3 * Copyright (c) 2002,2000 Apple Computer, Inc. All rights reserved.
5 * @APPLE_LICENSE_HEADER_START@
7 * The contents of this file constitute Original Code as defined in and
8 * are subject to the Apple Public Source License Version 1.1 (the
9 * "License"). You may not use this file except in compliance with the
10 * License. Please obtain a copy of the License at
11 * http://www.apple.com/publicsource and read it before using this file.
13 * This Original Code and all software distributed under the License are
14 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
15 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
16 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
18 * License for the specific language governing rights and limitations
21 * @APPLE_LICENSE_HEADER_END@
26 * File: vm/task_working_set.c
27 * Author: Chris Youngworth
30 * Working set detection and maintainence module
34 #include <mach/mach_types.h>
35 #include <mach/shared_memory_server.h>
36 #include <vm/task_working_set.h>
37 #include <vm/vm_kern.h>
38 #include <vm/vm_map.h>
39 #include <vm/vm_page.h>
40 #include <vm/vm_pageout.h>
41 #include <kern/sched.h>
43 extern unsigned sched_tick
;
44 extern zone_t lsf_zone
;
46 /* declarations for internal use only routines */
49 tws_create_startup_list(
53 tws_startup_list_lookup(
54 tws_startup_t startup
,
58 tws_internal_startup_send(
62 tws_traverse_address_hash_list (
65 vm_offset_t page_addr
,
67 vm_object_offset_t offset
,
69 tws_hash_ptr_t
*target_ele
,
70 tws_hash_ptr_t
**previous_ptr
,
71 tws_hash_ptr_t
**free_list
,
72 unsigned int exclusive_addr
);
75 tws_traverse_object_hash_list (
79 vm_object_offset_t offset
,
80 unsigned int page_mask
,
81 tws_hash_ptr_t
*target_ele
,
82 tws_hash_ptr_t
**previous_ptr
,
83 tws_hash_ptr_t
**free_list
);
97 int tws_test_for_community(
100 vm_object_offset_t offset
,
101 unsigned int threshold
,
102 unsigned int *page_mask
);
104 /* Note: all of the routines below depend on the associated map lock for */
105 /* synchronization, the map lock will be on when the routines are called */
106 /* and on when they return */
118 if ((style
!= TWS_HASH_STYLE_BASIC
) &&
119 (style
!= TWS_HASH_STYLE_BASIC
)) {
120 return((tws_hash_t
)NULL
);
124 tws
= (tws_hash_t
)(kalloc(sizeof(struct tws_hash
)));
125 if(tws
== (tws_hash_t
)NULL
)
128 if((tws
->table
[0] = (tws_hash_ptr_t
*)
129 kalloc(sizeof(tws_hash_ptr_t
) * lines
* rows
))
131 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
132 return (tws_hash_t
)NULL
;
134 if((tws
->table_ele
[0] = (tws_hash_ptr_t
)
135 kalloc(sizeof(struct tws_hash_ptr
) * lines
* rows
))
137 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ele_t
)
139 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
140 return (tws_hash_t
)NULL
;
142 if((tws
->alt_ele
[0] = (tws_hash_ptr_t
)
143 kalloc(sizeof(struct tws_hash_ptr
) * lines
* rows
))
145 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ptr_t
)
147 kfree((vm_offset_t
)tws
->table_ele
[0],
148 sizeof(struct tws_hash_ptr
)
150 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
151 return (tws_hash_t
)NULL
;
153 if((tws
->cache
[0] = (struct tws_hash_line
*)
154 kalloc(sizeof(struct tws_hash_line
) * lines
))
156 kfree((vm_offset_t
)tws
->table
[0], sizeof(tws_hash_ptr_t
)
158 kfree((vm_offset_t
)tws
->table_ele
[0],
159 sizeof(struct tws_hash_ptr
)
161 kfree((vm_offset_t
)tws
->alt_ele
[0], sizeof(struct tws_hash_ptr
)
163 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
164 return (tws_hash_t
)NULL
;
166 tws
->free_hash_ele
[0] = (tws_hash_ptr_t
)0;
167 tws
->obj_free_count
[0] = 0;
168 tws
->addr_free_count
[0] = 0;
170 /* most defaults are such that a bzero will initialize */
171 bzero((char *)tws
->table
[0],sizeof(tws_hash_ptr_t
)
173 bzero((char *)tws
->table_ele
[0],sizeof(struct tws_hash_ptr
)
175 bzero((char *)tws
->alt_ele
[0],sizeof(struct tws_hash_ptr
)
177 bzero((char *)tws
->cache
[0], sizeof(struct tws_hash_line
)
180 mutex_init(&tws
->lock
, ETAP_VM_MAP
);
182 tws
->current_line
= 0;
183 tws
->pageout_count
= 0;
185 tws
->startup_cache
= NULL
;
186 tws
->startup_name
= NULL
;
