ia64/xen-unstable

view linux-2.6-xen-sparse/mm/page_alloc.c @ 9167:cb5abeaabd1a

[IA64] fix print out in ia64 setup_guest()

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
author awilliam@xenbuild.aw
date Fri Mar 10 09:19:54 2006 -0700 (2006-03-10)
parents 3b74edc512b4
children ea67b8a9c7e0
line source
1 /*
2 * linux/mm/page_alloc.c
3 *
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
6 *
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
15 */
17 #include <linux/config.h>
18 #include <linux/stddef.h>
19 #include <linux/mm.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
41 #include <asm/tlbflush.h>
42 #include "internal.h"
44 /*
45 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
46 * initializer cleaner
47 */
48 nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
49 EXPORT_SYMBOL(node_online_map);
50 nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
51 EXPORT_SYMBOL(node_possible_map);
52 struct pglist_data *pgdat_list __read_mostly;
53 unsigned long totalram_pages __read_mostly;
54 unsigned long totalhigh_pages __read_mostly;
55 long nr_swap_pages;
56 int percpu_pagelist_fraction;
58 static void fastcall free_hot_cold_page(struct page *page, int cold);
59 static void __free_pages_ok(struct page *page, unsigned int order);
61 /*
62 * results with 256, 32 in the lowmem_reserve sysctl:
63 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
64 * 1G machine -> (16M dma, 784M normal, 224M high)
65 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
66 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
67 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
68 *
69 * TBD: should special case ZONE_DMA32 machines here - in those we normally
70 * don't need any ZONE_NORMAL reservation
71 */
72 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 };
74 EXPORT_SYMBOL(totalram_pages);
76 /*
77 * Used by page_zone() to look up the address of the struct zone whose
78 * id is encoded in the upper bits of page->flags
79 */
80 struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
81 EXPORT_SYMBOL(zone_table);
83 static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" };
84 int min_free_kbytes = 1024;
86 unsigned long __initdata nr_kernel_pages;
87 unsigned long __initdata nr_all_pages;
89 #ifdef CONFIG_DEBUG_VM
90 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
91 {
92 int ret = 0;
93 unsigned seq;
94 unsigned long pfn = page_to_pfn(page);
96 do {
97 seq = zone_span_seqbegin(zone);
98 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
99 ret = 1;
100 else if (pfn < zone->zone_start_pfn)
101 ret = 1;
102 } while (zone_span_seqretry(zone, seq));
104 return ret;
105 }
107 static int page_is_consistent(struct zone *zone, struct page *page)
108 {
109 #ifdef CONFIG_HOLES_IN_ZONE
110 if (!pfn_valid(page_to_pfn(page)))
111 return 0;
112 #endif
113 if (zone != page_zone(page))
114 return 0;
116 return 1;
117 }
118 /*
119 * Temporary debugging check for pages not lying within a given zone.
120 */
121 static int bad_range(struct zone *zone, struct page *page)
122 {
123 if (page_outside_zone_boundaries(zone, page))
124 return 1;
125 if (!page_is_consistent(zone, page))
126 return 1;
128 return 0;
129 }
131 #else
132 static inline int bad_range(struct zone *zone, struct page *page)
133 {
134 return 0;
135 }
136 #endif
138 static void bad_page(struct page *page)
139 {
140 printk(KERN_EMERG "Bad page state in process '%s'\n"
141 KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
142 KERN_EMERG "Trying to fix it up, but a reboot is needed\n"
143 KERN_EMERG "Backtrace:\n",
144 current->comm, page, (int)(2*sizeof(unsigned long)),
145 (unsigned long)page->flags, page->mapping,
146 page_mapcount(page), page_count(page));
147 dump_stack();
148 page->flags &= ~(1 << PG_lru |
149 1 << PG_private |
150 1 << PG_locked |
151 1 << PG_active |
152 1 << PG_dirty |
153 1 << PG_reclaim |
154 1 << PG_slab |
155 1 << PG_swapcache |
156 1 << PG_writeback );
157 set_page_count(page, 0);
158 reset_page_mapcount(page);
159 page->mapping = NULL;
160 add_taint(TAINT_BAD_PAGE);
161 }
163 /*
164 * Higher-order pages are called "compound pages". They are structured thusly:
165 *
166 * The first PAGE_SIZE page is called the "head page".
167 *
168 * The remaining PAGE_SIZE pages are called "tail pages".
169 *
170 * All pages have PG_compound set. All pages have their ->private pointing at
171 * the head page (even the head page has this).
172 *
173 * The first tail page's ->lru.next holds the address of the compound page's
174 * put_page() function. Its ->lru.prev holds the order of allocation.
175 * This usage means that zero-order pages may not be compound.
176 */
178 static void free_compound_page(struct page *page)
179 {
180 __free_pages_ok(page, (unsigned long)page[1].lru.prev);
181 }
183 static void prep_compound_page(struct page *page, unsigned long order)
184 {
185 int i;
186 int nr_pages = 1 << order;
188 page[1].lru.next = (void *)free_compound_page; /* set dtor */
189 page[1].lru.prev = (void *)order;
190 for (i = 0; i < nr_pages; i++) {
191 struct page *p = page + i;
193 SetPageCompound(p);
194 set_page_private(p, (unsigned long)page);
195 }
196 }
198 static void destroy_compound_page(struct page *page, unsigned long order)
199 {
200 int i;
201 int nr_pages = 1 << order;
203 if (unlikely((unsigned long)page[1].lru.prev != order))
204 bad_page(page);
206 for (i = 0; i < nr_pages; i++) {
207 struct page *p = page + i;
209 if (unlikely(!PageCompound(p) |
210 (page_private(p) != (unsigned long)page)))
211 bad_page(page);
212 ClearPageCompound(p);
213 }
214 }
216 /*
217 * function for dealing with page's order in buddy system.
218 * zone->lock is already acquired when we use these.
219 * So, we don't need atomic page->flags operations here.
220 */
221 static inline unsigned long page_order(struct page *page) {
222 return page_private(page);
223 }
225 static inline void set_page_order(struct page *page, int order) {
226 set_page_private(page, order);
227 __SetPagePrivate(page);
228 }
230 static inline void rmv_page_order(struct page *page)
231 {
232 __ClearPagePrivate(page);
233 set_page_private(page, 0);
234 }
236 /*
237 * Locate the struct page for both the matching buddy in our
238 * pair (buddy1) and the combined O(n+1) page they form (page).
239 *
240 * 1) Any buddy B1 will have an order O twin B2 which satisfies
241 * the following equation:
242 * B2 = B1 ^ (1 << O)
243 * For example, if the starting buddy (buddy2) is #8 its order
244 * 1 buddy is #10:
245 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
246 *
247 * 2) Any buddy B will have an order O+1 parent P which
248 * satisfies the following equation:
249 * P = B & ~(1 << O)
250 *
251 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
252 */
253 static inline struct page *
254 __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
255 {
256 unsigned long buddy_idx = page_idx ^ (1 << order);
258 return page + (buddy_idx - page_idx);
259 }
261 static inline unsigned long
262 __find_combined_index(unsigned long page_idx, unsigned int order)
263 {
264 return (page_idx & ~(1 << order));
265 }
267 /*
268 * This function checks whether a page is free && is the buddy
269 * we can do coalesce a page and its buddy if
270 * (a) the buddy is not in a hole &&
271 * (b) the buddy is free &&
272 * (c) the buddy is on the buddy system &&
273 * (d) a page and its buddy have the same order.
274 * for recording page's order, we use page_private(page) and PG_private.
275 *
276 */
277 static inline int page_is_buddy(struct page *page, int order)
278 {
279 #ifdef CONFIG_HOLES_IN_ZONE
280 if (!pfn_valid(page_to_pfn(page)))
281 return 0;
282 #endif
284 if (PagePrivate(page) &&
285 (page_order(page) == order) &&
286 page_count(page) == 0)
287 return 1;
288 return 0;
289 }
291 /*
292 * Freeing function for a buddy system allocator.
293 *
294 * The concept of a buddy system is to maintain direct-mapped table
295 * (containing bit values) for memory blocks of various "orders".
296 * The bottom level table contains the map for the smallest allocatable
297 * units of memory (here, pages), and each level above it describes
298 * pairs of units from the levels below, hence, "buddies".
299 * At a high level, all that happens here is marking the table entry
300 * at the bottom level available, and propagating the changes upward
301 * as necessary, plus some accounting needed to play nicely with other
302 * parts of the VM system.
303 * At each level, we keep a list of pages, which are heads of continuous
304 * free pages of length of (1 << order) and marked with PG_Private.Page's
305 * order is recorded in page_private(page) field.
306 * So when we are allocating or freeing one, we can derive the state of the
307 * other. That is, if we allocate a small block, and both were
308 * free, the remainder of the region must be split into blocks.
309 * If a block is freed, and its buddy is also free, then this
310 * triggers coalescing into a block of larger size.
