
))>>12);
953 if(ask_for_startup_cache_release
)
954 return KERN_OPERATION_TIMED_OUT
;
960 * lengthen the cluster of pages by the number of pages encountered in the
961 * working set up to the limit requested by the caller. The object needs
962 * to be locked on entry. The map does not because the tws_lookup function
963 * is used only to find if their is an entry in the cache. No transient
964 * data from the cache is de-referenced.
969 * MACH page map - an optional optimization where a bit map is maintained
970 * by the VM subsystem for internal objects to indicate which pages of
971 * the object currently reside on backing store. This existence map
972 * duplicates information maintained by the vnode pager. It is
973 * created at the time of the first pageout against the object, i.e.
974 * at the same time pager for the object is created. The optimization
975 * is designed to eliminate pager interaction overhead, if it is
976 * 'known' that the page does not exist on backing store.
978 * LOOK_FOR() evaluates to TRUE if the page specified by object/offset is
979 * either marked as paged out in the existence map for the object or no
980 * existence map exists for the object. LOOK_FOR() is one of the
981 * criteria in the decision to invoke the pager. It is also used as one
982 * of the criteria to terminate the scan for adjacent pages in a clustered
983 * pagein operation. Note that LOOK_FOR() always evaluates to TRUE for
984 * permanent objects. Note also that if the pager for an internal object
985 * has not been created, the pager is not invoked regardless of the value
986 * of LOOK_FOR() and that clustered pagein scans are only done on an object
987 * for which a pager has been created.
989 * PAGED_OUT() evaluates to TRUE if the page specified by the object/offset
990 * is marked as paged out in the existence map for the object. PAGED_OUT()
991 * PAGED_OUT() is used to determine if a page has already been pushed
992 * into a copy object in order to avoid a redundant page out operation.
994 #define LOOK_FOR(o, f) (vm_external_state_get((o)->existence_map, (f)) \
995 != VM_EXTERNAL_STATE_ABSENT)
996 #define PAGED_OUT(o, f) (vm_external_state_get((o)->existence_map, (f)) \
997 == VM_EXTERNAL_STATE_EXISTS)
998 #else /* MACH_PAGEMAP */
1000 * If the MACH page map optimization is not enabled,
1001 * LOOK_FOR() always evaluates to TRUE. The pager will always be
1002 * invoked to resolve missing pages in an object, assuming the pager
1003 * has been created for the object. In a clustered page operation, the
1004 * absence of a page on backing backing store cannot be used to terminate
1005 * a scan for adjacent pages since that information is available only in
1006 * the pager. Hence pages that may not be paged out are potentially
1007 * included in a clustered request. The vnode pager is coded to deal
1008 * with any combination of absent/present pages in a clustered
1009 * pagein request. PAGED_OUT() always evaluates to FALSE, i.e. the pager
1010 * will always be invoked to push a dirty page into a copy object assuming
1011 * a pager has been created. If the page has already been pushed, the
1012 * pager will ingore the new request.
1014 #define LOOK_FOR(o, f) TRUE
1015 #define PAGED_OUT(o, f) FALSE
1016 #endif /* MACH_PAGEMAP */
1022 vm_object_offset_t
*start
,
1023 vm_object_offset_t
*end
,
1024 vm_size_t max_length
)
1026 tws_hash_line_t line
;
1028 vm_object_offset_t before
= *start
;
1029 vm_object_offset_t after
= *end
;
1030 vm_object_offset_t original_start
= *start
;
1031 vm_object_offset_t original_end
= *end
;
1032 vm_size_t length
= (vm_size_t
)(*end
- *start
);
1035 vm_object_offset_t object_size
;
1038 unsigned int ele_cache
;
1039 unsigned int end_cache
= 0;
1040 unsigned int start_cache
= 0;
1042 if((object
->private) || !(object
->pager
))
1045 if (!object
->internal
) {
1046 kret
= vnode_pager_get_object_size(
1050 object_size
= object
->size
;
1053 if((!tws
) || (!tws_lock_try(tws
))) {
1057 age_of_cache
= ((sched_tick