187 tws
->number_of_lines
= lines
;
188 tws
->number_of_elements
= rows
;
189 tws
->expansion_count
= 1;
190 tws
->lookup_count
= 0;
191 tws
->insert_count
= 0;
192 tws
->time_of_creation
= sched_tick
;
201 tws_hash_line_t hash_line
,
204 struct tws_hash_ele
*hash_ele
;
205 struct tws_hash_ptr
**trailer
;
206 struct tws_hash_ptr
**free_list
;
207 tws_hash_ele_t addr_ele
;
209 unsigned int i
, j
, k
;
214 if(tws
->line_count
< tws
->number_of_lines
) {
218 if(tws
->pageout_count
!= vm_pageout_scan_event_counter
) {
220 vm_pageout_scan_event_counter
;
227 hash_line
->ele_count
= 0;
229 for (i
=0; i
<tws
->number_of_elements
; i
++) {
231 hash_ele
= &(hash_line
->list
[i
]);
232 if(hash_ele
->object
!= 0) {
234 vm_object_offset_t local_off
= 0;
235 tws_hash_ptr_t cache_ele
;
237 index
= alt_tws_hash(
238 hash_ele
->page_addr
& TWS_HASH_OFF_MASK
,
239 tws
->number_of_elements
,
240 tws
->number_of_lines
);
242 tws_traverse_address_hash_list(tws
, index
,
243 hash_ele
->page_addr
, hash_ele
->object
,
244 hash_ele
->offset
, hash_ele
->map
,
245 &cache_ele
, &trailer
, &free_list
, 0);
246 if(cache_ele
!= NULL
) {
247 addr_ele
= (tws_hash_ele_t
)((unsigned int)
248 (cache_ele
->element
) & ~TWS_ADDR_HASH
);
249 if(addr_ele
!= hash_ele
)
250 panic("tws_hash_line_clear:"
252 cache_ele
->element
= 0;
253 *trailer
= cache_ele
->next
;
254 cache_ele
->next
= *free_list
;
255 *free_list
= cache_ele
;
258 index
= alt_tws_hash(
259 (hash_ele
->page_addr
- 0x1f000)
261 tws
->number_of_elements
,
262 tws
->number_of_lines
);
264 tws_traverse_address_hash_list(tws
, index
,
265 hash_ele
->page_addr
, hash_ele
->object
,
266 hash_ele
->offset
, hash_ele
->map
,
267 &cache_ele
, &trailer
, &free_list
, 0);
269 if(cache_ele
!= NULL
) {
270 addr_ele
= (tws_hash_ele_t
)((unsigned int)
271 (cache_ele
->element
) & ~TWS_ADDR_HASH
);
272 if(addr_ele
!= hash_ele
)
273 panic("tws_hash_line_clear: "
275 cache_ele
->element
= 0;
276 *trailer
= cache_ele
->next
;
277 cache_ele
->next
= *free_list
;
278 *free_list
= cache_ele
;
282 if((hash_ele
->map
!= NULL
) && (live
)) {
285 for (j
= 0x1; j
!= 0; j
= j
<<1) {
286 if(j
& hash_ele
->page_cache
) {
287 p
= vm_page_lookup(hash_ele
->object
,
288 hash_ele
->offset
+ local_off
);
289 if((p
!= NULL
) && (p
->wire_count
== 0)
290 && (dump_pmap
== 1)) {
291 pmap_remove_some_phys((pmap_t
)
297 local_off
+= PAGE_SIZE_64
;
301 if(tws
->style
== TWS_HASH_STYLE_SIGNAL
) {
302 vm_object_deallocate(hash_ele
->object
);
303 vm_map_deallocate(hash_ele
->map
);
306 index
= do_tws_hash(hash_ele
->object
, hash_ele
->offset
,
307 tws
->number_of_elements
,
308 tws
->number_of_lines
);
310 tws_traverse_object_hash_list(tws
,
311 index
, hash_ele
->object
, hash_ele
->offset
,
312 0xFFFFFFFF, &cache_ele
, &trailer
, &free_list
);
313 if((cache_ele
!= NULL
) && (cache_ele
->element
== hash_ele
)) {
314 cache_ele
->element
= 0;
315 *trailer
= cache_ele
->next
;
316 cache_ele
->next
= *free_list
;
317 *free_list
= cache_ele
;
319 hash_ele
->object
= 0;
327 vm_object_offset_t offset
,
329 tws_hash_line_t
*line
)
336 tws_hash_ptr_t cache_ele
;
337 tws_hash_ptr_t
*trailer
;
338 tws_hash_ptr_t
*free_list
;
340 /* don't cache private objects */
344 index
= do_tws_hash(object
, offset
,
345 tws
->number_of_elements
, tws
->number_of_lines
);
349 if(tws
->lookup_count
== 0)
350 tws
->insert_count
= 0;
351 if(tws
->startup_name
!= NULL
) {
353 age_of_cache
= ((sched_tick
354 - tws
->time_of_creation
) >> SCHED_TICK_SHIFT
);
355 if (age_of_cache
> 35) {
356 return KERN_OPERATION_TIMED_OUT
;
360 if(tws
->lookup_count
> (4 * tws
->expansion_count
361 * tws
->number_of_elements
* tws
->number_of_lines
) &&
362 (tws
->lookup_count
> (2 * tws
->insert_count
))) {
363 if(tws
->startup_cache
) {
365 age_of_cache
= ((sched_tick
366 - tws
->time_of_creation
) >> SCHED_TICK_SHIFT
);
367 if (age_of_cache
> 60) {
368 return KERN_OPERATION_TIMED_OUT
;
373 pagenum
= (vm_offset_t
)(offset
& TWS_INDEX_MASK
);
374 pagenum
= pagenum
>> 12;
375 pagenum
= 1 << pagenum
; /* get the appropriate page in 32 page block */
376 tws_traverse_object_hash_list(tws
, index
, object