311 *
312 * -- wli
313 */
315 static inline void __free_one_page(struct page *page,
316 struct zone *zone, unsigned int order)
317 {
318 unsigned long page_idx;
319 int order_size = 1 << order;
321 if (unlikely(PageCompound(page)))
322 destroy_compound_page(page, order);
324 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
326 BUG_ON(page_idx & (order_size - 1));
327 BUG_ON(bad_range(zone, page));
329 zone->free_pages += order_size;
330 while (order < MAX_ORDER-1) {
331 unsigned long combined_idx;
332 struct free_area *area;
333 struct page *buddy;
335 buddy = __page_find_buddy(page, page_idx, order);
336 if (!page_is_buddy(buddy, order))
337 break; /* Move the buddy up one level. */
339 list_del(&buddy->lru);
340 area = zone->free_area + order;
341 area->nr_free--;
342 rmv_page_order(buddy);
343 combined_idx = __find_combined_index(page_idx, order);
344 page = page + (combined_idx - page_idx);
345 page_idx = combined_idx;
346 order++;
347 }
348 set_page_order(page, order);
349 list_add(&page->lru, &zone->free_area[order].free_list);
350 zone->free_area[order].nr_free++;
351 }
353 static inline int free_pages_check(struct page *page)
354 {
355 if (unlikely(page_mapcount(page) |
356 (page->mapping != NULL) |
357 (page_count(page) != 0) |
358 (page->flags & (
359 1 << PG_lru |
360 1 << PG_private |
361 1 << PG_locked |
362 1 << PG_active |
363 1 << PG_reclaim |
364 1 << PG_slab |
365 1 << PG_swapcache |
366 1 << PG_writeback |
367 1 << PG_reserved ))))
368 bad_page(page);
369 if (PageDirty(page))
370 __ClearPageDirty(page);
371 /*
372 * For now, we report if PG_reserved was found set, but do not
373 * clear it, and do not free the page. But we shall soon need
374 * to do more, for when the ZERO_PAGE count wraps negative.
375 */
376 return PageReserved(page);
377 }
379 /*
380 * Frees a list of pages.
381 * Assumes all pages on list are in same zone, and of same order.
382 * count is the number of pages to free.
383 *
384 * If the zone was previously in an "all pages pinned" state then look to
385 * see if this freeing clears that state.
386 *
387 * And clear the zone's pages_scanned counter, to hold off the "all pages are
388 * pinned" detection logic.
389 */
390 static void free_pages_bulk(struct zone *zone, int count,
391 struct list_head *list, int order)
392 {
393 spin_lock(&zone->lock);
394 zone->all_unreclaimable = 0;
395 zone->pages_scanned = 0;
396 while (count--) {
397 struct page *page;
399 BUG_ON(list_empty(list));
400 page = list_entry(list->prev, struct page, lru);
401 /* have to delete it as __free_one_page list manipulates */
402 list_del(&page->lru);
403 __free_one_page(page, zone, order);
404 }
405 spin_unlock(&zone->lock);
406 }
408 static void free_one_page(struct zone *zone, struct page *page, int order)
409 {
410 LIST_HEAD(list);
411 list_add(&page->lru, &list);
412 free_pages_bulk(zone, 1, &list, order);
413 }
415 static void __free_pages_ok(struct page *page, unsigned int order)
416 {
417 unsigned long flags;
418 int i;
419 int reserved = 0;
421 if (arch_free_page(page, order))
422 return;
423 if (!PageHighMem(page))
424 mutex_debug_check_no_locks_freed(page_address(page),
425 PAGE_SIZE<<order);
427 #ifndef CONFIG_MMU
428 for (i = 1 ; i < (1 << order) ; ++i)
429 __put_page(page + i);
430 #endif
432 for (i = 0 ; i < (1 << order) ; ++i)
433 reserved += free_pages_check(page + i);
434 if (reserved)
435 return;
437 kernel_map_pages(page, 1 << order, 0);
438 local_irq_save(flags);
439 __mod_page_state(pgfree, 1 << order);
440 free_one_page(page_zone(page), page, order);
441 local_irq_restore(flags);
442 }
444 /*
445 * permit the bootmem allocator to evade page validation on high-order frees
446 */
447 void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order)
448 {
449 if (order == 0) {
450 __ClearPageReserved(page);
451 set_page_count(page, 0);
453 free_hot_cold_page(page, 0);
454 } else {
455 LIST_HEAD(list);
456 int loop;
458 for (loop = 0; loop < BITS_PER_LONG; loop++) {
459 struct page *p = &page[loop];
461 if (loop + 16 < BITS_PER_LONG)
462 prefetchw(p + 16);
463 __ClearPageReserved(p);
464 set_page_count(p, 0);
465 }
467 arch_free_page(page, order);
469 mod_page_state(pgfree, 1 << order);
471 list_add(&page->lru, &list);
472 kernel_map_pages(page, 1 << order, 0);
473 free_pages_bulk(page_zone(page), 1, &list, order);
474 }
475 }
478 /*
479 * The order of subdivision here is critical for the IO subsystem.
480 * Please do not alter this order without good reasons and regression
481 * testing. Specifically, as large blocks of memory are subdivided,
482 * the order in which smaller blocks are delivered depends on the order
483 * they're subdivided in this function. This is the primary factor
484 * influencing the order in which pages are delivered to the IO
485 * subsystem according to empirical testing, and this is also justified
486 * by considering the behavior of a buddy system containing a single
487 * large block of memory acted on by a series of small allocations.
488 * This behavior is a critical factor in sglist merging's success.
489 *
490 * -- wli
491 */
492 static inline void expand(struct zone *zone, struct page *page,
493 int low, int high, struct free_area *area)
494 {
495 unsigned long size = 1 << high;
497 while (high > low) {
498 area--;
499 high--;
500 size >>= 1;
501 BUG_ON(bad_range(zone, &page[size]));
502 list_add(&page[size].lru, &area->free_list);
503 area->nr_free++;
504 set_page_order(&page[size], high);
505 }
506 }
508 /*
509 * This page is about to be returned from the page allocator
510 */
511 static int prep_new_page(struct page *page, int order)
512 {
513 if (unlikely(page_mapcount(page) |
514 (page->mapping != NULL) |
515 (page_count(page) != 0) |
516 (page->flags & (
517 1 << PG_lru |
518 1 << PG_private |
519 1 << PG_locked |
520 1 << PG_active |
521 1 << PG_dirty |
522 1 << PG_reclaim |
523 1 << PG_slab |
524 1 << PG_swapcache |
525 1 << PG_writeback |
526 1 << PG_reserved ))))
527 bad_page(page);
529 /*
530 * For now, we report if PG_reserved was found set, but do not
531 * clear it, and do not allocate the page: as a safety net.
532 */
533 if (PageReserved(page))
534 return 1;
536 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
537 1 << PG_referenced | 1 << PG_arch_1 |
538 1 << PG_checked | 1 << PG_mappedtodisk);
539 set_page_private(page, 0);
540 set_page_refs(page, order);
541 kernel_map_pages(page, 1 << order, 1);
542 return 0;
543 }
545 /*
546 * Do the hard work of removing an element from the buddy allocator.
547 * Call me with the zone->lock already held.
548 */
549 static struct page *__rmqueue(struct zone *zone, unsigned int order)
550 {
551 struct free_area * area;
552 unsigned int current_order;
553 struct page *page;
555 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
556 area = zone->free_area + current_order;
557 if (list_empty(&area->free_list))
558 continue;
560 page = list_entry(area->free_list.next, struct page, lru);
561 list_del(&page->lru);
562 rmv_page_order(page);
563 area->nr_free--;
564 zone->free_pages -= 1UL << order;
565 expand(zone, page, order, current_order, area);
566 return page;
567 }
569 return NULL;
570 }
572 /*
573 * Obtain a specified number of elements from the buddy allocator, all under
574 * a single hold of the lock, for efficiency. Add them to the supplied list.
575 * Returns the number of new pages which were placed at *list.
576 */
577 static int rmqueue_bulk(struct zone *zone, unsigned int order,
578 unsigned long count, struct list_head *list)
579 {
580 int i;
582 spin_lock(&zone->lock);
583 for (i = 0; i < count; ++i) {
584 struct page *page = __rmqueue(zone, order);
585 if (unlikely(page == NULL))
586 break;
587 list_add_tail(&page->lru, list);
588 }
589 spin_unlock(&zone->lock);
590 return i;
591 }
593 #ifdef CONFIG_NUMA
594 /* Called from the slab reaper to drain remote pagesets */
595 void drain_remote_pages(void)
596 {
597 struct zone *zone;
598 int i;
599 unsigned long flags;
601 local_irq_save(flags);
602 for_each_zone(zone) {
603 struct per_cpu_pageset *pset;
605 /* Do not drain local pagesets */
606 if (zone->zone_pgdat->node_id == numa_node_id())
607 continue;
609 pset = zone_pcp(zone, smp_processor_id());
610 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
611 struct per_cpu_pages *pcp;
613 pcp = &pset->pcp[i];
614 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
615 pcp->count = 0;
616 }
617 }
618 local_irq_restore(flags);
619 }
620 #endif
622 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
623 static void __drain_pages(unsigned int cpu)
624 {
625 unsigned long flags;
626 struct zone *zone;
627 int i;
629 for_each_zone(zone) {
630 struct per_cpu_pageset *pset;
632 pset = zone_pcp(zone, cpu);
633 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
634 struct per_cpu_pages *pcp;
636 pcp = &pset->pcp[i];
637 local_irq_save(flags);
638 free_pages_bulk(zone, pcp->count, &pcp->list, 0);
639 pcp->count = 0;
640 local_irq_restore(flags);
641 }
642 }
643 }
644 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
646 #ifdef CONFIG_PM
648 void mark_free_pages(struct zone *zone)
649 {
650 unsigned long zone_pfn, flags;
651 int order;
652 struct list_head *curr;
654 if (!zone->spanned_pages)
655 return;
657 spin_lock_irqsave(&zone->lock, flags);
658 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
659 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
661 for (order = MAX_ORDER - 1; order >= 0; --order)
662 list_for_each(curr, &zone->free_area[order].free_list) {
663 unsigned long start_pfn, i;
665 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
667 for (i=0; i < (1<<order); i++)
668 SetPageNosaveFree(pfn_to_page(start_pfn+i));
669 }
670 spin_unlock_irqrestore(&zone->lock, flags);
671 }
673 /*
674 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
675 */
676 void drain_local_pages(void)
677 {
678 unsigned long flags;
680 local_irq_save(flags);
681 __drain_pages(smp_processor_id());
682 local_irq_restore(flags);
683 }
684 #endif /* CONFIG_PM */
686 static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu)
687 {
688 #ifdef CONFIG_NUMA
689 pg_data_t *pg = z->zone_pgdat;
690 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
691 struct per_cpu_pageset *p;
693 p = zone_pcp(z, cpu);
694 if (pg == orig) {
695 p->numa_hit++;
696 } else {
697 p->numa_miss++;
698 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
699 }
700 if (pg == NODE_DATA(numa_node_id()))
701 p->local_node++;
702 else
703 p->other_node++;
704 #endif
705 }
707 /*
708 * Free a 0-order page
709 */
710 static void fastcall free_hot_cold_page(struct page *page, int cold)
711 {
712 struct zone *zone = page_zone(page);
713 struct per_cpu_pages *pcp;
714 unsigned long flags;
716 if (arch_free_page(page, 0))
717 return;
719 if (PageAnon(page))
720 page->mapping = NULL;
721 if (free_pages_check(page))
722 return;
724 kernel_map_pages(page, 1, 0);
726 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
727 local_irq_save(flags);
728 __inc_page_state(pgfree);
729 list_add(&page->lru, &pcp->list);
730 pcp->count++;
731 if (pcp->count >= pcp->high) {
732 free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
733 pcp->count -= pcp->batch;
734 }
735 local_irq_restore(flags);
736 put_cpu();
737 }
739 void fastcall free_hot_page(struct page *page)
740 {
741 free_hot_cold_page(page, 0);
742 }
744 void fastcall free_cold_page(struct page *page)
745 {
746 free_hot_cold_page(page, 1);
747 }
749 static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
750 {
751 int i;
753 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
754 for(i = 0; i < (1 << order); i++)
755 clear_highpage(page + i);
756 }
758 /*
759 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
760 * we cheat by calling it from here, in the order > 0 path. Saves a branch
761 * or two.
