]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
17 | #include <linux/stddef.h> | |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/interrupt.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/jiffies.h> | |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/compiler.h> | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/kmemcheck.h> | |
27 | #include <linux/module.h> | |
28 | #include <linux/suspend.h> | |
29 | #include <linux/pagevec.h> | |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/slab.h> | |
32 | #include <linux/oom.h> | |
33 | #include <linux/notifier.h> | |
34 | #include <linux/topology.h> | |
35 | #include <linux/sysctl.h> | |
36 | #include <linux/cpu.h> | |
37 | #include <linux/cpuset.h> | |
38 | #include <linux/memory_hotplug.h> | |
39 | #include <linux/nodemask.h> | |
40 | #include <linux/vmalloc.h> | |
41 | #include <linux/mempolicy.h> | |
42 | #include <linux/stop_machine.h> | |
43 | #include <linux/sort.h> | |
44 | #include <linux/pfn.h> | |
45 | #include <linux/backing-dev.h> | |
46 | #include <linux/fault-inject.h> | |
47 | #include <linux/page-isolation.h> | |
48 | #include <linux/page_cgroup.h> | |
49 | #include <linux/debugobjects.h> | |
50 | #include <linux/kmemleak.h> | |
51 | #include <linux/memory.h> | |
52 | #include <trace/events/kmem.h> | |
53 | ||
54 | #include <asm/tlbflush.h> | |
55 | #include <asm/div64.h> | |
56 | #include "internal.h" | |
57 | ||
58 | /* | |
59 | * Array of node states. | |
60 | */ | |
61 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { | |
62 | [N_POSSIBLE] = NODE_MASK_ALL, | |
63 | [N_ONLINE] = { { [0] = 1UL } }, | |
64 | #ifndef CONFIG_NUMA | |
65 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, | |
66 | #ifdef CONFIG_HIGHMEM | |
67 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, | |
68 | #endif | |
69 | [N_CPU] = { { [0] = 1UL } }, | |
70 | #endif /* NUMA */ | |
71 | }; | |
72 | EXPORT_SYMBOL(node_states); | |
73 | ||
74 | unsigned long totalram_pages __read_mostly; | |
75 | unsigned long totalreserve_pages __read_mostly; | |
76 | int percpu_pagelist_fraction; | |
77 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; | |
78 | ||
79 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE | |
80 | int pageblock_order __read_mostly; | |
81 | #endif | |
82 | ||
83 | static void __free_pages_ok(struct page *page, unsigned int order); | |
84 | ||
85 | /* | |
86 | * results with 256, 32 in the lowmem_reserve sysctl: | |
87 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
88 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
89 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
90 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
91 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
92 | * | |
93 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
94 | * don't need any ZONE_NORMAL reservation | |
95 | */ | |
96 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { | |
97 | #ifdef CONFIG_ZONE_DMA | |
98 | 256, | |
99 | #endif | |
100 | #ifdef CONFIG_ZONE_DMA32 | |
101 | 256, | |
102 | #endif | |
103 | #ifdef CONFIG_HIGHMEM | |
104 | 32, | |
105 | #endif | |
106 | 32, | |
107 | }; | |
108 | ||
109 | EXPORT_SYMBOL(totalram_pages); | |
110 | ||
111 | static char * const zone_names[MAX_NR_ZONES] = { | |
112 | #ifdef CONFIG_ZONE_DMA | |
113 | "DMA", | |
114 | #endif | |
115 | #ifdef CONFIG_ZONE_DMA32 | |
116 | "DMA32", | |
117 | #endif | |
118 | "Normal", | |
119 | #ifdef CONFIG_HIGHMEM | |
120 | "HighMem", | |
121 | #endif | |
122 | "Movable", | |
123 | }; | |
124 | ||
125 | int min_free_kbytes = 1024; | |
126 | ||
127 | static unsigned long __meminitdata nr_kernel_pages; | |
128 | static unsigned long __meminitdata nr_all_pages; | |
129 | static unsigned long __meminitdata dma_reserve; | |
130 | ||
131 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
132 | /* | |
133 | * MAX_ACTIVE_REGIONS determines the maximum number of distinct | |
134 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
135 | * Ranges passed to add_active_range() will be merged if possible | |
136 | * so the number of times add_active_range() can be called is | |
137 | * related to the number of nodes and the number of holes | |
138 | */ | |
139 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
140 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
141 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
142 | #else | |
143 | #if MAX_NUMNODES >= 32 | |
144 | /* If there can be many nodes, allow up to 50 holes per node */ | |
145 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
146 | #else | |
147 | /* By default, allow up to 256 distinct regions */ | |
148 | #define MAX_ACTIVE_REGIONS 256 | |
149 | #endif | |
150 | #endif | |
151 | ||
152 | static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS]; | |
153 | static int __meminitdata nr_nodemap_entries; | |
154 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
155 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
156 | static unsigned long __initdata required_kernelcore; | |
157 | static unsigned long __initdata required_movablecore; | |
158 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; | |
159 | ||
160 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ | |
161 | int movable_zone; | |
162 | EXPORT_SYMBOL(movable_zone); | |
163 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
164 | ||
165 | #if MAX_NUMNODES > 1 | |
166 | int nr_node_ids __read_mostly = MAX_NUMNODES; | |
167 | int nr_online_nodes __read_mostly = 1; | |
168 | EXPORT_SYMBOL(nr_node_ids); | |
169 | EXPORT_SYMBOL(nr_online_nodes); | |