1058 - tws
->time_of_creation
) >> SCHED_TICK_SHIFT
);
1060 /* When pre-heat files are not available, resort to speculation */
1061 /* based on size of file */
1063 if(tws
->startup_cache
|| object
->internal
|| age_of_cache
> 15 ||
1064 (age_of_cache
> 5 &&
1065 vm_page_free_count
< (vm_page_free_target
* 2) )) {
1068 if (object_size
> (vm_object_offset_t
)(1024 * 1024))
1069 pre_heat_size
= 8 * PAGE_SIZE
;
1070 else if (object_size
> (vm_object_offset_t
)(128 * 1024))
1071 pre_heat_size
= 4 * PAGE_SIZE
;
1073 pre_heat_size
= 2 * PAGE_SIZE
;
1076 if ((age_of_cache
< 10) && (tws
->startup_cache
)) {
1077 if ((max_length
>= ((*end
- *start
)
1078 + (32 * PAGE_SIZE
))) &&
1079 (tws_test_for_community(tws
, object
,
1080 *start
, 3, &ele_cache
))) {
1082 start_cache
= ele_cache
;
1083 *start
= *start
& TWS_HASH_OFF_MASK
;
1084 *end
= *start
+ (32 * PAGE_SIZE_64
);
1085 if(*end
> object_size
) {
1086 *end
= trunc_page_64(object_size
);
1088 if(before
>= *end
) {
1091 end_cache
= ele_cache
;
1094 end_cache
= ele_cache
;
1096 while (max_length
> ((*end
- *start
)
1097 + (32 * PAGE_SIZE
))) {
1100 before
= *start
- PAGE_SIZE_64
;
1101 if((*end
<= (object
->size
1102 + (32 * PAGE_SIZE_64
))) &&
1103 (tws_test_for_community(tws
,
1107 (32 * PAGE_SIZE_64
);
1108 if(*end
> object_size
) {
1109 *end
= trunc_page_64(object_size
);
1111 if(*start
>= *end
) {
1115 end_cache
= ele_cache
;
1118 if (max_length
> ((*end
- *start
)
1119 + (32 * PAGE_SIZE_64
))) {
1122 if((*start
>= (32 * PAGE_SIZE_64
)) &&
1123 (tws_test_for_community(tws
, object
,
1124 before
, 5, &ele_cache
))) {
1126 start_cache
= ele_cache
;
1133 if(start_cache
!= 0) {
1136 for (mask
= 1; mask
!= 0; mask
= mask
<< 1) {
1137 if (*start
== original_start
)
1139 if (!(start_cache
& mask
))
1140 *start
+= PAGE_SIZE_64
;
1145 if(end_cache
!= 0) {
1148 for (mask
= 0x80000000;
1149 mask
!= 0; mask
= mask
>> 1) {
1150 if (*end
== original_end
)
1152 if(!(end_cache
& mask
))
1153 *end
-= PAGE_SIZE_64
;
1160 panic("bad clipping occurred\n");
1167 while ((length
< max_length
) &&
1169 (after
+ PAGE_SIZE_64
))) {
1170 if(length
>= pre_heat_size
) {
1171 if(tws_internal_lookup(tws
, after
, object
,
1172 &line
) != KERN_SUCCESS
) {
1173 vm_object_offset_t extend
;
1175 extend
= after
+ PAGE_SIZE_64
;
1176 if(tws_internal_lookup(tws
, extend
, object
,
1177 &line
) != KERN_SUCCESS
) {
1183 if ((object
->existence_map
!= NULL
)
1184 && (!LOOK_FOR(object
, after
))) {
1188 if (vm_page_lookup(object
, after
) != VM_PAGE_NULL
) {
1190 * don't bridge resident pages
1195 if (object
->internal
) {
1197 * need to acquire a real page in
1198 * advance because this acts as
1199 * a throttling mechanism for
1200 * data_requests to the default
1201 * pager. If this fails, give up
1202 * trying to find any more pages
1203 * in the cluster and send off the
1204 * request for what we already have.
1206 if ((m
= vm_page_grab()) == VM_PAGE_NULL
) {
1209 } else if ((m
= vm_page_grab_fictitious())
1215 m
->clustered
= TRUE
;
1216 m
->list_req_pending
= TRUE
;
1218 vm_page_insert(m
, object
, after
);
1219 object
->absent_count
++;
1220 after
+= PAGE_SIZE_64
;
1221 length
+= PAGE_SIZE
;
1224 while (length
< max_length
) {
1227 before
-= PAGE_SIZE_64
;
1229 if(length
>= pre_heat_size
) {
1230 if(tws_internal_lookup(tws
, before
, object
,
1231 &line
) != KERN_SUCCESS
) {
1232 vm_object_offset_t extend
;
1237 extend
-= PAGE_SIZE_64
;
1238 if(tws_internal_lookup(tws
, extend
, object
,
1239 &line
) != KERN_SUCCESS
) {
1244 if ((object
->existence_map
!= NULL
)
1245 && (!LOOK_FOR(object
, before
))) {
1249 if (vm_page_lookup(object
, before
) != VM_PAGE_NULL
) {
1251 * don't bridge resident pages
1256 if (object
->internal
) {
1258 * need to acquire a real page in
1259 * advance because this acts as
1260 * a throttling mechanism for
1261 * data_requests to the default
1262 * pager. If this fails, give up
1263 * trying to find any more pages
1264 * in the cluster and send off the
1265 * request for what we already have.
