blob: 3974fd81d27c0de38a12a299e6f6b6dd2645cbf7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 */
16
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/module.h>
26#include <linux/suspend.h>
27#include <linux/pagevec.h>
28#include <linux/blkdev.h>
29#include <linux/slab.h>
30#include <linux/notifier.h>
31#include <linux/topology.h>
32#include <linux/sysctl.h>
33#include <linux/cpu.h>
34#include <linux/cpuset.h>
35#include <linux/nodemask.h>
36#include <linux/vmalloc.h>
37
38#include <asm/tlbflush.h>
39#include "internal.h"
40
41/*
42 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
43 * initializer cleaner
44 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070045nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070046EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070047nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070048EXPORT_SYMBOL(node_possible_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049struct pglist_data *pgdat_list __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070050unsigned long totalram_pages __read_mostly;
51unsigned long totalhigh_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052long nr_swap_pages;
53
54/*
55 * results with 256, 32 in the lowmem_reserve sysctl:
56 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
57 * 1G machine -> (16M dma, 784M normal, 224M high)
58 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
59 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
60 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
61 */
62int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 32 };
63
64EXPORT_SYMBOL(totalram_pages);
65EXPORT_SYMBOL(nr_swap_pages);
66
67/*
68 * Used by page_zone() to look up the address of the struct zone whose
69 * id is encoded in the upper bits of page->flags
70 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070071struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070072EXPORT_SYMBOL(zone_table);
73
74static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
75int min_free_kbytes = 1024;
76
77unsigned long __initdata nr_kernel_pages;
78unsigned long __initdata nr_all_pages;
79
80/*
81 * Temporary debugging check for pages not lying within a given zone.
82 */
83static int bad_range(struct zone *zone, struct page *page)
84{
85 if (page_to_pfn(page) >= zone->zone_start_pfn + zone->spanned_pages)
86 return 1;
87 if (page_to_pfn(page) < zone->zone_start_pfn)
88 return 1;
89#ifdef CONFIG_HOLES_IN_ZONE
90 if (!pfn_valid(page_to_pfn(page)))
91 return 1;
92#endif
93 if (zone != page_zone(page))
94 return 1;
95 return 0;
96}
97
98static void bad_page(const char *function, struct page *page)
99{
100 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
101 function, current->comm, page);
102 printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
103 (int)(2*sizeof(page_flags_t)), (unsigned long)page->flags,
104 page->mapping, page_mapcount(page), page_count(page));
105 printk(KERN_EMERG "Backtrace:\n");
106 dump_stack();
107 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
Hugh Dickins334795e2005-06-21 17:15:08 -0700108 page->flags &= ~(1 << PG_lru |
109 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 1 << PG_active |
112 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700113 1 << PG_reclaim |
114 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 1 << PG_swapcache |
116 1 << PG_writeback);
117 set_page_count(page, 0);
118 reset_page_mapcount(page);
119 page->mapping = NULL;
120 tainted |= TAINT_BAD_PAGE;
121}
122
123#ifndef CONFIG_HUGETLB_PAGE
124#define prep_compound_page(page, order) do { } while (0)
125#define destroy_compound_page(page, order) do { } while (0)
126#else
127/*
128 * Higher-order pages are called "compound pages". They are structured thusly:
129 *
130 * The first PAGE_SIZE page is called the "head page".
131 *
132 * The remaining PAGE_SIZE pages are called "tail pages".
133 *
134 * All pages have PG_compound set. All pages have their ->private pointing at
135 * the head page (even the head page has this).
136 *
137 * The first tail page's ->mapping, if non-zero, holds the address of the
138 * compound page's put_page() function.
139 *
140 * The order of the allocation is stored in the first tail page's ->index
141 * This is only for debug at present. This usage means that zero-order pages
142 * may not be compound.
143 */
144static void prep_compound_page(struct page *page, unsigned long order)
145{
146 int i;
147 int nr_pages = 1 << order;
148
149 page[1].mapping = NULL;
150 page[1].index = order;
151 for (i = 0; i < nr_pages; i++) {
152 struct page *p = page + i;
153
154 SetPageCompound(p);
155 p->private = (unsigned long)page;
156 }
157}
158
159static void destroy_compound_page(struct page *page, unsigned long order)
160{
161 int i;
162 int nr_pages = 1 << order;
163
164 if (!PageCompound(page))
165 return;
166
167 if (page[1].index != order)
168 bad_page(__FUNCTION__, page);
169
170 for (i = 0; i < nr_pages; i++) {
171 struct page *p = page + i;
172
173 if (!PageCompound(p))
174 bad_page(__FUNCTION__, page);
175 if (p->private != (unsigned long)page)
176 bad_page(__FUNCTION__, page);
177 ClearPageCompound(p);
178 }
179}
180#endif /* CONFIG_HUGETLB_PAGE */
181
182/*
183 * function for dealing with page's order in buddy system.
184 * zone->lock is already acquired when we use these.
185 * So, we don't need atomic page->flags operations here.
186 */
187static inline unsigned long page_order(struct page *page) {
188 return page->private;
189}
190
191static inline void set_page_order(struct page *page, int order) {
192 page->private = order;
193 __SetPagePrivate(page);
194}
195
196static inline void rmv_page_order(struct page *page)
197{
198 __ClearPagePrivate(page);
199 page->private = 0;
200}
201
202/*
203 * Locate the struct page for both the matching buddy in our
204 * pair (buddy1) and the combined O(n+1) page they form (page).
205 *
206 * 1) Any buddy B1 will have an order O twin B2 which satisfies
207 * the following equation:
208 * B2 = B1 ^ (1 << O)
209 * For example, if the starting buddy (buddy2) is #8 its order
210 * 1 buddy is #10:
211 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
212 *
213 * 2) Any buddy B will have an order O+1 parent P which
214 * satisfies the following equation:
215 * P = B & ~(1 << O)
216 *
217 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
218 */
219static inline struct page *
220__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
221{
222 unsigned long buddy_idx = page_idx ^ (1 << order);
223
224 return page + (buddy_idx - page_idx);
225}
226
227static inline unsigned long
228__find_combined_index(unsigned long page_idx, unsigned int order)
229{
230 return (page_idx & ~(1 << order));
231}
232
233/*
234 * This function checks whether a page is free && is the buddy
235 * we can do coalesce a page and its buddy if
236 * (a) the buddy is free &&
237 * (b) the buddy is on the buddy system &&
238 * (c) a page and its buddy have the same order.
239 * for recording page's order, we use page->private and PG_private.
240 *
241 */
242static inline int page_is_buddy(struct page *page, int order)
243{
244 if (PagePrivate(page) &&
245 (page_order(page) == order) &&
246 !PageReserved(page) &&
247 page_count(page) == 0)
248 return 1;
249 return 0;
250}
251
252/*
253 * Freeing function for a buddy system allocator.
254 *
255 * The concept of a buddy system is to maintain direct-mapped table
256 * (containing bit values) for memory blocks of various "orders".
257 * The bottom level table contains the map for the smallest allocatable
258 * units of memory (here, pages), and each level above it describes
259 * pairs of units from the levels below, hence, "buddies".
260 * At a high level, all that happens here is marking the table entry
261 * at the bottom level available, and propagating the changes upward
262 * as necessary, plus some accounting needed to play nicely with other
263 * parts of the VM system.
264 * At each level, we keep a list of pages, which are heads of continuous
265 * free pages of length of (1 << order) and marked with PG_Private.Page's
266 * order is recorded in page->private field.
267 * So when we are allocating or freeing one, we can derive the state of the
268 * other. That is, if we allocate a small block, and both were
269 * free, the remainder of the region must be split into blocks.
270 * If a block is freed, and its buddy is also free, then this
271 * triggers coalescing into a block of larger size.
272 *
273 * -- wli
274 */
275
276static inline void __free_pages_bulk (struct page *page,
277 struct zone *zone, unsigned int order)
278{
279 unsigned long page_idx;
280 int order_size = 1 << order;
281
282 if (unlikely(order))
283 destroy_compound_page(page, order);
284
285 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
286
287 BUG_ON(page_idx & (order_size - 1));
288 BUG_ON(bad_range(zone, page));
289
290 zone->free_pages += order_size;
291 while (order < MAX_ORDER-1) {
292 unsigned long combined_idx;
293 struct free_area *area;
294 struct page *buddy;
295
296 combined_idx = __find_combined_index(page_idx, order);
297 buddy = __page_find_buddy(page, page_idx, order);
298
299 if (bad_range(zone, buddy))
300 break;
301 if (!page_is_buddy(buddy, order))
302 break; /* Move the buddy up one level. */
303 list_del(&buddy->lru);
304 area = zone->free_area + order;
305 area->nr_free--;
306 rmv_page_order(buddy);
307 page = page + (combined_idx - page_idx);
308 page_idx = combined_idx;
309 order++;
310 }
311 set_page_order(page, order);
312 list_add(&page->lru, &zone->free_area[order].free_list);
313 zone->free_area[order].nr_free++;
314}
315
316static inline void free_pages_check(const char *function, struct page *page)
317{
318 if ( page_mapcount(page) ||
319 page->mapping != NULL ||
320 page_count(page) != 0 ||
321 (page->flags & (
322 1 << PG_lru |
323 1 << PG_private |
324 1 << PG_locked |
325 1 << PG_active |
326 1 << PG_reclaim |
327 1 << PG_slab |
328 1 << PG_swapcache |
329 1 << PG_writeback )))
330 bad_page(function, page);
331 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700332 __ClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
335/*
336 * Frees a list of pages.
337 * Assumes all pages on list are in same zone, and of same order.
338 * count is the number of pages to free, or 0 for all on the list.