762 */
763 static struct page *buffered_rmqueue(struct zonelist *zonelist,
764 struct zone *zone, int order, gfp_t gfp_flags)
765 {
766 unsigned long flags;
767 struct page *page;
768 int cold = !!(gfp_flags & __GFP_COLD);
769 int cpu;
771 again:
772 cpu = get_cpu();
773 if (likely(order == 0)) {
774 struct per_cpu_pages *pcp;
776 pcp = &zone_pcp(zone, cpu)->pcp[cold];
777 local_irq_save(flags);
778 if (!pcp->count) {
779 pcp->count += rmqueue_bulk(zone, 0,
780 pcp->batch, &pcp->list);
781 if (unlikely(!pcp->count))
782 goto failed;
783 }
784 page = list_entry(pcp->list.next, struct page, lru);
785 list_del(&page->lru);
786 pcp->count--;
787 } else {
788 spin_lock_irqsave(&zone->lock, flags);
789 page = __rmqueue(zone, order);
790 spin_unlock(&zone->lock);
791 if (!page)
792 goto failed;
793 }
795 __mod_page_state_zone(zone, pgalloc, 1 << order);
796 zone_statistics(zonelist, zone, cpu);
797 local_irq_restore(flags);
798 put_cpu();
800 BUG_ON(bad_range(zone, page));
801 if (prep_new_page(page, order))
802 goto again;
804 if (gfp_flags & __GFP_ZERO)
805 prep_zero_page(page, order, gfp_flags);
807 if (order && (gfp_flags & __GFP_COMP))
808 prep_compound_page(page, order);
809 return page;
811 failed:
812 local_irq_restore(flags);
813 put_cpu();
814 return NULL;
815 }
817 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
818 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
819 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
820 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
821 #define ALLOC_HARDER 0x10 /* try to alloc harder */
822 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
823 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
825 /*
826 * Return 1 if free pages are above 'mark'. This takes into account the order
827 * of the allocation.
828 */
829 int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
830 int classzone_idx, int alloc_flags)
831 {
832 /* free_pages my go negative - that's OK */
833 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
834 int o;
836 if (alloc_flags & ALLOC_HIGH)
837 min -= min / 2;
838 if (alloc_flags & ALLOC_HARDER)
839 min -= min / 4;
841 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
842 return 0;
843 for (o = 0; o < order; o++) {
844 /* At the next order, this order's pages become unavailable */
845 free_pages -= z->free_area[o].nr_free << o;
847 /* Require fewer higher order pages to be free */
848 min >>= 1;
850 if (free_pages <= min)
851 return 0;
852 }
853 return 1;
854 }
856 /*
857 * get_page_from_freeliest goes through the zonelist trying to allocate
858 * a page.
859 */
860 static struct page *
861 get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
862 struct zonelist *zonelist, int alloc_flags)
863 {
864 struct zone **z = zonelist->zones;
865 struct page *page = NULL;
866 int classzone_idx = zone_idx(*z);
868 /*
869 * Go through the zonelist once, looking for a zone with enough free.
870 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
871 */
872 do {
873 if ((alloc_flags & ALLOC_CPUSET) &&
874 !cpuset_zone_allowed(*z, gfp_mask))
875 continue;
877 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
878 unsigned long mark;
879 if (alloc_flags & ALLOC_WMARK_MIN)
880 mark = (*z)->pages_min;
881 else if (alloc_flags & ALLOC_WMARK_LOW)
882 mark = (*z)->pages_low;
883 else
884 mark = (*z)->pages_high;
885 if (!zone_watermark_ok(*z, order, mark,
886 classzone_idx, alloc_flags))
887 if (!zone_reclaim_mode ||
888 !zone_reclaim(*z, gfp_mask, order))
889 continue;
890 }
892 page = buffered_rmqueue(zonelist, *z, order, gfp_mask);
893 if (page) {
894 break;
895 }
896 } while (*(++z) != NULL);
897 return page;
898 }
900 /*
901 * This is the 'heart' of the zoned buddy allocator.
902 */
903 struct page * fastcall
904 __alloc_pages(gfp_t gfp_mask, unsigned int order,
905 struct zonelist *zonelist)
906 {
907 const gfp_t wait = gfp_mask & __GFP_WAIT;
908 struct zone **z;
909 struct page *page;
910 struct reclaim_state reclaim_state;
911 struct task_struct *p = current;
912 int do_retry;
913 int alloc_flags;
914 int did_some_progress;
916 might_sleep_if(wait);
918 restart:
919 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
921 if (unlikely(*z == NULL)) {
922 /* Should this ever happen?? */
923 return NULL;
924 }
926 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
927 zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET);
928 if (page)
929 goto got_pg;
931 do {
932 wakeup_kswapd(*z, order);
933 } while (*(++z));
935 /*
936 * OK, we're below the kswapd watermark and have kicked background
937 * reclaim. Now things get more complex, so set up alloc_flags according
938 * to how we want to proceed.
939 *
940 * The caller may dip into page reserves a bit more if the caller
941 * cannot run direct reclaim, or if the caller has realtime scheduling
942 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
943 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
944 */
945 alloc_flags = ALLOC_WMARK_MIN;
946 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
947 alloc_flags |= ALLOC_HARDER;
948 if (gfp_mask & __GFP_HIGH)
949 alloc_flags |= ALLOC_HIGH;
950 alloc_flags |= ALLOC_CPUSET;
952 /*
953 * Go through the zonelist again. Let __GFP_HIGH and allocations
954 * coming from realtime tasks go deeper into reserves.
955 *
956 * This is the last chance, in general, before the goto nopage.
957 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
958 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
959 */
960 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
961 if (page)
962 goto got_pg;
964 /* This allocation should allow future memory freeing. */
966 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
967 && !in_interrupt()) {
968 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
969 nofail_alloc:
970 /* go through the zonelist yet again, ignoring mins */
971 page = get_page_from_freelist(gfp_mask, order,
972 zonelist, ALLOC_NO_WATERMARKS);
973 if (page)
974 goto got_pg;
975 if (gfp_mask & __GFP_NOFAIL) {
976 blk_congestion_wait(WRITE, HZ/50);
977 goto nofail_alloc;
978 }
979 }
980 goto nopage;
981 }
983 /* Atomic allocations - we can't balance anything */
984 if (!wait)
985 goto nopage;
987 rebalance:
988 cond_resched();
990 /* We now go into synchronous reclaim */
991 cpuset_memory_pressure_bump();
992 p->flags |= PF_MEMALLOC;
993 reclaim_state.reclaimed_slab = 0;
994 p->reclaim_state = &reclaim_state;
996 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
998 p->reclaim_state = NULL;
999 p->flags &= ~PF_MEMALLOC;
1001 cond_resched();
1003 if (likely(did_some_progress)) {
1004 page = get_page_from_freelist(gfp_mask, order,
1005 zonelist, alloc_flags);
1006 if (page)
1007 goto got_pg;
1008 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1009 /*
1010 * Go through the zonelist yet one more time, keep
1011 * very high watermark here, this is only to catch
1012 * a parallel oom killing, we must fail if we're still
1013 * under heavy pressure.
1014 */
1015 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
1016 zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1017 if (page)
1018 goto got_pg;
1020 out_of_memory(zonelist, gfp_mask, order);
1021 goto restart;
1024 /*
1025 * Don't let big-order allocations loop unless the caller explicitly
1026 * requests that. Wait for some write requests to complete then retry.
1028 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1029 * <= 3, but that may not be true in other implementations.
