blob: 3c5cf664abd2eca14d2613a57023dcfcac6b6fde [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>
Randy Dunlap9f158332005-09-13 01:25:16 -070025#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#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>
Dave Hansenbdc8cb92005-10-29 18:16:53 -070036#include <linux/memory_hotplug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/nodemask.h>
38#include <linux/vmalloc.h>
39
40#include <asm/tlbflush.h>
41#include "internal.h"
42
43/*
44 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
45 * initializer cleaner
46 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070047nodemask_t node_online_map __read_mostly = { { [0] = 1UL } };
Dean Nelson7223a932005-03-23 19:00:00 -070048EXPORT_SYMBOL(node_online_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070049nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL;
Dean Nelson7223a932005-03-23 19:00:00 -070050EXPORT_SYMBOL(node_possible_map);
Christoph Lameterc3d8c142005-09-06 15:16:33 -070051struct pglist_data *pgdat_list __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070052unsigned long totalram_pages __read_mostly;
53unsigned long totalhigh_pages __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054long nr_swap_pages;
55
56/*
57 * results with 256, 32 in the lowmem_reserve sysctl:
58 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
59 * 1G machine -> (16M dma, 784M normal, 224M high)
60 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
61 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
62 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
63 */
64int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 32 };
65
66EXPORT_SYMBOL(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68/*
69 * Used by page_zone() to look up the address of the struct zone whose
70 * id is encoded in the upper bits of page->flags
71 */
Christoph Lameterc3d8c142005-09-06 15:16:33 -070072struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073EXPORT_SYMBOL(zone_table);
74
75static char *zone_names[MAX_NR_ZONES] = { "DMA", "Normal", "HighMem" };
76int min_free_kbytes = 1024;
77
78unsigned long __initdata nr_kernel_pages;
79unsigned long __initdata nr_all_pages;
80
Dave Hansenc6a57e12005-10-29 18:16:52 -070081static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
Dave Hansenbdc8cb92005-10-29 18:16:53 -070083 int ret = 0;
84 unsigned seq;
85 unsigned long pfn = page_to_pfn(page);
Dave Hansenc6a57e12005-10-29 18:16:52 -070086
Dave Hansenbdc8cb92005-10-29 18:16:53 -070087 do {
88 seq = zone_span_seqbegin(zone);
89 if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
90 ret = 1;
91 else if (pfn < zone->zone_start_pfn)
92 ret = 1;
93 } while (zone_span_seqretry(zone, seq));
94
95 return ret;
Dave Hansenc6a57e12005-10-29 18:16:52 -070096}
97
98static int page_is_consistent(struct zone *zone, struct page *page)
99{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100#ifdef CONFIG_HOLES_IN_ZONE
101 if (!pfn_valid(page_to_pfn(page)))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700102 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103#endif
104 if (zone != page_zone(page))
Dave Hansenc6a57e12005-10-29 18:16:52 -0700105 return 0;
106
107 return 1;
108}
109/*
110 * Temporary debugging check for pages not lying within a given zone.
111 */
112static int bad_range(struct zone *zone, struct page *page)
113{
114 if (page_outside_zone_boundaries(zone, page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 return 1;
Dave Hansenc6a57e12005-10-29 18:16:52 -0700116 if (!page_is_consistent(zone, page))
117 return 1;
118
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 return 0;
120}
121
122static void bad_page(const char *function, struct page *page)
123{
124 printk(KERN_EMERG "Bad page state at %s (in process '%s', page %p)\n",
125 function, current->comm, page);
126 printk(KERN_EMERG "flags:0x%0*lx mapping:%p mapcount:%d count:%d\n",
127 (int)(2*sizeof(page_flags_t)), (unsigned long)page->flags,
128 page->mapping, page_mapcount(page), page_count(page));
129 printk(KERN_EMERG "Backtrace:\n");
130 dump_stack();
131 printk(KERN_EMERG "Trying to fix it up, but a reboot is needed\n");
Hugh Dickins334795e2005-06-21 17:15:08 -0700132 page->flags &= ~(1 << PG_lru |
133 1 << PG_private |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 1 << PG_active |
136 1 << PG_dirty |
Hugh Dickins334795e2005-06-21 17:15:08 -0700137 1 << PG_reclaim |
138 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700140 1 << PG_writeback |
141 1 << PG_reserved );
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 set_page_count(page, 0);
143 reset_page_mapcount(page);
144 page->mapping = NULL;
Randy Dunlap9f158332005-09-13 01:25:16 -0700145 add_taint(TAINT_BAD_PAGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
148#ifndef CONFIG_HUGETLB_PAGE
149#define prep_compound_page(page, order) do { } while (0)
150#define destroy_compound_page(page, order) do { } while (0)
151#else
152/*
153 * Higher-order pages are called "compound pages". They are structured thusly:
154 *
155 * The first PAGE_SIZE page is called the "head page".
156 *
157 * The remaining PAGE_SIZE pages are called "tail pages".
158 *
159 * All pages have PG_compound set. All pages have their ->private pointing at
160 * the head page (even the head page has this).
161 *
162 * The first tail page's ->mapping, if non-zero, holds the address of the
163 * compound page's put_page() function.
164 *
165 * The order of the allocation is stored in the first tail page's ->index
166 * This is only for debug at present. This usage means that zero-order pages
167 * may not be compound.
168 */
169static void prep_compound_page(struct page *page, unsigned long order)
170{
171 int i;
172 int nr_pages = 1 << order;
173
174 page[1].mapping = NULL;
175 page[1].index = order;
176 for (i = 0; i < nr_pages; i++) {
177 struct page *p = page + i;
178
179 SetPageCompound(p);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700180 set_page_private(p, (unsigned long)page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 }
182}
183
184static void destroy_compound_page(struct page *page, unsigned long order)
185{
186 int i;
187 int nr_pages = 1 << order;
188
189 if (!PageCompound(page))
190 return;
191
192 if (page[1].index != order)
193 bad_page(__FUNCTION__, page);
194
195 for (i = 0; i < nr_pages; i++) {
196 struct page *p = page + i;
197
198 if (!PageCompound(p))
199 bad_page(__FUNCTION__, page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700200 if (page_private(p) != (unsigned long)page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 bad_page(__FUNCTION__, page);
202 ClearPageCompound(p);
203 }
204}
205#endif /* CONFIG_HUGETLB_PAGE */
206
207/*
208 * function for dealing with page's order in buddy system.
209 * zone->lock is already acquired when we use these.
210 * So, we don't need atomic page->flags operations here.
211 */
212static inline unsigned long page_order(struct page *page) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700213 return page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
216static inline void set_page_order(struct page *page, int order) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700217 set_page_private(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 __SetPagePrivate(page);
219}
220
221static inline void rmv_page_order(struct page *page)
222{
223 __ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700224 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
227/*
228 * Locate the struct page for both the matching buddy in our
229 * pair (buddy1) and the combined O(n+1) page they form (page).
230 *
231 * 1) Any buddy B1 will have an order O twin B2 which satisfies
232 * the following equation:
233 * B2 = B1 ^ (1 << O)
234 * For example, if the starting buddy (buddy2) is #8 its order
235 * 1 buddy is #10:
236 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
237 *
238 * 2) Any buddy B will have an order O+1 parent P which
239 * satisfies the following equation:
240 * P = B & ~(1 << O)
241 *
242 * Assumption: *_mem_map is contigious at least up to MAX_ORDER
243 */
244static inline struct page *
245__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
246{
247 unsigned long buddy_idx = page_idx ^ (1 << order);
248
249 return page + (buddy_idx - page_idx);
250}
251
252static inline unsigned long
253__find_combined_index(unsigned long page_idx, unsigned int order)
254{
255 return (page_idx & ~(1 << order));
256}
257
258/*
259 * This function checks whether a page is free && is the buddy
260 * we can do coalesce a page and its buddy if
261 * (a) the buddy is free &&
262 * (b) the buddy is on the buddy system &&
263 * (c) a page and its buddy have the same order.
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700264 * for recording page's order, we use page_private(page) and PG_private.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 *
266 */
267static inline int page_is_buddy(struct page *page, int order)
268{
269 if (PagePrivate(page) &&
270 (page_order(page) == order) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 page_count(page) == 0)
272 return 1;
273 return 0;
274}
275
276/*
277 * Freeing function for a buddy system allocator.
278 *
279 * The concept of a buddy system is to maintain direct-mapped table
280 * (containing bit values) for memory blocks of various "orders".
281 * The bottom level table contains the map for the smallest allocatable
282 * units of memory (here, pages), and each level above it describes
283 * pairs of units from the levels below, hence, "buddies".
284 * At a high level, all that happens here is marking the table entry
285 * at the bottom level available, and propagating the changes upward
286 * as necessary, plus some accounting needed to play nicely with other
287 * parts of the VM system.
288 * At each level, we keep a list of pages, which are heads of continuous
289 * free pages of length of (1 << order) and marked with PG_Private.Page's
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700290 * order is recorded in page_private(page) field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 * So when we are allocating or freeing one, we can derive the state of the
292 * other. That is, if we allocate a small block, and both were
293 * free, the remainder of the region must be split into blocks.
294 * If a block is freed, and its buddy is also free, then this
295 * triggers coalescing into a block of larger size.
296 *
297 * -- wli
298 */
299
300static inline void __free_pages_bulk (struct page *page,
301 struct zone *zone, unsigned int order)
302{
303 unsigned long page_idx;
304 int order_size = 1 << order;
305
306 if (unlikely(order))
307 destroy_compound_page(page, order);
308
309 page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
310
311 BUG_ON(page_idx & (order_size - 1));
312 BUG_ON(bad_range(zone, page));
313
314 zone->free_pages += order_size;
315 while (order < MAX_ORDER-1) {
316 unsigned long combined_idx;
317 struct free_area *area;
318 struct page *buddy;
319
320 combined_idx = __find_combined_index(page_idx, order);
321 buddy = __page_find_buddy(page, page_idx, order);
322
323 if (bad_range(zone, buddy))
324 break;
325 if (!page_is_buddy(buddy, order))
326 break; /* Move the buddy up one level. */
327 list_del(&buddy->lru);
328 area = zone->free_area + order;
329 area->nr_free--;
330 rmv_page_order(buddy);
331 page = page + (combined_idx - page_idx);
332 page_idx = combined_idx;
333 order++;
334 }
335 set_page_order(page, order);
336 list_add(&page->lru, &zone->free_area[order].free_list);
337 zone->free_area[order].nr_free++;
338}
339
340static inline void free_pages_check(const char *function, struct page *page)
341{
342 if ( page_mapcount(page) ||
343 page->mapping != NULL ||
344 page_count(page) != 0 ||
345 (page->flags & (
346 1 << PG_lru |
347 1 << PG_private |
348 1 << PG_locked |
349 1 << PG_active |
350 1 << PG_reclaim |
351 1 << PG_slab |
352 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700353 1 << PG_writeback |
354 1 << PG_reserved )))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 bad_page(function, page);
356 if (PageDirty(page))
Nick Piggin242e5462005-09-03 15:54:50 -0700357 __ClearPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358}
359
360/*
361 * Frees a list of pages.