170 | #endif | |
171 | ||
172 | int page_group_by_mobility_disabled __read_mostly; | |
173 | ||
174 | static void set_pageblock_migratetype(struct page *page, int migratetype) | |
175 | { | |
176 | ||
177 | if (unlikely(page_group_by_mobility_disabled)) | |
178 | migratetype = MIGRATE_UNMOVABLE; | |
179 | ||
180 | set_pageblock_flags_group(page, (unsigned long)migratetype, | |
181 | PB_migrate, PB_migrate_end); | |
182 | } | |
183 | ||
184 | bool oom_killer_disabled __read_mostly; | |
185 | ||
186 | #ifdef CONFIG_DEBUG_VM | |
187 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) | |
188 | { | |
189 | int ret = 0; | |
190 | unsigned seq; | |
191 | unsigned long pfn = page_to_pfn(page); | |
192 | ||
193 | do { | |
194 | seq = zone_span_seqbegin(zone); | |
195 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
196 | ret = 1; | |
197 | else if (pfn < zone->zone_start_pfn) | |
198 | ret = 1; | |
199 | } while (zone_span_seqretry(zone, seq)); | |
200 | ||
201 | return ret; | |
202 | } | |
203 | ||
204 | static int page_is_consistent(struct zone *zone, struct page *page) | |
205 | { | |
206 | if (!pfn_valid_within(page_to_pfn(page))) | |
207 | return 0; | |
208 | if (zone != page_zone(page)) | |
209 | return 0; | |
210 | ||
211 | return 1; | |
212 | } | |
213 | /* | |
214 | * Temporary debugging check for pages not lying within a given zone. | |
215 | */ | |
216 | static int bad_range(struct zone *zone, struct page *page) | |
217 | { | |
218 | if (page_outside_zone_boundaries(zone, page)) | |
219 | return 1; | |
220 | if (!page_is_consistent(zone, page)) | |
221 | return 1; | |
222 | ||
223 | return 0; | |
224 | } | |
225 | #else | |
226 | static inline int bad_range(struct zone *zone, struct page *page) | |
227 | { | |
228 | return 0; | |
229 | } | |
230 | #endif | |
231 | ||
232 | static void bad_page(struct page *page) | |
233 | { | |
234 | static unsigned long resume; | |
235 | static unsigned long nr_shown; | |
236 | static unsigned long nr_unshown; | |
237 | ||
238 | /* Don't complain about poisoned pages */ | |
239 | if (PageHWPoison(page)) { | |
240 | __ClearPageBuddy(page); | |
241 | return; | |
242 | } | |
243 | ||
244 | /* | |
245 | * Allow a burst of 60 reports, then keep quiet for that minute; | |
246 | * or allow a steady drip of one report per second. | |
247 | */ | |
248 | if (nr_shown == 60) { | |
249 | if (time_before(jiffies, resume)) { | |
250 | nr_unshown++; | |
251 | goto out; | |
252 | } | |
253 | if (nr_unshown) { | |
254 | printk(KERN_ALERT | |
255 | "BUG: Bad page state: %lu messages suppressed\n", | |
256 | nr_unshown); | |
257 | nr_unshown = 0; | |
258 | } | |
259 | nr_shown = 0; | |
260 | } | |
261 | if (nr_shown++ == 0) | |
262 | resume = jiffies + 60 * HZ; | |
263 | ||
264 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", | |
265 | current->comm, page_to_pfn(page)); | |
266 | printk(KERN_ALERT | |
267 | "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n", | |
268 | page, (void *)page->flags, page_count(page), | |
269 | page_mapcount(page), page->mapping, page->index); | |
270 | ||
271 | dump_stack(); | |
272 | out: | |
273 | /* Leave bad fields for debug, except PageBuddy could make trouble */ | |
274 | __ClearPageBuddy(page); | |
275 | add_taint(TAINT_BAD_PAGE); | |
276 | } | |
277 | ||
278 | /* | |
279 | * Higher-order pages are called "compound pages". They are structured thusly: | |
280 | * | |
281 | * The first PAGE_SIZE page is called the "head page". | |
282 | * | |
283 | * The remaining PAGE_SIZE pages are called "tail pages". | |
284 | * | |
285 | * All pages have PG_compound set. All pages have their ->private pointing at | |
286 | * the head page (even the head page has this). | |
287 | * | |
288 | * The first tail page's ->lru.next holds the address of the compound page's | |
289 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
290 | * This usage means that zero-order pages may not be compound. | |
291 | */ | |
292 | ||
293 | static void free_compound_page(struct page *page) | |
294 | { | |
295 | __free_pages_ok(page, compound_order(page)); | |
296 | } | |
297 | ||
298 | void prep_compound_page(struct page *page, unsigned long order) | |
299 | { | |
300 | int i; | |
301 | int nr_pages = 1 << order; | |
302 | ||
303 | set_compound_page_dtor(page, free_compound_page); | |
304 | set_compound_order(page, order); | |
305 | __SetPageHead(page); | |
306 | for (i = 1; i < nr_pages; i++) { | |
307 | struct page *p = page + i; | |
308 | ||
309 | __SetPageTail(p); | |
310 | p->first_page = page; | |
311 | } | |
312 | } | |
313 | ||
314 | static int destroy_compound_page(struct page *page, unsigned long order) | |
315 | { | |
316 | int i; | |
317 | int nr_pages = 1 << order; | |
318 | int bad = 0; | |
319 | ||
320 | if (unlikely(compound_order(page) != order) || | |
321 | unlikely(!PageHead(page))) { | |
322 | bad_page(page); | |
323 | bad++; | |
324 | } | |
325 | ||
326 | __ClearPageHead(page); | |
327 | ||
328 | for (i = 1; i < nr_pages; i++) { | |
329 | struct page *p = page + i; | |
330 | ||
331 | if (unlikely(!PageTail(p) || (p->first_page != page))) { | |
332 | bad_page(page); | |
333 | bad++; | |
334 | } | |
335 | __ClearPageTail(p); | |
336 | } | |
337 | ||
338 | return bad; | |
339 | } | |
340 | ||
341 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) | |
342 | { | |
343 | int i; | |
344 | ||
345 | /* | |
346 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
347 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
348 | */ | |
349 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); | |
350 | for (i = 0; i < (1 << order); i++) | |