339 *
340 * If the zone was previously in an "all pages pinned" state then look to
341 * see if this freeing clears that state.
342 *
343 * And clear the zone's pages_scanned counter, to hold off the "all pages are
344 * pinned" detection logic.
345 */
346static int
347free_pages_bulk(struct zone *zone, int count,
348 struct list_head *list, unsigned int order)
349{
350 unsigned long flags;
351 struct page *page = NULL;
352 int ret = 0;
353
354 spin_lock_irqsave(&zone->lock, flags);
355 zone->all_unreclaimable = 0;
356 zone->pages_scanned = 0;
357 while (!list_empty(list) && count--) {
358 page = list_entry(list->prev, struct page, lru);
359 /* have to delete it as __free_pages_bulk list manipulates */
360 list_del(&page->lru);
361 __free_pages_bulk(page, zone, order);
362 ret++;
363 }
364 spin_unlock_irqrestore(&zone->lock, flags);
365 return ret;
366}
367
368void __free_pages_ok(struct page *page, unsigned int order)
369{
370 LIST_HEAD(list);
371 int i;
372
373 arch_free_page(page, order);
374
375 mod_page_state(pgfree, 1 << order);
376
377#ifndef CONFIG_MMU
378 if (order > 0)
379 for (i = 1 ; i < (1 << order) ; ++i)
380 __put_page(page + i);
381#endif
382
383 for (i = 0 ; i < (1 << order) ; ++i)
384 free_pages_check(__FUNCTION__, page + i);
385 list_add(&page->lru, &list);
386 kernel_map_pages(page, 1<<order, 0);
387 free_pages_bulk(page_zone(page), 1, &list, order);
388}
389
390
391/*
392 * The order of subdivision here is critical for the IO subsystem.
393 * Please do not alter this order without good reasons and regression
394 * testing. Specifically, as large blocks of memory are subdivided,
395 * the order in which smaller blocks are delivered depends on the order
396 * they're subdivided in this function. This is the primary factor
397 * influencing the order in which pages are delivered to the IO
398 * subsystem according to empirical testing, and this is also justified
399 * by considering the behavior of a buddy system containing a single
400 * large block of memory acted on by a series of small allocations.
401 * This behavior is a critical factor in sglist merging's success.
402 *
403 * -- wli
404 */
405static inline struct page *
406expand(struct zone *zone, struct page *page,
407 int low, int high, struct free_area *area)
408{
409 unsigned long size = 1 << high;
410
411 while (high > low) {
412 area--;
413 high--;
414 size >>= 1;
415 BUG_ON(bad_range(zone, &page[size]));
416 list_add(&page[size].lru, &area->free_list);
417 area->nr_free++;
418 set_page_order(&page[size], high);
419 }
420 return page;
421}
422
423void set_page_refs(struct page *page, int order)
424{
425#ifdef CONFIG_MMU
426 set_page_count(page, 1);
427#else
428 int i;
429
430 /*
431 * We need to reference all the pages for this order, otherwise if
432 * anyone accesses one of the pages with (get/put) it will be freed.
433 * - eg: access_process_vm()
434 */
435 for (i = 0; i < (1 << order); i++)
436 set_page_count(page + i, 1);
437#endif /* CONFIG_MMU */
438}
439
440/*
441 * This page is about to be returned from the page allocator
442 */
443static void prep_new_page(struct page *page, int order)
444{
Hugh Dickins334795e2005-06-21 17:15:08 -0700445 if ( page_mapcount(page) ||
446 page->mapping != NULL ||
447 page_count(page) != 0 ||
448 (page->flags & (
449 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 1 << PG_private |
451 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 1 << PG_active |
453 1 << PG_dirty |
454 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700455 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 1 << PG_swapcache |
457 1 << PG_writeback )))
458 bad_page(__FUNCTION__, page);
459
460 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
461 1 << PG_referenced | 1 << PG_arch_1 |
462 1 << PG_checked | 1 << PG_mappedtodisk);
463 page->private = 0;
464 set_page_refs(page, order);
465 kernel_map_pages(page, 1 << order, 1);
466}
467
468/*
469 * Do the hard work of removing an element from the buddy allocator.
470 * Call me with the zone->lock already held.
471 */
472static struct page *__rmqueue(struct zone *zone, unsigned int order)
473{
474 struct free_area * area;
475 unsigned int current_order;
476 struct page *page;
477
478 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
479 area = zone->free_area + current_order;
480 if (list_empty(&area->free_list))
481 continue;
482
483 page = list_entry(area->free_list.next, struct page, lru);
484 list_del(&page->lru);
485 rmv_page_order(page);
486 area->nr_free--;
487 zone->free_pages -= 1UL << order;
488 return expand(zone, page, order, current_order, area);
489 }
490
491 return NULL;
492}
493
494/*
495 * Obtain a specified number of elements from the buddy allocator, all under
496 * a single hold of the lock, for efficiency. Add them to the supplied list.
497 * Returns the number of new pages which were placed at *list.
498 */
499static int rmqueue_bulk(struct zone *zone, unsigned int order,
500 unsigned long count, struct list_head *list)
501{
502 unsigned long flags;
503 int i;
504 int allocated = 0;
505 struct page *page;
506
507 spin_lock_irqsave(&zone->lock, flags);
508 for (i = 0; i < count; ++i) {
509 page = __rmqueue(zone, order);
510 if (page == NULL)
511 break;
512 allocated++;
513 list_add_tail(&page->lru, list);
514 }
515 spin_unlock_irqrestore(&zone->lock, flags);
516 return allocated;
517}
518
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700519#ifdef CONFIG_NUMA
520/* Called from the slab reaper to drain remote pagesets */
521void drain_remote_pages(void)
522{
523 struct zone *zone;
524 int i;
525 unsigned long flags;
526
527 local_irq_save(flags);
528 for_each_zone(zone) {
529 struct per_cpu_pageset *pset;
530
531 /* Do not drain local pagesets */
532 if (zone->zone_pgdat->node_id == numa_node_id())
533 continue;
534
535 pset = zone->pageset[smp_processor_id()];
536 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
537 struct per_cpu_pages *pcp;
538
539 pcp = &pset->pcp[i];
540 if (pcp->count)
541 pcp->count -= free_pages_bulk(zone, pcp->count,
542 &pcp->list, 0);
543 }
544 }
545 local_irq_restore(flags);
546}
547#endif
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
550static void __drain_pages(unsigned int cpu)
551{
552 struct zone *zone;
553 int i;
554
555 for_each_zone(zone) {
556 struct per_cpu_pageset *pset;
557
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700558 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
560 struct per_cpu_pages *pcp;
561
562 pcp = &pset->pcp[i];
563 pcp->count -= free_pages_bulk(zone, pcp->count,
564 &pcp->list, 0);
565 }
566 }
567}
568#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
569
570#ifdef CONFIG_PM
571
572void mark_free_pages(struct zone *zone)
573{
574 unsigned long zone_pfn, flags;
575 int order;
576 struct list_head *curr;
577
578 if (!zone->spanned_pages)
579 return;
580
581 spin_lock_irqsave(&zone->lock, flags);
582 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
583 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
584
585 for (order = MAX_ORDER - 1; order >= 0; --order)
586 list_for_each(curr, &zone->free_area[order].free_list) {
587 unsigned long start_pfn, i;
588
589 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
590
591 for (i=0; i < (1<<order); i++)
592 SetPageNosaveFree(pfn_to_page(start_pfn+i));
593 }
594 spin_unlock_irqrestore(&zone->lock, flags);
595}
596
597/*
598 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
599 */
600void drain_local_pages(void)
601{
602 unsigned long flags;
603
604 local_irq_save(flags);
605 __drain_pages(smp_processor_id());
606 local_irq_restore(flags);
607}
608#endif /* CONFIG_PM */
609
610static void zone_statistics(struct zonelist *zonelist, struct zone *z)
611{
612#ifdef CONFIG_NUMA
613 unsigned long flags;
614 int cpu;
615 pg_data_t *pg = z->zone_pgdat;
616 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
617 struct per_cpu_pageset *p;
618
619 local_irq_save(flags);
620 cpu = smp_processor_id();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700621 p = zone_pcp(z,cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700623 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 } else {
625 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700626 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 }
628 if (pg == NODE_DATA(numa_node_id()))
629 p->local_node++;
630 else
631 p->other_node++;
632 local_irq_restore(flags);
633#endif
634}
635
636/*
637 * Free a 0-order page
638 */
639static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
640static void fastcall free_hot_cold_page(struct page *page, int cold)
641{
642 struct zone *zone = page_zone(page);
643 struct per_cpu_pages *pcp;
644 unsigned long flags;
645
646 arch_free_page(page, 0);
647
648 kernel_map_pages(page, 1, 0);
649 inc_page_state(pgfree);
650 if (PageAnon(page))
651 page->mapping = NULL;
652 free_pages_check(__FUNCTION__, page);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700653 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 list_add(&page->lru, &pcp->list);
656 pcp->count++;
Christoph Lameter2caaad42005-06-21 17:15:00 -0700657 if (pcp->count >= pcp->high)
658 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 local_irq_restore(flags);
660 put_cpu();
661}
662
663void fastcall free_hot_page(struct page *page)
664{
665 free_hot_cold_page(page, 0);
666}
667
668void fastcall free_cold_page(struct page *page)
669{
670 free_hot_cold_page(page, 1);
671}
672
673static inline void prep_zero_page(struct page *page, int order, unsigned int __nocast gfp_flags)
674{
675 int i;
676
677 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
678 for(i = 0; i < (1 << order); i++)
679 clear_highpage(page + i);
680}
681
682/*
683 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
684 * we cheat by calling it from here, in the order > 0 path. Saves a branch
685 * or two.