1267 if ((m
= vm_page_grab()) == VM_PAGE_NULL
) {
1270 } else if ((m
= vm_page_grab_fictitious())
1276 m
->clustered
= TRUE
;
1277 m
->list_req_pending
= TRUE
;
1279 vm_page_insert(m
, object
, before
);
1280 object
->absent_count
++;
1281 *start
-= PAGE_SIZE_64
;
1282 length
+= PAGE_SIZE
;
1290 tws_hash_line_t hash_line
,
1291 vm_offset_t target_page
)
1295 vm_object_offset_t offset
;
1296 vm_object_offset_t before
;
1297 vm_object_offset_t after
;
1298 struct tws_hash_ele
*element
;
1302 if(tws
->style
!= TWS_HASH_STYLE_SIGNAL
)
1306 for (i
=0; i
<tws
->number_of_elements
; i
++) {
1308 vm_object_offset_t local_off
= 0;
1310 if(hash_line
->list
[i
].object
== 0)
1313 element
= &hash_line
->list
[i
];
1315 if (element
->page_addr
== target_page
)
1320 if(j
& element
->page_cache
)
1323 local_off
+= PAGE_SIZE_64
;
1325 object
= element
->object
;
1326 offset
= element
->offset
+ local_off
;
1328 /* first try a fast test to speed up no-op signal */
1329 if (((p
= vm_page_lookup(object
, offset
)) != NULL
)
1330 || (object
->pager
== NULL
)
1331 || (object
->shadow_severed
)) {
1335 if((!object
->alive
) ||
1336 (!object
->pager_created
) || (!object
->pager_ready
))
1339 if (object
->internal
) {
1340 if (object
->existence_map
== NULL
) {
1344 if(!LOOK_FOR(object
, offset
))
1349 vm_object_reference(object
);
1352 if(object
->internal
) {
1355 m
= vm_page_grab_fictitious();
1359 vm_object_deallocate(object
);
1364 vm_object_lock(object
);
1365 if (((p
= vm_page_lookup(object
, offset
)) != NULL
)
1366 || (object
->pager
== NULL
)
1367 || (object
->shadow_severed
)) {
1369 vm_object_unlock(object
);
1370 vm_object_deallocate(object
);
1375 vm_page_insert(m
, object
, offset
);
1377 if (object
->absent_count
> vm_object_absent_max
) {
1379 vm_object_unlock(object
);
1380 vm_object_deallocate(object
);
1384 m
->list_req_pending
= TRUE
;
1387 object
->absent_count
++;
1390 after
= offset
+ PAGE_SIZE_64
;
1391 tws_build_cluster(tws
, object
, &before
, &after
, 0x16000);
1392 vm_object_unlock(object
);
1394 rc
= memory_object_data_request(object
->pager
,
1395 before
+ object
->paging_offset
,
1396 (vm_size_t
)(after
- before
), VM_PROT_READ
);
1397 if (rc
!= KERN_SUCCESS
) {
1399 vm_object_lock(object
);
1400 while (offset
< after
) {
1401 m
= vm_page_lookup(object
, offset
);
1402 if(m
&& m
->absent
&& m
->busy
)
1404 offset
+= PAGE_SIZE
;
1406 vm_object_unlock(object
);
1407 vm_object_deallocate(object
);
1409 vm_object_deallocate(object
);
1417 /* tws locked on entry */
1420 tws_create_startup_list(
1424 tws_startup_t startup
;
1426 unsigned int total_elements
;
1427 unsigned int startup_size
;
1428 unsigned int sindex
;
1429 unsigned int hash_index
;
1430 tws_startup_ptr_t element
;
1432 total_elements
= tws
->expansion_count
*
1433 (tws
->number_of_lines
* tws
->number_of_elements
);
1435 startup_size
= sizeof(struct tws_startup
)
1436 + (total_elements
* sizeof(tws_startup_ptr_t
*))
1437 + (total_elements
* sizeof(struct tws_startup_ptr
))
1438 + (total_elements
* sizeof(struct tws_startup_ele
));
1439 startup
= (tws_startup_t
)(kalloc(startup_size
));
1444 bzero((char *) startup
, startup_size
);
1446 startup
->table
= (tws_startup_ptr_t
*)
1447 (((int)startup
) + (sizeof(struct tws_startup
)));
1448 startup
->ele
= (struct tws_startup_ptr
*)
1449 (((vm_offset_t
)startup
->table
) +
1450 (total_elements
* sizeof(tws_startup_ptr_t
)));
1452 startup
->array
= (struct tws_startup_ele
*)
1453 (((vm_offset_t
)startup
->ele
) +
1454 (total_elements
* sizeof(struct tws_startup_ptr
)));
1456 startup
->tws_hash_size
= startup_size
;
1457 startup
->ele_count
= 0; /* burn first hash ele, else we can't tell from zero */
1458 startup
->array_size
= total_elements
;
1459 startup
->hash_count
= 1;
1464 for(i
= 0; i
<tws
->number_of_lines
; i
++) {
1465 for(j
= 0; j
<tws
->number_of_elements