351 | clear_highpage(page + i); | |
352 | } | |
353 | ||
354 | static inline void set_page_order(struct page *page, int order) | |
355 | { | |
356 | set_page_private(page, order); | |
357 | __SetPageBuddy(page); | |
358 | } | |
359 | ||
360 | static inline void rmv_page_order(struct page *page) | |
361 | { | |
362 | __ClearPageBuddy(page); | |
363 | set_page_private(page, 0); | |
364 | } | |
365 | ||
366 | /* | |
367 | * Locate the struct page for both the matching buddy in our | |
368 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
369 | * | |
370 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
371 | * the following equation: | |
372 | * B2 = B1 ^ (1 << O) | |
373 | * For example, if the starting buddy (buddy2) is #8 its order | |
374 | * 1 buddy is #10: | |
375 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
376 | * | |
377 | * 2) Any buddy B will have an order O+1 parent P which | |
378 | * satisfies the following equation: | |
379 | * P = B & ~(1 << O) | |
380 | * | |
381 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER | |
382 | */ | |
383 | static inline struct page * | |
384 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
385 | { | |
386 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
387 | ||
388 | return page + (buddy_idx - page_idx); | |
389 | } | |
390 | ||
391 | static inline unsigned long | |
392 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
393 | { | |
394 | return (page_idx & ~(1 << order)); | |
395 | } | |
396 | ||
397 | /* | |
398 | * This function checks whether a page is free && is the buddy | |
399 | * we can do coalesce a page and its buddy if | |
400 | * (a) the buddy is not in a hole && | |
401 | * (b) the buddy is in the buddy system && | |
402 | * (c) a page and its buddy have the same order && | |
403 | * (d) a page and its buddy are in the same zone. | |
404 | * | |
405 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
406 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
407 | * | |
408 | * For recording page's order, we use page_private(page). | |
409 | */ | |
410 | static inline int page_is_buddy(struct page *page, struct page *buddy, | |
411 | int order) | |
412 | { | |
413 | if (!pfn_valid_within(page_to_pfn(buddy))) | |
414 | return 0; | |
415 | ||
416 | if (page_zone_id(page) != page_zone_id(buddy)) | |
417 | return 0; | |
418 | ||
419 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
420 | VM_BUG_ON(page_count(buddy) != 0); | |
421 | return 1; | |
422 | } | |
423 | return 0; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Freeing function for a buddy system allocator. | |
428 | * | |
429 | * The concept of a buddy system is to maintain direct-mapped table | |
430 | * (containing bit values) for memory blocks of various "orders". | |
431 | * The bottom level table contains the map for the smallest allocatable | |
432 | * units of memory (here, pages), and each level above it describes | |
433 | * pairs of units from the levels below, hence, "buddies". | |
434 | * At a high level, all that happens here is marking the table entry | |
435 | * at the bottom level available, and propagating the changes upward | |
436 | * as necessary, plus some accounting needed to play nicely with other | |
437 | * parts of the VM system. | |
438 | * At each level, we keep a list of pages, which are heads of continuous | |
439 | * free pages of length of (1 << order) and marked with PG_buddy. Page's | |
440 | * order is recorded in page_private(page) field. | |
441 | * So when we are allocating or freeing one, we can derive the state of the | |
442 | * other. That is, if we allocate a small block, and both were | |
443 | * free, the remainder of the region must be split into blocks. | |
444 | * If a block is freed, and its buddy is also free, then this | |
445 | * triggers coalescing into a block of larger size. | |
446 | * | |
447 | * -- wli | |
448 | */ | |
449 | ||
450 | static inline void __free_one_page(struct page *page, | |
451 | struct zone *zone, unsigned int order, | |
452 | int migratetype) | |
453 | { | |
454 | unsigned long page_idx; | |
455 | ||
456 | if (unlikely(PageCompound(page))) | |
457 | if (unlikely(destroy_compound_page(page, order))) | |
458 | return; | |
459 | ||
460 | VM_BUG_ON(migratetype == -1); | |
461 | ||
462 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
463 | ||
464 | VM_BUG_ON(page_idx & ((1 << order) - 1)); | |
465 | VM_BUG_ON(bad_range(zone, page)); | |
466 | ||
467 | while (order < MAX_ORDER-1) { | |
468 | unsigned long combined_idx; | |
469 | struct page *buddy; | |
470 | ||
471 | buddy = __page_find_buddy(page, page_idx, order); | |
472 | if (!page_is_buddy(page, buddy, order)) | |
473 | break; | |
474 | ||
475 | /* Our buddy is free, merge with it and move up one order. */ | |
476 | list_del(&buddy->lru); | |
477 | zone->free_area[order].nr_free--; | |
478 | rmv_page_order(buddy); | |
479 | combined_idx = __find_combined_index(page_idx, order); | |
480 | page = page + (combined_idx - page_idx); | |
481 | page_idx = combined_idx; | |
482 | order++; | |
483 | } | |
484 | set_page_order(page, order); | |
485 | list_add(&page->lru, | |
486 | &zone->free_area[order].free_list[migratetype]); | |
487 | zone->free_area[order].nr_free++; | |
488 | } | |
489 | ||
490 | /* | |
491 | * free_page_mlock() -- clean up attempts to free and mlocked() page. | |
492 | * Page should not be on lru, so no need to fix that up. | |
493 | * free_pages_check() will verify... | |
494 | */ | |
495 | static inline void free_page_mlock(struct page *page) | |
496 | { | |