686 */
687static struct page *
688buffered_rmqueue(struct zone *zone, int order, unsigned int __nocast gfp_flags)
689{
690 unsigned long flags;
691 struct page *page = NULL;
692 int cold = !!(gfp_flags & __GFP_COLD);
693
694 if (order == 0) {
695 struct per_cpu_pages *pcp;
696
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700697 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 local_irq_save(flags);
699 if (pcp->count <= pcp->low)
700 pcp->count += rmqueue_bulk(zone, 0,
701 pcp->batch, &pcp->list);
702 if (pcp->count) {
703 page = list_entry(pcp->list.next, struct page, lru);
704 list_del(&page->lru);
705 pcp->count--;
706 }
707 local_irq_restore(flags);
708 put_cpu();
709 }
710
711 if (page == NULL) {
712 spin_lock_irqsave(&zone->lock, flags);
713 page = __rmqueue(zone, order);
714 spin_unlock_irqrestore(&zone->lock, flags);
715 }
716
717 if (page != NULL) {
718 BUG_ON(bad_range(zone, page));
719 mod_page_state_zone(zone, pgalloc, 1 << order);
720 prep_new_page(page, order);
721
722 if (gfp_flags & __GFP_ZERO)
723 prep_zero_page(page, order, gfp_flags);
724
725 if (order && (gfp_flags & __GFP_COMP))
726 prep_compound_page(page, order);
727 }
728 return page;
729}
730
731/*
732 * Return 1 if free pages are above 'mark'. This takes into account the order
733 * of the allocation.
734 */
735int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
736 int classzone_idx, int can_try_harder, int gfp_high)
737{
738 /* free_pages my go negative - that's OK */
739 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
740 int o;
741
742 if (gfp_high)
743 min -= min / 2;
744 if (can_try_harder)
745 min -= min / 4;
746
747 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
748 return 0;
749 for (o = 0; o < order; o++) {
750 /* At the next order, this order's pages become unavailable */
751 free_pages -= z->free_area[o].nr_free << o;
752
753 /* Require fewer higher order pages to be free */
754 min >>= 1;
755
756 if (free_pages <= min)
757 return 0;
758 }
759 return 1;
760}
761
Martin Hicks753ee722005-06-21 17:14:41 -0700762static inline int
763should_reclaim_zone(struct zone *z, unsigned int gfp_mask)
764{
765 if (!z->reclaim_pages)
766 return 0;
Martin Hicks0c35bba2005-06-21 17:14:42 -0700767 if (gfp_mask & __GFP_NORECLAIM)
768 return 0;
Martin Hicks753ee722005-06-21 17:14:41 -0700769 return 1;
770}
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772/*
773 * This is the 'heart' of the zoned buddy allocator.
774 */
775struct page * fastcall
776__alloc_pages(unsigned int __nocast gfp_mask, unsigned int order,
777 struct zonelist *zonelist)
778{
779 const int wait = gfp_mask & __GFP_WAIT;
780 struct zone **zones, *z;
781 struct page *page;
782 struct reclaim_state reclaim_state;
783 struct task_struct *p = current;
784 int i;
785 int classzone_idx;
786 int do_retry;
787 int can_try_harder;
788 int did_some_progress;
789
790 might_sleep_if(wait);
791
792 /*
793 * The caller may dip into page reserves a bit more if the caller
794 * cannot run direct reclaim, or is the caller has realtime scheduling
795 * policy
796 */
797 can_try_harder = (unlikely(rt_task(p)) && !in_interrupt()) || !wait;
798
799 zones = zonelist->zones; /* the list of zones suitable for gfp_mask */
800
801 if (unlikely(zones[0] == NULL)) {
802 /* Should this ever happen?? */
803 return NULL;
804 }
805
806 classzone_idx = zone_idx(zones[0]);
807
Martin Hicks753ee722005-06-21 17:14:41 -0700808restart:
Paul Jackson9bf22292005-09-06 15:18:12 -0700809 /*
810 * Go through the zonelist once, looking for a zone with enough free.
811 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
812 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 for (i = 0; (z = zones[i]) != NULL; i++) {
Martin Hicks753ee722005-06-21 17:14:41 -0700814 int do_reclaim = should_reclaim_zone(z, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Paul Jackson9bf22292005-09-06 15:18:12 -0700816 if (!cpuset_zone_allowed(z, __GFP_HARDWALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 continue;
818
Martin Hicks753ee722005-06-21 17:14:41 -0700819 /*
820 * If the zone is to attempt early page reclaim then this loop
821 * will try to reclaim pages and check the watermark a second
822 * time before giving up and falling back to the next zone.
823 */
824zone_reclaim_retry:
825 if (!zone_watermark_ok(z, order, z->pages_low,
826 classzone_idx, 0, 0)) {
827 if (!do_reclaim)
828 continue;
829 else {
830 zone_reclaim(z, gfp_mask, order);
831 /* Only try reclaim once */
832 do_reclaim = 0;
833 goto zone_reclaim_retry;
834 }
835 }
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 page = buffered_rmqueue(z, order, gfp_mask);
838 if (page)
839 goto got_pg;
840 }
841
842 for (i = 0; (z = zones[i]) != NULL; i++)
843 wakeup_kswapd(z, order);
844
845 /*
846 * Go through the zonelist again. Let __GFP_HIGH and allocations
847 * coming from realtime tasks to go deeper into reserves
848 *
849 * This is the last chance, in general, before the goto nopage.
850 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700851 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 */
853 for (i = 0; (z = zones[i]) != NULL; i++) {
854 if (!zone_watermark_ok(z, order, z->pages_min,
855 classzone_idx, can_try_harder,
856 gfp_mask & __GFP_HIGH))
857 continue;
858
Paul Jackson9bf22292005-09-06 15:18:12 -0700859 if (wait && !cpuset_zone_allowed(z, gfp_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 continue;
861
862 page = buffered_rmqueue(z, order, gfp_mask);
863 if (page)
864 goto got_pg;
865 }
866
867 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700868
869 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
870 && !in_interrupt()) {
871 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
872 /* go through the zonelist yet again, ignoring mins */
873 for (i = 0; (z = zones[i]) != NULL; i++) {
Paul Jackson9bf22292005-09-06 15:18:12 -0700874 if (!cpuset_zone_allowed(z, gfp_mask))
Nick Pigginb84a35b2005-05-01 08:58:36 -0700875 continue;
876 page = buffered_rmqueue(z, order, gfp_mask);
877 if (page)
878 goto got_pg;
879 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 }
881 goto nopage;
882 }
883
884 /* Atomic allocations - we can't balance anything */
885 if (!wait)
886 goto nopage;
887
888rebalance:
889 cond_resched();
890
891 /* We now go into synchronous reclaim */
892 p->flags |= PF_MEMALLOC;
893 reclaim_state.reclaimed_slab = 0;
894 p->reclaim_state = &reclaim_state;
895
Darren Hart1ad539b2005-06-21 17:14:53 -0700896 did_some_progress = try_to_free_pages(zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 p->reclaim_state = NULL;
899 p->flags &= ~PF_MEMALLOC;
900
901 cond_resched();
902
903 if (likely(did_some_progress)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 for (i = 0; (z = zones[i]) != NULL; i++) {
905 if (!zone_watermark_ok(z, order, z->pages_min,
906 classzone_idx, can_try_harder,
907 gfp_mask & __GFP_HIGH))
908 continue;
909
Paul Jackson9bf22292005-09-06 15:18:12 -0700910 if (!cpuset_zone_allowed(z, gfp_mask))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 continue;
912
913 page = buffered_rmqueue(z, order, gfp_mask);
914 if (page)
915 goto got_pg;
916 }
917 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
918 /*
919 * Go through the zonelist yet one more time, keep
920 * very high watermark here, this is only to catch
921 * a parallel oom killing, we must fail if we're still
922 * under heavy pressure.
923 */
924 for (i = 0; (z = zones[i]) != NULL; i++) {
925 if (!zone_watermark_ok(z, order, z->pages_high,
926 classzone_idx, 0, 0))
927 continue;
928
Paul Jackson9bf22292005-09-06 15:18:12 -0700929 if (!cpuset_zone_allowed(z, __GFP_HARDWALL))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 continue;
931
932 page = buffered_rmqueue(z, order, gfp_mask);
933 if (page)
934 goto got_pg;
935 }
936
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -0700937 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 goto restart;
939 }
940
941 /*
942 * Don't let big-order allocations loop unless the caller explicitly
943 * requests that. Wait for some write requests to complete then retry.
944 *
945 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
946 * <= 3, but that may not be true in other implementations.
947 */
948 do_retry = 0;
949 if (!(gfp_mask & __GFP_NORETRY)) {
950 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
951 do_retry = 1;
952 if (gfp_mask & __GFP_NOFAIL)
953 do_retry = 1;
954 }
955 if (do_retry) {
956 blk_congestion_wait(WRITE, HZ/50);
957 goto rebalance;
958 }
959
960nopage:
961 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
962 printk(KERN_WARNING "%s: page allocation failure."
963 " order:%d, mode:0x%x\n",
964 p->comm, order, gfp_mask);
965 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -0700966 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
968 return NULL;
969got_pg:
970 zone_statistics(zonelist, z);
971 return page;
972}
973
974EXPORT_SYMBOL(__alloc_pages);
975
976/*
977 * Common helper functions.