362 * Assumes all pages on list are in same zone, and of same order.
Renaud Lienhart207f36e2005-09-10 00:26:59 -0700363 * count is the number of pages to free.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 *
365 * If the zone was previously in an "all pages pinned" state then look to
366 * see if this freeing clears that state.
367 *
368 * And clear the zone's pages_scanned counter, to hold off the "all pages are
369 * pinned" detection logic.
370 */
371static int
372free_pages_bulk(struct zone *zone, int count,
373 struct list_head *list, unsigned int order)
374{
375 unsigned long flags;
376 struct page *page = NULL;
377 int ret = 0;
378
379 spin_lock_irqsave(&zone->lock, flags);
380 zone->all_unreclaimable = 0;
381 zone->pages_scanned = 0;
382 while (!list_empty(list) && count--) {
383 page = list_entry(list->prev, struct page, lru);
384 /* have to delete it as __free_pages_bulk list manipulates */
385 list_del(&page->lru);
386 __free_pages_bulk(page, zone, order);
387 ret++;
388 }
389 spin_unlock_irqrestore(&zone->lock, flags);
390 return ret;
391}
392
393void __free_pages_ok(struct page *page, unsigned int order)
394{
395 LIST_HEAD(list);
396 int i;
397
398 arch_free_page(page, order);
399
400 mod_page_state(pgfree, 1 << order);
401
402#ifndef CONFIG_MMU
403 if (order > 0)
404 for (i = 1 ; i < (1 << order) ; ++i)
405 __put_page(page + i);
406#endif
407
408 for (i = 0 ; i < (1 << order) ; ++i)
409 free_pages_check(__FUNCTION__, page + i);
410 list_add(&page->lru, &list);
411 kernel_map_pages(page, 1<<order, 0);
412 free_pages_bulk(page_zone(page), 1, &list, order);
413}
414
415
416/*
417 * The order of subdivision here is critical for the IO subsystem.
418 * Please do not alter this order without good reasons and regression
419 * testing. Specifically, as large blocks of memory are subdivided,
420 * the order in which smaller blocks are delivered depends on the order
421 * they're subdivided in this function. This is the primary factor
422 * influencing the order in which pages are delivered to the IO
423 * subsystem according to empirical testing, and this is also justified
424 * by considering the behavior of a buddy system containing a single
425 * large block of memory acted on by a series of small allocations.
426 * This behavior is a critical factor in sglist merging's success.
427 *
428 * -- wli
429 */
430static inline struct page *
431expand(struct zone *zone, struct page *page,
432 int low, int high, struct free_area *area)
433{
434 unsigned long size = 1 << high;
435
436 while (high > low) {
437 area--;
438 high--;
439 size >>= 1;
440 BUG_ON(bad_range(zone, &page[size]));
441 list_add(&page[size].lru, &area->free_list);
442 area->nr_free++;
443 set_page_order(&page[size], high);
444 }
445 return page;
446}
447
448void set_page_refs(struct page *page, int order)
449{
450#ifdef CONFIG_MMU
451 set_page_count(page, 1);
452#else
453 int i;
454
455 /*
456 * We need to reference all the pages for this order, otherwise if
457 * anyone accesses one of the pages with (get/put) it will be freed.
458 * - eg: access_process_vm()
459 */
460 for (i = 0; i < (1 << order); i++)
461 set_page_count(page + i, 1);
462#endif /* CONFIG_MMU */
463}
464
465/*
466 * This page is about to be returned from the page allocator
467 */
468static void prep_new_page(struct page *page, int order)
469{
Hugh Dickins334795e2005-06-21 17:15:08 -0700470 if ( page_mapcount(page) ||
471 page->mapping != NULL ||
472 page_count(page) != 0 ||
473 (page->flags & (
474 1 << PG_lru |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 1 << PG_private |
476 1 << PG_locked |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 1 << PG_active |
478 1 << PG_dirty |
479 1 << PG_reclaim |
Hugh Dickins334795e2005-06-21 17:15:08 -0700480 1 << PG_slab |
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 1 << PG_swapcache |
Nick Pigginb5810032005-10-29 18:16:12 -0700482 1 << PG_writeback |
483 1 << PG_reserved )))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 bad_page(__FUNCTION__, page);
485
486 page->flags &= ~(1 << PG_uptodate | 1 << PG_error |
487 1 << PG_referenced | 1 << PG_arch_1 |
488 1 << PG_checked | 1 << PG_mappedtodisk);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700489 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 set_page_refs(page, order);
491 kernel_map_pages(page, 1 << order, 1);
492}
493
494/*
495 * Do the hard work of removing an element from the buddy allocator.
496 * Call me with the zone->lock already held.
497 */
498static struct page *__rmqueue(struct zone *zone, unsigned int order)
499{
500 struct free_area * area;
501 unsigned int current_order;
502 struct page *page;
503
504 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
505 area = zone->free_area + current_order;
506 if (list_empty(&area->free_list))
507 continue;
508
509 page = list_entry(area->free_list.next, struct page, lru);
510 list_del(&page->lru);
511 rmv_page_order(page);
512 area->nr_free--;
513 zone->free_pages -= 1UL << order;
514 return expand(zone, page, order, current_order, area);
515 }
516
517 return NULL;
518}
519
520/*
521 * Obtain a specified number of elements from the buddy allocator, all under
522 * a single hold of the lock, for efficiency. Add them to the supplied list.
523 * Returns the number of new pages which were placed at *list.
524 */
525static int rmqueue_bulk(struct zone *zone, unsigned int order,
526 unsigned long count, struct list_head *list)
527{
528 unsigned long flags;
529 int i;
530 int allocated = 0;
531 struct page *page;
532
533 spin_lock_irqsave(&zone->lock, flags);
534 for (i = 0; i < count; ++i) {
535 page = __rmqueue(zone, order);
536 if (page == NULL)
537 break;
538 allocated++;
539 list_add_tail(&page->lru, list);
540 }
541 spin_unlock_irqrestore(&zone->lock, flags);
542 return allocated;
543}
544
Christoph Lameter4ae7c032005-06-21 17:14:57 -0700545#ifdef CONFIG_NUMA
546/* Called from the slab reaper to drain remote pagesets */
547void drain_remote_pages(void)
548{
549 struct zone *zone;
550 int i;
551 unsigned long flags;
552
553 local_irq_save(flags);
554 for_each_zone(zone) {
555 struct per_cpu_pageset *pset;
556
557 /* Do not drain local pagesets */
558 if (zone->zone_pgdat->node_id == numa_node_id())
559 continue;
560
561 pset = zone->pageset[smp_processor_id()];
562 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
563 struct per_cpu_pages *pcp;
564
565 pcp = &pset->pcp[i];
566 if (pcp->count)
567 pcp->count -= free_pages_bulk(zone, pcp->count,
568 &pcp->list, 0);
569 }
570 }
571 local_irq_restore(flags);
572}
573#endif
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575#if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
576static void __drain_pages(unsigned int cpu)
577{
578 struct zone *zone;
579 int i;
580
581 for_each_zone(zone) {
582 struct per_cpu_pageset *pset;
583
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700584 pset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) {
586 struct per_cpu_pages *pcp;
587
588 pcp = &pset->pcp[i];
589 pcp->count -= free_pages_bulk(zone, pcp->count,
590 &pcp->list, 0);
591 }
592 }
593}
594#endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
595
596#ifdef CONFIG_PM
597
598void mark_free_pages(struct zone *zone)
599{
600 unsigned long zone_pfn, flags;
601 int order;
602 struct list_head *curr;
603
604 if (!zone->spanned_pages)
605 return;
606
607 spin_lock_irqsave(&zone->lock, flags);
608 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
609 ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn));
610
611 for (order = MAX_ORDER - 1; order >= 0; --order)
612 list_for_each(curr, &zone->free_area[order].free_list) {
613 unsigned long start_pfn, i;
614
615 start_pfn = page_to_pfn(list_entry(curr, struct page, lru));
616
617 for (i=0; i < (1<<order); i++)
618 SetPageNosaveFree(pfn_to_page(start_pfn+i));
619 }
620 spin_unlock_irqrestore(&zone->lock, flags);
621}
622
623/*
624 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
625 */
626void drain_local_pages(void)
627{
628 unsigned long flags;
629
630 local_irq_save(flags);
631 __drain_pages(smp_processor_id());
632 local_irq_restore(flags);
633}
634#endif /* CONFIG_PM */
635
636static void zone_statistics(struct zonelist *zonelist, struct zone *z)
637{
638#ifdef CONFIG_NUMA
639 unsigned long flags;
640 int cpu;
641 pg_data_t *pg = z->zone_pgdat;
642 pg_data_t *orig = zonelist->zones[0]->zone_pgdat;
643 struct per_cpu_pageset *p;
644
645 local_irq_save(flags);
646 cpu = smp_processor_id();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700647 p = zone_pcp(z,cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 if (pg == orig) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700649 p->numa_hit++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 } else {
651 p->numa_miss++;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700652 zone_pcp(zonelist->zones[0], cpu)->numa_foreign++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 }
654 if (pg == NODE_DATA(numa_node_id()))
655 p->local_node++;
656 else
657 p->other_node++;
658 local_irq_restore(flags);
659#endif
660}
661
662/*
663 * Free a 0-order page
664 */
665static void FASTCALL(free_hot_cold_page(struct page *page, int cold));
666static void fastcall free_hot_cold_page(struct page *page, int cold)
667{
668 struct zone *zone = page_zone(page);
669 struct per_cpu_pages *pcp;
670 unsigned long flags;
671
672 arch_free_page(page, 0);
673
674 kernel_map_pages(page, 1, 0);
675 inc_page_state(pgfree);
676 if (PageAnon(page))
677 page->mapping = NULL;
678 free_pages_check(__FUNCTION__, page);
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700679 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 local_irq_save(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 list_add(&page->lru, &pcp->list);
682 pcp->count++;
Christoph Lameter2caaad42005-06-21 17:15:00 -0700683 if (pcp->count >= pcp->high)
684 pcp->count -= free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 local_irq_restore(flags);
686 put_cpu();
687}
688
689void fastcall free_hot_page(struct page *page)
690{
691 free_hot_cold_page(page, 0);
692}
693
694void fastcall free_cold_page(struct page *page)
695{
696 free_hot_cold_page(page, 1);
697}
698
Al Virodd0fc662005-10-07 07:46:04 +0100699static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
701 int i;
702
703 BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM);
704 for(i = 0; i < (1 << order); i++)
705 clear_highpage(page + i);
706}
707
708/*
709 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
710 * we cheat by calling it from here, in the order > 0 path. Saves a branch
711 * or two.