1030 */
1031 do_retry = 0;
1032 if (!(gfp_mask & __GFP_NORETRY)) {
1033 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
1034 do_retry = 1;
1035 if (gfp_mask & __GFP_NOFAIL)
1036 do_retry = 1;
1038 if (do_retry) {
1039 blk_congestion_wait(WRITE, HZ/50);
1040 goto rebalance;
1043 nopage:
1044 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1045 printk(KERN_WARNING "%s: page allocation failure."
1046 " order:%d, mode:0x%x\n",
1047 p->comm, order, gfp_mask);
1048 dump_stack();
1049 show_mem();
1051 got_pg:
1052 return page;
1055 EXPORT_SYMBOL(__alloc_pages);
1057 /*
1058 * Common helper functions.
1059 */
1060 fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1062 struct page * page;
1063 page = alloc_pages(gfp_mask, order);
1064 if (!page)
1065 return 0;
1066 return (unsigned long) page_address(page);
1069 EXPORT_SYMBOL(__get_free_pages);
1071 fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
1073 struct page * page;
1075 /*
1076 * get_zeroed_page() returns a 32-bit address, which cannot represent
1077 * a highmem page
1078 */
1079 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1081 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1082 if (page)
1083 return (unsigned long) page_address(page);
1084 return 0;
1087 EXPORT_SYMBOL(get_zeroed_page);
1089 void __pagevec_free(struct pagevec *pvec)
1091 int i = pagevec_count(pvec);
1093 while (--i >= 0)
1094 free_hot_cold_page(pvec->pages[i], pvec->cold);
1097 fastcall void __free_pages(struct page *page, unsigned int order)
1099 if (put_page_testzero(page)) {
1100 if (order == 0)
1101 free_hot_page(page);
1102 else
1103 __free_pages_ok(page, order);
1107 EXPORT_SYMBOL(__free_pages);
1109 fastcall void free_pages(unsigned long addr, unsigned int order)
1111 if (addr != 0) {
1112 BUG_ON(!virt_addr_valid((void *)addr));
1113 __free_pages(virt_to_page((void *)addr), order);
1117 EXPORT_SYMBOL(free_pages);
1119 /*
1120 * Total amount of free (allocatable) RAM:
1121 */
1122 unsigned int nr_free_pages(void)
1124 unsigned int sum = 0;
1125 struct zone *zone;
1127 for_each_zone(zone)
1128 sum += zone->free_pages;
1130 return sum;
1133 EXPORT_SYMBOL(nr_free_pages);
1135 #ifdef CONFIG_NUMA
1136 unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1138 unsigned int i, sum = 0;
1140 for (i = 0; i < MAX_NR_ZONES; i++)
1141 sum += pgdat->node_zones[i].free_pages;
1143 return sum;
1145 #endif
1147 static unsigned int nr_free_zone_pages(int offset)
1149 /* Just pick one node, since fallback list is circular */
1150 pg_data_t *pgdat = NODE_DATA(numa_node_id());
1151 unsigned int sum = 0;
1153 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1154 struct zone **zonep = zonelist->zones;
1155 struct zone *zone;
1157 for (zone = *zonep++; zone; zone = *zonep++) {
1158 unsigned long size = zone->present_pages;
1159 unsigned long high = zone->pages_high;
1160 if (size > high)
1161 sum += size - high;
1164 return sum;
1167 /*
1168 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1169 */
1170 unsigned int nr_free_buffer_pages(void)
1172 return nr_free_zone_pages(gfp_zone(GFP_USER));
1175 /*
1176 * Amount of free RAM allocatable within all zones
1177 */
1178 unsigned int nr_free_pagecache_pages(void)
1180 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
1183 #ifdef CONFIG_HIGHMEM
1184 unsigned int nr_free_highpages (void)
1186 pg_data_t *pgdat;
1187 unsigned int pages = 0;
1189 for_each_pgdat(pgdat)
1190 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1192 return pages;
1194 #endif
1196 #ifdef CONFIG_NUMA
1197 static void show_node(struct zone *zone)
1199 printk("Node %d ", zone->zone_pgdat->node_id);
1201 #else
1202 #define show_node(zone) do { } while (0)
1203 #endif
1205 /*
1206 * Accumulate the page_state information across all CPUs.
1207 * The result is unavoidably approximate - it can change
1208 * during and after execution of this function.
1209 */
1210 static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1212 atomic_t nr_pagecache = ATOMIC_INIT(0);
1213 EXPORT_SYMBOL(nr_pagecache);
1214 #ifdef CONFIG_SMP
1215 DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1216 #endif
1218 static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
1220 int cpu = 0;
1222 memset(ret, 0, nr * sizeof(unsigned long));
1223 cpus_and(*cpumask, *cpumask, cpu_online_map);
1225 cpu = first_cpu(*cpumask);
1226 while (cpu < NR_CPUS) {
1227 unsigned long *in, *out, off;
1229 if (!cpu_isset(cpu, *cpumask))
1230 continue;
1232 in = (unsigned long *)&per_cpu(page_states, cpu);
1234 cpu = next_cpu(cpu, *cpumask);
1236 if (likely(cpu < NR_CPUS))
1237 prefetch(&per_cpu(page_states, cpu));
1239 out = (unsigned long *)ret;
1240 for (off = 0; off < nr; off++)
1241 *out++ += *in++;
1245 void get_page_state_node(struct page_state *ret, int node)
1247 int nr;
1248 cpumask_t mask = node_to_cpumask(node);
1250 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1251 nr /= sizeof(unsigned long);
1253 __get_page_state(ret, nr+1, &mask);
1256 void get_page_state(struct page_state *ret)
1258 int nr;
1259 cpumask_t mask = CPU_MASK_ALL;
1261 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1262 nr /= sizeof(unsigned long);
1264 __get_page_state(ret, nr + 1, &mask);
1267 void get_full_page_state(struct page_state *ret)
1269 cpumask_t mask = CPU_MASK_ALL;
1271 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
1274 unsigned long read_page_state_offset(unsigned long offset)
1276 unsigned long ret = 0;
1277 int cpu;
1279 for_each_online_cpu(cpu) {
1280 unsigned long in;
1282 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1283 ret += *((unsigned long *)in);
1285 return ret;
1288 void __mod_page_state_offset(unsigned long offset, unsigned long delta)
1290 void *ptr;
1292 ptr = &__get_cpu_var(page_states);
1293 *(unsigned long *)(ptr + offset) += delta;
1295 EXPORT_SYMBOL(__mod_page_state_offset);
1297 void mod_page_state_offset(unsigned long offset, unsigned long delta)
1299 unsigned long flags;
1300 void *ptr;
1302 local_irq_save(flags);
1303 ptr = &__get_cpu_var(page_states);
1304 *(unsigned long *)(ptr + offset) += delta;
1305 local_irq_restore(flags);
1307 EXPORT_SYMBOL(mod_page_state_offset);
1309 void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1310 unsigned long *free, struct pglist_data *pgdat)
1312 struct zone *zones = pgdat->node_zones;
1313 int i;
1315 *active = 0;
1316 *inactive = 0;
1317 *free = 0;
1318 for (i = 0; i < MAX_NR_ZONES; i++) {
1319 *active += zones[i].nr_active;
1320 *inactive += zones[i].nr_inactive;
1321 *free += zones[i].free_pages;
1325 void get_zone_counts(unsigned long *active,
1326 unsigned long *inactive, unsigned long *free)
1328 struct pglist_data *pgdat;
1330 *active = 0;
1331 *inactive = 0;
1332 *free = 0;
1333 for_each_pgdat(pgdat) {
1334 unsigned long l, m, n;
1335 __get_zone_counts(&l, &m, &n, pgdat);
1336 *active += l;
1337 *inactive += m;
1338 *free += n;
1342 void si_meminfo(struct sysinfo *val)
1344 val->totalram = totalram_pages;
1345 val->sharedram = 0;
1346 val->freeram = nr_free_pages();
1347 val->bufferram = nr_blockdev_pages();
1348 #ifdef CONFIG_HIGHMEM
1349 val->totalhigh = totalhigh_pages;
1350 val->freehigh = nr_free_highpages();
1351 #else
1352 val->totalhigh = 0;
1353 val->freehigh = 0;
1354 #endif
1355 val->mem_unit = PAGE_SIZE;
1358 EXPORT_SYMBOL(si_meminfo);
1360 #ifdef CONFIG_NUMA
1361 void si_meminfo_node(struct sysinfo *val, int nid)
1363 pg_data_t *pgdat = NODE_DATA(nid);
1365 val->totalram = pgdat->node_present_pages;
1366 val->freeram = nr_free_pages_pgdat(pgdat);
1367 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1368 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1369 val->mem_unit = PAGE_SIZE;
1371 #endif
1373 #define K(x) ((x) << (PAGE_SHIFT-10))
1375 /*
1376 * Show free area list (used inside shift_scroll-lock stuff)
1377 * We also calculate the percentage fragmentation. We do this by counting the
1378 * memory on each free list with the exception of the first item on the list.