978 */
979fastcall unsigned long __get_free_pages(unsigned int __nocast gfp_mask, unsigned int order)
980{
981 struct page * page;
982 page = alloc_pages(gfp_mask, order);
983 if (!page)
984 return 0;
985 return (unsigned long) page_address(page);
986}
987
988EXPORT_SYMBOL(__get_free_pages);
989
990fastcall unsigned long get_zeroed_page(unsigned int __nocast gfp_mask)
991{
992 struct page * page;
993
994 /*
995 * get_zeroed_page() returns a 32-bit address, which cannot represent
996 * a highmem page
997 */
998 BUG_ON(gfp_mask & __GFP_HIGHMEM);
999
1000 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1001 if (page)
1002 return (unsigned long) page_address(page);
1003 return 0;
1004}
1005
1006EXPORT_SYMBOL(get_zeroed_page);
1007
1008void __pagevec_free(struct pagevec *pvec)
1009{
1010 int i = pagevec_count(pvec);
1011
1012 while (--i >= 0)
1013 free_hot_cold_page(pvec->pages[i], pvec->cold);
1014}
1015
1016fastcall void __free_pages(struct page *page, unsigned int order)
1017{
1018 if (!PageReserved(page) && put_page_testzero(page)) {
1019 if (order == 0)
1020 free_hot_page(page);
1021 else
1022 __free_pages_ok(page, order);
1023 }
1024}
1025
1026EXPORT_SYMBOL(__free_pages);
1027
1028fastcall void free_pages(unsigned long addr, unsigned int order)
1029{
1030 if (addr != 0) {
1031 BUG_ON(!virt_addr_valid((void *)addr));
1032 __free_pages(virt_to_page((void *)addr), order);
1033 }
1034}
1035
1036EXPORT_SYMBOL(free_pages);
1037
1038/*
1039 * Total amount of free (allocatable) RAM:
1040 */
1041unsigned int nr_free_pages(void)
1042{
1043 unsigned int sum = 0;
1044 struct zone *zone;
1045
1046 for_each_zone(zone)
1047 sum += zone->free_pages;
1048
1049 return sum;
1050}
1051
1052EXPORT_SYMBOL(nr_free_pages);
1053
1054#ifdef CONFIG_NUMA
1055unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1056{
1057 unsigned int i, sum = 0;
1058
1059 for (i = 0; i < MAX_NR_ZONES; i++)
1060 sum += pgdat->node_zones[i].free_pages;
1061
1062 return sum;
1063}
1064#endif
1065
1066static unsigned int nr_free_zone_pages(int offset)
1067{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001068 /* Just pick one node, since fallback list is circular */
1069 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 unsigned int sum = 0;
1071
Martin J. Blighe310fd42005-07-29 22:59:18 -07001072 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1073 struct zone **zonep = zonelist->zones;
1074 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
Martin J. Blighe310fd42005-07-29 22:59:18 -07001076 for (zone = *zonep++; zone; zone = *zonep++) {
1077 unsigned long size = zone->present_pages;
1078 unsigned long high = zone->pages_high;
1079 if (size > high)
1080 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
1082
1083 return sum;
1084}
1085
1086/*
1087 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1088 */
1089unsigned int nr_free_buffer_pages(void)
1090{
1091 return nr_free_zone_pages(GFP_USER & GFP_ZONEMASK);
1092}
1093
1094/*
1095 * Amount of free RAM allocatable within all zones
1096 */
1097unsigned int nr_free_pagecache_pages(void)
1098{
1099 return nr_free_zone_pages(GFP_HIGHUSER & GFP_ZONEMASK);
1100}
1101
1102#ifdef CONFIG_HIGHMEM
1103unsigned int nr_free_highpages (void)
1104{
1105 pg_data_t *pgdat;
1106 unsigned int pages = 0;
1107
1108 for_each_pgdat(pgdat)
1109 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1110
1111 return pages;
1112}
1113#endif
1114
1115#ifdef CONFIG_NUMA
1116static void show_node(struct zone *zone)
1117{
1118 printk("Node %d ", zone->zone_pgdat->node_id);
1119}
1120#else
1121#define show_node(zone) do { } while (0)
1122#endif
1123
1124/*
1125 * Accumulate the page_state information across all CPUs.
1126 * The result is unavoidably approximate - it can change
1127 * during and after execution of this function.
1128 */
1129static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1130
1131atomic_t nr_pagecache = ATOMIC_INIT(0);
1132EXPORT_SYMBOL(nr_pagecache);
1133#ifdef CONFIG_SMP
1134DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1135#endif
1136
Martin Hicksc07e02d2005-09-03 15:55:11 -07001137void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
1139 int cpu = 0;
1140
1141 memset(ret, 0, sizeof(*ret));
Martin Hicksc07e02d2005-09-03 15:55:11 -07001142 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Martin Hicksc07e02d2005-09-03 15:55:11 -07001144 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 while (cpu < NR_CPUS) {
1146 unsigned long *in, *out, off;
1147
1148 in = (unsigned long *)&per_cpu(page_states, cpu);
1149
Martin Hicksc07e02d2005-09-03 15:55:11 -07001150 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152 if (cpu < NR_CPUS)
1153 prefetch(&per_cpu(page_states, cpu));
1154
1155 out = (unsigned long *)ret;
1156 for (off = 0; off < nr; off++)
1157 *out++ += *in++;
1158 }
1159}
1160
Martin Hicksc07e02d2005-09-03 15:55:11 -07001161void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162{
1163 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001164 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165
1166 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1167 nr /= sizeof(unsigned long);
1168
Martin Hicksc07e02d2005-09-03 15:55:11 -07001169 __get_page_state(ret, nr+1, &mask);
1170}
1171
1172void get_page_state(struct page_state *ret)
1173{
1174 int nr;
1175 cpumask_t mask = CPU_MASK_ALL;
1176
1177 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1178 nr /= sizeof(unsigned long);
1179
1180 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
1183void get_full_page_state(struct page_state *ret)
1184{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001185 cpumask_t mask = CPU_MASK_ALL;
1186
1187 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
Benjamin LaHaisec2f29ea2005-06-21 17:14:55 -07001190unsigned long __read_page_state(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191{
1192 unsigned long ret = 0;
1193 int cpu;
1194
1195 for_each_online_cpu(cpu) {
1196 unsigned long in;
1197
1198 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1199 ret += *((unsigned long *)in);
1200 }
1201 return ret;
1202}
1203
Benjamin LaHaise83e5d8f2005-06-21 17:14:54 -07001204void __mod_page_state(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205{
1206 unsigned long flags;
1207 void* ptr;
1208
1209 local_irq_save(flags);
1210 ptr = &__get_cpu_var(page_states);
1211 *(unsigned long*)(ptr + offset) += delta;
1212 local_irq_restore(flags);
1213}
1214
1215EXPORT_SYMBOL(__mod_page_state);
1216
1217void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1218 unsigned long *free, struct pglist_data *pgdat)
1219{
1220 struct zone *zones = pgdat->node_zones;
1221 int i;
1222
1223 *active = 0;
1224 *inactive = 0;
1225 *free = 0;
1226 for (i = 0; i < MAX_NR_ZONES; i++) {
1227 *active += zones[i].nr_active;
1228 *inactive += zones[i].nr_inactive;
1229 *free += zones[i].free_pages;
1230 }
1231}
1232
1233void get_zone_counts(unsigned long *active,
1234 unsigned long *inactive, unsigned long *free)
1235{
1236 struct pglist_data *pgdat;
1237
1238 *active = 0;
1239 *inactive = 0;
1240 *free = 0;
1241 for_each_pgdat(pgdat) {
1242 unsigned long l, m, n;
1243 __get_zone_counts(&l, &m, &n, pgdat);
1244 *active += l;
1245 *inactive += m;
1246 *free += n;
1247 }
1248}
1249
1250void si_meminfo(struct sysinfo *val)
1251{
1252 val->totalram = totalram_pages;
1253 val->sharedram = 0;
1254 val->freeram = nr_free_pages();
1255 val->bufferram = nr_blockdev_pages();
1256#ifdef CONFIG_HIGHMEM
1257 val->totalhigh = totalhigh_pages;
1258 val->freehigh = nr_free_highpages();
1259#else
1260 val->totalhigh = 0;
1261 val->freehigh = 0;
1262#endif
1263 val->mem_unit = PAGE_SIZE;
1264}
1265
1266EXPORT_SYMBOL(si_meminfo);
1267
1268#ifdef CONFIG_NUMA
1269void si_meminfo_node(struct sysinfo *val, int nid)
1270{
1271 pg_data_t *pgdat = NODE_DATA(nid);
1272
1273 val->totalram = pgdat->node_present_pages;
1274 val->freeram = nr_free_pages_pgdat(pgdat);
1275 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1276 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1277 val->mem_unit = PAGE_SIZE;
1278}
1279#endif
1280
1281#define K(x) ((x) << (PAGE_SHIFT-10))
1282
1283/*
1284 * Show free area list (used inside shift_scroll-lock stuff)
1285 * We also calculate the percentage fragmentation. We do this by counting the
1286 * memory on each free list with the exception of the first item on the list.