712 */
713static struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100714buffered_rmqueue(struct zone *zone, int order, gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715{
716 unsigned long flags;
717 struct page *page = NULL;
718 int cold = !!(gfp_flags & __GFP_COLD);
719
720 if (order == 0) {
721 struct per_cpu_pages *pcp;
722
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700723 pcp = &zone_pcp(zone, get_cpu())->pcp[cold];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 local_irq_save(flags);
725 if (pcp->count <= pcp->low)
726 pcp->count += rmqueue_bulk(zone, 0,
727 pcp->batch, &pcp->list);
728 if (pcp->count) {
729 page = list_entry(pcp->list.next, struct page, lru);
730 list_del(&page->lru);
731 pcp->count--;
732 }
733 local_irq_restore(flags);
734 put_cpu();
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800735 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 spin_lock_irqsave(&zone->lock, flags);
737 page = __rmqueue(zone, order);
738 spin_unlock_irqrestore(&zone->lock, flags);
739 }
740
741 if (page != NULL) {
742 BUG_ON(bad_range(zone, page));
743 mod_page_state_zone(zone, pgalloc, 1 << order);
744 prep_new_page(page, order);
745
746 if (gfp_flags & __GFP_ZERO)
747 prep_zero_page(page, order, gfp_flags);
748
749 if (order && (gfp_flags & __GFP_COMP))
750 prep_compound_page(page, order);
751 }
752 return page;
753}
754
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800755#define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
756#define ALLOC_HARDER 0x02 /* try to alloc harder */
757#define ALLOC_HIGH 0x04 /* __GFP_HIGH set */
758#define ALLOC_CPUSET 0x08 /* check for correct cpuset */
759
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760/*
761 * Return 1 if free pages are above 'mark'. This takes into account the order
762 * of the allocation.
763 */
764int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800765 int classzone_idx, int alloc_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766{
767 /* free_pages my go negative - that's OK */
768 long min = mark, free_pages = z->free_pages - (1 << order) + 1;
769 int o;
770
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800771 if (alloc_flags & ALLOC_HIGH)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 min -= min / 2;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800773 if (alloc_flags & ALLOC_HARDER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 min -= min / 4;
775
776 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
777 return 0;
778 for (o = 0; o < order; o++) {
779 /* At the next order, this order's pages become unavailable */
780 free_pages -= z->free_area[o].nr_free << o;
781
782 /* Require fewer higher order pages to be free */
783 min >>= 1;
784
785 if (free_pages <= min)
786 return 0;
787 }
788 return 1;
789}
790
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800791/*
792 * get_page_from_freeliest goes through the zonelist trying to allocate
793 * a page.
794 */
795static struct page *
796get_page_from_freelist(gfp_t gfp_mask, unsigned int order,
797 struct zonelist *zonelist, int alloc_flags)
Martin Hicks753ee722005-06-21 17:14:41 -0700798{
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800799 struct zone **z = zonelist->zones;
800 struct page *page = NULL;
801 int classzone_idx = zone_idx(*z);
802
803 /*
804 * Go through the zonelist once, looking for a zone with enough free.
805 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
806 */
807 do {
808 if ((alloc_flags & ALLOC_CPUSET) &&
809 !cpuset_zone_allowed(*z, gfp_mask))
810 continue;
811
812 if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
813 if (!zone_watermark_ok(*z, order, (*z)->pages_low,
814 classzone_idx, alloc_flags))
815 continue;
816 }
817
818 page = buffered_rmqueue(*z, order, gfp_mask);
819 if (page) {
820 zone_statistics(zonelist, *z);
821 break;
822 }
823 } while (*(++z) != NULL);
824 return page;
Martin Hicks753ee722005-06-21 17:14:41 -0700825}
826
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827/*
828 * This is the 'heart' of the zoned buddy allocator.
829 */
830struct page * fastcall
Al Virodd0fc662005-10-07 07:46:04 +0100831__alloc_pages(gfp_t gfp_mask, unsigned int order,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 struct zonelist *zonelist)
833{
Al Viro260b2362005-10-21 03:22:44 -0400834 const gfp_t wait = gfp_mask & __GFP_WAIT;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800835 struct zone **z;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 struct page *page;
837 struct reclaim_state reclaim_state;
838 struct task_struct *p = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 int do_retry;
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800840 int alloc_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 int did_some_progress;
842
843 might_sleep_if(wait);
844
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800845 z = zonelist->zones; /* the list of zones suitable for gfp_mask */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800847 if (unlikely(*z == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 /* Should this ever happen?? */
849 return NULL;
850 }
Martin Hicks753ee722005-06-21 17:14:41 -0700851restart:
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800852 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
853 zonelist, ALLOC_CPUSET);
854 if (page)
855 goto got_pg;
856
857 do
858 wakeup_kswapd(*z, order);
859 while (*(++z));
860
Paul Jackson9bf22292005-09-06 15:18:12 -0700861 /*
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800862 * OK, we're below the kswapd watermark and have kicked background
863 * reclaim. Now things get more complex, so set up alloc_flags according
864 * to how we want to proceed.
865 *
866 * The caller may dip into page reserves a bit more if the caller
867 * cannot run direct reclaim, or if the caller has realtime scheduling
868 * policy.
Paul Jackson9bf22292005-09-06 15:18:12 -0700869 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800870 alloc_flags = 0;
871 if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
872 alloc_flags |= ALLOC_HARDER;
873 if (gfp_mask & __GFP_HIGH)
874 alloc_flags |= ALLOC_HIGH;
875 if (wait)
876 alloc_flags |= ALLOC_CPUSET;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878 /*
879 * Go through the zonelist again. Let __GFP_HIGH and allocations
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800880 * coming from realtime tasks go deeper into reserves.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 *
882 * This is the last chance, in general, before the goto nopage.
883 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
Paul Jackson9bf22292005-09-06 15:18:12 -0700884 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800886 page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags);
887 if (page)
888 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 /* This allocation should allow future memory freeing. */
Nick Pigginb84a35b2005-05-01 08:58:36 -0700891
892 if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
893 && !in_interrupt()) {
894 if (!(gfp_mask & __GFP_NOMEMALLOC)) {
Kirill Korotaev885036d2005-11-13 16:06:41 -0800895nofail_alloc:
Nick Pigginb84a35b2005-05-01 08:58:36 -0700896 /* go through the zonelist yet again, ignoring mins */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800897 page = get_page_from_freelist(gfp_mask, order,
898 zonelist, ALLOC_NO_WATERMARKS|ALLOC_CPUSET);
899 if (page)
900 goto got_pg;
Kirill Korotaev885036d2005-11-13 16:06:41 -0800901 if (gfp_mask & __GFP_NOFAIL) {
902 blk_congestion_wait(WRITE, HZ/50);
903 goto nofail_alloc;
904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 }
906 goto nopage;
907 }
908
909 /* Atomic allocations - we can't balance anything */
910 if (!wait)
911 goto nopage;
912
913rebalance:
914 cond_resched();
915
916 /* We now go into synchronous reclaim */
917 p->flags |= PF_MEMALLOC;
918 reclaim_state.reclaimed_slab = 0;
919 p->reclaim_state = &reclaim_state;
920
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800921 did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 p->reclaim_state = NULL;
924 p->flags &= ~PF_MEMALLOC;
925
926 cond_resched();
927
928 if (likely(did_some_progress)) {
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800929 page = get_page_from_freelist(gfp_mask, order,
930 zonelist, alloc_flags);
931 if (page)
932 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
934 /*
935 * Go through the zonelist yet one more time, keep
936 * very high watermark here, this is only to catch
937 * a parallel oom killing, we must fail if we're still
938 * under heavy pressure.
939 */
Rohit Seth7fb1d9f2005-11-13 16:06:43 -0800940 page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order,
941 zonelist, ALLOC_CPUSET);
942 if (page)
943 goto got_pg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Marcelo Tosatti79b9ce32005-07-07 17:56:04 -0700945 out_of_memory(gfp_mask, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 goto restart;
947 }
948
949 /*
950 * Don't let big-order allocations loop unless the caller explicitly
951 * requests that. Wait for some write requests to complete then retry.
952 *
953 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
954 * <= 3, but that may not be true in other implementations.
955 */
956 do_retry = 0;
957 if (!(gfp_mask & __GFP_NORETRY)) {
958 if ((order <= 3) || (gfp_mask & __GFP_REPEAT))
959 do_retry = 1;
960 if (gfp_mask & __GFP_NOFAIL)
961 do_retry = 1;
962 }
963 if (do_retry) {
964 blk_congestion_wait(WRITE, HZ/50);
965 goto rebalance;
966 }
967
968nopage:
969 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
970 printk(KERN_WARNING "%s: page allocation failure."
971 " order:%d, mode:0x%x\n",
972 p->comm, order, gfp_mask);
973 dump_stack();
Janet Morgan578c2fd2005-06-21 17:14:56 -0700974 show_mem();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976got_pg:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 return page;
978}
979
980EXPORT_SYMBOL(__alloc_pages);
981
982/*
983 * Common helper functions.