1379 */
1380 void show_free_areas(void)
1382 struct page_state ps;
1383 int cpu, temperature;
1384 unsigned long active;
1385 unsigned long inactive;
1386 unsigned long free;
1387 struct zone *zone;
1389 for_each_zone(zone) {
1390 show_node(zone);
1391 printk("%s per-cpu:", zone->name);
1393 if (!populated_zone(zone)) {
1394 printk(" empty\n");
1395 continue;
1396 } else
1397 printk("\n");
1399 for_each_online_cpu(cpu) {
1400 struct per_cpu_pageset *pageset;
1402 pageset = zone_pcp(zone, cpu);
1404 for (temperature = 0; temperature < 2; temperature++)
1405 printk("cpu %d %s: high %d, batch %d used:%d\n",
1406 cpu,
1407 temperature ? "cold" : "hot",
1408 pageset->pcp[temperature].high,
1409 pageset->pcp[temperature].batch,
1410 pageset->pcp[temperature].count);
1414 get_page_state(&ps);
1415 get_zone_counts(&active, &inactive, &free);
1417 printk("Free pages: %11ukB (%ukB HighMem)\n",
1418 K(nr_free_pages()),
1419 K(nr_free_highpages()));
1421 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1422 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1423 active,
1424 inactive,
1425 ps.nr_dirty,
1426 ps.nr_writeback,
1427 ps.nr_unstable,
1428 nr_free_pages(),
1429 ps.nr_slab,
1430 ps.nr_mapped,
1431 ps.nr_page_table_pages);
1433 for_each_zone(zone) {
1434 int i;
1436 show_node(zone);
1437 printk("%s"
1438 " free:%lukB"
1439 " min:%lukB"
1440 " low:%lukB"
1441 " high:%lukB"
1442 " active:%lukB"
1443 " inactive:%lukB"
1444 " present:%lukB"
1445 " pages_scanned:%lu"
1446 " all_unreclaimable? %s"
1447 "\n",
1448 zone->name,
1449 K(zone->free_pages),
1450 K(zone->pages_min),
1451 K(zone->pages_low),
1452 K(zone->pages_high),
1453 K(zone->nr_active),
1454 K(zone->nr_inactive),
1455 K(zone->present_pages),
1456 zone->pages_scanned,
1457 (zone->all_unreclaimable ? "yes" : "no")
1458 );
1459 printk("lowmem_reserve[]:");
1460 for (i = 0; i < MAX_NR_ZONES; i++)
1461 printk(" %lu", zone->lowmem_reserve[i]);
1462 printk("\n");
1465 for_each_zone(zone) {
1466 unsigned long nr, flags, order, total = 0;
1468 show_node(zone);
1469 printk("%s: ", zone->name);
1470 if (!populated_zone(zone)) {
1471 printk("empty\n");
1472 continue;
1475 spin_lock_irqsave(&zone->lock, flags);
1476 for (order = 0; order < MAX_ORDER; order++) {
1477 nr = zone->free_area[order].nr_free;
1478 total += nr << order;
1479 printk("%lu*%lukB ", nr, K(1UL) << order);
1481 spin_unlock_irqrestore(&zone->lock, flags);
1482 printk("= %lukB\n", K(total));
1485 show_swap_cache_info();
1488 /*
1489 * Builds allocation fallback zone lists.
1491 * Add all populated zones of a node to the zonelist.
1492 */
1493 static int __init build_zonelists_node(pg_data_t *pgdat,
1494 struct zonelist *zonelist, int nr_zones, int zone_type)
1496 struct zone *zone;
1498 BUG_ON(zone_type > ZONE_HIGHMEM);
1500 do {
1501 zone = pgdat->node_zones + zone_type;
1502 if (populated_zone(zone)) {
1503 #ifndef CONFIG_HIGHMEM
1504 BUG_ON(zone_type > ZONE_NORMAL);
1505 #endif
1506 zonelist->zones[nr_zones++] = zone;
1507 check_highest_zone(zone_type);
1509 zone_type--;
1511 } while (zone_type >= 0);
1512 return nr_zones;
1515 static inline int highest_zone(int zone_bits)
1517 int res = ZONE_NORMAL;
1518 if (zone_bits & (__force int)__GFP_HIGHMEM)
1519 res = ZONE_HIGHMEM;
1520 if (zone_bits & (__force int)__GFP_DMA32)
1521 res = ZONE_DMA32;
1522 if (zone_bits & (__force int)__GFP_DMA)
1523 res = ZONE_DMA;
1524 return res;
1527 #ifdef CONFIG_NUMA
1528 #define MAX_NODE_LOAD (num_online_nodes())
1529 static int __initdata node_load[MAX_NUMNODES];
1530 /**
1531 * find_next_best_node - find the next node that should appear in a given node's fallback list
1532 * @node: node whose fallback list we're appending
1533 * @used_node_mask: nodemask_t of already used nodes
1535 * We use a number of factors to determine which is the next node that should
1536 * appear on a given node's fallback list. The node should not have appeared
1537 * already in @node's fallback list, and it should be the next closest node
1538 * according to the distance array (which contains arbitrary distance values
1539 * from each node to each node in the system), and should also prefer nodes
1540 * with no CPUs, since presumably they'll have very little allocation pressure
1541 * on them otherwise.
1542 * It returns -1 if no node is found.
1543 */
1544 static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1546 int n, val;
1547 int min_val = INT_MAX;
1548 int best_node = -1;
1550 /* Use the local node if we haven't already */
1551 if (!node_isset(node, *used_node_mask)) {
1552 node_set(node, *used_node_mask);
1553 return node;
1556 for_each_online_node(n) {
1557 cpumask_t tmp;
1559 /* Don't want a node to appear more than once */
1560 if (node_isset(n, *used_node_mask))
1561 continue;
1563 /* Use the distance array to find the distance */
1564 val = node_distance(node, n);
1566 /* Penalize nodes under us ("prefer the next node") */
1567 val += (n < node);
1569 /* Give preference to headless and unused nodes */
1570 tmp = node_to_cpumask(n);
1571 if (!cpus_empty(tmp))
1572 val += PENALTY_FOR_NODE_WITH_CPUS;
1574 /* Slight preference for less loaded node */
1575 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1576 val += node_load[n];
1578 if (val < min_val) {
1579 min_val = val;
1580 best_node = n;
1584 if (best_node >= 0)
1585 node_set(best_node, *used_node_mask);
1587 return best_node;
1590 static void __init build_zonelists(pg_data_t *pgdat)
1592 int i, j, k, node, local_node;
1593 int prev_node, load;
1594 struct zonelist *zonelist;
1595 nodemask_t used_mask;
1597 /* initialize zonelists */
1598 for (i = 0; i < GFP_ZONETYPES; i++) {
1599 zonelist = pgdat->node_zonelists + i;
1600 zonelist->zones[0] = NULL;
1603 /* NUMA-aware ordering of nodes */
1604 local_node = pgdat->node_id;
1605 load = num_online_nodes();
1606 prev_node = local_node;
1607 nodes_clear(used_mask);
1608 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1609 int distance = node_distance(local_node, node);
1611 /*
1612 * If another node is sufficiently far away then it is better
1613 * to reclaim pages in a zone before going off node.
1614 */
1615 if (distance > RECLAIM_DISTANCE)
1616 zone_reclaim_mode = 1;
1618 /*
1619 * We don't want to pressure a particular node.
1620 * So adding penalty to the first node in same
1621 * distance group to make it round-robin.
1622 */
1624 if (distance != node_distance(local_node, prev_node))
1625 node_load[node] += load;
1626 prev_node = node;
1627 load--;
1628 for (i = 0; i < GFP_ZONETYPES; i++) {
1629 zonelist = pgdat->node_zonelists + i;
1630 for (j = 0; zonelist->zones[j] != NULL; j++);
1632 k = highest_zone(i);
1634 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1635 zonelist->zones[j] = NULL;
1640 #else /* CONFIG_NUMA */
1642 static void __init build_zonelists(pg_data_t *pgdat)
1644 int i, j, k, node, local_node;
1646 local_node = pgdat->node_id;
1647 for (i = 0; i < GFP_ZONETYPES; i++) {
1648 struct zonelist *zonelist;
1650 zonelist = pgdat->node_zonelists + i;
1652 j = 0;
1653 k = highest_zone(i);
1654 j = build_zonelists_node(pgdat, zonelist, j, k);
1655 /*
1656 * Now we build the zonelist so that it contains the zones
1657 * of all the other nodes.
1658 * We don't want to pressure a particular node, so when
1659 * building the zones for node N, we make sure that the
1660 * zones coming right after the local ones are those from
1661 * node N+1 (modulo N)
1662 */
1663 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1664 if (!node_online(node))
1665 continue;
1666 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1668 for (node = 0; node < local_node; node++) {
1669 if (!node_online(node))
1670 continue;
1671 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1674 zonelist->zones[j] = NULL;
1678 #endif /* CONFIG_NUMA */
1680 void __init build_all_zonelists(void)
1682 int i;
1684 for_each_online_node(i)
1685 build_zonelists(NODE_DATA(i));
1686 printk("Built %i zonelists\n", num_online_nodes());
1687 cpuset_init_current_mems_allowed();
1690 /*
1691 * Helper functions to size the waitqueue hash table.
1692 * Essentially these want to choose hash table sizes sufficiently
1693 * large so that collisions trying to wait on pages are rare.
1694 * But in fact, the number of active page waitqueues on typical
1695 * systems is ridiculously low, less than 200. So this is even
1696 * conservative, even though it seems large.
1698 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1699 * waitqueues, i.e. the size of the waitq table given the number of pages.
1700 */
1701 #define PAGES_PER_WAITQUEUE 256
1703 static inline unsigned long wait_table_size(unsigned long pages)
1705 unsigned long size = 1;
1707 pages /= PAGES_PER_WAITQUEUE;
1709 while (size < pages)
1710 size <<= 1;
1712 /*
1713 * Once we have dozens or even hundreds of threads sleeping
1714 * on IO we've got bigger problems than wait queue collision.
1715 * Limit the size of the wait table to a reasonable size.
1716 */
1717 size = min(size, 4096UL);
1719 return max(size, 4UL);
1722 /*
1723 * This is an integer logarithm so that shifts can be used later
1724 * to extract the more random high bits from the multiplicative
1725 * hash function before the remainder is taken.