1287 */
1288void show_free_areas(void)
1289{
1290 struct page_state ps;
1291 int cpu, temperature;
1292 unsigned long active;
1293 unsigned long inactive;
1294 unsigned long free;
1295 struct zone *zone;
1296
1297 for_each_zone(zone) {
1298 show_node(zone);
1299 printk("%s per-cpu:", zone->name);
1300
1301 if (!zone->present_pages) {
1302 printk(" empty\n");
1303 continue;
1304 } else
1305 printk("\n");
1306
1307 for (cpu = 0; cpu < NR_CPUS; ++cpu) {
1308 struct per_cpu_pageset *pageset;
1309
1310 if (!cpu_possible(cpu))
1311 continue;
1312
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001313 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
1315 for (temperature = 0; temperature < 2; temperature++)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001316 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 cpu,
1318 temperature ? "cold" : "hot",
1319 pageset->pcp[temperature].low,
1320 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001321 pageset->pcp[temperature].batch,
1322 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 }
1324 }
1325
1326 get_page_state(&ps);
1327 get_zone_counts(&active, &inactive, &free);
1328
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001329 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 K(nr_free_pages()),
1331 K(nr_free_highpages()));
1332
1333 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1334 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1335 active,
1336 inactive,
1337 ps.nr_dirty,
1338 ps.nr_writeback,
1339 ps.nr_unstable,
1340 nr_free_pages(),
1341 ps.nr_slab,
1342 ps.nr_mapped,
1343 ps.nr_page_table_pages);
1344
1345 for_each_zone(zone) {
1346 int i;
1347
1348 show_node(zone);
1349 printk("%s"
1350 " free:%lukB"
1351 " min:%lukB"
1352 " low:%lukB"
1353 " high:%lukB"
1354 " active:%lukB"
1355 " inactive:%lukB"
1356 " present:%lukB"
1357 " pages_scanned:%lu"
1358 " all_unreclaimable? %s"
1359 "\n",
1360 zone->name,
1361 K(zone->free_pages),
1362 K(zone->pages_min),
1363 K(zone->pages_low),
1364 K(zone->pages_high),
1365 K(zone->nr_active),
1366 K(zone->nr_inactive),
1367 K(zone->present_pages),
1368 zone->pages_scanned,
1369 (zone->all_unreclaimable ? "yes" : "no")
1370 );
1371 printk("lowmem_reserve[]:");
1372 for (i = 0; i < MAX_NR_ZONES; i++)
1373 printk(" %lu", zone->lowmem_reserve[i]);
1374 printk("\n");
1375 }
1376
1377 for_each_zone(zone) {
1378 unsigned long nr, flags, order, total = 0;
1379
1380 show_node(zone);
1381 printk("%s: ", zone->name);
1382 if (!zone->present_pages) {
1383 printk("empty\n");
1384 continue;
1385 }
1386
1387 spin_lock_irqsave(&zone->lock, flags);
1388 for (order = 0; order < MAX_ORDER; order++) {
1389 nr = zone->free_area[order].nr_free;
1390 total += nr << order;
1391 printk("%lu*%lukB ", nr, K(1UL) << order);
1392 }
1393 spin_unlock_irqrestore(&zone->lock, flags);
1394 printk("= %lukB\n", K(total));
1395 }
1396
1397 show_swap_cache_info();
1398}
1399
1400/*
1401 * Builds allocation fallback zone lists.
1402 */
1403static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1404{
1405 switch (k) {
1406 struct zone *zone;
1407 default:
1408 BUG();
1409 case ZONE_HIGHMEM:
1410 zone = pgdat->node_zones + ZONE_HIGHMEM;
1411 if (zone->present_pages) {
1412#ifndef CONFIG_HIGHMEM
1413 BUG();
1414#endif
1415 zonelist->zones[j++] = zone;
1416 }
1417 case ZONE_NORMAL:
1418 zone = pgdat->node_zones + ZONE_NORMAL;
1419 if (zone->present_pages)
1420 zonelist->zones[j++] = zone;
1421 case ZONE_DMA:
1422 zone = pgdat->node_zones + ZONE_DMA;
1423 if (zone->present_pages)
1424 zonelist->zones[j++] = zone;
1425 }
1426
1427 return j;
1428}
1429
1430#ifdef CONFIG_NUMA
1431#define MAX_NODE_LOAD (num_online_nodes())
1432static int __initdata node_load[MAX_NUMNODES];
1433/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001434 * find_next_best_node - find the next node that should appear in a given node's fallback list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 * @node: node whose fallback list we're appending
1436 * @used_node_mask: nodemask_t of already used nodes
1437 *
1438 * We use a number of factors to determine which is the next node that should
1439 * appear on a given node's fallback list. The node should not have appeared
1440 * already in @node's fallback list, and it should be the next closest node
1441 * according to the distance array (which contains arbitrary distance values
1442 * from each node to each node in the system), and should also prefer nodes
1443 * with no CPUs, since presumably they'll have very little allocation pressure
1444 * on them otherwise.
1445 * It returns -1 if no node is found.
1446 */
1447static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1448{
1449 int i, n, val;
1450 int min_val = INT_MAX;
1451 int best_node = -1;
1452
1453 for_each_online_node(i) {
1454 cpumask_t tmp;
1455
1456 /* Start from local node */
1457 n = (node+i) % num_online_nodes();
1458
1459 /* Don't want a node to appear more than once */
1460 if (node_isset(n, *used_node_mask))
1461 continue;
1462
1463 /* Use the local node if we haven't already */
1464 if (!node_isset(node, *used_node_mask)) {
1465 best_node = node;
1466 break;
1467 }
1468
1469 /* Use the distance array to find the distance */
1470 val = node_distance(node, n);
1471
1472 /* Give preference to headless and unused nodes */
1473 tmp = node_to_cpumask(n);
1474 if (!cpus_empty(tmp))
1475 val += PENALTY_FOR_NODE_WITH_CPUS;
1476
1477 /* Slight preference for less loaded node */
1478 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1479 val += node_load[n];
1480
1481 if (val < min_val) {
1482 min_val = val;
1483 best_node = n;
1484 }
1485 }
1486
1487 if (best_node >= 0)
1488 node_set(best_node, *used_node_mask);
1489
1490 return best_node;
1491}
1492
1493static void __init build_zonelists(pg_data_t *pgdat)
1494{
1495 int i, j, k, node, local_node;
1496 int prev_node, load;
1497 struct zonelist *zonelist;
1498 nodemask_t used_mask;
1499
1500 /* initialize zonelists */
1501 for (i = 0; i < GFP_ZONETYPES; i++) {
1502 zonelist = pgdat->node_zonelists + i;
1503 zonelist->zones[0] = NULL;
1504 }
1505
1506 /* NUMA-aware ordering of nodes */
1507 local_node = pgdat->node_id;
1508 load = num_online_nodes();
1509 prev_node = local_node;
1510 nodes_clear(used_mask);
1511 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1512 /*
1513 * We don't want to pressure a particular node.
1514 * So adding penalty to the first node in same
1515 * distance group to make it round-robin.
1516 */
1517 if (node_distance(local_node, node) !=
1518 node_distance(local_node, prev_node))
1519 node_load[node] += load;
1520 prev_node = node;
1521 load--;
1522 for (i = 0; i < GFP_ZONETYPES; i++) {
1523 zonelist = pgdat->node_zonelists + i;
1524 for (j = 0; zonelist->zones[j] != NULL; j++);
1525
1526 k = ZONE_NORMAL;
1527 if (i & __GFP_HIGHMEM)
1528 k = ZONE_HIGHMEM;
1529 if (i & __GFP_DMA)
1530 k = ZONE_DMA;
1531
1532 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1533 zonelist->zones[j] = NULL;
1534 }
1535 }
1536}
1537
1538#else /* CONFIG_NUMA */
1539
1540static void __init build_zonelists(pg_data_t *pgdat)
1541{
1542 int i, j, k, node, local_node;
1543
1544 local_node = pgdat->node_id;
1545 for (i = 0; i < GFP_ZONETYPES; i++) {
1546 struct zonelist *zonelist;
1547
1548 zonelist = pgdat->node_zonelists + i;
1549
1550 j = 0;
1551 k = ZONE_NORMAL;
1552 if (i & __GFP_HIGHMEM)
1553 k = ZONE_HIGHMEM;
1554 if (i & __GFP_DMA)
1555 k = ZONE_DMA;
1556
1557 j = build_zonelists_node(pgdat, zonelist, j, k);
1558 /*
1559 * Now we build the zonelist so that it contains the zones
1560 * of all the other nodes.
1561 * We don't want to pressure a particular node, so when
1562 * building the zones for node N, we make sure that the
1563 * zones coming right after the local ones are those from
1564 * node N+1 (modulo N)
1565 */
1566 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1567 if (!node_online(node))
1568 continue;
1569 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1570 }
1571 for (node = 0; node < local_node; node++) {
1572 if (!node_online(node))
1573 continue;
1574 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1575 }
1576
1577 zonelist->zones[j] = NULL;
1578 }
1579}
1580
1581#endif /* CONFIG_NUMA */
1582
1583void __init build_all_zonelists(void)
1584{
1585 int i;
1586
1587 for_each_online_node(i)
1588 build_zonelists(NODE_DATA(i));
1589 printk("Built %i zonelists\n", num_online_nodes());
1590 cpuset_init_current_mems_allowed();
1591}
1592
1593/*
1594 * Helper functions to size the waitqueue hash table.
1595 * Essentially these want to choose hash table sizes sufficiently
1596 * large so that collisions trying to wait on pages are rare.
1597 * But in fact, the number of active page waitqueues on typical
1598 * systems is ridiculously low, less than 200. So this is even
1599 * conservative, even though it seems large.
1600 *
1601 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1602 * waitqueues, i.e. the size of the waitq table given the number of pages.
1603 */
1604#define PAGES_PER_WAITQUEUE 256
1605
1606static inline unsigned long wait_table_size(unsigned long pages)
1607{
1608 unsigned long size = 1;
1609
1610 pages /= PAGES_PER_WAITQUEUE;
1611
1612 while (size < pages)
1613 size <<= 1;
1614
1615 /*
1616 * Once we have dozens or even hundreds of threads sleeping
1617 * on IO we've got bigger problems than wait queue collision.
1618 * Limit the size of the wait table to a reasonable size.
1619 */
1620 size = min(size, 4096UL);
1621
1622 return max(size, 4UL);
1623}
1624
1625/*
1626 * This is an integer logarithm so that shifts can be used later
1627 * to extract the more random high bits from the multiplicative
1628 * hash function before the remainder is taken.