, offset
, pagenum
,
377 &cache_ele
, &trailer
, &free_list
);
378 if(cache_ele
!= NULL
) {
379 set
= cache_ele
->element
->line
/tws
->number_of_lines
;
380 ele_line
= cache_ele
->element
->line
- set
;
381 *line
= &tws
->cache
[set
][ele_line
];
393 vm_object_offset_t offset
,
395 tws_hash_line_t
*line
)
399 if(!tws_lock_try(tws
)) {
402 kr
= tws_internal_lookup(tws
,
403 offset
, object
, line
);
409 tws_expand_working_set(
417 struct tws_hash temp
;
419 old_tws
= (tws_hash_t
)tws
;
421 /* Note we do an elaborate dance to preserve the header that */
422 /* task is pointing to. In this way we can avoid taking a task */
423 /* lock every time we want to access the tws */
425 if (old_tws
->number_of_lines
>= line_count
) {
428 if((new_tws
= tws_hash_create(line_count
,
429 old_tws
->number_of_elements
, old_tws
->style
)) == 0) {
430 return(KERN_NO_SPACE
);
435 for(i
= 0; i
<old_tws
->number_of_lines
; i
++) {
436 for(j
= 0; j
<old_tws
->number_of_elements
; j
++) {
437 for(k
= 0; k
<old_tws
->expansion_count
; k
++) {
438 tws_hash_ele_t entry
;
439 vm_object_offset_t paddr
;
440 unsigned int page_index
;
441 entry
= &old_tws
->cache
[k
][i
].list
[j
];
442 if(entry
->object
!= 0) {
444 for(page_index
= 1; page_index
!= 0;
445 page_index
= page_index
<< 1); {
446 if (entry
->page_cache
& page_index
) {
450 entry
->page_addr
+paddr
,
462 temp
.style
= new_tws
->style
;
463 temp
.current_line
= new_tws
->current_line
;
464 temp
.pageout_count
= new_tws
->pageout_count
;
465 temp
.line_count
= new_tws
->line_count
;
466 temp
.number_of_lines
= new_tws
->number_of_lines
;
467 temp
.number_of_elements
= new_tws
->number_of_elements
;
468 temp
.expansion_count
= new_tws
->expansion_count
;
469 temp
.lookup_count
= new_tws
->lookup_count
;
470 temp
.insert_count
= new_tws
->insert_count
;
471 for(i
= 0; i
<new_tws
->expansion_count
; i
++) {
472 temp
.obj_free_count
[i
] = new_tws
->obj_free_count
[i
];
473 temp
.addr_free_count
[i
] = new_tws
->addr_free_count
[i
];
474 temp
.free_hash_ele
[i
] = new_tws
->free_hash_ele
[i
];
475 temp
.table
[i
] = new_tws
->table
[i
];
476 temp
.table_ele
[i
] = new_tws
->table_ele
[i
];
477 temp
.alt_ele
[i
] = new_tws
->alt_ele
[i
];
478 temp
.cache
[i
] = new_tws
->cache
[i
];
481 new_tws
->style
= old_tws
->style
;
482 new_tws
->current_line
= old_tws
->current_line
;
483 new_tws
->pageout_count
= old_tws
->pageout_count
;
484 new_tws
->line_count
= old_tws
->line_count
;
485 new_tws
->number_of_lines
= old_tws
->number_of_lines
;
486 new_tws
->number_of_elements
= old_tws
->number_of_elements
;
487 new_tws
->expansion_count
= old_tws
->expansion_count
;
488 new_tws
->lookup_count
= old_tws
->lookup_count
;
489 new_tws
->insert_count
= old_tws
->insert_count
;
490 for(i
= 0; i
<old_tws
->expansion_count
; i
++) {
491 new_tws
->obj_free_count
[i
] = old_tws
->obj_free_count
[i
];
492 new_tws
->addr_free_count
[i
] = old_tws
->addr_free_count
[i
];
493 new_tws
->free_hash_ele
[i
] = old_tws
->free_hash_ele
[i
];
494 new_tws
->table
[i
] = old_tws
->table
[i
];
495 new_tws
->table_ele
[i
] = old_tws
->table_ele
[i
];
496 new_tws
->alt_ele
[i
] = old_tws
->alt_ele
[i
];
497 new_tws
->cache
[i
] = old_tws
->cache
[i
];
500 old_tws
->style
= temp
.style
;
501 old_tws
->current_line
= temp
.current_line
;
502 old_tws
->pageout_count
= temp
.pageout_count
;
503 old_tws
->line_count
= temp
.line_count
;
504 old_tws
->number_of_lines
= temp
.number_of_lines
;
505 old_tws
->number_of_elements
= temp
.number_of_elements
;
506 old_tws
->expansion_count
= temp
.expansion_count
;
507 old_tws
->lookup_count
= temp
.lookup_count
;
508 old_tws
->insert_count
= temp
.insert_count
;
509 for(i
= 0; i
<temp
.expansion_count
; i
++) {
510 old_tws
->obj_free_count
[i
] = temp
.obj_free_count
[i
];;
511 old_tws
->addr_free_count
[i
] = temp
.addr_free_count
[i
];;
512 old_tws
->free_hash_ele
[i
] = NULL
;
513 old_tws
->table
[i
] = temp
.table
[i
];
514 old_tws
->table_ele
[i
] = temp
.table_ele
[i
];
515 old_tws
->alt_ele
[i
] = temp
.alt_ele
[i
];
516 old_tws
->cache
[i
] = temp
.cache
[i
];
519 tws_hash_destroy(new_tws
);
524 tws_hash_t test_tws