; j
++) {
1466 for(k
= 0; k
<tws
->expansion_count
; k
++) {
1467 tws_hash_ele_t entry
;
1468 unsigned int hash_retry
;
1471 entry
= &tws
->cache
[k
][i
].list
[j
];
1472 addr
= entry
->page_addr
;
1474 if(entry
->object
!= 0) {
1475 /* get a hash element */
1476 hash_index
= do_startup_hash(addr
,
1477 startup
->array_size
);
1479 if(startup
->hash_count
< total_elements
) {
1480 element
= &(startup
->ele
[startup
->hash_count
]);
1481 startup
->hash_count
+= 1;
1483 /* exit we're out of elements */
1486 /* place the hash element */
1487 element
->next
= startup
->table
[hash_index
];
1488 startup
->table
[hash_index
] = (tws_startup_ptr_t
)
1489 ((int)element
- (int)&startup
->ele
[0]);
1491 /* set entry OFFSET in hash element */
1492 element
->element
= (tws_startup_ele_t
)
1493 ((int)&startup
->array
[sindex
] -
1494 (int)&startup
->array
[0]);
1496 startup
->array
[sindex
].page_addr
= entry
->page_addr
;
1497 startup
->array
[sindex
].page_cache
= entry
->page_cache
;
1498 startup
->ele_count
++;
1511 * Returns an entire cache line. The line is deleted from the startup
1512 * cache on return. The caller can check startup->ele_count for an empty
1513 * list. Access synchronization is the responsibility of the caller.
1517 tws_startup_list_lookup(
1518 tws_startup_t startup
,
1521 unsigned int hash_index
;
1522 unsigned int page_cache_bits
;
1523 unsigned int startup_shift
;
1524 tws_startup_ele_t entry
;
1525 vm_offset_t next_addr
;
1526 tws_startup_ptr_t element
;
1527 tws_startup_ptr_t base_ele
;
1528 tws_startup_ptr_t
*previous_ptr
;
1530 page_cache_bits
= 0;
1532 hash_index
= do_startup_hash(addr
, startup
->array_size
);
1534 if(((unsigned int)&(startup
->table
[hash_index
])) >= startup
->tws_hash_size
) {
1535 return page_cache_bits
= 0;
1537 element
= (tws_startup_ptr_t
)((int)startup
->table
[hash_index
] +
1538 (int)&startup
->ele
[0]);
1540 previous_ptr
= &(startup
->table
[hash_index
]);
1541 while(element
> &startup
->ele
[0]) {
1542 if (((int)element
+ sizeof(struct tws_startup_ptr
))
1543 > ((int)startup
+ startup
->tws_hash_size
)) {
1544 return page_cache_bits
;
1546 entry
= (tws_startup_ele_t
)
1547 ((int)element
->element
1548 + (int)&startup
->array
[0]);
1549 if((((int)entry
+ sizeof(struct tws_startup_ele
))
1550 > ((int)startup
+ startup
->tws_hash_size
))
1551 || ((int)entry
< (int)startup
)) {
1552 return page_cache_bits
;
1554 if ((addr
>= entry
->page_addr
) &&
1555 (addr
<= (entry
->page_addr
+ 0x1F000))) {
1556 startup_shift
= (addr
- entry
->page_addr
)>>12;
1557 page_cache_bits
|= entry
->page_cache
>> startup_shift
;
1558 /* don't dump the pages, unless the addresses */
1559 /* line up perfectly. The cache may be used */
1560 /* by other mappings */
1561 entry
->page_cache
&= (1 << startup_shift
) - 1;
1562 if(addr
== entry
->page_addr
) {
1563 if(base_ele
== element
) {
1564 base_ele
= (tws_startup_ptr_t
)
1566 + (int)&startup
->ele
[0]);
1567 startup
->table
[hash_index
] = element
->next
;
1570 *previous_ptr
= element
->next
;
1571 element
= (tws_startup_ptr_t
)
1573 + (int)&startup
->ele
[0]);
1575 entry
->page_addr
= 0;
1576 startup
->ele_count
--;
1580 next_addr
= addr
+ 0x1F000;
1581 if ((next_addr
>= entry
->page_addr
) &&
1582 (next_addr
<= (entry
->page_addr
+ 0x1F000))) {
1583 startup_shift
= (next_addr
- entry
->page_addr
)>>12;
1584 page_cache_bits
|= entry
->page_cache
<< (0x1F - startup_shift
);
1585 entry
->page_cache
&= ~((1 << (startup_shift
+ 1)) - 1);
1586 if(entry
->page_cache
== 0) {
1587 if(base_ele
== element
) {
1588 base_ele
= (tws_startup_ptr_t
)
1590 + (int)&startup
->ele
[0]);
1591 startup
->table
[hash_index
] = element
->next
;
1594 *previous_ptr
= element
->next
;
1595 element
= (tws_startup_ptr_t
)
1597 + (int)&startup
->ele
[0]);
1599 entry
->page_addr
= 0;
1600 startup
->ele_count
--;
1604 previous_ptr