497 | __dec_zone_page_state(page, NR_MLOCK); | |
498 | __count_vm_event(UNEVICTABLE_MLOCKFREED); | |
499 | } | |
500 | ||
501 | static inline int free_pages_check(struct page *page) | |
502 | { | |
503 | if (unlikely(page_mapcount(page) | | |
504 | (page->mapping != NULL) | | |
505 | (atomic_read(&page->_count) != 0) | | |
506 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) { | |
507 | bad_page(page); | |
508 | return 1; | |
509 | } | |
510 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
511 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; | |
512 | return 0; | |
513 | } | |
514 | ||
515 | /* | |
516 | * Frees a number of pages from the PCP lists | |
517 | * Assumes all pages on list are in same zone, and of same order. | |
518 | * count is the number of pages to free. | |
519 | * | |
520 | * If the zone was previously in an "all pages pinned" state then look to | |
521 | * see if this freeing clears that state. | |
522 | * | |
523 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
524 | * pinned" detection logic. | |
525 | */ | |
526 | static void free_pcppages_bulk(struct zone *zone, int count, | |
527 | struct per_cpu_pages *pcp) | |
528 | { | |
529 | int migratetype = 0; | |
530 | int batch_free = 0; | |
531 | ||
532 | spin_lock(&zone->lock); | |
533 | zone->all_unreclaimable = 0; | |
534 | zone->pages_scanned = 0; | |
535 | ||
536 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); | |
537 | while (count) { | |
538 | struct page *page; | |
539 | struct list_head *list; | |
540 | ||
541 | /* | |
542 | * Remove pages from lists in a round-robin fashion. A | |
543 | * batch_free count is maintained that is incremented when an | |
544 | * empty list is encountered. This is so more pages are freed | |
545 | * off fuller lists instead of spinning excessively around empty | |
546 | * lists | |
547 | */ | |
548 | do { | |
549 | batch_free++; | |
550 | if (++migratetype == MIGRATE_PCPTYPES) | |
551 | migratetype = 0; | |
552 | list = &pcp->lists[migratetype]; | |
553 | } while (list_empty(list)); | |
554 | ||
555 | do { | |
556 | page = list_entry(list->prev, struct page, lru); | |
557 | /* must delete as __free_one_page list manipulates */ | |
558 | list_del(&page->lru); | |
559 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ | |
560 | __free_one_page(page, zone, 0, page_private(page)); | |
561 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); | |
562 | } while (--count && --batch_free && !list_empty(list)); | |
563 | } | |
564 | spin_unlock(&zone->lock); | |
565 | } | |
566 | ||
567 | static void free_one_page(struct zone *zone, struct page *page, int order, | |
568 | int migratetype) | |
569 | { | |
570 | spin_lock(&zone->lock); | |
571 | zone->all_unreclaimable = 0; | |
572 | zone->pages_scanned = 0; | |
573 | ||
574 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); | |
575 | __free_one_page(page, zone, order, migratetype); | |
576 | spin_unlock(&zone->lock); | |
577 | } | |
578 | ||
579 | static void __free_pages_ok(struct page *page, unsigned int order) | |
580 | { | |
581 | unsigned long flags; | |
582 | int i; | |
583 | int bad = 0; | |
584 | int wasMlocked = __TestClearPageMlocked(page); | |
585 | ||
586 | trace_mm_page_free_direct(page, order); | |
587 | kmemcheck_free_shadow(page, order); | |
588 | ||
589 | for (i = 0 ; i < (1 << order) ; ++i) | |
590 | bad += free_pages_check(page + i); | |
591 | if (bad) | |
592 | return; | |
593 | ||
594 | if (!PageHighMem(page)) { | |
595 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
596 | debug_check_no_obj_freed(page_address(page), | |
597 | PAGE_SIZE << order); | |
598 | } | |
599 | arch_free_page(page, order); | |
600 | kernel_map_pages(page, 1 << order, 0); | |
601 | ||
602 | local_irq_save(flags); | |
603 | if (unlikely(wasMlocked)) | |
604 | free_page_mlock(page); | |
605 | __count_vm_events(PGFREE, 1 << order); | |
606 | free_one_page(page_zone(page), page, order, | |
607 | get_pageblock_migratetype(page)); | |
608 | local_irq_restore(flags); | |
609 | } | |
610 | ||
611 | /* | |
612 | * permit the bootmem allocator to evade page validation on high-order frees | |
613 | */ | |
614 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) | |
615 | { | |
616 | if (order == 0) { | |
617 | __ClearPageReserved(page); | |
618 | set_page_count(page, 0); | |
619 | set_page_refcounted(page); | |
620 | __free_page(page); | |
621 | } else { | |
622 | int loop; | |
623 | ||
624 | prefetchw(page); | |
625 | for (loop = 0; loop < BITS_PER_LONG; loop++) { | |
626 | struct page *p = &page[loop]; | |
627 | ||
628 | if (loop + 1 < BITS_PER_LONG) | |
629 | prefetchw(p + 1); | |
630 | __ClearPageReserved(p); | |
631 | set_page_count(p, 0); | |
632 | } | |
633 | ||
634 | set_page_refcounted(page); | |
635 | __free_pages(page, order); | |
636 | } | |
637 | } | |
638 | ||
639 | ||
640 | /* | |
641 | * The order of subdivision here is critical for the IO subsystem. | |
642 | * Please do not alter this order without good reasons and regression | |
643 | * testing. Specifically, as large blocks of memory are subdivided, | |
644 | * the order in which smaller blocks are delivered depends on the order | |
645 | * they're subdivided in this function. This is the primary factor | |
646 | * influencing the order in which pages are delivered to the IO | |
647 | * subsystem according to empirical testing, and this is also justified | |
648 | * by considering the behavior of a buddy system containing a single | |
649 | * large block of memory acted on by a series of small allocations. | |