1629 */
1630static inline unsigned long wait_table_bits(unsigned long size)
1631{
1632 return ffz(~size);
1633}
1634
1635#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1636
1637static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1638 unsigned long *zones_size, unsigned long *zholes_size)
1639{
1640 unsigned long realtotalpages, totalpages = 0;
1641 int i;
1642
1643 for (i = 0; i < MAX_NR_ZONES; i++)
1644 totalpages += zones_size[i];
1645 pgdat->node_spanned_pages = totalpages;
1646
1647 realtotalpages = totalpages;
1648 if (zholes_size)
1649 for (i = 0; i < MAX_NR_ZONES; i++)
1650 realtotalpages -= zholes_size[i];
1651 pgdat->node_present_pages = realtotalpages;
1652 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1653}
1654
1655
1656/*
1657 * Initially all pages are reserved - free ones are freed
1658 * up by free_all_bootmem() once the early boot process is
1659 * done. Non-atomic initialization, single-pass.
1660 */
1661void __init memmap_init_zone(unsigned long size, int nid, unsigned long zone,
1662 unsigned long start_pfn)
1663{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001665 unsigned long end_pfn = start_pfn + size;
1666 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001668 for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1669 if (!early_pfn_valid(pfn))
1670 continue;
Andy Whitcroft641c7672005-06-23 00:07:59 -07001671 if (!early_pfn_in_nid(pfn, nid))
1672 continue;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001673 page = pfn_to_page(pfn);
1674 set_page_links(page, zone, nid, pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 set_page_count(page, 0);
1676 reset_page_mapcount(page);
1677 SetPageReserved(page);
1678 INIT_LIST_HEAD(&page->lru);
1679#ifdef WANT_PAGE_VIRTUAL
1680 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1681 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001682 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
1685}
1686
1687void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1688 unsigned long size)
1689{
1690 int order;
1691 for (order = 0; order < MAX_ORDER ; order++) {
1692 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1693 zone->free_area[order].nr_free = 0;
1694 }
1695}
1696
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001697#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1698void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1699 unsigned long size)
1700{
1701 unsigned long snum = pfn_to_section_nr(pfn);
1702 unsigned long end = pfn_to_section_nr(pfn + size);
1703
1704 if (FLAGS_HAS_NODE)
1705 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1706 else
1707 for (; snum <= end; snum++)
1708 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1709}
1710
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711#ifndef __HAVE_ARCH_MEMMAP_INIT
1712#define memmap_init(size, nid, zone, start_pfn) \
1713 memmap_init_zone((size), (nid), (zone), (start_pfn))
1714#endif
1715
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001716static int __devinit zone_batchsize(struct zone *zone)
1717{
1718 int batch;
1719
1720 /*
1721 * The per-cpu-pages pools are set to around 1000th of the
1722 * size of the zone. But no more than 1/4 of a meg - there's
1723 * no point in going beyond the size of L2 cache.
1724 *
1725 * OK, so we don't know how big the cache is. So guess.
1726 */
1727 batch = zone->present_pages / 1024;
1728 if (batch * PAGE_SIZE > 256 * 1024)
1729 batch = (256 * 1024) / PAGE_SIZE;
1730 batch /= 4; /* We effectively *= 4 below */
1731 if (batch < 1)
1732 batch = 1;
1733
1734 /*
1735 * Clamp the batch to a 2^n - 1 value. Having a power
1736 * of 2 value was found to be more likely to have
1737 * suboptimal cache aliasing properties in some cases.
1738 *
1739 * For example if 2 tasks are alternately allocating
1740 * batches of pages, one task can end up with a lot
1741 * of pages of one half of the possible page colors
1742 * and the other with pages of the other colors.
1743 */
1744 batch = (1 << fls(batch + batch/2)) - 1;
1745 return batch;
1746}
1747
Christoph Lameter2caaad42005-06-21 17:15:00 -07001748inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1749{
1750 struct per_cpu_pages *pcp;
1751
1752 pcp = &p->pcp[0]; /* hot */
1753 pcp->count = 0;
1754 pcp->low = 2 * batch;
1755 pcp->high = 6 * batch;
1756 pcp->batch = max(1UL, 1 * batch);
1757 INIT_LIST_HEAD(&pcp->list);
1758
1759 pcp = &p->pcp[1]; /* cold*/
1760 pcp->count = 0;
1761 pcp->low = 0;
1762 pcp->high = 2 * batch;
1763 pcp->batch = max(1UL, 1 * batch);
1764 INIT_LIST_HEAD(&pcp->list);
1765}
1766
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001767#ifdef CONFIG_NUMA
1768/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001769 * Boot pageset table. One per cpu which is going to be used for all
1770 * zones and all nodes. The parameters will be set in such a way
1771 * that an item put on a list will immediately be handed over to
1772 * the buddy list. This is safe since pageset manipulation is done
1773 * with interrupts disabled.
1774 *
1775 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001776 *
1777 * The boot_pagesets must be kept even after bootup is complete for
1778 * unused processors and/or zones. They do play a role for bootstrapping
1779 * hotplugged processors.
1780 *
1781 * zoneinfo_show() and maybe other functions do
1782 * not check if the processor is online before following the pageset pointer.
1783 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001784 */
1785static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001786 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001787
1788/*
1789 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001790 * per cpu pageset array in struct zone.
1791 */
1792static int __devinit process_zones(int cpu)
1793{
1794 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001795
1796 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001797
Christoph Lameter2caaad42005-06-21 17:15:00 -07001798 zone->pageset[cpu] = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001799 GFP_KERNEL, cpu_to_node(cpu));
Christoph Lameter2caaad42005-06-21 17:15:00 -07001800 if (!zone->pageset[cpu])
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001801 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001802
Christoph Lameter2caaad42005-06-21 17:15:00 -07001803 setup_pageset(zone->pageset[cpu], zone_batchsize(zone));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001804 }
1805
1806 return 0;
1807bad:
1808 for_each_zone(dzone) {
1809 if (dzone == zone)
1810 break;
1811 kfree(dzone->pageset[cpu]);
1812 dzone->pageset[cpu] = NULL;
1813 }
1814 return -ENOMEM;
1815}
1816
1817static inline void free_zone_pagesets(int cpu)
1818{
1819#ifdef CONFIG_NUMA
1820 struct zone *zone;
1821
1822 for_each_zone(zone) {
1823 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1824
1825 zone_pcp(zone, cpu) = NULL;
1826 kfree(pset);
1827 }
1828#endif
1829}
1830
1831static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1832 unsigned long action,
1833 void *hcpu)
1834{
1835 int cpu = (long)hcpu;
1836 int ret = NOTIFY_OK;
1837
1838 switch (action) {
1839 case CPU_UP_PREPARE:
1840 if (process_zones(cpu))
1841 ret = NOTIFY_BAD;
1842 break;
1843#ifdef CONFIG_HOTPLUG_CPU
1844 case CPU_DEAD:
1845 free_zone_pagesets(cpu);
1846 break;
1847#endif
1848 default:
1849 break;
1850 }
1851 return ret;
1852}
1853
1854static struct notifier_block pageset_notifier =
1855 { &pageset_cpuup_callback, NULL, 0 };
1856
1857void __init setup_per_cpu_pageset()
1858{
1859 int err;
1860
1861 /* Initialize per_cpu_pageset for cpu 0.
1862 * A cpuup callback will do this for every cpu
1863 * as it comes online
1864 */
1865 err = process_zones(smp_processor_id());
1866 BUG_ON(err);
1867 register_cpu_notifier(&pageset_notifier);
1868}
1869
1870#endif
1871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872/*
1873 * Set up the zone data structures:
1874 * - mark all pages reserved
1875 * - mark all memory queues empty
1876 * - clear the memory bitmaps
1877 */
1878static void __init free_area_init_core(struct pglist_data *pgdat,
1879 unsigned long *zones_size, unsigned long *zholes_size)
1880{
1881 unsigned long i, j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 int cpu, nid = pgdat->node_id;
1883 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1884
1885 pgdat->nr_zones = 0;
1886 init_waitqueue_head(&pgdat->kswapd_wait);
1887 pgdat->kswapd_max_order = 0;
1888
1889 for (j = 0; j < MAX_NR_ZONES; j++) {
1890 struct zone *zone = pgdat->node_zones + j;
1891 unsigned long size, realsize;
1892 unsigned long batch;
1893
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 realsize = size = zones_size[j];
1895 if (zholes_size)
1896 realsize -= zholes_size[j];
1897
1898 if (j == ZONE_DMA || j == ZONE_NORMAL)
1899 nr_kernel_pages += realsize;
1900 nr_all_pages += realsize;
1901
1902 zone->spanned_pages = size;
1903 zone->present_pages = realsize;
1904 zone->name = zone_names[j];
1905 spin_lock_init(&zone->lock);
1906 spin_lock_init(&zone->lru_lock);
1907 zone->zone_pgdat = pgdat;
1908 zone->free_pages = 0;
1909
1910 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1911
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001912 batch = zone_batchsize(zone);
Nick Piggin8e30f272005-05-01 08:58:36 -07001913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 for (cpu = 0; cpu < NR_CPUS; cpu++) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001915#ifdef CONFIG_NUMA
Christoph Lameter2caaad42005-06-21 17:15:00 -07001916 /* Early boot. Slab allocator not functional yet */
1917 zone->pageset[cpu] = &boot_pageset[cpu];
1918 setup_pageset(&boot_pageset[cpu],0);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001919#else
Christoph Lameter2caaad42005-06-21 17:15:00 -07001920 setup_pageset(zone_pcp(zone,cpu), batch);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001921#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 }
1923 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1924 zone_names[j], realsize, batch);
1925 INIT_LIST_HEAD(&zone->active_list);
1926 INIT_LIST_HEAD(&zone->inactive_list);
1927 zone->nr_scan_active = 0;
1928 zone->nr_scan_inactive = 0;
1929 zone->nr_active = 0;
1930 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07001931 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 if (!size)
1933 continue;
1934
1935 /*
1936 * The per-page waitqueue mechanism uses hashed waitqueues
1937 * per zone.