= &(element
->next
);
1605 element
= (tws_startup_ptr_t
)
1606 ((int) element
->next
+ (int) &startup
->ele
[0]);
1609 return page_cache_bits
;
1613 tws_send_startup_info(
1618 tws_startup_t scache
;
1621 tws
= (tws_hash_t
)task
->dynamic_working_set
;
1624 return KERN_FAILURE
;
1626 return tws_internal_startup_send(tws
);
1631 tws_internal_startup_send(
1635 tws_startup_t scache
;
1638 return KERN_FAILURE
;
1641 /* used to signal write or release depending on state of tws */
1642 if(tws
->startup_cache
) {
1643 vm_offset_t startup_buf
;
1645 startup_buf
= (vm_offset_t
)tws
->startup_cache
;
1646 size
= tws
->startup_cache
->tws_hash_size
;
1647 tws
->startup_cache
= 0;
1649 kmem_free(kernel_map
, startup_buf
, size
);
1650 return KERN_SUCCESS
;
1652 if(tws
->startup_name
== NULL
) {
1654 return KERN_FAILURE
;
1656 scache
= tws_create_startup_list(tws
);
1658 return KERN_FAILURE
;
1659 bsd_write_page_cache_file(tws
->uid
, tws
->startup_name
,
1660 scache
, scache
->tws_hash_size
,
1661 tws
->mod
, tws
->fid
);
1662 kfree((vm_offset_t
)scache
, scache
->tws_hash_size
);
1663 kfree((vm_offset_t
) tws
->startup_name
, tws
->startup_name_length
);
1664 tws
->startup_name
= NULL
;
1666 return KERN_SUCCESS
;
1670 tws_handle_startup_file(
1675 boolean_t
*new_info
)
1678 tws_startup_t startup
;
1679 vm_offset_t cache_size
;
1680 kern_return_t error
;
1685 /* don't pre-heat kernel task */
1686 if(task
== kernel_task
)
1687 return KERN_SUCCESS
;
1688 error
= bsd_read_page_cache_file(uid
, &fid
,
1693 return KERN_FAILURE
;
1695 if(startup
== NULL
) {
1696 /* Entry for app does not exist, make */
1698 /* we will want our own copy of the shared */
1699 /* regions to pick up a true picture of all */
1700 /* the pages we will touch. */
1701 if((lsf_zone
->count
* lsf_zone
->elem_size
)
1702 > (lsf_zone
->max_size
>> 1)) {
1703 /* We don't want to run out of shared memory */
1704 /* map entries by starting too many private versions */
1705 /* of the shared library structures */
1706 return KERN_SUCCESS
;
1710 error
= tws_write_startup_file(task
,
1711 fid
, mod
, app_name
, uid
);
1716 error
= tws_read_startup_file(task
,
1717 (tws_startup_t
)startup
,
1720 kmem_free(kernel_map
,
1721 (vm_offset_t
)startup
, cache_size
);
1725 return KERN_SUCCESS
;
1729 tws_write_startup_file(
1737 unsigned int string_length
;
1739 string_length
= strlen(name
);
1742 tws
= (tws_hash_t
)task
->dynamic_working_set
;
1746 /* create a dynamic working set of normal size */
1747 task_working_set_create(task
, 0,
1748 0, TWS_HASH_STYLE_DEFAULT
);
1752 if(tws
->startup_name
!= NULL
) {
1754 return KERN_FAILURE
;
1757 tws
->startup_name
= (char *)
1758 kalloc((string_length
+ 1) * (sizeof(char)));
1759 if(tws
->startup_name
== NULL
) {
1761 return KERN_FAILURE
;
1764 bcopy(name
, (char *)tws
->startup_name
, string_length
+ 1);
1765 tws
->startup_name_length
= (string_length
+ 1) * sizeof(char);
1771 return KERN_SUCCESS
;
1775 tws_read_startup_file(
1777 tws_startup_t startup
,
1778 vm_offset_t cache_size
)
1786 tws
= (tws_hash_t
)task
->dynamic_working_set
;
1788 if(cache_size
< sizeof(struct tws_hash
)) {
1790 kmem_free(kernel_map
, (vm_offset_t
)startup
, cache_size
);
1791 return(KERN_SUCCESS
);
1794 /* create a dynamic working set to match file size */
1795 lines
= (cache_size
- sizeof(struct tws_hash
))/TWS_ARRAY_SIZE
;
1796 /* we now need to divide out element size and word size */
1797 /* all fields are 4 bytes. There are 8 bytes in each hash element */
1798 /* entry, 4 bytes in each table ptr location and 8 bytes in each */
1799 /* page_cache entry, making a total of 20 bytes for each entry */
1800 lines
= (lines
/(20));
1801 if(lines
<= TWS_SMALL_HASH_LINE_COUNT
) {
1802 lines
= TWS_SMALL_HASH_LINE_COUNT
;
1804 kmem_free(kernel_map
, (vm_offset_t
)startup
, cache_size
);
1805 return(KERN_SUCCESS
);