= 0;
529 vm_object_offset_t offset
,
531 vm_offset_t page_addr
,
536 unsigned int alt_index
;
537 unsigned int index_enum
[2];
538 unsigned int ele_index
;
539 tws_hash_ptr_t cache_ele
;
540 tws_hash_ptr_t obj_ele
= NULL
;
541 tws_hash_ptr_t addr_ele
= NULL
;
542 tws_hash_ptr_t
*trailer
;
543 tws_hash_ptr_t
*free_list
;
544 tws_hash_ele_t target_element
= NULL
;
549 unsigned int startup_cache_line
;
550 vm_offset_t startup_page_addr
;
552 int ask_for_startup_cache_release
= 0;
555 if(!tws_lock_try(tws
)) {
559 current_line
= 0xFFFFFFFF;
561 startup_cache_line
= 0;
563 page_addr
- (offset
- (offset
& TWS_HASH_OFF_MASK
));
564 if(tws
->startup_cache
) {
566 age_of_cache
= ((sched_tick
- tws
->time_of_creation
)
567 >> SCHED_TICK_SHIFT
);
568 startup_cache_line
= tws_startup_list_lookup(
569 tws
->startup_cache
, startup_page_addr
);
570 if(tws
== test_tws
) {
571 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
);
573 if(age_of_cache
> 60) {
574 ask_for_startup_cache_release
= 1;
577 if((tws
->startup_name
!= NULL
) && (tws
->mod
== 0)) {
578 /* Ensure as good a working set as possible */
579 pmap_remove(map
->pmap
, 0, GLOBAL_SHARED_TEXT_SEGMENT
);
580 pmap_remove(map
->pmap
,
581 GLOBAL_SHARED_DATA_SEGMENT
582 + SHARED_DATA_REGION_SIZE
, 0xFFFFF000);
585 /* This next bit of code, the and alternate hash */
586 /* are all made necessary because of IPC COW */
588 /* Note: the use of page_addr modified by delta from offset */
589 /* frame base means we may miss some previous entries. However */
590 /* we will not miss the present entry. This is most important */
591 /* in avoiding duplication of entries against long lived non-cow */
593 index_enum
[0] = alt_tws_hash(
594 page_addr
& TWS_HASH_OFF_MASK
,
595 tws
->number_of_elements
, tws
->number_of_lines
);
597 index_enum
[1] = alt_tws_hash(
598 (page_addr
- 0x1f000) & TWS_HASH_OFF_MASK
,
599 tws
->number_of_elements
, tws
->number_of_lines
);
601 for(ctr
= 0; ctr
< 2;) {
602 tws_hash_ele_t resident
;
603 tws_traverse_address_hash_list(tws
,
604 index_enum
[ctr
], page_addr
, NULL
,
606 &cache_ele
, &trailer
, &free_list
, 1);
607 if(cache_ele
!= NULL
) {
609 resident
= (tws_hash_ele_t
)((unsigned int)
610 cache_ele
->element
& ~TWS_ADDR_HASH
);
611 if((object
== resident
->object
) &&
613 (offset
& TWS_HASH_OFF_MASK
)) {
614 /* This is our object/offset */
616 |= startup_cache_line
;
617 resident
->page_cache
|=
619 (offset
& TWS_INDEX_MASK
))>>12));
621 if(ask_for_startup_cache_release
)
622 return KERN_OPERATION_TIMED_OUT
;
625 if((object
->shadow
==
628 + object
->shadow_offset
)
629 == (offset
& TWS_HASH_OFF_MASK
))) {
630 /* if we just shadowed, inherit */
631 /* access pattern from parent */
632 startup_cache_line
|=
633 resident
->page_cache
;
634 /* thow out old entry */
635 resident
->page_cache
= 0;
638 resident
->page_cache
&=
639 ~(1<<(((vm_offset_t
)(page_addr
640 - resident
->page_addr
))
643 /* Throw out old entry if there are no */
644 /* more pages in cache */
645 if(resident
->page_cache
== 0) {
646 /* delete addr hash entry */
647 cache_ele
->element
= 0;
648 *trailer
= cache_ele
->next
;
649 cache_ele
->next
= *free_list
;
650 *free_list
= cache_ele
;
651 /* go after object hash */
655 tws
->number_of_elements
,
656 tws
->number_of_lines
);
657 tws_traverse_object_hash_list(tws
,
658 index
, resident
->object
,
660 0xFFFFFFFF, &cache_ele
,
661 &trailer
, &free_list
);
662 if(cache_ele
!= NULL
) {
664 TWS_HASH_STYLE_SIGNAL
) {
665 vm_object_deallocate(
666 cache_ele
->element
->object
);
668 cache_ele
->element
->map
);
671 cache_ele
->element
->line
;
673 /tws
->number_of_lines
;
674 current_line
-= set
*
675 tws
->number_of_lines
;
676 if(cache_ele
->element
->object
!= 0) {
677 cache_ele
->element
->object
= 0;
679 [current_line
].ele_count
--;
681 cache_ele
->element
= 0;
682 *trailer
= cache_ele
->next
;
683 cache_ele
->next
= *free_list
;
684 *free_list
= cache_ele
;
693 * We may or may not have a current line setting coming out of
694 * the code above. If we have a current line it means we can
695 * choose to back-fill the spot vacated by a previous entry.