650 | * This behavior is a critical factor in sglist merging's success. | |
651 | * | |
652 | * -- wli | |
653 | */ | |
654 | static inline void expand(struct zone *zone, struct page *page, | |
655 | int low, int high, struct free_area *area, | |
656 | int migratetype) | |
657 | { | |
658 | unsigned long size = 1 << high; | |
659 | ||
660 | while (high > low) { | |
661 | area--; | |
662 | high--; | |
663 | size >>= 1; | |
664 | VM_BUG_ON(bad_range(zone, &page[size])); | |
665 | list_add(&page[size].lru, &area->free_list[migratetype]); | |
666 | area->nr_free++; | |
667 | set_page_order(&page[size], high); | |
668 | } | |
669 | } | |
670 | ||
671 | /* | |
672 | * This page is about to be returned from the page allocator | |
673 | */ | |
674 | static inline int check_new_page(struct page *page) | |
675 | { | |
676 | if (unlikely(page_mapcount(page) | | |
677 | (page->mapping != NULL) | | |
678 | (atomic_read(&page->_count) != 0) | | |
679 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) { | |
680 | bad_page(page); | |
681 | return 1; | |
682 | } | |
683 | return 0; | |
684 | } | |
685 | ||
686 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
687 | { | |
688 | int i; | |
689 | ||
690 | for (i = 0; i < (1 << order); i++) { | |
691 | struct page *p = page + i; | |
692 | if (unlikely(check_new_page(p))) | |
693 | return 1; | |
694 | } | |
695 | ||
696 | set_page_private(page, 0); | |
697 | set_page_refcounted(page); | |
698 | ||
699 | arch_alloc_page(page, order); | |
700 | kernel_map_pages(page, 1 << order, 1); | |
701 | ||
702 | if (gfp_flags & __GFP_ZERO) | |
703 | prep_zero_page(page, order, gfp_flags); | |
704 | ||
705 | if (order && (gfp_flags & __GFP_COMP)) | |
706 | prep_compound_page(page, order); | |
707 | ||
708 | return 0; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Go through the free lists for the given migratetype and remove | |
713 | * the smallest available page from the freelists | |
714 | */ | |
715 | static inline | |
716 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, | |
717 | int migratetype) | |
718 | { | |
719 | unsigned int current_order; | |
720 | struct free_area * area; | |
721 | struct page *page; | |
722 | ||
723 | /* Find a page of the appropriate size in the preferred list */ | |
724 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
725 | area = &(zone->free_area[current_order]); | |
726 | if (list_empty(&area->free_list[migratetype])) | |
727 | continue; | |
728 | ||
729 | page = list_entry(area->free_list[migratetype].next, | |
730 | struct page, lru); | |
731 | list_del(&page->lru); | |
732 | rmv_page_order(page); | |
733 | area->nr_free--; | |
734 | expand(zone, page, order, current_order, area, migratetype); | |
735 | return page; | |
736 | } | |
737 | ||
738 | return NULL; | |
739 | } | |
740 | ||
741 | ||
742 | /* | |
743 | * This array describes the order lists are fallen back to when | |
744 | * the free lists for the desirable migrate type are depleted | |
745 | */ | |
746 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { | |
747 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
748 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, | |
749 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, | |
750 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ | |
751 | }; | |
752 | ||
753 | /* | |
754 | * Move the free pages in a range to the free lists of the requested type. | |
755 | * Note that start_page and end_pages are not aligned on a pageblock | |
756 | * boundary. If alignment is required, use move_freepages_block() | |
757 | */ | |
758 | static int move_freepages(struct zone *zone, | |
759 | struct page *start_page, struct page *end_page, | |
760 | int migratetype) | |
761 | { | |
762 | struct page *page; | |
763 | unsigned long order; | |
764 | int pages_moved = 0; | |
765 | ||
766 | #ifndef CONFIG_HOLES_IN_ZONE | |
767 | /* | |
768 | * page_zone is not safe to call in this context when | |
769 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant | |
770 | * anyway as we check zone boundaries in move_freepages_block(). | |
771 | * Remove at a later date when no bug reports exist related to | |
772 | * grouping pages by mobility | |
773 | */ | |
774 | BUG_ON(page_zone(start_page) != page_zone(end_page)); | |
775 | #endif | |
776 | ||
777 | for (page = start_page; page <= end_page;) { | |
778 | /* Make sure we are not inadvertently changing nodes */ | |
779 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); | |
780 | ||
781 | if (!pfn_valid_within(page_to_pfn(page))) { | |
782 | page++; | |
783 | continue; | |
784 | } | |
785 | ||
786 | if (!PageBuddy(page)) { | |
787 | page++; | |
788 | continue; | |
789 | } | |
790 | ||
791 | order = page_order(page); | |
792 | list_del(&page->lru); | |
793 | list_add(&page->lru, | |
794 | &zone->free_area[order].free_list[migratetype]); | |
795 | page += 1 << order; | |
796 | pages_moved += 1 << order; | |
797 | } | |
798 | ||
799 | return pages_moved; | |
800 | } | |
801 | ||
802 | static int move_freepages_block(struct zone *zone, struct page *page, | |
803 | int migratetype) | |
804 | { | |
805 | unsigned long start_pfn, end_pfn; | |
806 | struct page *start_page, *end_page; | |
807 | ||
808 | start_pfn = page_to_pfn(page); | |
809 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); | |
810 | start_page = pfn_to_page(start_pfn); | |
811 | end_page = start_page + pageblock_nr_pages - 1; | |
812 | end_pfn = start_pfn + pageblock_nr_pages - 1; | |
813 | ||
814 | /* Do not cross zone boundaries */ | |
815 | if (start_pfn < zone->zone_start_pfn) | |
816 | start_page = page; | |