984 */
Al Virodd0fc662005-10-07 07:46:04 +0100985fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986{
987 struct page * page;
988 page = alloc_pages(gfp_mask, order);
989 if (!page)
990 return 0;
991 return (unsigned long) page_address(page);
992}
993
994EXPORT_SYMBOL(__get_free_pages);
995
Al Virodd0fc662005-10-07 07:46:04 +0100996fastcall unsigned long get_zeroed_page(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997{
998 struct page * page;
999
1000 /*
1001 * get_zeroed_page() returns a 32-bit address, which cannot represent
1002 * a highmem page
1003 */
Al Viro260b2362005-10-21 03:22:44 -04001004 BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006 page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1007 if (page)
1008 return (unsigned long) page_address(page);
1009 return 0;
1010}
1011
1012EXPORT_SYMBOL(get_zeroed_page);
1013
1014void __pagevec_free(struct pagevec *pvec)
1015{
1016 int i = pagevec_count(pvec);
1017
1018 while (--i >= 0)
1019 free_hot_cold_page(pvec->pages[i], pvec->cold);
1020}
1021
1022fastcall void __free_pages(struct page *page, unsigned int order)
1023{
Nick Pigginb5810032005-10-29 18:16:12 -07001024 if (put_page_testzero(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (order == 0)
1026 free_hot_page(page);
1027 else
1028 __free_pages_ok(page, order);
1029 }
1030}
1031
1032EXPORT_SYMBOL(__free_pages);
1033
1034fastcall void free_pages(unsigned long addr, unsigned int order)
1035{
1036 if (addr != 0) {
1037 BUG_ON(!virt_addr_valid((void *)addr));
1038 __free_pages(virt_to_page((void *)addr), order);
1039 }
1040}
1041
1042EXPORT_SYMBOL(free_pages);
1043
1044/*
1045 * Total amount of free (allocatable) RAM:
1046 */
1047unsigned int nr_free_pages(void)
1048{
1049 unsigned int sum = 0;
1050 struct zone *zone;
1051
1052 for_each_zone(zone)
1053 sum += zone->free_pages;
1054
1055 return sum;
1056}
1057
1058EXPORT_SYMBOL(nr_free_pages);
1059
1060#ifdef CONFIG_NUMA
1061unsigned int nr_free_pages_pgdat(pg_data_t *pgdat)
1062{
1063 unsigned int i, sum = 0;
1064
1065 for (i = 0; i < MAX_NR_ZONES; i++)
1066 sum += pgdat->node_zones[i].free_pages;
1067
1068 return sum;
1069}
1070#endif
1071
1072static unsigned int nr_free_zone_pages(int offset)
1073{
Martin J. Blighe310fd42005-07-29 22:59:18 -07001074 /* Just pick one node, since fallback list is circular */
1075 pg_data_t *pgdat = NODE_DATA(numa_node_id());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 unsigned int sum = 0;
1077
Martin J. Blighe310fd42005-07-29 22:59:18 -07001078 struct zonelist *zonelist = pgdat->node_zonelists + offset;
1079 struct zone **zonep = zonelist->zones;
1080 struct zone *zone;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Martin J. Blighe310fd42005-07-29 22:59:18 -07001082 for (zone = *zonep++; zone; zone = *zonep++) {
1083 unsigned long size = zone->present_pages;
1084 unsigned long high = zone->pages_high;
1085 if (size > high)
1086 sum += size - high;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
1088
1089 return sum;
1090}
1091
1092/*
1093 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1094 */
1095unsigned int nr_free_buffer_pages(void)
1096{
Al Viroaf4ca452005-10-21 02:55:38 -04001097 return nr_free_zone_pages(gfp_zone(GFP_USER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
1100/*
1101 * Amount of free RAM allocatable within all zones
1102 */
1103unsigned int nr_free_pagecache_pages(void)
1104{
Al Viroaf4ca452005-10-21 02:55:38 -04001105 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106}
1107
1108#ifdef CONFIG_HIGHMEM
1109unsigned int nr_free_highpages (void)
1110{
1111 pg_data_t *pgdat;
1112 unsigned int pages = 0;
1113
1114 for_each_pgdat(pgdat)
1115 pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1116
1117 return pages;
1118}
1119#endif
1120
1121#ifdef CONFIG_NUMA
1122static void show_node(struct zone *zone)
1123{
1124 printk("Node %d ", zone->zone_pgdat->node_id);
1125}
1126#else
1127#define show_node(zone) do { } while (0)
1128#endif
1129
1130/*
1131 * Accumulate the page_state information across all CPUs.
1132 * The result is unavoidably approximate - it can change
1133 * during and after execution of this function.
1134 */
1135static DEFINE_PER_CPU(struct page_state, page_states) = {0};
1136
1137atomic_t nr_pagecache = ATOMIC_INIT(0);
1138EXPORT_SYMBOL(nr_pagecache);
1139#ifdef CONFIG_SMP
1140DEFINE_PER_CPU(long, nr_pagecache_local) = 0;
1141#endif
1142
Martin Hicksc07e02d2005-09-03 15:55:11 -07001143void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
1145 int cpu = 0;
1146
1147 memset(ret, 0, sizeof(*ret));
Martin Hicksc07e02d2005-09-03 15:55:11 -07001148 cpus_and(*cpumask, *cpumask, cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Martin Hicksc07e02d2005-09-03 15:55:11 -07001150 cpu = first_cpu(*cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 while (cpu < NR_CPUS) {
1152 unsigned long *in, *out, off;
1153
1154 in = (unsigned long *)&per_cpu(page_states, cpu);
1155
Martin Hicksc07e02d2005-09-03 15:55:11 -07001156 cpu = next_cpu(cpu, *cpumask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158 if (cpu < NR_CPUS)
1159 prefetch(&per_cpu(page_states, cpu));
1160
1161 out = (unsigned long *)ret;
1162 for (off = 0; off < nr; off++)
1163 *out++ += *in++;
1164 }
1165}
1166
Martin Hicksc07e02d2005-09-03 15:55:11 -07001167void get_page_state_node(struct page_state *ret, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168{
1169 int nr;
Martin Hicksc07e02d2005-09-03 15:55:11 -07001170 cpumask_t mask = node_to_cpumask(node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1173 nr /= sizeof(unsigned long);
1174
Martin Hicksc07e02d2005-09-03 15:55:11 -07001175 __get_page_state(ret, nr+1, &mask);
1176}
1177
1178void get_page_state(struct page_state *ret)
1179{
1180 int nr;
1181 cpumask_t mask = CPU_MASK_ALL;
1182
1183 nr = offsetof(struct page_state, GET_PAGE_STATE_LAST);
1184 nr /= sizeof(unsigned long);
1185
1186 __get_page_state(ret, nr + 1, &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187}
1188
1189void get_full_page_state(struct page_state *ret)
1190{
Martin Hicksc07e02d2005-09-03 15:55:11 -07001191 cpumask_t mask = CPU_MASK_ALL;
1192
1193 __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194}
1195
Benjamin LaHaisec2f29ea2005-06-21 17:14:55 -07001196unsigned long __read_page_state(unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197{
1198 unsigned long ret = 0;
1199 int cpu;
1200
1201 for_each_online_cpu(cpu) {
1202 unsigned long in;
1203
1204 in = (unsigned long)&per_cpu(page_states, cpu) + offset;
1205 ret += *((unsigned long *)in);
1206 }
1207 return ret;
1208}
1209
Benjamin LaHaise83e5d8f2005-06-21 17:14:54 -07001210void __mod_page_state(unsigned long offset, unsigned long delta)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 unsigned long flags;
1213 void* ptr;
1214
1215 local_irq_save(flags);
1216 ptr = &__get_cpu_var(page_states);
1217 *(unsigned long*)(ptr + offset) += delta;
1218 local_irq_restore(flags);
1219}
1220
1221EXPORT_SYMBOL(__mod_page_state);
1222
1223void __get_zone_counts(unsigned long *active, unsigned long *inactive,
1224 unsigned long *free, struct pglist_data *pgdat)
1225{
1226 struct zone *zones = pgdat->node_zones;
1227 int i;
1228
1229 *active = 0;
1230 *inactive = 0;
1231 *free = 0;
1232 for (i = 0; i < MAX_NR_ZONES; i++) {
1233 *active += zones[i].nr_active;
1234 *inactive += zones[i].nr_inactive;
1235 *free += zones[i].free_pages;
1236 }
1237}
1238
1239void get_zone_counts(unsigned long *active,
1240 unsigned long *inactive, unsigned long *free)
1241{
1242 struct pglist_data *pgdat;
1243
1244 *active = 0;
1245 *inactive = 0;
1246 *free = 0;
1247 for_each_pgdat(pgdat) {
1248 unsigned long l, m, n;
1249 __get_zone_counts(&l, &m, &n, pgdat);
1250 *active += l;
1251 *inactive += m;
1252 *free += n;
1253 }
1254}
1255
1256void si_meminfo(struct sysinfo *val)
1257{
1258 val->totalram = totalram_pages;
1259 val->sharedram = 0;
1260 val->freeram = nr_free_pages();
1261 val->bufferram = nr_blockdev_pages();
1262#ifdef CONFIG_HIGHMEM
1263 val->totalhigh = totalhigh_pages;
1264 val->freehigh = nr_free_highpages();
1265#else
1266 val->totalhigh = 0;
1267 val->freehigh = 0;
1268#endif
1269 val->mem_unit = PAGE_SIZE;
1270}
1271
1272EXPORT_SYMBOL(si_meminfo);
1273
1274#ifdef CONFIG_NUMA
1275void si_meminfo_node(struct sysinfo *val, int nid)
1276{
1277 pg_data_t *pgdat = NODE_DATA(nid);
1278
1279 val->totalram = pgdat->node_present_pages;
1280 val->freeram = nr_free_pages_pgdat(pgdat);
1281 val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1282 val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages;
1283 val->mem_unit = PAGE_SIZE;
1284}
1285#endif
1286
1287#define K(x) ((x) << (PAGE_SHIFT-10))
1288
1289/*
1290 * Show free area list (used inside shift_scroll-lock stuff)
1291 * We also calculate the percentage fragmentation. We do this by counting the
1292 * memory on each free list with the exception of the first item on the list.
1293 */
1294void show_free_areas(void)
1295{
1296 struct page_state ps;
1297 int cpu, temperature;
1298 unsigned long active;
1299 unsigned long inactive;
1300 unsigned long free;
1301 struct zone *zone;
1302
1303 for_each_zone(zone) {
1304 show_node(zone);
1305 printk("%s per-cpu:", zone->name);
1306
1307 if (!zone->present_pages) {
1308 printk(" empty\n");
1309 continue;
1310 } else
1311 printk("\n");
1312
Dave Jones6b482c62005-11-10 15:45:56 -05001313 for_each_online_cpu(cpu) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 struct per_cpu_pageset *pageset;
1315
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001316 pageset = zone_pcp(zone, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 for (temperature = 0; temperature < 2; temperature++)
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001319 printk("cpu %d %s: low %d, high %d, batch %d used:%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 cpu,
1321 temperature ? "cold" : "hot",
1322 pageset->pcp[temperature].low,
1323 pageset->pcp[temperature].high,
Christoph Lameter4ae7c032005-06-21 17:14:57 -07001324 pageset->pcp[temperature].batch,
1325 pageset->pcp[temperature].count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
1327 }
1328
1329 get_page_state(&ps);
1330 get_zone_counts(&active, &inactive, &free);
1331
Denis Vlasenkoc0d62212005-06-21 17:15:14 -07001332 printk("Free pages: %11ukB (%ukB HighMem)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 K(nr_free_pages()),
1334 K(nr_free_highpages()));
1335
1336 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1337 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1338 active,
1339 inactive,
1340 ps.nr_dirty,
1341 ps.nr_writeback,
1342 ps.nr_unstable,
1343 nr_free_pages(),
1344 ps.nr_slab,
1345 ps.nr_mapped,
1346 ps.nr_page_table_pages);
1347
1348 for_each_zone(zone) {
1349 int i;
1350
1351 show_node(zone);
1352 printk("%s"
1353 " free:%lukB"
1354 " min:%lukB"
1355 " low:%lukB"
1356 " high:%lukB"
1357 " active:%lukB"
1358 " inactive:%lukB"
1359 " present:%lukB"
1360 " pages_scanned:%lu"
1361 " all_unreclaimable? %s"
1362 "\n",
1363 zone->name,
1364 K(zone->free_pages),
1365 K(zone->pages_min),
1366 K(zone->pages_low),
1367 K(zone->pages_high),
1368 K(zone->nr_active),
1369 K(zone->nr_inactive),
1370 K(zone->present_pages),
1371 zone->pages_scanned,
1372 (zone->all_unreclaimable ? "yes" : "no")
1373 );
1374 printk("lowmem_reserve[]:");
1375 for (i = 0; i < MAX_NR_ZONES; i++)
1376 printk(" %lu", zone->lowmem_reserve[i]);
1377 printk("\n");
1378 }
1379
1380 for_each_zone(zone) {
1381 unsigned long nr, flags, order, total = 0;
1382
1383 show_node(zone);
1384 printk("%s: ", zone->name);
1385 if (!zone->present_pages) {
1386 printk("empty\n");
1387 continue;
1388 }
1389
1390 spin_lock_irqsave(&zone->lock, flags);
1391 for (order = 0; order < MAX_ORDER; order++) {
1392 nr = zone->free_area[order].nr_free;
1393 total += nr << order;
1394 printk("%lu*%lukB ", nr, K(1UL) << order);
1395 }
1396 spin_unlock_irqrestore(&zone->lock, flags);
1397 printk("= %lukB\n", K(total));
1398 }
1399
1400 show_swap_cache_info();
1401}
1402
1403/*
1404 * Builds allocation fallback zone lists.