696 * We have yet to do a definitive check using the original obj/off
697 * We will do that now and override the current line if we
701 index
= do_tws_hash(object
, offset
,
702 tws
->number_of_elements
, tws
->number_of_lines
);
704 alt_index
= index_enum
[0];
706 tws_traverse_object_hash_list(tws
, index
, object
, offset
,
707 0xFFFFFFFF, &cache_ele
, &trailer
, &free_list
);
708 if(cache_ele
!= NULL
) {
710 current_line
= cache_ele
->element
->line
;
711 set
= current_line
/tws
->number_of_lines
;
712 current_line
-= set
* tws
->number_of_lines
;
713 target_element
= cache_ele
->element
;
715 /* Now check to see if we have a hash addr for it */
716 tws_traverse_address_hash_list(tws
,
717 alt_index
, obj_ele
->element
->page_addr
,
718 obj_ele
->element
->object
,
719 obj_ele
->element
->offset
,
720 obj_ele
->element
->map
,
721 &cache_ele
, &trailer
, &free_list
, 0);
722 if(cache_ele
!= NULL
) {
723 addr_ele
= cache_ele
;
725 addr_ele
= new_addr_hash(tws
, set
, alt_index
);
726 /* if cannot allocate just do without */
727 /* we'll get it next time around */
733 if(tws
->style
== TWS_HASH_STYLE_SIGNAL
) {
734 vm_object_reference(object
);
735 vm_map_reference(map
);
738 if(current_line
== 0xFFFFFFFF) {
739 current_line
= tws
->current_line
;
740 set
= current_line
/tws
->number_of_lines
;
741 current_line
= current_line
- (set
* tws
->number_of_lines
);
745 tws
->current_line
= tws
->number_of_lines
- 1;
748 if(tws
->cache
[set
][current_line
].ele_count
749 >= tws
->number_of_elements
) {
752 if(current_line
== tws
->number_of_lines
) {
755 if (set
== tws
->expansion_count
) {
756 if((tws
->lookup_count
<
757 (2 * tws
->insert_count
)) &&
758 (set
<TWS_HASH_EXPANSION_MAX
)) {
759 tws
->lookup_count
= 0;
760 tws
->insert_count
= 0;
761 if(tws
->number_of_lines
762 < TWS_HASH_LINE_COUNT
) {
765 return KERN_NO_SPACE
;
767 if((tws
->table
[set
] = (tws_hash_ptr_t
*)
768 kalloc(sizeof(tws_hash_ptr_t
)
769 * tws
->number_of_lines
770 * tws
->number_of_elements
))
773 } else if((tws
->table_ele
[set
] =
775 kalloc(sizeof(struct tws_hash_ptr
)
776 * tws
->number_of_lines
777 * tws
->number_of_elements
))
779 kfree((vm_offset_t
)tws
->table
[set
],
780 sizeof(tws_hash_ptr_t
)
781 * tws
->number_of_lines
782 * tws
->number_of_elements
);
784 } else if((tws
->alt_ele
[set
] =
786 kalloc(sizeof(struct tws_hash_ptr
)
787 * tws
->number_of_lines
788 * tws
->number_of_elements
))
790 kfree((vm_offset_t
)tws
->table_ele
[set
],
791 sizeof(tws_hash_ptr_t
)
792 * tws
->number_of_lines
793 * tws
->number_of_elements
);
794 kfree((vm_offset_t
)tws
->table
[set
],
795 sizeof(struct tws_hash_ptr
)
796 * tws
->number_of_lines
797 * tws
->number_of_elements
);
798 tws
->table
[set
] = NULL
;
801 } else if((tws
->cache
[set
] =
802 (struct tws_hash_line
*)
804 (struct tws_hash_line
)
805 * tws
->number_of_lines
))
807 kfree((vm_offset_t
)tws
->table
[set
],
808 sizeof(tws_hash_ptr_t
)
809 * tws
->number_of_lines
810 * tws
->number_of_elements
);
811 kfree((vm_offset_t
)tws
->table_ele
[set
],
812 sizeof(struct tws_hash_ptr
)
813 * tws
->number_of_lines
814 * tws
->number_of_elements
);
815 kfree((vm_offset_t
)tws
->alt_ele
[set
],
816 sizeof(struct tws_hash_ptr
)
817 * tws
->number_of_lines
818 * tws
->number_of_elements
);
819 tws
->table
[set
] = NULL
;
823 tws
->free_hash_ele
[set
] =
825 tws
->obj_free_count
[set
] = 0;
826 tws
->addr_free_count
[set
] = 0;
827 bzero((char *)tws
->table
[set
],
828 sizeof(tws_hash_ptr_t
)
829 * tws
->number_of_lines
830 * tws
->number_of_elements
);
831 bzero((char *)tws
->table_ele
[set
],
832 sizeof(struct tws_hash_ptr
)
833 * tws
->number_of_lines
834 * tws
->number_of_elements
);
835 bzero((char *)tws
->alt_ele
[set
],
836 sizeof(struct tws_hash_ptr
)
837 * tws
->number_of_lines
838 * tws
->number_of_elements
);
839 bzero((char *)tws
->cache
[set
],
840 sizeof(struct tws_hash_line
)
841 * tws
->number_of_lines
);
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
= NULL
;
1040 unsigned int start_cache
= NULL
;
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(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(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
!= NULL
) {