817 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
818 | return 0; | |
819 | ||
820 | return move_freepages(zone, start_page, end_page, migratetype); | |
821 | } | |
822 | ||
823 | static void change_pageblock_range(struct page *pageblock_page, | |
824 | int start_order, int migratetype) | |
825 | { | |
826 | int nr_pageblocks = 1 << (start_order - pageblock_order); | |
827 | ||
828 | while (nr_pageblocks--) { | |
829 | set_pageblock_migratetype(pageblock_page, migratetype); | |
830 | pageblock_page += pageblock_nr_pages; | |
831 | } | |
832 | } | |
833 | ||
834 | /* Remove an element from the buddy allocator from the fallback list */ | |
835 | static inline struct page * | |
836 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) | |
837 | { | |
838 | struct free_area * area; | |
839 | int current_order; | |
840 | struct page *page; | |
841 | int migratetype, i; | |
842 | ||
843 | /* Find the largest possible block of pages in the other list */ | |
844 | for (current_order = MAX_ORDER-1; current_order >= order; | |
845 | --current_order) { | |
846 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { | |
847 | migratetype = fallbacks[start_migratetype][i]; | |
848 | ||
849 | /* MIGRATE_RESERVE handled later if necessary */ | |
850 | if (migratetype == MIGRATE_RESERVE) | |
851 | continue; | |
852 | ||
853 | area = &(zone->free_area[current_order]); | |
854 | if (list_empty(&area->free_list[migratetype])) | |
855 | continue; | |
856 | ||
857 | page = list_entry(area->free_list[migratetype].next, | |
858 | struct page, lru); | |
859 | area->nr_free--; | |
860 | ||
861 | /* | |
862 | * If breaking a large block of pages, move all free | |
863 | * pages to the preferred allocation list. If falling | |
864 | * back for a reclaimable kernel allocation, be more | |
865 | * agressive about taking ownership of free pages | |
866 | */ | |
867 | if (unlikely(current_order >= (pageblock_order >> 1)) || | |
868 | start_migratetype == MIGRATE_RECLAIMABLE || | |
869 | page_group_by_mobility_disabled) { | |
870 | unsigned long pages; | |
871 | pages = move_freepages_block(zone, page, | |
872 | start_migratetype); | |
873 | ||
874 | /* Claim the whole block if over half of it is free */ | |
875 | if (pages >= (1 << (pageblock_order-1)) || | |
876 | page_group_by_mobility_disabled) | |
877 | set_pageblock_migratetype(page, | |
878 | start_migratetype); | |
879 | ||
880 | migratetype = start_migratetype; | |
881 | } | |
882 | ||
883 | /* Remove the page from the freelists */ | |
884 | list_del(&page->lru); | |
885 | rmv_page_order(page); | |
886 | ||
887 | /* Take ownership for orders >= pageblock_order */ | |
888 | if (current_order >= pageblock_order) | |
889 | change_pageblock_range(page, current_order, | |
890 | start_migratetype); | |
891 | ||
892 | expand(zone, page, order, current_order, area, migratetype); | |
893 | ||
894 | trace_mm_page_alloc_extfrag(page, order, current_order, | |
895 | start_migratetype, migratetype); | |
896 | ||
897 | return page; | |
898 | } | |
899 | } | |
900 | ||
901 | return NULL; | |
902 | } | |
903 | ||
904 | /* | |
905 | * Do the hard work of removing an element from the buddy allocator. | |
906 | * Call me with the zone->lock already held. | |
907 | */ | |
908 | static struct page *__rmqueue(struct zone *zone, unsigned int order, | |
909 | int migratetype) | |
910 | { | |
911 | struct page *page; | |
912 | ||
913 | retry_reserve: | |
914 | page = __rmqueue_smallest(zone, order, migratetype); | |
915 | ||
916 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { | |
917 | page = __rmqueue_fallback(zone, order, migratetype); | |
918 | ||
919 | /* | |
920 | * Use MIGRATE_RESERVE rather than fail an allocation. goto | |
921 | * is used because __rmqueue_smallest is an inline function | |
922 | * and we want just one call site | |
923 | */ | |
924 | if (!page) { | |
925 | migratetype = MIGRATE_RESERVE; | |
926 | goto retry_reserve; | |
927 | } | |
928 | } | |
929 | ||
930 | trace_mm_page_alloc_zone_locked(page, order, migratetype); | |
931 | return page; | |
932 | } | |
933 | ||
934 | /* | |
935 | * Obtain a specified number of elements from the buddy allocator, all under | |
936 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
937 | * Returns the number of new pages which were placed at *list. | |
938 | */ | |
939 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
940 | unsigned long count, struct list_head *list, | |
941 | int migratetype, int cold) | |
942 | { | |
943 | int i; | |
944 | ||
945 | spin_lock(&zone->lock); | |
946 | for (i = 0; i < count; ++i) { | |
947 | struct page *page = __rmqueue(zone, order, migratetype); | |
948 | if (unlikely(page == NULL)) | |
949 | break; | |
950 | ||
951 | /* | |
952 | * Split buddy pages returned by expand() are received here | |
953 | * in physical page order. The page is added to the callers and | |
954 | * list and the list head then moves forward. From the callers | |
955 | * perspective, the linked list is ordered by page number in | |
956 | * some conditions. This is useful for IO devices that can | |
957 | * merge IO requests if the physical pages are ordered | |
958 | * properly. | |
959 | */ | |
960 | if (likely(cold == 0)) | |
961 | list_add(&page->lru, list); | |
962 | else | |
963 | list_add_tail(&page->lru, list); | |
964 | set_page_private(page, migratetype); | |
965 | list = &page->lru; | |
966 | } | |
967 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); | |
968 | spin_unlock(&zone->lock); | |
969 | return i; | |
970 | } | |
971 | ||
972 | #ifdef CONFIG_NUMA | |
973 | /* | |