1807 old_exp_count
= lines
/TWS_HASH_LINE_COUNT
;
1808 if((old_exp_count
* TWS_HASH_LINE_COUNT
) != lines
) {
1809 lines
= (old_exp_count
+ 1)
1810 * TWS_HASH_LINE_COUNT
;
1813 task_working_set_create(task
, lines
,
1814 0, TWS_HASH_STYLE_DEFAULT
);
1818 tws_expand_working_set(
1819 (vm_offset_t
)tws
, lines
, TRUE
);
1826 if(tws
->startup_cache
!= NULL
) {
1828 return KERN_FAILURE
;
1832 /* now need to fix up internal table pointers */
1833 startup
->table
= (tws_startup_ptr_t
*)
1834 (((int)startup
) + (sizeof(struct tws_startup
)));
1835 startup
->ele
= (struct tws_startup_ptr
*)
1836 (((vm_offset_t
)startup
->table
) +
1837 (startup
->array_size
* sizeof(tws_startup_ptr_t
)));
1838 startup
->array
= (struct tws_startup_ele
*)
1839 (((vm_offset_t
)startup
->ele
) +
1840 (startup
->array_size
* sizeof(struct tws_startup_ptr
)));
1841 /* the allocation size and file size should be the same */
1842 /* just in case their not, make sure we dealloc correctly */
1843 startup
->tws_hash_size
= cache_size
;
1845 tws
->startup_cache
= startup
;
1847 return KERN_SUCCESS
;
1852 tws_hash_ws_flush(tws_hash_t tws
) {
1853 tws_startup_t scache
;
1858 if(tws
->startup_name
!= NULL
) {
1859 scache
= tws_create_startup_list(tws
);
1860 if(scache
== NULL
) {
1861 /* dump the name cache, we'll */
1862 /* get it next time */
1865 tws
->startup_name_length
);
1866 tws
->startup_name
= NULL
;
1870 bsd_write_page_cache_file(tws
->uid
, tws
->startup_name
,
1871 scache
, scache
->tws_hash_size
,
1872 tws
->mod
, tws
->fid
);
1873 kfree((vm_offset_t
)scache
,
1874 scache
->tws_hash_size
);
1877 tws
->startup_name_length
);
1878 tws
->startup_name
= NULL
;
1885 tws_hash_destroy(tws_hash_t tws
)
1888 vm_size_t cache_size
;
1890 if(tws
->startup_cache
!= NULL
) {
1891 kmem_free(kernel_map
,
1892 (vm_offset_t
)tws
->startup_cache
,
1893 tws
->startup_cache
->tws_hash_size
);
1894 tws
->startup_cache
= NULL
;
1896 if(tws
->startup_name
!= NULL
) {
1897 tws_internal_startup_send(tws
);
1899 for (i
=0; i
<tws
->number_of_lines
; i
++) {
1900 for(k
=0; k
<tws
->expansion_count
; k
++) {
1901 /* clear the object refs */
1902 tws_hash_line_clear(tws
, &(tws
->cache
[k
][i
]), FALSE
);
1906 while (i
< tws
->expansion_count
) {
1908 kfree((vm_offset_t
)tws
->table
[i
], sizeof(tws_hash_ptr_t
)
1909 * tws
->number_of_lines
1910 * tws
->number_of_elements
);
1911 kfree((vm_offset_t
)tws
->table_ele
[i
],
1912 sizeof(struct tws_hash_ptr
)
1913 * tws
->number_of_lines
1914 * tws
->number_of_elements
);
1915 kfree((vm_offset_t
)tws
->alt_ele
[i
],
1916 sizeof(struct tws_hash_ptr
)
1917 * tws
->number_of_lines
1918 * tws
->number_of_elements
);
1919 kfree((vm_offset_t
)tws
->cache
[i
], sizeof(struct tws_hash_line
)
1920 * tws
->number_of_lines
);
1923 if(tws
->startup_name
!= NULL
) {
1924 kfree((vm_offset_t
)tws
->startup_name
,
1925 tws
->startup_name_length
);
1927 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
1931 tws_hash_clear(tws_hash_t tws
)
1935 for (i
=0; i
<tws
->number_of_lines
; i
++) {
1936 for(k
=0; k
<tws
->expansion_count
; k
++) {
1937 /* clear the object refs */
1938 tws_hash_line_clear(tws
, &(tws
->cache
[k
][i
]), FALSE
);
1944 task_working_set_create(
1952 lines
= TWS_HASH_LINE_COUNT
;
1955 rows
= TWS_ARRAY_SIZE
;
1957 if (style
== TWS_HASH_STYLE_DEFAULT
) {
1958 style
= TWS_HASH_STYLE_BASIC
;
1961 if(task
->dynamic_working_set
!= 0) {
1963 return(KERN_FAILURE
);
1964 } else if((task
->dynamic_working_set
1965 = (vm_offset_t
) tws_hash_create(lines
, rows
, style
)) == 0) {
1967 return(KERN_NO_SPACE
);
1970 return KERN_SUCCESS
;
1974 /* Internal use only routines */
1978 * internal sub-function for address space lookup
1979 * returns the target element and the address of the
1980 * previous pointer The previous pointer is the address
1981 * of the pointer pointing to the target element.