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
!= NULL
) {
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
) {
1189 /* we can bridge resident pages */
1190 after
+= PAGE_SIZE_64
;
1191 length
+= PAGE_SIZE
;
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
) {
1250 /* we can bridge resident pages */
1251 *start
-= PAGE_SIZE_64
;
1252 length
+= PAGE_SIZE
;
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
;
1709 error
= tws_write_startup_file(task
,
1710 fid
, mod
, app_name
, uid
);
1713 /* use the mod in the write case as an init */
1718 error
= tws_read_startup_file(task
,
1719 (tws_startup_t
)startup
,
1722 kmem_free(kernel_map
,
1723 (vm_offset_t
)startup
, cache_size
);
1727 return KERN_SUCCESS
;
1731 tws_write_startup_file(
1739 unsigned int string_length
;
1741 string_length
= strlen(name
);
1744 tws
= (tws_hash_t
)task
->dynamic_working_set
;
1748 /* create a dynamic working set of normal size */
1749 task_working_set_create(task
, 0,
1750 0, TWS_HASH_STYLE_DEFAULT
);
1754 if(tws
->startup_name
!= NULL
) {
1756 return KERN_FAILURE
;
1759 tws
->startup_name
= (char *)
1760 kalloc((string_length
+ 1) * (sizeof(char)));
1761 if(tws
->startup_name
== NULL
) {
1763 return KERN_FAILURE
;
1766 bcopy(name
, (char *)tws
->startup_name
, string_length
+ 1);
1767 tws
->startup_name_length
= (string_length
+ 1) * sizeof(char);
1773 return KERN_SUCCESS
;
1777 tws_read_startup_file(
1779 tws_startup_t startup
,
1780 vm_offset_t cache_size
)
1788 tws
= (tws_hash_t
)task
->dynamic_working_set
;
1790 if(cache_size
< sizeof(struct tws_hash
)) {
1792 kmem_free(kernel_map
, (vm_offset_t
)startup
, cache_size
);
1793 return(KERN_SUCCESS
);
1796 /* create a dynamic working set to match file size */
1797 lines
= (cache_size
- sizeof(struct tws_hash
))/TWS_ARRAY_SIZE
;
1798 /* we now need to divide out element size and word size */
1799 /* all fields are 4 bytes. There are 8 bytes in each hash element */
1800 /* entry, 4 bytes in each table ptr location and 8 bytes in each */
1801 /* page_cache entry, making a total of 20 bytes for each entry */
1802 lines
= (lines
/(20));
1803 if(lines
<= TWS_SMALL_HASH_LINE_COUNT
) {
1804 lines
= TWS_SMALL_HASH_LINE_COUNT
;
1806 kmem_free(kernel_map
, (vm_offset_t
)startup
, cache_size
);
1807 return(KERN_SUCCESS
);
1809 old_exp_count
= lines
/TWS_HASH_LINE_COUNT
;
1810 if((old_exp_count
* TWS_HASH_LINE_COUNT
) != lines
) {
1811 lines
= (old_exp_count
+ 1)
1812 * TWS_HASH_LINE_COUNT
;
1815 task_working_set_create(task
, lines
,
1816 0, TWS_HASH_STYLE_DEFAULT
);
1820 tws_expand_working_set(
1821 (vm_offset_t
)tws
, lines
, TRUE
);
1828 if(tws
->startup_cache
!= NULL
) {
1830 return KERN_FAILURE
;
1834 /* now need to fix up internal table pointers */
1835 startup
->table
= (tws_startup_ptr_t
*)
1836 (((int)startup
) + (sizeof(struct tws_startup
)));
1837 startup
->ele
= (struct tws_startup_ptr
*)
1838 (((vm_offset_t
)startup
->table
) +
1839 (startup
->array_size
* sizeof(tws_startup_ptr_t
)));
1840 startup
->array
= (struct tws_startup_ele
*)
1841 (((vm_offset_t
)startup
->ele
) +
1842 (startup
->array_size
* sizeof(struct tws_startup_ptr
)));
1843 /* the allocation size and file size should be the same */
1844 /* just in case their not, make sure we dealloc correctly */
1845 startup
->tws_hash_size
= cache_size
;
1848 tws
->startup_cache
= startup
;
1850 return KERN_SUCCESS
;
1855 tws_hash_ws_flush(tws_hash_t tws
) {
1856 tws_startup_t scache
;
1861 if(tws
->startup_name
!= NULL
) {
1862 scache
= tws_create_startup_list(tws
);
1863 if(scache
== NULL
) {
1864 /* dump the name cache, we'll */
1865 /* get it next time */
1868 tws
->startup_name_length
);
1869 tws
->startup_name
= NULL
;
1873 bsd_write_page_cache_file(tws
->uid
, tws
->startup_name
,
1874 scache
, scache
->tws_hash_size
,
1875 tws
->mod
, tws
->fid
);
1876 kfree((vm_offset_t
)scache
,
1877 scache
->tws_hash_size
);
1880 tws
->startup_name_length
);
1881 tws
->startup_name
= NULL
;
1888 tws_hash_destroy(tws_hash_t tws
)
1891 vm_size_t cache_size
;
1893 if(tws
->startup_cache
!= NULL
) {
1894 kmem_free(kernel_map
,
1895 (vm_offset_t
)tws
->startup_cache
,
1896 tws
->startup_cache
->tws_hash_size
);
1897 tws
->startup_cache
= NULL