974 | * Called from the vmstat counter updater to drain pagesets of this | |
975 | * currently executing processor on remote nodes after they have | |
976 | * expired. | |
977 | * | |
978 | * Note that this function must be called with the thread pinned to | |
979 | * a single processor. | |
980 | */ | |
981 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) | |
982 | { | |
983 | unsigned long flags; | |
984 | int to_drain; | |
985 | ||
986 | local_irq_save(flags); | |
987 | if (pcp->count >= pcp->batch) | |
988 | to_drain = pcp->batch; | |
989 | else | |
990 | to_drain = pcp->count; | |
991 | free_pcppages_bulk(zone, to_drain, pcp); | |
992 | pcp->count -= to_drain; | |
993 | local_irq_restore(flags); | |
994 | } | |
995 | #endif | |
996 | ||
997 | /* | |
998 | * Drain pages of the indicated processor. | |
999 | * | |
1000 | * The processor must either be the current processor and the | |
1001 | * thread pinned to the current processor or a processor that | |
1002 | * is not online. | |
1003 | */ | |
1004 | static void drain_pages(unsigned int cpu) | |
1005 | { | |
1006 | unsigned long flags; | |
1007 | struct zone *zone; | |
1008 | ||
1009 | for_each_populated_zone(zone) { | |
1010 | struct per_cpu_pageset *pset; | |
1011 | struct per_cpu_pages *pcp; | |
1012 | ||
1013 | local_irq_save(flags); | |
1014 | pset = per_cpu_ptr(zone->pageset, cpu); | |
1015 | ||
1016 | pcp = &pset->pcp; | |
1017 | free_pcppages_bulk(zone, pcp->count, pcp); | |
1018 | pcp->count = 0; | |
1019 | local_irq_restore(flags); | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
1025 | */ | |
1026 | void drain_local_pages(void *arg) | |
1027 | { | |
1028 | drain_pages(smp_processor_id()); | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator | |
1033 | */ | |
1034 | void drain_all_pages(void) | |
1035 | { | |
1036 | on_each_cpu(drain_local_pages, NULL, 1); | |
1037 | } | |
1038 | ||
1039 | #ifdef CONFIG_HIBERNATION | |
1040 | ||
1041 | void mark_free_pages(struct zone *zone) | |
1042 | { | |
1043 | unsigned long pfn, max_zone_pfn; | |
1044 | unsigned long flags; | |
1045 | int order, t; | |
1046 | struct list_head *curr; | |
1047 | ||
1048 | if (!zone->spanned_pages) | |
1049 | return; | |
1050 | ||
1051 | spin_lock_irqsave(&zone->lock, flags); | |
1052 | ||
1053 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
1054 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1055 | if (pfn_valid(pfn)) { | |
1056 | struct page *page = pfn_to_page(pfn); | |
1057 | ||
1058 | if (!swsusp_page_is_forbidden(page)) | |
1059 | swsusp_unset_page_free(page); | |
1060 | } | |
1061 | ||
1062 | for_each_migratetype_order(order, t) { | |
1063 | list_for_each(curr, &zone->free_area[order].free_list[t]) { | |
1064 | unsigned long i; | |
1065 | ||
1066 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
1067 | for (i = 0; i < (1UL << order); i++) | |
1068 | swsusp_set_page_free(pfn_to_page(pfn + i)); | |
1069 | } | |
1070 | } | |
1071 | spin_unlock_irqrestore(&zone->lock, flags); | |
1072 | } | |
1073 | #endif /* CONFIG_PM */ | |
1074 | ||
1075 | /* | |
1076 | * Free a 0-order page | |
1077 | * cold == 1 ? free a cold page : free a hot page | |
1078 | */ | |
1079 | void free_hot_cold_page(struct page *page, int cold) | |
1080 | { | |
1081 | struct zone *zone = page_zone(page); | |
1082 | struct per_cpu_pages *pcp; | |
1083 | unsigned long flags; | |
1084 | int migratetype; | |
1085 | int wasMlocked = __TestClearPageMlocked(page); | |
1086 | ||
1087 | trace_mm_page_free_direct(page, 0); | |
1088 | kmemcheck_free_shadow(page, 0); | |
1089 | ||
1090 | if (PageAnon(page)) | |
1091 | page->mapping = NULL; | |
1092 | if (free_pages_check(page)) | |
1093 | return; | |
1094 | ||
1095 | if (!PageHighMem(page)) { | |
1096 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
1097 | debug_check_no_obj_freed(page_address(page), PAGE_SIZE); | |
1098 | } | |
1099 | arch_free_page(page, 0); | |
1100 | kernel_map_pages(page, 1, 0); | |
1101 | ||
1102 | migratetype = get_pageblock_migratetype(page); | |
1103 | set_page_private(page, migratetype); | |
1104 | local_irq_save(flags); | |
1105 | if (unlikely(wasMlocked)) | |
1106 | free_page_mlock(page); | |
1107 | __count_vm_event(PGFREE); | |
1108 | ||
1109 | /* | |
1110 | * We only track unmovable, reclaimable and movable on pcp lists. | |
1111 | * Free ISOLATE pages back to the allocator because they are being | |
1112 | * offlined but treat RESERVE as movable pages so we can get those | |
1113 | * areas back if necessary. Otherwise, we may have to free | |
1114 | * excessively into the page allocator | |
1115 | */ | |
1116 | if (migratetype >= MIGRATE_PCPTYPES) { | |
1117 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { | |
1118 | free_one_page(zone, page, 0, migratetype); | |
1119 | goto out; | |
1120 | } | |
1121 | migratetype = MIGRATE_MOVABLE; | |
1122 | } | |
1123 | ||
1124 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
1125 | if (cold) | |
1126 | list_add_tail(&page->lru, &pcp->lists[migratetype]); | |
1127 | else | |
1128 | list_add(&page->lru, &pcp->lists[migratetype]); | |
1129 | pcp->count++; | |
1130 | if (pcp->count >= pcp->high) { | |
1131 | free_pcppages_bulk(zone, pcp->batch, pcp); | |
1132 | pcp->count -= pcp->batch; | |
1133 | } | |
1134 | ||
1135 | out: | |
1136 | local_irq_restore(flags); | |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * split_page takes a non-compound higher-order page, and splits it into | |
1141 | * n (1<<order) sub-pages: page[0..n] | |
1142 | * Each sub-page must be freed individually. | |
1143 | * | |