1405 */
1406static int __init build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, int j, int k)
1407{
1408 switch (k) {
1409 struct zone *zone;
1410 default:
1411 BUG();
1412 case ZONE_HIGHMEM:
1413 zone = pgdat->node_zones + ZONE_HIGHMEM;
1414 if (zone->present_pages) {
1415#ifndef CONFIG_HIGHMEM
1416 BUG();
1417#endif
1418 zonelist->zones[j++] = zone;
1419 }
1420 case ZONE_NORMAL:
1421 zone = pgdat->node_zones + ZONE_NORMAL;
1422 if (zone->present_pages)
1423 zonelist->zones[j++] = zone;
1424 case ZONE_DMA:
1425 zone = pgdat->node_zones + ZONE_DMA;
1426 if (zone->present_pages)
1427 zonelist->zones[j++] = zone;
1428 }
1429
1430 return j;
1431}
1432
Al Viro260b2362005-10-21 03:22:44 -04001433static inline int highest_zone(int zone_bits)
1434{
1435 int res = ZONE_NORMAL;
1436 if (zone_bits & (__force int)__GFP_HIGHMEM)
1437 res = ZONE_HIGHMEM;
1438 if (zone_bits & (__force int)__GFP_DMA)
1439 res = ZONE_DMA;
1440 return res;
1441}
1442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443#ifdef CONFIG_NUMA
1444#define MAX_NODE_LOAD (num_online_nodes())
1445static int __initdata node_load[MAX_NUMNODES];
1446/**
Pavel Pisa4dc3b162005-05-01 08:59:25 -07001447 * 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 -07001448 * @node: node whose fallback list we're appending
1449 * @used_node_mask: nodemask_t of already used nodes
1450 *
1451 * We use a number of factors to determine which is the next node that should
1452 * appear on a given node's fallback list. The node should not have appeared
1453 * already in @node's fallback list, and it should be the next closest node
1454 * according to the distance array (which contains arbitrary distance values
1455 * from each node to each node in the system), and should also prefer nodes
1456 * with no CPUs, since presumably they'll have very little allocation pressure
1457 * on them otherwise.
1458 * It returns -1 if no node is found.
1459 */
1460static int __init find_next_best_node(int node, nodemask_t *used_node_mask)
1461{
1462 int i, n, val;
1463 int min_val = INT_MAX;
1464 int best_node = -1;
1465
1466 for_each_online_node(i) {
1467 cpumask_t tmp;
1468
1469 /* Start from local node */
1470 n = (node+i) % num_online_nodes();
1471
1472 /* Don't want a node to appear more than once */
1473 if (node_isset(n, *used_node_mask))
1474 continue;
1475
1476 /* Use the local node if we haven't already */
1477 if (!node_isset(node, *used_node_mask)) {
1478 best_node = node;
1479 break;
1480 }
1481
1482 /* Use the distance array to find the distance */
1483 val = node_distance(node, n);
1484
1485 /* Give preference to headless and unused nodes */
1486 tmp = node_to_cpumask(n);
1487 if (!cpus_empty(tmp))
1488 val += PENALTY_FOR_NODE_WITH_CPUS;
1489
1490 /* Slight preference for less loaded node */
1491 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
1492 val += node_load[n];
1493
1494 if (val < min_val) {
1495 min_val = val;
1496 best_node = n;
1497 }
1498 }
1499
1500 if (best_node >= 0)
1501 node_set(best_node, *used_node_mask);
1502
1503 return best_node;
1504}
1505
1506static void __init build_zonelists(pg_data_t *pgdat)
1507{
1508 int i, j, k, node, local_node;
1509 int prev_node, load;
1510 struct zonelist *zonelist;
1511 nodemask_t used_mask;
1512
1513 /* initialize zonelists */
1514 for (i = 0; i < GFP_ZONETYPES; i++) {
1515 zonelist = pgdat->node_zonelists + i;
1516 zonelist->zones[0] = NULL;
1517 }
1518
1519 /* NUMA-aware ordering of nodes */
1520 local_node = pgdat->node_id;
1521 load = num_online_nodes();
1522 prev_node = local_node;
1523 nodes_clear(used_mask);
1524 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
1525 /*
1526 * We don't want to pressure a particular node.
1527 * So adding penalty to the first node in same
1528 * distance group to make it round-robin.
1529 */
1530 if (node_distance(local_node, node) !=
1531 node_distance(local_node, prev_node))
1532 node_load[node] += load;
1533 prev_node = node;
1534 load--;
1535 for (i = 0; i < GFP_ZONETYPES; i++) {
1536 zonelist = pgdat->node_zonelists + i;
1537 for (j = 0; zonelist->zones[j] != NULL; j++);
1538
Al Viro260b2362005-10-21 03:22:44 -04001539 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
1541 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1542 zonelist->zones[j] = NULL;
1543 }
1544 }
1545}
1546
1547#else /* CONFIG_NUMA */
1548
1549static void __init build_zonelists(pg_data_t *pgdat)
1550{
1551 int i, j, k, node, local_node;
1552
1553 local_node = pgdat->node_id;
1554 for (i = 0; i < GFP_ZONETYPES; i++) {
1555 struct zonelist *zonelist;
1556
1557 zonelist = pgdat->node_zonelists + i;
1558
1559 j = 0;
Al Viro260b2362005-10-21 03:22:44 -04001560 k = highest_zone(i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 j = build_zonelists_node(pgdat, zonelist, j, k);
1562 /*
1563 * Now we build the zonelist so that it contains the zones
1564 * of all the other nodes.
1565 * We don't want to pressure a particular node, so when
1566 * building the zones for node N, we make sure that the
1567 * zones coming right after the local ones are those from
1568 * node N+1 (modulo N)
1569 */
1570 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
1571 if (!node_online(node))
1572 continue;
1573 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1574 }
1575 for (node = 0; node < local_node; node++) {
1576 if (!node_online(node))
1577 continue;
1578 j = build_zonelists_node(NODE_DATA(node), zonelist, j, k);
1579 }
1580
1581 zonelist->zones[j] = NULL;
1582 }
1583}
1584
1585#endif /* CONFIG_NUMA */
1586
1587void __init build_all_zonelists(void)
1588{
1589 int i;
1590
1591 for_each_online_node(i)
1592 build_zonelists(NODE_DATA(i));
1593 printk("Built %i zonelists\n", num_online_nodes());
1594 cpuset_init_current_mems_allowed();
1595}
1596
1597/*
1598 * Helper functions to size the waitqueue hash table.
1599 * Essentially these want to choose hash table sizes sufficiently
1600 * large so that collisions trying to wait on pages are rare.
1601 * But in fact, the number of active page waitqueues on typical
1602 * systems is ridiculously low, less than 200. So this is even
1603 * conservative, even though it seems large.
1604 *
1605 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1606 * waitqueues, i.e. the size of the waitq table given the number of pages.
1607 */
1608#define PAGES_PER_WAITQUEUE 256
1609
1610static inline unsigned long wait_table_size(unsigned long pages)
1611{
1612 unsigned long size = 1;
1613
1614 pages /= PAGES_PER_WAITQUEUE;
1615
1616 while (size < pages)
1617 size <<= 1;
1618
1619 /*
1620 * Once we have dozens or even hundreds of threads sleeping
1621 * on IO we've got bigger problems than wait queue collision.
1622 * Limit the size of the wait table to a reasonable size.
1623 */
1624 size = min(size, 4096UL);
1625
1626 return max(size, 4UL);
1627}
1628
1629/*
1630 * This is an integer logarithm so that shifts can be used later
1631 * to extract the more random high bits from the multiplicative
1632 * hash function before the remainder is taken.
1633 */
1634static inline unsigned long wait_table_bits(unsigned long size)
1635{
1636 return ffz(~size);
1637}
1638
1639#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1640
1641static void __init calculate_zone_totalpages(struct pglist_data *pgdat,
1642 unsigned long *zones_size, unsigned long *zholes_size)
1643{
1644 unsigned long realtotalpages, totalpages = 0;
1645 int i;
1646
1647 for (i = 0; i < MAX_NR_ZONES; i++)
1648 totalpages += zones_size[i];
1649 pgdat->node_spanned_pages = totalpages;
1650
1651 realtotalpages = totalpages;
1652 if (zholes_size)
1653 for (i = 0; i < MAX_NR_ZONES; i++)
1654 realtotalpages -= zholes_size[i];
1655 pgdat->node_present_pages = realtotalpages;
1656 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1657}
1658
1659
1660/*
1661 * Initially all pages are reserved - free ones are freed
1662 * up by free_all_bootmem() once the early boot process is
1663 * done. Non-atomic initialization, single-pass.