;
1899 if(tws
->startup_name
!= NULL
) {
1900 tws_internal_startup_send(tws
);
1902 for (i
=0; i
<tws
->number_of_lines
; i
++) {
1903 for(k
=0; k
<tws
->expansion_count
; k
++) {
1904 /* clear the object refs */
1905 tws_hash_line_clear(tws
, &(tws
->cache
[k
][i
]), FALSE
);
1909 while (i
< tws
->expansion_count
) {
1911 kfree((vm_offset_t
)tws
->table
[i
], sizeof(tws_hash_ptr_t
)
1912 * tws
->number_of_lines
1913 * tws
->number_of_elements
);
1914 kfree((vm_offset_t
)tws
->table_ele
[i
],
1915 sizeof(struct tws_hash_ptr
)
1916 * tws
->number_of_lines
1917 * tws
->number_of_elements
);
1918 kfree((vm_offset_t
)tws
->alt_ele
[i
],
1919 sizeof(struct tws_hash_ptr
)
1920 * tws
->number_of_lines
1921 * tws
->number_of_elements
);
1922 kfree((vm_offset_t
)tws
->cache
[i
], sizeof(struct tws_hash_line
)
1923 * tws
->number_of_lines
);
1926 if(tws
->startup_name
!= NULL
) {
1927 kfree((vm_offset_t
)tws
->startup_name
,
1928 tws
->startup_name_length
);
1930 kfree((vm_offset_t
)tws
, sizeof(struct tws_hash
));
1934 tws_hash_clear(tws_hash_t tws
)
1938 for (i
=0; i
<tws
->number_of_lines
; i
++) {
1939 for(k
=0; k
<tws
->expansion_count
; k
++) {
1940 /* clear the object refs */
1941 tws_hash_line_clear(tws
, &(tws
->cache
[k
][i
]), FALSE
);
1947 task_working_set_create(
1955 lines
= TWS_HASH_LINE_COUNT
;
1958 rows
= TWS_ARRAY_SIZE
;
1960 if (style
== TWS_HASH_STYLE_DEFAULT
) {
1961 style
= TWS_HASH_STYLE_BASIC
;
1964 if(task
->dynamic_working_set
!= 0) {
1966 return(KERN_FAILURE
);
1967 } else if((task
->dynamic_working_set
1968 = (vm_offset_t
) tws_hash_create(lines
, rows
, style
)) == 0) {
1970 return(KERN_NO_SPACE
);
1973 return KERN_SUCCESS
;
1977 /* Internal use only routines */
1981 * internal sub-function for address space lookup
1982 * returns the target element and the address of the
1983 * previous pointer The previous pointer is the address
1984 * of the pointer pointing to the target element.
1985 * TWS must be locked
1989 tws_traverse_address_hash_list (
1992 vm_offset_t page_addr
,
1994 vm_object_offset_t offset
,
1996 tws_hash_ptr_t
*target_ele
,
1997 tws_hash_ptr_t
**previous_ptr
,
1998 tws_hash_ptr_t
**free_list
,
1999 unsigned int exclusive_addr
)
2002 tws_hash_ptr_t cache_ele
;
2003 tws_hash_ptr_t base_ele
;
2006 *previous_ptr
= NULL
;
2008 for(k
=0; k
<tws
->expansion_count
; k
++) {
2010 cache_ele
= tws
->table
[k
][index
];
2011 base_ele
= cache_ele
;
2012 *previous_ptr
= (tws_hash_ptr_t
*)&(tws
->table
[k
][index
]);
2013 while(cache_ele
!= NULL
) {
2015 cache_ele
->element
& TWS_ADDR_HASH
) == 0) {
2016 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2017 cache_ele
= cache_ele
->next
;
2020 ele
= (tws_hash_ele_t
)((unsigned int)
2021 cache_ele
->element
& ~TWS_ADDR_HASH
);
2022 if ((ele
== 0) || (ele
->object
== 0)) {
2023 /* A little clean-up of empty elements */
2024 cache_ele
->element
= 0;
2025 if(base_ele
== cache_ele
) {
2026 base_ele
= cache_ele
->next
;
2027 tws
->table
[k
][index
] = cache_ele
->next
;
2028 cache_ele
->next
= tws
->free_hash_ele
[k
];
2029 tws
->free_hash_ele
[k
] = cache_ele
;
2030 cache_ele
= base_ele
;
2032 **previous_ptr
= cache_ele
->next
;
2033 cache_ele
->next
= tws
->free_hash_ele
[k
];
2034 tws
->free_hash_ele
[k
] = cache_ele
;
2035 cache_ele
= **previous_ptr
;
2040 if ((ele
->page_addr
<= page_addr
)
2041 && (page_addr
<= (ele
->page_addr
+
2042 (vm_offset_t
)TWS_INDEX_MASK
))
2043 && ((object
== NULL
)
2044 || ((object
== ele
->object
)
2045 && (offset
== ele
->offset
)
2046 && (map
== ele
->map
)))) {
2047 if(exclusive_addr
) {
2049 delta
= ((page_addr
- ele
->page_addr
)
2051 if((1 << delta
) & ele
->page_cache
) {
2052 /* We've found a match */
2053 *target_ele
= cache_ele
;
2056 &(tws
->free_hash_ele
[k
]);
2060 /* We've found a match */
2061 *target_ele
= cache_ele
;
2062 *free_list
= (tws_hash_ptr_t
*)
2063 &(tws
->free_hash_ele
[k
]);
2067 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2068 cache_ele
= cache_ele
->next
;
2075 * internal sub-function for object space lookup
2076 * returns the target element and the address of the
2077 * previous pointer The previous pointer is the address
2078 * of the pointer pointing to the target element.