1726 */
1727 static inline unsigned long wait_table_bits(unsigned long size)
1729 return ffz(~size);
1732 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1734 static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1735 unsigned long *zones_size, unsigned long *zholes_size)
1737 unsigned long realtotalpages, totalpages = 0;
1738 int i;
1740 for (i = 0; i < MAX_NR_ZONES; i++)
1741 totalpages += zones_size[i];
1742 pgdat->node_spanned_pages = totalpages;
1744 realtotalpages = totalpages;
1745 if (zholes_size)
1746 for (i = 0; i < MAX_NR_ZONES; i++)
1747 realtotalpages -= zholes_size[i];
1748 pgdat->node_present_pages = realtotalpages;
1749 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1753 /*
1754 * Initially all pages are reserved - free ones are freed
1755 * up by free_all_bootmem() once the early boot process is
1756 * done. Non-atomic initialization, single-pass.
1757 */
1758 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1759 unsigned long start_pfn)
1761 struct page *page;
1762 unsigned long end_pfn = start_pfn + size;
1763 unsigned long pfn;
1765 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
1766 if (!early_pfn_valid(pfn))
1767 continue;
1768 page = pfn_to_page(pfn);
1769 set_page_links(page, zone, nid, pfn);
1770 set_page_count(page, 1);
1771 reset_page_mapcount(page);
1772 SetPageReserved(page);
1773 INIT_LIST_HEAD(&page->lru);
1774 #ifdef WANT_PAGE_VIRTUAL
1775 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1776 if (!is_highmem_idx(zone))
1777 set_page_address(page, __va(pfn << PAGE_SHIFT));
1778 #endif
1782 void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1783 unsigned long size)
1785 int order;
1786 for (order = 0; order < MAX_ORDER ; order++) {
1787 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1788 zone->free_area[order].nr_free = 0;
1792 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1793 void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1794 unsigned long size)
1796 unsigned long snum = pfn_to_section_nr(pfn);
1797 unsigned long end = pfn_to_section_nr(pfn + size);
1799 if (FLAGS_HAS_NODE)
1800 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1801 else
1802 for (; snum <= end; snum++)
1803 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1806 #ifndef __HAVE_ARCH_MEMMAP_INIT
1807 #define memmap_init(size, nid, zone, start_pfn) \
1808 memmap_init_zone((size), (nid), (zone), (start_pfn))
1809 #endif
1811 static int __cpuinit zone_batchsize(struct zone *zone)
1813 int batch;
1815 /*
1816 * The per-cpu-pages pools are set to around 1000th of the
1817 * size of the zone. But no more than 1/2 of a meg.
1819 * OK, so we don't know how big the cache is. So guess.
1820 */
1821 batch = zone->present_pages / 1024;
1822 if (batch * PAGE_SIZE > 512 * 1024)
1823 batch = (512 * 1024) / PAGE_SIZE;
1824 batch /= 4; /* We effectively *= 4 below */
1825 if (batch < 1)
1826 batch = 1;
1828 /*
1829 * Clamp the batch to a 2^n - 1 value. Having a power
1830 * of 2 value was found to be more likely to have
1831 * suboptimal cache aliasing properties in some cases.
1833 * For example if 2 tasks are alternately allocating
1834 * batches of pages, one task can end up with a lot
1835 * of pages of one half of the possible page colors
1836 * and the other with pages of the other colors.
1837 */
1838 batch = (1 << (fls(batch + batch/2)-1)) - 1;
1840 return batch;
1843 inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1845 struct per_cpu_pages *pcp;
1847 memset(p, 0, sizeof(*p));
1849 pcp = &p->pcp[0]; /* hot */
1850 pcp->count = 0;
1851 pcp->high = 6 * batch;
1852 pcp->batch = max(1UL, 1 * batch);
1853 INIT_LIST_HEAD(&pcp->list);
1855 pcp = &p->pcp[1]; /* cold*/
1856 pcp->count = 0;
1857 pcp->high = 2 * batch;
1858 pcp->batch = max(1UL, batch/2);
1859 INIT_LIST_HEAD(&pcp->list);
1862 /*
1863 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1864 * to the value high for the pageset p.
1865 */
1867 static void setup_pagelist_highmark(struct per_cpu_pageset *p,
1868 unsigned long high)
1870 struct per_cpu_pages *pcp;
1872 pcp = &p->pcp[0]; /* hot list */
1873 pcp->high = high;
1874 pcp->batch = max(1UL, high/4);
1875 if ((high/4) > (PAGE_SHIFT * 8))
1876 pcp->batch = PAGE_SHIFT * 8;
1880 #ifdef CONFIG_NUMA
1881 /*
1882 * Boot pageset table. One per cpu which is going to be used for all
1883 * zones and all nodes. The parameters will be set in such a way
1884 * that an item put on a list will immediately be handed over to
1885 * the buddy list. This is safe since pageset manipulation is done
1886 * with interrupts disabled.
1888 * Some NUMA counter updates may also be caught by the boot pagesets.
1890 * The boot_pagesets must be kept even after bootup is complete for
1891 * unused processors and/or zones. They do play a role for bootstrapping
1892 * hotplugged processors.
1894 * zoneinfo_show() and maybe other functions do
1895 * not check if the processor is online before following the pageset pointer.
1896 * Other parts of the kernel may not check if the zone is available.
1897 */
1898 static struct per_cpu_pageset boot_pageset[NR_CPUS];
1900 /*
1901 * Dynamically allocate memory for the
1902 * per cpu pageset array in struct zone.
1903 */
1904 static int __cpuinit process_zones(int cpu)
1906 struct zone *zone, *dzone;
1908 for_each_zone(zone) {
1910 zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
1911 GFP_KERNEL, cpu_to_node(cpu));
1912 if (!zone_pcp(zone, cpu))
1913 goto bad;
1915 setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
1917 if (percpu_pagelist_fraction)
1918 setup_pagelist_highmark(zone_pcp(zone, cpu),
1919 (zone->present_pages / percpu_pagelist_fraction));
1922 return 0;
1923 bad:
1924 for_each_zone(dzone) {
1925 if (dzone == zone)
1926 break;
1927 kfree(zone_pcp(dzone, cpu));
1928 zone_pcp(dzone, cpu) = NULL;
1930 return -ENOMEM;
1933 static inline void free_zone_pagesets(int cpu)
1935 struct zone *zone;
1937 for_each_zone(zone) {
1938 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1940 zone_pcp(zone, cpu) = NULL;
1941 kfree(pset);
1945 static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
1946 unsigned long action,
1947 void *hcpu)
1949 int cpu = (long)hcpu;
1950 int ret = NOTIFY_OK;
1952 switch (action) {
1953 case CPU_UP_PREPARE:
1954 if (process_zones(cpu))
1955 ret = NOTIFY_BAD;
1956 break;
1957 case CPU_UP_CANCELED:
1958 case CPU_DEAD:
1959 free_zone_pagesets(cpu);
1960 break;
1961 default:
1962 break;
1964 return ret;
1967 static struct notifier_block pageset_notifier =
1968 { &pageset_cpuup_callback, NULL, 0 };
1970 void __init setup_per_cpu_pageset(void)
1972 int err;
1974 /* Initialize per_cpu_pageset for cpu 0.
1975 * A cpuup callback will do this for every cpu
1976 * as it comes online
1977 */
1978 err = process_zones(smp_processor_id());
1979 BUG_ON(err);
1980 register_cpu_notifier(&pageset_notifier);
1983 #endif
1985 static __meminit
1986 void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1988 int i;
1989 struct pglist_data *pgdat = zone->zone_pgdat;
1991 /*
1992 * The per-page waitqueue mechanism uses hashed waitqueues
1993 * per zone.