1982 * TWS must be locked
1986 tws_traverse_address_hash_list (
1989 vm_offset_t page_addr
,
1991 vm_object_offset_t offset
,
1993 tws_hash_ptr_t
*target_ele
,
1994 tws_hash_ptr_t
**previous_ptr
,
1995 tws_hash_ptr_t
**free_list
,
1996 unsigned int exclusive_addr
)
1999 tws_hash_ptr_t cache_ele
;
2000 tws_hash_ptr_t base_ele
;
2003 *previous_ptr
= NULL
;
2005 for(k
=0; k
<tws
->expansion_count
; k
++) {
2007 cache_ele
= tws
->table
[k
][index
];
2008 base_ele
= cache_ele
;
2009 *previous_ptr
= (tws_hash_ptr_t
*)&(tws
->table
[k
][index
]);
2010 while(cache_ele
!= NULL
) {
2012 cache_ele
->element
& TWS_ADDR_HASH
) == 0) {
2013 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2014 cache_ele
= cache_ele
->next
;
2017 ele
= (tws_hash_ele_t
)((unsigned int)
2018 cache_ele
->element
& ~TWS_ADDR_HASH
);
2019 if ((ele
== 0) || (ele
->object
== 0)) {
2020 /* A little clean-up of empty elements */
2021 cache_ele
->element
= 0;
2022 if(base_ele
== cache_ele
) {
2023 base_ele
= cache_ele
->next
;
2024 tws
->table
[k
][index
] = cache_ele
->next
;
2025 cache_ele
->next
= tws
->free_hash_ele
[k
];
2026 tws
->free_hash_ele
[k
] = cache_ele
;
2027 cache_ele
= base_ele
;
2029 **previous_ptr
= cache_ele
->next
;
2030 cache_ele
->next
= tws
->free_hash_ele
[k
];
2031 tws
->free_hash_ele
[k
] = cache_ele
;
2032 cache_ele
= **previous_ptr
;
2037 if ((ele
->page_addr
<= page_addr
)
2038 && (page_addr
<= (ele
->page_addr
+
2039 (vm_offset_t
)TWS_INDEX_MASK
))
2040 && ((object
== NULL
)
2041 || ((object
== ele
->object
)
2042 && (offset
== ele
->offset
)
2043 && (map
== ele
->map
)))) {
2044 if(exclusive_addr
) {
2046 delta
= ((page_addr
- ele
->page_addr
)
2048 if((1 << delta
) & ele
->page_cache
) {
2049 /* We've found a match */
2050 *target_ele
= cache_ele
;
2053 &(tws
->free_hash_ele
[k
]);
2057 /* We've found a match */
2058 *target_ele
= cache_ele
;
2059 *free_list
= (tws_hash_ptr_t
*)
2060 &(tws
->free_hash_ele
[k
]);
2064 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2065 cache_ele
= cache_ele
->next
;
2072 * internal sub-function for object space lookup
2073 * returns the target element and the address of the
2074 * previous pointer The previous pointer is the address
2075 * of the pointer pointing to the target element.