1664 */
Dave Hansen3947be12005-10-29 18:16:54 -07001665void __devinit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 unsigned long start_pfn)
1667{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 struct page *page;
Andy Whitcroft29751f62005-06-23 00:08:00 -07001669 unsigned long end_pfn = start_pfn + size;
1670 unsigned long pfn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001672 for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1673 if (!early_pfn_valid(pfn))
1674 continue;
Andy Whitcroft641c7672005-06-23 00:07:59 -07001675 if (!early_pfn_in_nid(pfn, nid))
1676 continue;
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001677 page = pfn_to_page(pfn);
1678 set_page_links(page, zone, nid, pfn);
Nick Pigginb5810032005-10-29 18:16:12 -07001679 set_page_count(page, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 reset_page_mapcount(page);
1681 SetPageReserved(page);
1682 INIT_LIST_HEAD(&page->lru);
1683#ifdef WANT_PAGE_VIRTUAL
1684 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1685 if (!is_highmem_idx(zone))
Bob Picco3212c6b2005-06-27 14:36:28 -07001686 set_page_address(page, __va(pfn << PAGE_SHIFT));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 }
1689}
1690
1691void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone,
1692 unsigned long size)
1693{
1694 int order;
1695 for (order = 0; order < MAX_ORDER ; order++) {
1696 INIT_LIST_HEAD(&zone->free_area[order].free_list);
1697 zone->free_area[order].nr_free = 0;
1698 }
1699}
1700
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001701#define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1702void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn,
1703 unsigned long size)
1704{
1705 unsigned long snum = pfn_to_section_nr(pfn);
1706 unsigned long end = pfn_to_section_nr(pfn + size);
1707
1708 if (FLAGS_HAS_NODE)
1709 zone_table[ZONETABLE_INDEX(nid, zid)] = zone;
1710 else
1711 for (; snum <= end; snum++)
1712 zone_table[ZONETABLE_INDEX(snum, zid)] = zone;
1713}
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715#ifndef __HAVE_ARCH_MEMMAP_INIT
1716#define memmap_init(size, nid, zone, start_pfn) \
1717 memmap_init_zone((size), (nid), (zone), (start_pfn))
1718#endif
1719
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001720static int __devinit zone_batchsize(struct zone *zone)
1721{
1722 int batch;
1723
1724 /*
1725 * The per-cpu-pages pools are set to around 1000th of the
Seth, Rohitba56e912005-10-29 18:15:47 -07001726 * size of the zone. But no more than 1/2 of a meg.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001727 *
1728 * OK, so we don't know how big the cache is. So guess.
1729 */
1730 batch = zone->present_pages / 1024;
Seth, Rohitba56e912005-10-29 18:15:47 -07001731 if (batch * PAGE_SIZE > 512 * 1024)
1732 batch = (512 * 1024) / PAGE_SIZE;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001733 batch /= 4; /* We effectively *= 4 below */
1734 if (batch < 1)
1735 batch = 1;
1736
1737 /*
Seth, Rohitba56e912005-10-29 18:15:47 -07001738 * We will be trying to allcoate bigger chunks of contiguous
1739 * memory of the order of fls(batch). This should result in
1740 * better cache coloring.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001741 *
Seth, Rohitba56e912005-10-29 18:15:47 -07001742 * A sanity check also to ensure that batch is still in limits.
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001743 */
Seth, Rohitba56e912005-10-29 18:15:47 -07001744 batch = (1 << fls(batch + batch/2));
1745
1746 if (fls(batch) >= (PAGE_SHIFT + MAX_ORDER - 2))
1747 batch = PAGE_SHIFT + ((MAX_ORDER - 1 - PAGE_SHIFT)/2);
1748
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001749 return batch;
1750}
1751
Christoph Lameter2caaad42005-06-21 17:15:00 -07001752inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
1753{
1754 struct per_cpu_pages *pcp;
1755
Magnus Damm1c6fe942005-10-26 01:58:59 -07001756 memset(p, 0, sizeof(*p));
1757
Christoph Lameter2caaad42005-06-21 17:15:00 -07001758 pcp = &p->pcp[0]; /* hot */
1759 pcp->count = 0;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001760 pcp->low = 0;
Christoph Lameter2caaad42005-06-21 17:15:00 -07001761 pcp->high = 6 * batch;
1762 pcp->batch = max(1UL, 1 * batch);
1763 INIT_LIST_HEAD(&pcp->list);
1764
1765 pcp = &p->pcp[1]; /* cold*/
1766 pcp->count = 0;
1767 pcp->low = 0;
1768 pcp->high = 2 * batch;
Seth, Rohite46a5e22005-10-29 18:15:48 -07001769 pcp->batch = max(1UL, batch/2);
Christoph Lameter2caaad42005-06-21 17:15:00 -07001770 INIT_LIST_HEAD(&pcp->list);
1771}
1772
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001773#ifdef CONFIG_NUMA
1774/*
Christoph Lameter2caaad42005-06-21 17:15:00 -07001775 * Boot pageset table. One per cpu which is going to be used for all
1776 * zones and all nodes. The parameters will be set in such a way
1777 * that an item put on a list will immediately be handed over to
1778 * the buddy list. This is safe since pageset manipulation is done
1779 * with interrupts disabled.
1780 *
1781 * Some NUMA counter updates may also be caught by the boot pagesets.
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001782 *
1783 * The boot_pagesets must be kept even after bootup is complete for
1784 * unused processors and/or zones. They do play a role for bootstrapping
1785 * hotplugged processors.
1786 *
1787 * zoneinfo_show() and maybe other functions do
1788 * not check if the processor is online before following the pageset pointer.
1789 * Other parts of the kernel may not check if the zone is available.
Christoph Lameter2caaad42005-06-21 17:15:00 -07001790 */
1791static struct per_cpu_pageset
Christoph Lameterb7c84c62005-06-22 20:26:07 -07001792 boot_pageset[NR_CPUS];
Christoph Lameter2caaad42005-06-21 17:15:00 -07001793
1794/*
1795 * Dynamically allocate memory for the
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001796 * per cpu pageset array in struct zone.
1797 */
1798static int __devinit process_zones(int cpu)
1799{
1800 struct zone *zone, *dzone;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001801
1802 for_each_zone(zone) {
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001803
Christoph Lameter2caaad42005-06-21 17:15:00 -07001804 zone->pageset[cpu] = kmalloc_node(sizeof(struct per_cpu_pageset),
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001805 GFP_KERNEL, cpu_to_node(cpu));
Christoph Lameter2caaad42005-06-21 17:15:00 -07001806 if (!zone->pageset[cpu])
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001807 goto bad;
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001808
Christoph Lameter2caaad42005-06-21 17:15:00 -07001809 setup_pageset(zone->pageset[cpu], zone_batchsize(zone));
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07001810 }
1811
1812 return 0;
1813bad:
1814 for_each_zone(dzone) {
1815 if (dzone == zone)
1816 break;
1817 kfree(dzone->pageset[cpu]);
1818 dzone->pageset[cpu] = NULL;
1819 }
1820 return -ENOMEM;
1821}
1822
1823static inline void free_zone_pagesets(int cpu)
1824{
1825#ifdef CONFIG_NUMA
1826 struct zone *zone;
1827
1828 for_each_zone(zone) {
1829 struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
1830
1831 zone_pcp(zone, cpu) = NULL;
1832 kfree(pset);
1833 }
1834#endif
1835}
1836
1837static int __devinit pageset_cpuup_callback(struct notifier_block *nfb,
1838 unsigned long action,
1839 void *hcpu)
1840{
1841 int cpu = (long)hcpu;
1842 int ret = NOTIFY_OK;
1843
1844 switch (action) {
1845 case CPU_UP_PREPARE:
1846 if (process_zones(cpu))
1847 ret = NOTIFY_BAD;
1848 break;
1849#ifdef CONFIG_HOTPLUG_CPU
1850 case CPU_DEAD:
1851 free_zone_pagesets(cpu);
1852 break;
1853#endif
1854 default:
1855 break;
1856 }
1857 return ret;
1858}
1859
1860static struct notifier_block pageset_notifier =
1861 { &pageset_cpuup_callback, NULL, 0 };
1862
1863void __init setup_per_cpu_pageset()
1864{
1865 int err;
1866
1867 /* Initialize per_cpu_pageset for cpu 0.
1868 * A cpuup callback will do this for every cpu
1869 * as it comes online
1870 */
1871 err = process_zones(smp_processor_id());
1872 BUG_ON(err);
1873 register_cpu_notifier(&pageset_notifier);
1874}
1875
1876#endif
1877
Dave Hansened8ece22005-10-29 18:16:50 -07001878static __devinit
1879void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
1880{
1881 int i;
1882 struct pglist_data *pgdat = zone->zone_pgdat;
1883
1884 /*
1885 * The per-page waitqueue mechanism uses hashed waitqueues
1886 * per zone.