1994 */
1995 zone->wait_table_size = wait_table_size(zone_size_pages);
1996 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1997 zone->wait_table = (wait_queue_head_t *)
1998 alloc_bootmem_node(pgdat, zone->wait_table_size
1999 * sizeof(wait_queue_head_t));
2001 for(i = 0; i < zone->wait_table_size; ++i)
2002 init_waitqueue_head(zone->wait_table + i);
2005 static __meminit void zone_pcp_init(struct zone *zone)
2007 int cpu;
2008 unsigned long batch = zone_batchsize(zone);
2010 for (cpu = 0; cpu < NR_CPUS; cpu++) {
2011 #ifdef CONFIG_NUMA
2012 /* Early boot. Slab allocator not functional yet */
2013 zone_pcp(zone, cpu) = &boot_pageset[cpu];
2014 setup_pageset(&boot_pageset[cpu],0);
2015 #else
2016 setup_pageset(zone_pcp(zone,cpu), batch);
2017 #endif
2019 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
2020 zone->name, zone->present_pages, batch);
2023 static __meminit void init_currently_empty_zone(struct zone *zone,
2024 unsigned long zone_start_pfn, unsigned long size)
2026 struct pglist_data *pgdat = zone->zone_pgdat;
2028 zone_wait_table_init(zone, size);
2029 pgdat->nr_zones = zone_idx(zone) + 1;
2031 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
2032 zone->zone_start_pfn = zone_start_pfn;
2034 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
2036 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
2039 /*
2040 * Set up the zone data structures:
2041 * - mark all pages reserved
2042 * - mark all memory queues empty
2043 * - clear the memory bitmaps
2044 */
2045 static void __init free_area_init_core(struct pglist_data *pgdat,
2046 unsigned long *zones_size, unsigned long *zholes_size)
2048 unsigned long j;
2049 int nid = pgdat->node_id;
2050 unsigned long zone_start_pfn = pgdat->node_start_pfn;
2052 pgdat_resize_init(pgdat);
2053 pgdat->nr_zones = 0;
2054 init_waitqueue_head(&pgdat->kswapd_wait);
2055 pgdat->kswapd_max_order = 0;
2057 for (j = 0; j < MAX_NR_ZONES; j++) {
2058 struct zone *zone = pgdat->node_zones + j;
2059 unsigned long size, realsize;
2061 realsize = size = zones_size[j];
2062 if (zholes_size)
2063 realsize -= zholes_size[j];
2065 if (j < ZONE_HIGHMEM)
2066 nr_kernel_pages += realsize;
2067 nr_all_pages += realsize;
2069 zone->spanned_pages = size;
2070 zone->present_pages = realsize;
2071 zone->name = zone_names[j];
2072 spin_lock_init(&zone->lock);
2073 spin_lock_init(&zone->lru_lock);
2074 zone_seqlock_init(zone);
2075 zone->zone_pgdat = pgdat;
2076 zone->free_pages = 0;
2078 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
2080 zone_pcp_init(zone);
2081 INIT_LIST_HEAD(&zone->active_list);
2082 INIT_LIST_HEAD(&zone->inactive_list);
2083 zone->nr_scan_active = 0;
2084 zone->nr_scan_inactive = 0;
2085 zone->nr_active = 0;
2086 zone->nr_inactive = 0;
2087 atomic_set(&zone->reclaim_in_progress, 0);
2088 if (!size)
2089 continue;
2091 zonetable_add(zone, nid, j, zone_start_pfn, size);
2092 init_currently_empty_zone(zone, zone_start_pfn, size);
2093 zone_start_pfn += size;
2097 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
2099 /* Skip empty nodes */
2100 if (!pgdat->node_spanned_pages)
2101 return;
2103 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2104 /* ia64 gets its own node_mem_map, before this, without bootmem */
2105 if (!pgdat->node_mem_map) {
2106 unsigned long size;
2107 struct page *map;
2109 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
2110 map = alloc_remap(pgdat->node_id, size);
2111 if (!map)
2112 map = alloc_bootmem_node(pgdat, size);
2113 pgdat->node_mem_map = map;
2115 #ifdef CONFIG_FLATMEM
2116 /*
2117 * With no DISCONTIG, the global mem_map is just set as node 0's
2118 */
2119 if (pgdat == NODE_DATA(0))
2120 mem_map = NODE_DATA(0)->node_mem_map;
2121 #endif
2122 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2125 void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2126 unsigned long *zones_size, unsigned long node_start_pfn,
2127 unsigned long *zholes_size)
2129 pgdat->node_id = nid;
2130 pgdat->node_start_pfn = node_start_pfn;
2131 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2133 alloc_node_mem_map(pgdat);
2135 free_area_init_core(pgdat, zones_size, zholes_size);
2138 #ifndef CONFIG_NEED_MULTIPLE_NODES
2139 static bootmem_data_t contig_bootmem_data;
2140 struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2142 EXPORT_SYMBOL(contig_page_data);
2143 #endif
2145 void __init free_area_init(unsigned long *zones_size)
2147 free_area_init_node(0, NODE_DATA(0), zones_size,
2148 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2151 #ifdef CONFIG_PROC_FS
2153 #include <linux/seq_file.h>
2155 static void *frag_start(struct seq_file *m, loff_t *pos)
2157 pg_data_t *pgdat;
2158 loff_t node = *pos;
2160 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2161 --node;
2163 return pgdat;
2166 static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2168 pg_data_t *pgdat = (pg_data_t *)arg;
2170 (*pos)++;
2171 return pgdat->pgdat_next;
2174 static void frag_stop(struct seq_file *m, void *arg)
2178 /*
2179 * This walks the free areas for each zone.
2180 */
2181 static int frag_show(struct seq_file *m, void *arg)
2183 pg_data_t *pgdat = (pg_data_t *)arg;
2184 struct zone *zone;
2185 struct zone *node_zones = pgdat->node_zones;
2186 unsigned long flags;
2187 int order;
2189 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2190 if (!populated_zone(zone))
2191 continue;
2193 spin_lock_irqsave(&zone->lock, flags);
2194 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2195 for (order = 0; order < MAX_ORDER; ++order)
2196 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2197 spin_unlock_irqrestore(&zone->lock, flags);
2198 seq_putc(m, '\n');
2200 return 0;
2203 struct seq_operations fragmentation_op = {
2204 .start = frag_start,
2205 .next = frag_next,
2206 .stop = frag_stop,
2207 .show = frag_show,
2208 };
2210 /*
2211 * Output information about zones in @pgdat.
2212 */
2213 static int zoneinfo_show(struct seq_file *m, void *arg)
2215 pg_data_t *pgdat = arg;
2216 struct zone *zone;
2217 struct zone *node_zones = pgdat->node_zones;
2218 unsigned long flags;
2220 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2221 int i;
2223 if (!populated_zone(zone))
2224 continue;
2226 spin_lock_irqsave(&zone->lock, flags);
2227 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2228 seq_printf(m,
2229 "\n pages free %lu"
2230 "\n min %lu"
2231 "\n low %lu"
2232 "\n high %lu"
2233 "\n active %lu"
2234 "\n inactive %lu"
2235 "\n scanned %lu (a: %lu i: %lu)"
2236 "\n spanned %lu"
2237 "\n present %lu",
2238 zone->free_pages,
2239 zone->pages_min,
2240 zone->pages_low,
2241 zone->pages_high,
2242 zone->nr_active,
2243 zone->nr_inactive,
2244 zone->pages_scanned,
2245 zone->nr_scan_active, zone->nr_scan_inactive,
2246 zone->spanned_pages,
2247 zone->present_pages);
2248 seq_printf(m,
2249 "\n protection: (%lu",
2250 zone->lowmem_reserve[0]);
2251 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2252 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2253 seq_printf(m,
2254 ")"
2255 "\n pagesets");
2256 for_each_online_cpu(i) {
2257 struct per_cpu_pageset *pageset;
2258 int j;
2260 pageset = zone_pcp(zone, i);
2261 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2262 if (pageset->pcp[j].count)
2263 break;
2265 if (j == ARRAY_SIZE(pageset->pcp))
2266 continue;
2267 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2268 seq_printf(m,
2269 "\n cpu: %i pcp: %i"
2270 "\n count: %i"
2271 "\n high: %i"
2272 "\n batch: %i",
2273 i, j,
2274 pageset->pcp[j].count,
2275 pageset->pcp[j].high,
2276 pageset->pcp[j].batch);
2278 #ifdef CONFIG_NUMA
2279 seq_printf(m,
2280 "\n numa_hit: %lu"
2281 "\n numa_miss: %lu"
2282 "\n numa_foreign: %lu"
2283 "\n interleave_hit: %lu"
2284 "\n local_node: %lu"
2285 "\n other_node: %lu",
2286 pageset->numa_hit,
2287 pageset->numa_miss,
2288 pageset->numa_foreign,
2289 pageset->interleave_hit,
2290 pageset->local_node,
2291 pageset->other_node);
2292 #endif
2294 seq_printf(m,
2295 "\n all_unreclaimable: %u"
2296 "\n prev_priority: %i"
2297 "\n temp_priority: %i"
2298 "\n start_pfn: %lu",
2299 zone->all_unreclaimable,
2300 zone->prev_priority,
2301 zone->temp_priority,
2302 zone->zone_start_pfn);
2303 spin_unlock_irqrestore(&zone->lock, flags);
2304 seq_putc(m, '\n');
2306 return 0;
2309 struct seq_operations zoneinfo_op = {
2310 .start = frag_start, /* iterate over all zones. The same as in
2311 * fragmentation. */
2312 .next = frag_next,
2313 .stop = frag_stop,
2314 .show = zoneinfo_show,
2315 };
2317 static char *vmstat_text[] = {
2318 "nr_dirty",
2319 "nr_writeback",
2320 "nr_unstable",
2321 "nr_page_table_pages",
2322 "nr_mapped",
2323 "nr_slab",
2325 "pgpgin",
2326 "pgpgout",
2327 "pswpin",
2328 "pswpout",
2330 "pgalloc_high",
2331 "pgalloc_normal",
2332 "pgalloc_dma32",
2333 "pgalloc_dma",
2335 "pgfree",
2336 "pgactivate",
2337 "pgdeactivate",
2339 "pgfault",
2340 "pgmajfault",
2342 "pgrefill_high",
2343 "pgrefill_normal",
2344 "pgrefill_dma32",
2345 "pgrefill_dma",
2347 "pgsteal_high",
2348 "pgsteal_normal",
2349 "pgsteal_dma32",
2350 "pgsteal_dma",
2352 "pgscan_kswapd_high",
2353 "pgscan_kswapd_normal",
2354 "pgscan_kswapd_dma32",
2355 "pgscan_kswapd_dma",
2357 "pgscan_direct_high",
2358 "pgscan_direct_normal",
2359 "pgscan_direct_dma32",
2360 "pgscan_direct_dma",
2362 "pginodesteal",
2363 "slabs_scanned",
2364 "kswapd_steal",
2365 "kswapd_inodesteal",
2366 "pageoutrun",
2367 "allocstall",
2369 "pgrotated",
2370 "nr_bounce",
2371 };
2373 static void *vmstat_start(struct seq_file *m, loff_t *pos)
2375 struct page_state *ps;
2377 if (*pos >= ARRAY_SIZE(vmstat_text))
2378 return NULL;
2380 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2381 m->private = ps;
2382 if (!ps)
2383 return ERR_PTR(-ENOMEM);
2384 get_full_page_state(ps);
2385 ps->pgpgin /= 2; /* sectors -> kbytes */
2386 ps->pgpgout /= 2;
2387 return (unsigned long *)ps + *pos;
2390 static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2392 (*pos)++;
2393 if (*pos >= ARRAY_SIZE(vmstat_text))
2394 return NULL;
2395 return (unsigned long *)m->private + *pos;
2398 static int vmstat_show(struct seq_file *m, void *arg)
2400 unsigned long *l = arg;
2401 unsigned long off = l - (unsigned long *)m->private;
2403 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2404 return 0;
2407 static void vmstat_stop(struct seq_file *m, void *arg)
2409 kfree(m->private);
2410 m->private = NULL;
2413 struct seq_operations vmstat_op = {
2414 .start = vmstat_start,
2415 .next = vmstat_next,
2416 .stop = vmstat_stop,
2417 .show = vmstat_show,
2418 };
2420 #endif /* CONFIG_PROC_FS */
2422 #ifdef CONFIG_HOTPLUG_CPU
2423 static int page_alloc_cpu_notify(struct notifier_block *self,
2424 unsigned long action, void *hcpu)
2426 int cpu = (unsigned long)hcpu;
2427 long *count;
2428 unsigned long *src, *dest;
2430 if (action == CPU_DEAD) {
2431 int i;
2433 /* Drain local pagecache count. */
2434 count = &per_cpu(nr_pagecache_local, cpu);
2435 atomic_add(*count, &nr_pagecache);
2436 *count = 0;
2437 local_irq_disable();
2438 __drain_pages(cpu);
2440 /* Add dead cpu's page_states to our own. */
2441 dest = (unsigned long *)&__get_cpu_var(page_states);
2442 src = (unsigned long *)&per_cpu(page_states, cpu);
2444 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2445 i++) {
2446 dest[i] += src[i];
2447 src[i] = 0;
2450 local_irq_enable();
2452 return NOTIFY_OK;
2454 #endif /* CONFIG_HOTPLUG_CPU */
2456 void __init page_alloc_init(void)
2458 hotcpu_notifier(page_alloc_cpu_notify, 0);
2461 /*
2462 * setup_per_zone_lowmem_reserve - called whenever
2463 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2464 * has a correct pages reserved value, so an adequate number of
2465 * pages are left in the zone after a successful __alloc_pages().