2079 * TWS must be locked
2084 tws_traverse_object_hash_list (
2088 vm_object_offset_t offset
,
2089 unsigned int page_mask
,
2090 tws_hash_ptr_t
*target_ele
,
2091 tws_hash_ptr_t
**previous_ptr
,
2092 tws_hash_ptr_t
**free_list
)
2095 tws_hash_ptr_t cache_ele
;
2096 tws_hash_ptr_t base_ele
;
2099 *previous_ptr
= NULL
;
2101 for(k
=0; k
<tws
->expansion_count
; k
++) {
2102 cache_ele
= tws
->table
[k
][index
];
2103 base_ele
= cache_ele
;
2104 *previous_ptr
= &(tws
->table
[k
][index
]);
2105 while(cache_ele
!= NULL
) {
2106 if((((unsigned int)cache_ele
->element
)
2107 & TWS_ADDR_HASH
) != 0) {
2108 *previous_ptr
= &(cache_ele
->next
);
2109 cache_ele
= cache_ele
->next
;
2112 if ((cache_ele
->element
== 0) ||
2113 (cache_ele
->element
->object
== 0)) {
2114 /* A little clean-up of empty elements */
2115 cache_ele
->element
= 0;
2116 if(base_ele
== cache_ele
) {
2117 base_ele
= cache_ele
->next
;
2118 tws
->table
[k
][index
] = cache_ele
->next
;
2119 cache_ele
->next
= tws
->free_hash_ele
[k
];
2120 tws
->free_hash_ele
[k
] = cache_ele
;
2121 cache_ele
= tws
->table
[k
][index
];
2123 **previous_ptr
= cache_ele
->next
;
2124 cache_ele
->next
= tws
->free_hash_ele
[k
];
2125 tws
->free_hash_ele
[k
] = cache_ele
;
2126 cache_ele
= **previous_ptr
;
2130 if ((cache_ele
->element
->object
== object
)
2131 && (cache_ele
->element
->offset
==
2132 (offset
- (offset
& ~TWS_HASH_OFF_MASK
)))) {
2133 if((cache_ele
->element
->page_cache
& page_mask
)
2134 || (page_mask
== 0xFFFFFFFF)) {
2135 /* We've found a match */
2136 *target_ele
= cache_ele
;
2137 *free_list
= &(tws
->free_hash_ele
[k
]);
2141 *previous_ptr
= (tws_hash_ptr_t
*)&(cache_ele
->next
);
2142 cache_ele
= cache_ele
->next
;
2149 * For a given object/offset, discover whether the indexed 32 page frame
2150 * containing the object/offset exists and if their are at least threshold
2151 * pages present. Returns true if population meets threshold.
2154 tws_test_for_community(
2157 vm_object_offset_t offset
,
2158 unsigned int threshold
,
2159 unsigned int *page_mask
)
2162 tws_hash_ptr_t cache_ele
;
2163 tws_hash_ptr_t
*trailer
;
2164 tws_hash_ptr_t
*free_list
;
2167 index
= do_tws_hash(object
, offset
,
2168 tws
->number_of_elements
, tws
->number_of_lines
);
2169 tws_traverse_object_hash_list(tws
, index
, object
, offset
, 0xFFFFFFFF,
2170 &cache_ele
, &trailer
, &free_list
);
2172 if(cache_ele
!= NULL
) {
2176 for(i
=1; i
!=0; i
=i
<<1) {
2177 if(i
& cache_ele
->element
->page_cache
)
2179 if(ctr
== threshold
) {
2181 *page_mask
= cache_ele
->element
->page_cache
;
2193 * Gets new hash element for object hash from free pools
2194 * TWS must be locked
2203 tws_hash_ptr_t element
;
2205 if(tws
->obj_free_count
[set
] < tws
->number_of_lines
* tws
->number_of_elements
) {
2206 element
= &(tws
->table_ele
[set
][tws
->obj_free_count
[set
]]);
2207 tws
->obj_free_count
[set
]+=1;
2208 } else if(tws
->free_hash_ele
[set
] == NULL
) {
2211 element
= tws
->free_hash_ele
[set
];
2214 tws
->free_hash_ele
[set
] = tws
->free_hash_ele
[set
]->next
;
2216 element
->element
= 0;
2217 element
->next
= tws
->table
[set
][index
];
2218 tws
->table
[set
][index
] = element
;
2223 * Gets new hash element for addr hash from free pools
2224 * TWS must be locked
2233 tws_hash_ptr_t element
;
2235 if(tws
->addr_free_count
[set
]
2236 < tws
->number_of_lines
* tws
->number_of_elements
) {
2237 element
= &(tws
->alt_ele
[set
][tws
->addr_free_count
[set
]]);
2238 tws
->addr_free_count
[set
]+=1;
2239 } else if(tws
->free_hash_ele
[set
] == NULL
) {
2242 element
= tws
->free_hash_ele
[set
];
2245 tws
->free_hash_ele
[set
] = tws
->free_hash_ele
[set
]->next
;
2247 element
->element
= (tws_hash_ele_t
)TWS_ADDR_HASH
;
2248 element
->next
= tws
->table
[set
][index
];
2249 tws
->table
[set
][index
] = element
;