1887 */
1888 zone->wait_table_size = wait_table_size(zone_size_pages);
1889 zone->wait_table_bits = wait_table_bits(zone->wait_table_size);
1890 zone->wait_table = (wait_queue_head_t *)
1891 alloc_bootmem_node(pgdat, zone->wait_table_size
1892 * sizeof(wait_queue_head_t));
1893
1894 for(i = 0; i < zone->wait_table_size; ++i)
1895 init_waitqueue_head(zone->wait_table + i);
1896}
1897
1898static __devinit void zone_pcp_init(struct zone *zone)
1899{
1900 int cpu;
1901 unsigned long batch = zone_batchsize(zone);
1902
1903 for (cpu = 0; cpu < NR_CPUS; cpu++) {
1904#ifdef CONFIG_NUMA
1905 /* Early boot. Slab allocator not functional yet */
1906 zone->pageset[cpu] = &boot_pageset[cpu];
1907 setup_pageset(&boot_pageset[cpu],0);
1908#else
1909 setup_pageset(zone_pcp(zone,cpu), batch);
1910#endif
1911 }
1912 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n",
1913 zone->name, zone->present_pages, batch);
1914}
1915
1916static __devinit void init_currently_empty_zone(struct zone *zone,
1917 unsigned long zone_start_pfn, unsigned long size)
1918{
1919 struct pglist_data *pgdat = zone->zone_pgdat;
1920
1921 zone_wait_table_init(zone, size);
1922 pgdat->nr_zones = zone_idx(zone) + 1;
1923
1924 zone->zone_mem_map = pfn_to_page(zone_start_pfn);
1925 zone->zone_start_pfn = zone_start_pfn;
1926
1927 memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn);
1928
1929 zone_init_free_lists(pgdat, zone, zone->spanned_pages);
1930}
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932/*
1933 * Set up the zone data structures:
1934 * - mark all pages reserved
1935 * - mark all memory queues empty
1936 * - clear the memory bitmaps
1937 */
1938static void __init free_area_init_core(struct pglist_data *pgdat,
1939 unsigned long *zones_size, unsigned long *zholes_size)
1940{
Dave Hansened8ece22005-10-29 18:16:50 -07001941 unsigned long j;
1942 int nid = pgdat->node_id;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 unsigned long zone_start_pfn = pgdat->node_start_pfn;
1944
Dave Hansen208d54e2005-10-29 18:16:52 -07001945 pgdat_resize_init(pgdat);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 pgdat->nr_zones = 0;
1947 init_waitqueue_head(&pgdat->kswapd_wait);
1948 pgdat->kswapd_max_order = 0;
1949
1950 for (j = 0; j < MAX_NR_ZONES; j++) {
1951 struct zone *zone = pgdat->node_zones + j;
1952 unsigned long size, realsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 realsize = size = zones_size[j];
1955 if (zholes_size)
1956 realsize -= zholes_size[j];
1957
1958 if (j == ZONE_DMA || j == ZONE_NORMAL)
1959 nr_kernel_pages += realsize;
1960 nr_all_pages += realsize;
1961
1962 zone->spanned_pages = size;
1963 zone->present_pages = realsize;
1964 zone->name = zone_names[j];
1965 spin_lock_init(&zone->lock);
1966 spin_lock_init(&zone->lru_lock);
Dave Hansenbdc8cb92005-10-29 18:16:53 -07001967 zone_seqlock_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 zone->zone_pgdat = pgdat;
1969 zone->free_pages = 0;
1970
1971 zone->temp_priority = zone->prev_priority = DEF_PRIORITY;
1972
Dave Hansened8ece22005-10-29 18:16:50 -07001973 zone_pcp_init(zone);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 INIT_LIST_HEAD(&zone->active_list);
1975 INIT_LIST_HEAD(&zone->inactive_list);
1976 zone->nr_scan_active = 0;
1977 zone->nr_scan_inactive = 0;
1978 zone->nr_active = 0;
1979 zone->nr_inactive = 0;
Martin Hicks53e9a612005-09-03 15:54:51 -07001980 atomic_set(&zone->reclaim_in_progress, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (!size)
1982 continue;
1983
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001984 zonetable_add(zone, nid, j, zone_start_pfn, size);
Dave Hansened8ece22005-10-29 18:16:50 -07001985 init_currently_empty_zone(zone, zone_start_pfn, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 zone_start_pfn += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 }
1988}
1989
1990static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1991{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 /* Skip empty nodes */
1993 if (!pgdat->node_spanned_pages)
1994 return;
1995
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001996#ifdef CONFIG_FLAT_NODE_MEM_MAP
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 /* ia64 gets its own node_mem_map, before this, without bootmem */
1998 if (!pgdat->node_mem_map) {
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001999 unsigned long size;
2000 struct page *map;
2001
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 size = (pgdat->node_spanned_pages + 1) * sizeof(struct page);
Dave Hansen6f167ec2005-06-23 00:07:39 -07002003 map = alloc_remap(pgdat->node_id, size);
2004 if (!map)
2005 map = alloc_bootmem_node(pgdat, size);
2006 pgdat->node_mem_map = map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002008#ifdef CONFIG_FLATMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 /*
2010 * With no DISCONTIG, the global mem_map is just set as node 0's
2011 */
2012 if (pgdat == NODE_DATA(0))
2013 mem_map = NODE_DATA(0)->node_mem_map;
2014#endif
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002015#endif /* CONFIG_FLAT_NODE_MEM_MAP */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016}
2017
2018void __init free_area_init_node(int nid, struct pglist_data *pgdat,
2019 unsigned long *zones_size, unsigned long node_start_pfn,
2020 unsigned long *zholes_size)
2021{
2022 pgdat->node_id = nid;
2023 pgdat->node_start_pfn = node_start_pfn;
2024 calculate_zone_totalpages(pgdat, zones_size, zholes_size);
2025
2026 alloc_node_mem_map(pgdat);
2027
2028 free_area_init_core(pgdat, zones_size, zholes_size);
2029}
2030
Dave Hansen93b75042005-06-23 00:07:47 -07002031#ifndef CONFIG_NEED_MULTIPLE_NODES
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032static bootmem_data_t contig_bootmem_data;
2033struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data };
2034
2035EXPORT_SYMBOL(contig_page_data);
Dave Hansen93b75042005-06-23 00:07:47 -07002036#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038void __init free_area_init(unsigned long *zones_size)
2039{
Dave Hansen93b75042005-06-23 00:07:47 -07002040 free_area_init_node(0, NODE_DATA(0), zones_size,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
2042}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044#ifdef CONFIG_PROC_FS
2045
2046#include <linux/seq_file.h>
2047
2048static void *frag_start(struct seq_file *m, loff_t *pos)
2049{
2050 pg_data_t *pgdat;
2051 loff_t node = *pos;
2052
2053 for (pgdat = pgdat_list; pgdat && node; pgdat = pgdat->pgdat_next)
2054 --node;
2055
2056 return pgdat;
2057}
2058
2059static void *frag_next(struct seq_file *m, void *arg, loff_t *pos)
2060{
2061 pg_data_t *pgdat = (pg_data_t *)arg;
2062
2063 (*pos)++;
2064 return pgdat->pgdat_next;
2065}
2066
2067static void frag_stop(struct seq_file *m, void *arg)
2068{
2069}
2070
2071/*
2072 * This walks the free areas for each zone.
2073 */
2074static int frag_show(struct seq_file *m, void *arg)
2075{
2076 pg_data_t *pgdat = (pg_data_t *)arg;
2077 struct zone *zone;
2078 struct zone *node_zones = pgdat->node_zones;
2079 unsigned long flags;
2080 int order;
2081
2082 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) {
2083 if (!zone->present_pages)
2084 continue;
2085
2086 spin_lock_irqsave(&zone->lock, flags);
2087 seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name);
2088 for (order = 0; order < MAX_ORDER; ++order)
2089 seq_printf(m, "%6lu ", zone->free_area[order].nr_free);
2090 spin_unlock_irqrestore(&zone->lock, flags);
2091 seq_putc(m, '\n');
2092 }
2093 return 0;
2094}
2095
2096struct seq_operations fragmentation_op = {
2097 .start = frag_start,
2098 .next = frag_next,
2099 .stop = frag_stop,
2100 .show = frag_show,
2101};
2102
Nikita Danilov295ab932005-06-21 17:14:38 -07002103/*
2104 * Output information about zones in @pgdat.
2105 */
2106static int zoneinfo_show(struct seq_file *m, void *arg)
2107{
2108 pg_data_t *pgdat = arg;
2109 struct zone *zone;
2110 struct zone *node_zones = pgdat->node_zones;
2111 unsigned long flags;
2112
2113 for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) {
2114 int i;
2115
2116 if (!zone->present_pages)
2117 continue;
2118
2119 spin_lock_irqsave(&zone->lock, flags);
2120 seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name);
2121 seq_printf(m,
2122 "\n pages free %lu"
2123 "\n min %lu"
2124 "\n low %lu"
2125 "\n high %lu"
2126 "\n active %lu"
2127 "\n inactive %lu"
2128 "\n scanned %lu (a: %lu i: %lu)"
2129 "\n spanned %lu"
2130 "\n present %lu",
2131 zone->free_pages,
2132 zone->pages_min,
2133 zone->pages_low,
2134 zone->pages_high,
2135 zone->nr_active,
2136 zone->nr_inactive,
2137 zone->pages_scanned,
2138 zone->nr_scan_active, zone->nr_scan_inactive,
2139 zone->spanned_pages,
2140 zone->present_pages);
2141 seq_printf(m,
2142 "\n protection: (%lu",
2143 zone->lowmem_reserve[0]);
2144 for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++)
2145 seq_printf(m, ", %lu", zone->lowmem_reserve[i]);
2146 seq_printf(m,
2147 ")"
2148 "\n pagesets");
2149 for (i = 0; i < ARRAY_SIZE(zone->pageset); i++) {
2150 struct per_cpu_pageset *pageset;
2151 int j;
2152
Christoph Lametere7c8d5c2005-06-21 17:14:47 -07002153 pageset = zone_pcp(zone, i);
Nikita Danilov295ab932005-06-21 17:14:38 -07002154 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2155 if (pageset->pcp[j].count)
2156 break;
2157 }
2158 if (j == ARRAY_SIZE(pageset->pcp))
2159 continue;
2160 for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) {
2161 seq_printf(m,
2162 "\n cpu: %i pcp: %i"
2163 "\n count: %i"
2164 "\n low: %i"
2165 "\n high: %i"
2166 "\n batch: %i",
2167 i, j,
2168 pageset->pcp[j].count,
2169 pageset->pcp[j].low,
2170 pageset->pcp[j].high,
2171 pageset->pcp[j].batch);
2172 }
2173#ifdef CONFIG_NUMA
2174 seq_printf(m,
2175 "\n numa_hit: %lu"
2176 "\n numa_miss: %lu"
2177 "\n numa_foreign: %lu"
2178 "\n interleave_hit: %lu"
2179 "\n local_node: %lu"
2180 "\n other_node: %lu",
2181 pageset->numa_hit,
2182 pageset->numa_miss,
2183 pageset->numa_foreign,
2184 pageset->interleave_hit,
2185 pageset->local_node,
2186 pageset->other_node);
2187#endif
2188 }
2189 seq_printf(m,
2190 "\n all_unreclaimable: %u"
2191 "\n prev_priority: %i"
2192 "\n temp_priority: %i"
2193 "\n start_pfn: %lu",
2194 zone->all_unreclaimable,
2195 zone->prev_priority,
2196 zone->temp_priority,
2197 zone->zone_start_pfn);
2198 spin_unlock_irqrestore(&zone->lock, flags);
2199 seq_putc(m, '\n');
2200 }
2201 return 0;
2202}
2203
2204struct seq_operations zoneinfo_op = {
2205 .start = frag_start, /* iterate over all zones. The same as in
2206 * fragmentation. */
2207 .next = frag_next,
2208 .stop = frag_stop,
2209 .show = zoneinfo_show,
2210};
2211
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212static char *vmstat_text[] = {
2213 "nr_dirty",
2214 "nr_writeback",
2215 "nr_unstable",
2216 "nr_page_table_pages",
2217 "nr_mapped",
2218 "nr_slab",
2219
2220 "pgpgin",
2221 "pgpgout",
2222 "pswpin",
2223 "pswpout",
2224 "pgalloc_high",
2225
2226 "pgalloc_normal",
2227 "pgalloc_dma",
2228 "pgfree",
2229 "pgactivate",
2230 "pgdeactivate",
2231
2232 "pgfault",
2233 "pgmajfault",
2234 "pgrefill_high",
2235 "pgrefill_normal",
2236 "pgrefill_dma",
2237
2238 "pgsteal_high",
2239 "pgsteal_normal",
2240 "pgsteal_dma",
2241 "pgscan_kswapd_high",
2242 "pgscan_kswapd_normal",
2243
2244 "pgscan_kswapd_dma",
2245 "pgscan_direct_high",
2246 "pgscan_direct_normal",
2247 "pgscan_direct_dma",
2248 "pginodesteal",
2249
2250 "slabs_scanned",
2251 "kswapd_steal",
2252 "kswapd_inodesteal",
2253 "pageoutrun",
2254 "allocstall",
2255
2256 "pgrotated",
KAMEZAWA Hiroyukiedfbe2b2005-05-01 08:58:37 -07002257 "nr_bounce",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258};
2259
2260static void *vmstat_start(struct seq_file *m, loff_t *pos)
2261{
2262 struct page_state *ps;
2263
2264 if (*pos >= ARRAY_SIZE(vmstat_text))
2265 return NULL;
2266
2267 ps = kmalloc(sizeof(*ps), GFP_KERNEL);
2268 m->private = ps;
2269 if (!ps)
2270 return ERR_PTR(-ENOMEM);
2271 get_full_page_state(ps);
2272 ps->pgpgin /= 2; /* sectors -> kbytes */
2273 ps->pgpgout /= 2;
2274 return (unsigned long *)ps + *pos;
2275}
2276
2277static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos)
2278{
2279 (*pos)++;
2280 if (*pos >= ARRAY_SIZE(vmstat_text))
2281 return NULL;
2282 return (unsigned long *)m->private + *pos;
2283}
2284
2285static int vmstat_show(struct seq_file *m, void *arg)
2286{
2287 unsigned long *l = arg;
2288 unsigned long off = l - (unsigned long *)m->private;
2289
2290 seq_printf(m, "%s %lu\n", vmstat_text[off], *l);
2291 return 0;
2292}
2293
2294static void vmstat_stop(struct seq_file *m, void *arg)
2295{
2296 kfree(m->private);
2297 m->private = NULL;
2298}
2299
2300struct seq_operations vmstat_op = {
2301 .start = vmstat_start,
2302 .next = vmstat_next,
2303 .stop = vmstat_stop,
2304 .show = vmstat_show,
2305};
2306
2307#endif /* CONFIG_PROC_FS */
2308
2309#ifdef CONFIG_HOTPLUG_CPU
2310static int page_alloc_cpu_notify(struct notifier_block *self,
2311 unsigned long action, void *hcpu)
2312{
2313 int cpu = (unsigned long)hcpu;
2314 long *count;
2315 unsigned long *src, *dest;
2316
2317 if (action == CPU_DEAD) {
2318 int i;
2319
2320 /* Drain local pagecache count. */
2321 count = &per_cpu(nr_pagecache_local, cpu);
2322 atomic_add(*count, &nr_pagecache);
2323 *count = 0;
2324 local_irq_disable();
2325 __drain_pages(cpu);
2326
2327 /* Add dead cpu's page_states to our own. */
2328 dest = (unsigned long *)&__get_cpu_var(page_states);
2329 src = (unsigned long *)&per_cpu(page_states, cpu);
2330
2331 for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long);
2332 i++) {
2333 dest[i] += src[i];
2334 src[i] = 0;
2335 }
2336
2337 local_irq_enable();
2338 }
2339 return NOTIFY_OK;
2340}
2341#endif /* CONFIG_HOTPLUG_CPU */
2342
2343void __init page_alloc_init(void)
2344{
2345 hotcpu_notifier(page_alloc_cpu_notify, 0);
2346}
2347
2348/*
2349 * setup_per_zone_lowmem_reserve - called whenever
2350 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2351 * has a correct pages reserved value, so an adequate number of
2352 * pages are left in the zone after a successful __alloc_pages().