2076 * TWS must be locked
2081 tws_traverse_object_hash_list (
2085 vm_object_offset_t offset
,
2086 unsigned int page_mask
,
2087 tws_hash_ptr_t
*target_ele
,
2088 tws_hash_ptr_t
**previous_ptr
,
2089 tws_hash_ptr_t
**free_list
)
2092 tws_hash_ptr_t cache_ele
;
2093 tws_hash_ptr_t base_ele
;
2096 *previous_ptr
= NULL
;
2098 for(k
=0; k
<tws
->expansion_count
; k
++) {
2099 cache_ele
= tws
->table
[k
][index
];
2100 base_ele
= cache_ele
;
2101 *previous_ptr
= &(tws
->table
[k
][index
]);
2102 while(cache_ele
!= NULL
) {
2103 if((((unsigned int)cache_ele
->element
)
2104 & TWS_ADDR_HASH
) != 0) {
2105 *previous_ptr
= &(cache_ele
->next
);
2106 cache_ele
= cache_ele
->next
;
2109 if ((cache_ele
->element
== 0) ||
2110 (cache_ele
->element
->object
== 0)) {
2111 /* A little clean-up of empty elements */
2112 cache_ele
->element
= 0;
2113 if(base_ele
== cache_ele
) {
2114 base_ele
= cache_ele
->next
;
2115 tws
->table
[k
][index
] = cache_ele
->next
;
2116 cache_ele
->next
= tws
->free_hash_ele
[k
];
2117 tws
->free_hash_ele
[k
] = cache_ele
;
2118 cache_ele
= tws
->table
[k
][index
];
2120 **previous_ptr
= cache_ele
->next
;
2121 cache_ele
->next
= tws
->free_hash_ele
[k
];
2122 tws
->free_hash_ele
[k
] = cache_ele
;
2123 cache_ele
= **previous_ptr
;
2127 if ((cache_ele
->element
->object
== object
)
2128 && (cache_ele
->element
->offset
==
2129 (offset
- (offset
& ~TWS_HASH_OFF_MASK
)))) {
2130 if((cache_ele
->element
->page_cache
& page_mask
)
2131 || (page_mask
== 0xFFFFFFFF)) {
2132 /* We've found a match */
2133 *target_ele
= cache_ele
;
2134 *free_list
= &(tws
->free_hash_ele
[k
]);
2138 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2139 cache_ele
= cache_ele
->next
;
2146 * For a given object/offset, discover whether the indexed 32 page frame
2147 * containing the object/offset exists and if their are at least threshold
2148 * pages present. Returns true if population meets threshold.
2151 tws_test_for_community(
2154 vm_object_offset_t offset
,
2155 unsigned int threshold
,
2156 unsigned int *page_mask
)
2159 tws_hash_ptr_t cache_ele
;
2160 tws_hash_ptr_t
*trailer
;
2161 tws_hash_ptr_t
*free_list
;
2164 index
= do_tws_hash(object
, offset
,
2165 tws
->number_of_elements
, tws
->number_of_lines
);
2166 tws_traverse_object_hash_list(tws
, index
, object
, offset
, 0xFFFFFFFF,
2167 &cache_ele
, &trailer
, &free_list
);
2169 if(cache_ele
!= NULL
) {
2173 for(i
=1; i
!=0; i
=i
<<1) {
2174 if(i
& cache_ele
->element
->page_cache
)
2176 if(ctr
== threshold
) {
2178 *page_mask
= cache_ele
->element
->page_cache
;
2190 * Gets new hash element for object hash from free pools
2191 * TWS must be locked
2200 tws_hash_ptr_t element
;
2202 if(tws
->obj_free_count
[set
] < tws
->number_of_lines
* tws
->number_of_elements
) {
2203 element
= &(tws
->table_ele
[set
][tws
->obj_free_count
[set
]]);
2204 tws
->obj_free_count
[set
]+=1;
2205 } else if(tws
->free_hash_ele
[set
] == NULL
) {
2208 element
= tws
->free_hash_ele
[set
];
2211 tws
->free_hash_ele
[set
] = tws
->free_hash_ele
[set
]->next
;
2213 element
->element
= 0;
2214 element
->next
= tws
->table
[set
][index
];
2215 tws
->table
[set
][index
] = element
;
2220 * Gets new hash element for addr hash from free pools
2221 * TWS must be locked
2230 tws_hash_ptr_t element
;
2232 if(tws
->addr_free_count
[set
]
2233 < tws
->number_of_lines
* tws
->number_of_elements
) {
2234 element
= &(tws
->alt_ele
[set
][tws
->addr_free_count
[set
]]);
2235 tws
->addr_free_count
[set
]+=1;
2236 } else if(tws
->free_hash_ele
[set
] == NULL
) {
2239 element
= tws
->free_hash_ele
[set
];
2242 tws
->free_hash_ele
[set
] = tws
->free_hash_ele
[set
]->next
;
2244 element
->element
= (tws_hash_ele_t
)TWS_ADDR_HASH
;
2245 element
->next
= tws
->table
[set
][index
];
2246 tws
->table
[set
][index
] = element
;