1938 */
1939 zone->wait_table_size = wait_table_size(size);
1940 zone->wait_table_bits =
1941 wait_table_bits(zone->wait_table_size);
1942 zone->wait_table = (wait_queue_head_t *)
1943 alloc_bootmem_node(pgdat, zone->wait_table_size
1944 * sizeof(wait_queue_head_t));
1945
1946 for(i = 0; i < zone->wait_table_size; ++i)
1947 init_waitqueue_head(zone->wait_table + i);
1948
1949 pgdat->nr_zones = j+1;
1950
1951 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1952 zone->zone_start_pfn = zone_start_pfn;
1953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 memmap_init(size, nid, j, zone_start_pfn);
1955
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001956 zonetable_add(zone, nid, j, zone_start_pfn, size);
1957
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 zone_start_pfn += size;
1959
1960 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1961 }
1962}
1963
1964static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1965{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 /* Skip empty nodes */
1967 if (!pgdat->node_spanned_pages)
1968 return;
1969
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001970#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 /* ia64 gets its own node_mem_map, before this, without bootmem */
1972 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001973 unsigned long size;
1974 struct page *map;
1975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07001977 map = alloc_remap(pgdat->node_id, size);
1978 if (!map)
1979 map = alloc_bootmem_node(pgdat, size);
1980 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001982#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 /*
1984 * With no DISCONTIG, the global mem_map is just set as node 0's
1985 */
1986 if (pgdat == NODE_DATA(0))
1987 mem_map = NODE_DATA(0)->node_mem_map;
1988#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001989#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990}
1991
1992void __init free_area_init_node(int nid, struct pglist_data *pgdat,
1993 unsigned long *zones_size, unsigned long node_start_pfn,
1994 unsigned long *zholes_size)
1995{
1996 pgdat->node_id = nid;
1997 pgdat->node_start_pfn = node_start_pfn;
1998 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
1999
2000 alloc_node_mem_map(pgdat);
2001
2002 free_area_init_core(pgdat, zones_size, zholes_size);
2003}
2004
Dave Hansen93b75042005-06-23 00:07:47 -07002005#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006static bootmem_data_t contig_bootmem_data;
2007struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2008
2009EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002010#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
2012void __init free_area_init(unsigned long *zones_size)
2013{
Dave Hansen93b75042005-06-23 00:07:47 -07002014 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2016}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
2018#ifdef CONFIG_PROC_FS
2019
2020#include <linux/seq_file.h>
2021
2022static void *frag_start(struct seq_file *m, loff_t *pos)
2023{
2024 pg_data_t *pgdat;
2025 loff_t node = *pos;
2026
2027 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2028 --node;
2029
2030 return pgdat;
2031}
2032
2033static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2034{
2035 pg_data_t *pgdat = (pg_data_t *)arg;
2036
2037 (*pos)++;
2038 return pgdat->pgdat_next;
2039}
2040
2041static void frag_stop(struct seq_file *m, void *arg)
2042{
2043}
2044
2045/*
2046 * This walks the free areas for each zone.
2047 */
2048static int frag_show(struct seq_file *m, void *arg)
2049{
2050 pg_data_t *pgdat = (pg_data_t *)arg;
2051 struct zone *zone;
2052 struct zone *node_zones = pgdat->node_zones;
2053 unsigned long flags;
2054 int order;
2055
2056 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2057 if (!zone->present_pages)
2058 continue;
2059
2060 spin_lock_irqsave(&zone->lock, flags);
2061 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2062 for (order = 0; order < MAX_ORDER; ++order)
2063 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2064 spin_unlock_irqrestore(&zone->lock, flags);
2065 seq_putc(m, '\n');
2066 }
2067 return 0;
2068}
2069
2070struct seq_operations fragmentation_op = {
2071 .start = frag_start,
2072 .next = frag_next,
2073 .stop = frag_stop,
2074 .show = frag_show,
2075};
2076
Nikita Danilov295ab932005-06-21 17:14:38 -07002077/*
2078 * Output information about zones in @pgdat.
2079 */
2080static int zoneinfo_show(struct seq_file *m, void *arg)
2081{
2082 pg_data_t *pgdat = arg;
2083 struct zone *zone;
2084 struct zone *node_zones = pgdat->node_zones;
2085 unsigned long flags;
2086
2087 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2088 int i;
2089
2090 if (!zone->present_pages)
2091 continue;
2092
2093 spin_lock_irqsave(&zone->lock, flags);
2094 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2095 seq_printf(m,
2096 "\n pages free %lu"
2097 "\n min %lu"
2098 "\n low %lu"
2099 "\n high %lu"
2100 "\n active %lu"
2101 "\n inactive %lu"
2102 "\n scanned %lu (a: %lu i: %lu)"
2103 "\n spanned %lu"
2104 "\n present %lu",
2105 zone->free_pages,
2106 zone->pages_min,
2107 zone->pages_low,
2108 zone->pages_high,
2109 zone->nr_active,
2110 zone->nr_inactive,
2111 zone->pages_scanned,
2112 zone->nr_scan_active, zone->nr_scan_inactive,
2113 zone->spanned_pages,
2114 zone->present_pages);
2115 seq_printf(m,
2116 "\n protection: (%lu",
2117 zone->lowmem_reserve[0]);
2118 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2119 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2120 seq_printf(m,
2121 ")"
2122 "\n pagesets");
2123 for (i = 0; i < ARRAY_SIZE(zone->pageset); i++) {
2124 struct per_cpu_pageset *pageset;
2125 int j;
2126
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002127 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002128 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2129 if (pageset->pcp[j].count)
2130 break;
2131 }
2132 if (j == ARRAY_SIZE(pageset->pcp))
2133 continue;
2134 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2135 seq_printf(m,
2136 "\n cpu: %i pcp: %i"
2137 "\n count: %i"
2138 "\n low: %i"
2139 "\n high: %i"
2140 "\n batch: %i",
2141 i, j,
2142 pageset->pcp[j].count,
2143 pageset->pcp[j].low,
2144 pageset->pcp[j].high,
2145 pageset->pcp[j].batch);
2146 }
2147#ifdef CONFIG_NUMA
2148 seq_printf(m,
2149 "\n numa_hit: %lu"
2150 "\n numa_miss: %lu"
2151 "\n numa_foreign: %lu"
2152 "\n interleave_hit: %lu"
2153 "\n local_node: %lu"
2154 "\n other_node: %lu",
2155 pageset->numa_hit,
2156 pageset->numa_miss,
2157 pageset->numa_foreign,
2158 pageset->interleave_hit,
2159 pageset->local_node,
2160 pageset->other_node);
2161#endif
2162 }
2163 seq_printf(m,
2164 "\n all_unreclaimable: %u"
2165 "\n prev_priority: %i"
2166 "\n temp_priority: %i"
2167 "\n start_pfn: %lu",
2168 zone->all_unreclaimable,
2169 zone->prev_priority,
2170 zone->temp_priority,
2171 zone->zone_start_pfn);
2172 spin_unlock_irqrestore(&zone->lock, flags);
2173 seq_putc(m, '\n');
2174 }
2175 return 0;
2176}
2177
2178struct seq_operations zoneinfo_op = {
2179 .start = frag_start, /* iterate over all zones. The same as in
2180 * fragmentation. */
2181 .next = frag_next,
2182 .stop = frag_stop,
2183 .show = zoneinfo_show,
2184};
2185
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186static char *vmstat_text[] = {
2187 "nr_dirty",
2188 "nr_writeback",
2189 "nr_unstable",
2190 "nr_page_table_pages",
2191 "nr_mapped",
2192 "nr_slab",
2193
2194 "pgpgin",
2195 "pgpgout",
2196 "pswpin",
2197 "pswpout",
2198 "pgalloc_high",
2199
2200 "pgalloc_normal",
2201 "pgalloc_dma",
2202 "pgfree",
2203 "pgactivate",
2204 "pgdeactivate",
2205
2206 "pgfault",
2207 "pgmajfault",
2208 "pgrefill_high",
2209 "pgrefill_normal",
2210 "pgrefill_dma",
2211
2212 "pgsteal_high",
2213 "pgsteal_normal",
2214 "pgsteal_dma",
2215 "pgscan_kswapd_high",
2216 "pgscan_kswapd_normal",
2217
2218 "pgscan_kswapd_dma",
2219 "pgscan_direct_high",
2220 "pgscan_direct_normal",
2221 "pgscan_direct_dma",
2222 "pginodesteal",
2223
2224 "slabs_scanned",
2225 "kswapd_steal",
2226 "kswapd_inodesteal",
2227 "pageoutrun",
2228 "allocstall",
2229
2230 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002231 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232};
2233
2234static void *vmstat_start(struct seq_file *m, loff_t *pos)
2235{
2236 struct page_state *ps;
2237
2238 if (*pos >= ARRAY_SIZE(vmstat_text))
2239 return NULL;
2240
2241 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2242 m->private = ps;
2243 if (!ps)
2244 return ERR_PTR(-ENOMEM);
2245 get_full_page_state(ps);
2246 ps->pgpgin /= 2; /* sectors -> kbytes */
2247 ps->pgpgout /= 2;
2248 return (unsigned long *)ps + *pos;
2249}
2250
2251static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2252{
2253 (*pos)++;
2254 if (*pos >= ARRAY_SIZE(vmstat_text))
2255 return NULL;
2256 return (unsigned long *)m->private + *pos;
2257}
2258
2259static int vmstat_show(struct seq_file *m, void *arg)
2260{
2261 unsigned long *l = arg;
2262 unsigned long off = l - (unsigned long *)m->private;
2263
2264 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2265 return 0;
2266}
2267
2268static void vmstat_stop(struct seq_file *m, void *arg)
2269{
2270 kfree(m->private);
2271 m->private = NULL;
2272}
2273
2274struct seq_operations vmstat_op = {
2275 .start = vmstat_start,
2276 .next = vmstat_next,
2277 .stop = vmstat_stop,
2278 .show = vmstat_show,
2279};
2280
2281#endif /* CONFIG_PROC_FS */
2282
2283#ifdef CONFIG_HOTPLUG_CPU
2284static int page_alloc_cpu_notify(struct notifier_block *self,
2285 unsigned long action, void *hcpu)
2286{
2287 int cpu = (unsigned long)hcpu;
2288 long *count;
2289 unsigned long *src, *dest;
2290
2291 if (action == CPU_DEAD) {
2292 int i;
2293
2294 /* Drain local pagecache count. */
2295 count = &per_cpu(nr_pagecache_local, cpu);
2296 atomic_add(*count, &nr_pagecache);
2297 *count = 0;
2298 local_irq_disable();
2299 __drain_pages(cpu);
2300
2301 /* Add dead cpu's page_states to our own. */
2302 dest = (unsigned long *)&__get_cpu_var(page_states);
2303 src = (unsigned long *)&per_cpu(page_states, cpu);
2304
2305 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2306 i++) {
2307 dest[i] += src[i];
2308 src[i] = 0;
2309 }
2310
2311 local_irq_enable();
2312 }
2313 return NOTIFY_OK;
2314}
2315#endif /* CONFIG_HOTPLUG_CPU */
2316
2317void __init page_alloc_init(void)
2318{
2319 hotcpu_notifier(page_alloc_cpu_notify, 0);
2320}
2321
2322/*
2323 * setup_per_zone_lowmem_reserve - called whenever
2324 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2325 * has a correct pages reserved value, so an adequate number of
2326 * pages are left in the zone after a successful __alloc_pages().