2353 */
2354static void setup_per_zone_lowmem_reserve(void)
2355{
2356 struct pglist_data *pgdat;
2357 int j, idx;
2358
2359 for_each_pgdat(pgdat) {
2360 for (j = 0; j < MAX_NR_ZONES; j++) {
2361 struct zone *zone = pgdat->node_zones + j;
2362 unsigned long present_pages = zone->present_pages;
2363
2364 zone->lowmem_reserve[j] = 0;
2365
2366 for (idx = j-1; idx >= 0; idx--) {
2367 struct zone *lower_zone;
2368
2369 if (sysctl_lowmem_reserve_ratio[idx] < 1)
2370 sysctl_lowmem_reserve_ratio[idx] = 1;
2371
2372 lower_zone = pgdat->node_zones + idx;
2373 lower_zone->lowmem_reserve[j] = present_pages /
2374 sysctl_lowmem_reserve_ratio[idx];
2375 present_pages += lower_zone->present_pages;
2376 }
2377 }
2378 }
2379}
2380
2381/*
2382 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2383 * that the pages_{min,low,high} values for each zone are set correctly
2384 * with respect to min_free_kbytes.
2385 */
Dave Hansen3947be12005-10-29 18:16:54 -07002386void setup_per_zone_pages_min(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387{
2388 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
2389 unsigned long lowmem_pages = 0;
2390 struct zone *zone;
2391 unsigned long flags;
2392
2393 /* Calculate total number of !ZONE_HIGHMEM pages */
2394 for_each_zone(zone) {
2395 if (!is_highmem(zone))
2396 lowmem_pages += zone->present_pages;
2397 }
2398
2399 for_each_zone(zone) {
Nick Piggin669ed172005-11-13 16:06:45 -08002400 unsigned long tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 spin_lock_irqsave(&zone->lru_lock, flags);
Nick Piggin669ed172005-11-13 16:06:45 -08002402 tmp = (pages_min * zone->present_pages) / lowmem_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 if (is_highmem(zone)) {
2404 /*
Nick Piggin669ed172005-11-13 16:06:45 -08002405 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2406 * need highmem pages, so cap pages_min to a small
2407 * value here.
2408 *
2409 * The (pages_high-pages_low) and (pages_low-pages_min)
2410 * deltas controls asynch page reclaim, and so should
2411 * not be capped for highmem.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 */
2413 int min_pages;
2414
2415 min_pages = zone->present_pages / 1024;
2416 if (min_pages < SWAP_CLUSTER_MAX)
2417 min_pages = SWAP_CLUSTER_MAX;
2418 if (min_pages > 128)
2419 min_pages = 128;
2420 zone->pages_min = min_pages;
2421 } else {
Nick Piggin669ed172005-11-13 16:06:45 -08002422 /*
2423 * If it's a lowmem zone, reserve a number of pages
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 * proportionate to the zone's size.
2425 */
Nick Piggin669ed172005-11-13 16:06:45 -08002426 zone->pages_min = tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 }
2428
Nick Piggin669ed172005-11-13 16:06:45 -08002429 zone->pages_low = zone->pages_min + tmp / 4;
2430 zone->pages_high = zone->pages_min + tmp / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 spin_unlock_irqrestore(&zone->lru_lock, flags);
2432 }
2433}
2434
2435/*
2436 * Initialise min_free_kbytes.
2437 *
2438 * For small machines we want it small (128k min). For large machines
2439 * we want it large (64MB max). But it is not linear, because network
2440 * bandwidth does not increase linearly with machine size. We use
2441 *
2442 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2443 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2444 *
2445 * which yields
2446 *
2447 * 16MB: 512k
2448 * 32MB: 724k
2449 * 64MB: 1024k
2450 * 128MB: 1448k
2451 * 256MB: 2048k
2452 * 512MB: 2896k
2453 * 1024MB: 4096k
2454 * 2048MB: 5792k
2455 * 4096MB: 8192k
2456 * 8192MB: 11584k
2457 * 16384MB: 16384k
2458 */
2459static int __init init_per_zone_pages_min(void)
2460{
2461 unsigned long lowmem_kbytes;
2462
2463 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
2464
2465 min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
2466 if (min_free_kbytes < 128)
2467 min_free_kbytes = 128;
2468 if (min_free_kbytes > 65536)
2469 min_free_kbytes = 65536;
2470 setup_per_zone_pages_min();
2471 setup_per_zone_lowmem_reserve();
2472 return 0;
2473}
2474module_init(init_per_zone_pages_min)
2475
2476/*
2477 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2478 * that we can call two helper functions whenever min_free_kbytes
2479 * changes.
2480 */
2481int min_free_kbytes_sysctl_handler(ctl_table *table, int write,
2482 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2483{
2484 proc_dointvec(table, write, file, buffer, length, ppos);
2485 setup_per_zone_pages_min();
2486 return 0;
2487}
2488
2489/*
2490 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2491 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2492 * whenever sysctl_lowmem_reserve_ratio changes.
2493 *
2494 * The reserve ratio obviously has absolutely no relation with the
2495 * pages_min watermarks. The lowmem reserve ratio can only make sense
2496 * if in function of the boot time zone sizes.
2497 */
2498int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
2499 struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
2500{
2501 proc_dointvec_minmax(table, write, file, buffer, length, ppos);
2502 setup_per_zone_lowmem_reserve();
2503 return 0;
2504}
2505
2506__initdata int hashdist = HASHDIST_DEFAULT;
2507
2508#ifdef CONFIG_NUMA
2509static int __init set_hashdist(char *str)
2510{
2511 if (!str)
2512 return 0;
2513 hashdist = simple_strtoul(str, &str, 0);
2514 return 1;
2515}
2516__setup("hashdist=", set_hashdist);
2517#endif
2518
2519/*
2520 * allocate a large system hash table from bootmem
2521 * - it is assumed that the hash table must contain an exact power-of-2
2522 * quantity of entries
2523 * - limit is the number of hash buckets, not the total allocation size
2524 */
2525void *__init alloc_large_system_hash(const char *tablename,
2526 unsigned long bucketsize,
2527 unsigned long numentries,
2528 int scale,
2529 int flags,
2530 unsigned int *_hash_shift,
2531 unsigned int *_hash_mask,
2532 unsigned long limit)
2533{
2534 unsigned long long max = limit;
2535 unsigned long log2qty, size;
2536 void *table = NULL;
2537
2538 /* allow the kernel cmdline to have a say */
2539 if (!numentries) {
2540 /* round applicable memory size up to nearest megabyte */
2541 numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages;
2542 numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
2543 numentries >>= 20 - PAGE_SHIFT;
2544 numentries <<= 20 - PAGE_SHIFT;
2545
2546 /* limit to 1 bucket per 2^scale bytes of low memory */
2547 if (scale > PAGE_SHIFT)
2548 numentries >>= (scale - PAGE_SHIFT);
2549 else
2550 numentries <<= (PAGE_SHIFT - scale);
2551 }
2552 /* rounded up to nearest power of 2 in size */
2553 numentries = 1UL << (long_log2(numentries) + 1);
2554
2555 /* limit allocation size to 1/16 total memory by default */
2556 if (max == 0) {
2557 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2558 do_div(max, bucketsize);
2559 }
2560
2561 if (numentries > max)
2562 numentries = max;
2563
2564 log2qty = long_log2(numentries);
2565
2566 do {
2567 size = bucketsize << log2qty;
2568 if (flags & HASH_EARLY)
2569 table = alloc_bootmem(size);
2570 else if (hashdist)
2571 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
2572 else {
2573 unsigned long order;
2574 for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++)
2575 ;
2576 table = (void*) __get_free_pages(GFP_ATOMIC, order);
2577 }
2578 } while (!table && size > PAGE_SIZE && --log2qty);
2579
2580 if (!table)
2581 panic("Failed to allocate %s hash table\n", tablename);
2582
2583 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2584 tablename,
2585 (1U << log2qty),
2586 long_log2(size) - PAGE_SHIFT,
2587 size);
2588
2589 if (_hash_shift)
2590 *_hash_shift = log2qty;
2591 if (_hash_mask)
2592 *_hash_mask = (1 << log2qty) - 1;
2593
2594 return table;
2595}