2466 */
2467 static void setup_per_zone_lowmem_reserve(void)
2469 struct pglist_data *pgdat;
2470 int j, idx;
2472 for_each_pgdat(pgdat) {
2473 for (j = 0; j < MAX_NR_ZONES; j++) {
2474 struct zone *zone = pgdat->node_zones + j;
2475 unsigned long present_pages = zone->present_pages;
2477 zone->lowmem_reserve[j] = 0;
2479 for (idx = j-1; idx >= 0; idx--) {
2480 struct zone *lower_zone;
2482 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2483 sysctl_lowmem_reserve_ratio[idx] = 1;
2485 lower_zone = pgdat->node_zones + idx;
2486 lower_zone->lowmem_reserve[j] = present_pages /
2487 sysctl_lowmem_reserve_ratio[idx];
2488 present_pages += lower_zone->present_pages;
2494 /*
2495 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2496 * that the pages_{min,low,high} values for each zone are set correctly
2497 * with respect to min_free_kbytes.
2498 */
2499 void setup_per_zone_pages_min(void)
2501 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2502 unsigned long lowmem_pages = 0;
2503 struct zone *zone;
2504 unsigned long flags;
2506 /* Calculate total number of !ZONE_HIGHMEM pages */
2507 for_each_zone(zone) {
2508 if (!is_highmem(zone))
2509 lowmem_pages += zone->present_pages;
2512 for_each_zone(zone) {
2513 unsigned long tmp;
2514 spin_lock_irqsave(&zone->lru_lock, flags);
2515 tmp = (pages_min * zone->present_pages) / lowmem_pages;
2516 if (is_highmem(zone)) {
2517 /*
2518 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2519 * need highmem pages, so cap pages_min to a small
2520 * value here.
2522 * The (pages_high-pages_low) and (pages_low-pages_min)
2523 * deltas controls asynch page reclaim, and so should
2524 * not be capped for highmem.
2525 */
2526 int min_pages;
2528 min_pages = zone->present_pages / 1024;
2529 if (min_pages < SWAP_CLUSTER_MAX)
2530 min_pages = SWAP_CLUSTER_MAX;
2531 if (min_pages > 128)
2532 min_pages = 128;
2533 zone->pages_min = min_pages;
2534 } else {
2535 /*
2536 * If it's a lowmem zone, reserve a number of pages
2537 * proportionate to the zone's size.
2538 */
2539 zone->pages_min = tmp;
2542 zone->pages_low = zone->pages_min + tmp / 4;
2543 zone->pages_high = zone->pages_min + tmp / 2;
2544 spin_unlock_irqrestore(&zone->lru_lock, flags);
2548 /*
2549 * Initialise min_free_kbytes.
2551 * For small machines we want it small (128k min). For large machines
2552 * we want it large (64MB max). But it is not linear, because network
2553 * bandwidth does not increase linearly with machine size. We use
2555 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2556 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2558 * which yields
2560 * 16MB: 512k
2561 * 32MB: 724k
2562 * 64MB: 1024k
2563 * 128MB: 1448k
2564 * 256MB: 2048k
2565 * 512MB: 2896k
2566 * 1024MB: 4096k
2567 * 2048MB: 5792k
2568 * 4096MB: 8192k
2569 * 8192MB: 11584k
2570 * 16384MB: 16384k
2571 */
2572 static int __init init_per_zone_pages_min(void)
2574 unsigned long lowmem_kbytes;
2576 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2578 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2579 if (min_free_kbytes < 128)
2580 min_free_kbytes = 128;
2581 if (min_free_kbytes > 65536)
2582 min_free_kbytes = 65536;
2583 setup_per_zone_pages_min();
2584 setup_per_zone_lowmem_reserve();
2585 return 0;
2587 module_init(init_per_zone_pages_min)
2589 /*
2590 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2591 * that we can call two helper functions whenever min_free_kbytes
2592 * changes.
2593 */
2594 int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2595 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2597 proc_dointvec(table, write, file, buffer, length, ppos);
2598 setup_per_zone_pages_min();
2599 return 0;
2602 /*
2603 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2604 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2605 * whenever sysctl_lowmem_reserve_ratio changes.
2607 * The reserve ratio obviously has absolutely no relation with the
2608 * pages_min watermarks. The lowmem reserve ratio can only make sense
2609 * if in function of the boot time zone sizes.
2610 */
2611 int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2612 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2614 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2615 setup_per_zone_lowmem_reserve();
2616 return 0;
2619 /*
2620 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2621 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2622 * can have before it gets flushed back to buddy allocator.
2623 */
2625 int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
2626 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2628 struct zone *zone;
2629 unsigned int cpu;
2630 int ret;
2632 ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2633 if (!write || (ret == -EINVAL))
2634 return ret;
2635 for_each_zone(zone) {
2636 for_each_online_cpu(cpu) {
2637 unsigned long high;
2638 high = zone->present_pages / percpu_pagelist_fraction;
2639 setup_pagelist_highmark(zone_pcp(zone, cpu), high);
2642 return 0;
2645 __initdata int hashdist = HASHDIST_DEFAULT;
2647 #ifdef CONFIG_NUMA
2648 static int __init set_hashdist(char *str)
2650 if (!str)
2651 return 0;
2652 hashdist = simple_strtoul(str, &str, 0);
2653 return 1;
2655 __setup("hashdist=", set_hashdist);
2656 #endif
2658 /*
2659 * allocate a large system hash table from bootmem
2660 * - it is assumed that the hash table must contain an exact power-of-2
2661 * quantity of entries
2662 * - limit is the number of hash buckets, not the total allocation size
2663 */
2664 void *__init alloc_large_system_hash(const char *tablename,
2665 unsigned long bucketsize,
2666 unsigned long numentries,
2667 int scale,
2668 int flags,
2669 unsigned int *_hash_shift,
2670 unsigned int *_hash_mask,
2671 unsigned long limit)
2673 unsigned long long max = limit;
2674 unsigned long log2qty, size;
2675 void *table = NULL;
2677 /* allow the kernel cmdline to have a say */
2678 if (!numentries) {
2679 /* round applicable memory size up to nearest megabyte */
2680 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2681 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2682 numentries >>= 20 - PAGE_SHIFT;
2683 numentries <<= 20 - PAGE_SHIFT;
2685 /* limit to 1 bucket per 2^scale bytes of low memory */
2686 if (scale > PAGE_SHIFT)
2687 numentries >>= (scale - PAGE_SHIFT);
2688 else
2689 numentries <<= (PAGE_SHIFT - scale);
2691 /* rounded up to nearest power of 2 in size */
2692 numentries = 1UL << (long_log2(numentries) + 1);
2694 /* limit allocation size to 1/16 total memory by default */
2695 if (max == 0) {
2696 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2697 do_div(max, bucketsize);
2700 if (numentries > max)
2701 numentries = max;
2703 log2qty = long_log2(numentries);
2705 do {
2706 size = bucketsize << log2qty;
2707 if (flags & HASH_EARLY)
2708 table = alloc_bootmem(size);
2709 else if (hashdist)
2710 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2711 else {
2712 unsigned long order;
2713 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2715 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2717 } while (!table && size > PAGE_SIZE && --log2qty);
2719 if (!table)
2720 panic("Failed to allocate %s hash table\n", tablename);
2722 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2723 tablename,
2724 (1U << log2qty),
2725 long_log2(size) - PAGE_SHIFT,
2726 size);
2728 if (_hash_shift)
2729 *_hash_shift = log2qty;
2730 if (_hash_mask)
2731 *_hash_mask = (1 << log2qty) - 1;
2733 return table;