2327 */
2328static void setup_per_zone_lowmem_reserve(void)
2329{
2330 struct pglist_data *pgdat;
2331 int j, idx;
2332
2333 for_each_pgdat(pgdat) {
2334 for (j = 0; j < MAX_NR_ZONES; j++) {
2335 struct zone *zone = pgdat->node_zones + j;
2336 unsigned long present_pages = zone->present_pages;
2337
2338 zone->lowmem_reserve[j] = 0;
2339
2340 for (idx = j-1; idx >= 0; idx--) {
2341 struct zone *lower_zone;
2342
2343 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2344 sysctl_lowmem_reserve_ratio[idx] = 1;
2345
2346 lower_zone = pgdat->node_zones + idx;
2347 lower_zone->lowmem_reserve[j] = present_pages /
2348 sysctl_lowmem_reserve_ratio[idx];
2349 present_pages += lower_zone->present_pages;
2350 }
2351 }
2352 }
2353}
2354
2355/*
2356 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2357 * that the pages_{min,low,high} values for each zone are set correctly
2358 * with respect to min_free_kbytes.
2359 */
2360static void setup_per_zone_pages_min(void)
2361{
2362 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2363 unsigned long lowmem_pages = 0;
2364 struct zone *zone;
2365 unsigned long flags;
2366
2367 /* Calculate total number of !ZONE_HIGHMEM pages */
2368 for_each_zone(zone) {
2369 if (!is_highmem(zone))
2370 lowmem_pages += zone->present_pages;
2371 }
2372
2373 for_each_zone(zone) {
2374 spin_lock_irqsave(&zone->lru_lock, flags);
2375 if (is_highmem(zone)) {
2376 /*
2377 * Often, highmem doesn't need to reserve any pages.
2378 * But the pages_min/low/high values are also used for
2379 * batching up page reclaim activity so we need a
2380 * decent value here.
2381 */
2382 int min_pages;
2383
2384 min_pages = zone->present_pages / 1024;
2385 if (min_pages < SWAP_CLUSTER_MAX)
2386 min_pages = SWAP_CLUSTER_MAX;
2387 if (min_pages > 128)
2388 min_pages = 128;
2389 zone->pages_min = min_pages;
2390 } else {
Nikita Danilov295ab932005-06-21 17:14:38 -07002391 /* if it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 * proportionate to the zone's size.
2393 */
Nikita Danilov295ab932005-06-21 17:14:38 -07002394 zone->pages_min = (pages_min * zone->present_pages) /
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 lowmem_pages;
2396 }
2397
2398 /*
2399 * When interpreting these watermarks, just keep in mind that:
2400 * zone->pages_min == (zone->pages_min * 4) / 4;
2401 */
2402 zone->pages_low = (zone->pages_min * 5) / 4;
2403 zone->pages_high = (zone->pages_min * 6) / 4;
2404 spin_unlock_irqrestore(&zone->lru_lock, flags);
2405 }
2406}
2407
2408/*
2409 * Initialise min_free_kbytes.
2410 *
2411 * For small machines we want it small (128k min). For large machines
2412 * we want it large (64MB max). But it is not linear, because network
2413 * bandwidth does not increase linearly with machine size. We use
2414 *
2415 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2416 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2417 *
2418 * which yields
2419 *
2420 * 16MB: 512k
2421 * 32MB: 724k
2422 * 64MB: 1024k
2423 * 128MB: 1448k
2424 * 256MB: 2048k
2425 * 512MB: 2896k
2426 * 1024MB: 4096k
2427 * 2048MB: 5792k
2428 * 4096MB: 8192k
2429 * 8192MB: 11584k
2430 * 16384MB: 16384k
2431 */
2432static int __init init_per_zone_pages_min(void)
2433{
2434 unsigned long lowmem_kbytes;
2435
2436 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2437
2438 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2439 if (min_free_kbytes < 128)
2440 min_free_kbytes = 128;
2441 if (min_free_kbytes > 65536)
2442 min_free_kbytes = 65536;
2443 setup_per_zone_pages_min();
2444 setup_per_zone_lowmem_reserve();
2445 return 0;
2446}
2447module_init(init_per_zone_pages_min)
2448
2449/*
2450 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2451 * that we can call two helper functions whenever min_free_kbytes
2452 * changes.
2453 */
2454int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2455 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2456{
2457 proc_dointvec(table, write, file, buffer, length, ppos);
2458 setup_per_zone_pages_min();
2459 return 0;
2460}
2461
2462/*
2463 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2464 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2465 * whenever sysctl_lowmem_reserve_ratio changes.
2466 *
2467 * The reserve ratio obviously has absolutely no relation with the
2468 * pages_min watermarks. The lowmem reserve ratio can only make sense
2469 * if in function of the boot time zone sizes.
2470 */
2471int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2472 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2473{
2474 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2475 setup_per_zone_lowmem_reserve();
2476 return 0;
2477}
2478
2479__initdata int hashdist = HASHDIST_DEFAULT;
2480
2481#ifdef CONFIG_NUMA
2482static int __init set_hashdist(char *str)
2483{
2484 if (!str)
2485 return 0;
2486 hashdist = simple_strtoul(str, &str, 0);
2487 return 1;
2488}
2489__setup("hashdist=", set_hashdist);
2490#endif
2491
2492/*
2493 * allocate a large system hash table from bootmem
2494 * - it is assumed that the hash table must contain an exact power-of-2
2495 * quantity of entries
2496 * - limit is the number of hash buckets, not the total allocation size
2497 */
2498void *__init alloc_large_system_hash(const char *tablename,
2499 unsigned long bucketsize,
2500 unsigned long numentries,
2501 int scale,
2502 int flags,
2503 unsigned int *_hash_shift,
2504 unsigned int *_hash_mask,
2505 unsigned long limit)
2506{
2507 unsigned long long max = limit;
2508 unsigned long log2qty, size;
2509 void *table = NULL;
2510
2511 /* allow the kernel cmdline to have a say */
2512 if (!numentries) {
2513 /* round applicable memory size up to nearest megabyte */
2514 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2515 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2516 numentries >>= 20 - PAGE_SHIFT;
2517 numentries <<= 20 - PAGE_SHIFT;
2518
2519 /* limit to 1 bucket per 2^scale bytes of low memory */
2520 if (scale > PAGE_SHIFT)
2521 numentries >>= (scale - PAGE_SHIFT);
2522 else
2523 numentries <<= (PAGE_SHIFT - scale);
2524 }
2525 /* rounded up to nearest power of 2 in size */
2526 numentries = 1UL << (long_log2(numentries) + 1);
2527
2528 /* limit allocation size to 1/16 total memory by default */
2529 if (max == 0) {
2530 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2531 do_div(max, bucketsize);
2532 }
2533
2534 if (numentries > max)
2535 numentries = max;
2536
2537 log2qty = long_log2(numentries);
2538
2539 do {
2540 size = bucketsize << log2qty;
2541 if (flags & HASH_EARLY)
2542 table = alloc_bootmem(size);
2543 else if (hashdist)
2544 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2545 else {
2546 unsigned long order;
2547 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2548 ;
2549 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2550 }
2551 } while (!table && size > PAGE_SIZE && --log2qty);
2552
2553 if (!table)
2554 panic("Failed to allocate %s hash table\n", tablename);
2555
2556 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2557 tablename,
2558 (1U << log2qty),
2559 long_log2(size) - PAGE_SHIFT,
2560 size);
2561
2562 if (_hash_shift)
2563 *_hash_shift = log2qty;
2564 if (_hash_mask)
2565 *_hash_mask = (1 << log2qty) - 1;
2566
2567 return table;
2568}