1144 | * Note: this is probably too low level an operation for use in drivers. | |
1145 | * Please consult with lkml before using this in your driver. | |
1146 | */ | |
1147 | void split_page(struct page *page, unsigned int order) | |
1148 | { | |
1149 | int i; | |
1150 | ||
1151 | VM_BUG_ON(PageCompound(page)); | |
1152 | VM_BUG_ON(!page_count(page)); | |
1153 | ||
1154 | #ifdef CONFIG_KMEMCHECK | |
1155 | /* | |
1156 | * Split shadow pages too, because free(page[0]) would | |
1157 | * otherwise free the whole shadow. | |
1158 | */ | |
1159 | if (kmemcheck_page_is_tracked(page)) | |
1160 | split_page(virt_to_page(page[0].shadow), order); | |
1161 | #endif | |
1162 | ||
1163 | for (i = 1; i < (1 << order); i++) | |
1164 | set_page_refcounted(page + i); | |
1165 | } | |
1166 | ||
1167 | /* | |
1168 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
1169 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
1170 | * or two. | |
1171 | */ | |
1172 | static inline | |
1173 | struct page *buffered_rmqueue(struct zone *preferred_zone, | |
1174 | struct zone *zone, int order, gfp_t gfp_flags, | |
1175 | int migratetype) | |
1176 | { | |
1177 | unsigned long flags; | |
1178 | struct page *page; | |
1179 | int cold = !!(gfp_flags & __GFP_COLD); | |
1180 | ||
1181 | again: | |
1182 | if (likely(order == 0)) { | |
1183 | struct per_cpu_pages *pcp; | |
1184 | struct list_head *list; | |
1185 | ||
1186 | local_irq_save(flags); | |
1187 | pcp = &this_cpu_ptr(zone->pageset)->pcp; | |
1188 | list = &pcp->lists[migratetype]; | |
1189 | if (list_empty(list)) { | |
1190 | pcp->count += rmqueue_bulk(zone, 0, | |
1191 | pcp->batch, list, | |
1192 | migratetype, cold); | |
1193 | if (unlikely(list_empty(list))) | |
1194 | goto failed; | |
1195 | } | |
1196 | ||
1197 | if (cold) | |
1198 | page = list_entry(list->prev, struct page, lru); | |
1199 | else | |
1200 | page = list_entry(list->next, struct page, lru); | |
1201 | ||
1202 | list_del(&page->lru); | |
1203 | pcp->count--; | |
1204 | } else { | |
1205 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { | |
1206 | /* | |
1207 | * __GFP_NOFAIL is not to be used in new code. | |
1208 | * | |
1209 | * All __GFP_NOFAIL callers should be fixed so that they | |
1210 | * properly detect and handle allocation failures. | |
1211 | * | |
1212 | * We most definitely don't want callers attempting to | |
1213 | * allocate greater than order-1 page units with | |
1214 | * __GFP_NOFAIL. | |
1215 | */ | |
1216 | WARN_ON_ONCE(order > 1); | |
1217 | } | |
1218 | spin_lock_irqsave(&zone->lock, flags); | |
1219 | page = __rmqueue(zone, order, migratetype); | |
1220 | spin_unlock(&zone->lock); | |
1221 | if (!page) | |
1222 | goto failed; | |
1223 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); | |
1224 | } | |
1225 | ||
1226 | __count_zone_vm_events(PGALLOC, zone, 1 << order); | |
1227 | zone_statistics(preferred_zone, zone); | |
1228 | local_irq_restore(flags); | |
1229 | ||
1230 | VM_BUG_ON(bad_range(zone, page)); | |
1231 | if (prep_new_page(page, order, gfp_flags)) | |
1232 | goto again; | |
1233 | return page; | |
1234 | ||
1235 | failed: | |
1236 | local_irq_restore(flags); | |
1237 | return NULL; | |
1238 | } | |
1239 | ||
1240 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ | |
1241 | #define ALLOC_WMARK_MIN WMARK_MIN | |
1242 | #define ALLOC_WMARK_LOW WMARK_LOW | |
1243 | #define ALLOC_WMARK_HIGH WMARK_HIGH | |
1244 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ | |
1245 | ||
1246 | /* Mask to get the watermark bits */ | |
1247 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) | |
1248 | ||
1249 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
1250 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
1251 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
1252 | ||
1253 | #ifdef CONFIG_FAIL_PAGE_ALLOC | |
1254 | ||
1255 | static struct fail_page_alloc_attr { | |
1256 | struct fault_attr attr; | |
1257 | ||
1258 | u32 ignore_gfp_highmem; | |
1259 | u32 ignore_gfp_wait; | |
1260 | u32 min_order; | |
1261 | ||
1262 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1263 | ||
1264 | struct dentry *ignore_gfp_highmem_file; | |
1265 | struct dentry *ignore_gfp_wait_file; | |
1266 | struct dentry *min_order_file; | |
1267 | ||
1268 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1269 | ||
1270 | } fail_page_alloc = { | |
1271 | .attr = FAULT_ATTR_INITIALIZER, | |
1272 | .ignore_gfp_wait = 1, | |
1273 | .ignore_gfp_highmem = 1, | |
1274 | .min_order = 1, | |
1275 | }; | |
1276 | ||
1277 | static int __init setup_fail_page_alloc(char *str) | |
1278 | { | |
1279 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
1280 | } | |
1281 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
1282 | ||
1283 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1284 | { | |
1285 | if (order < fail_page_alloc.min_order) | |
1286 | return 0; | |
1287 | if (gfp_mask & __GFP_NOFAIL) | |
1288 | return 0; | |
1289 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
1290 | return 0; | |
1291 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
1292 | return 0; | |
1293 | ||
1294 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
1295 | } | |
1296 | ||
1297 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
1298 | ||
1299 | static int __init fail_page_alloc_debugfs(void) | |
1300 | { | |
1301 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
1302 | struct dentry *dir; | |
1303 | int err; | |
1304 | ||
1305 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
1306 | "fail_page_alloc"); | |
1307 | if (err) | |
1308 | return err; | |
1309 | dir = fail_page_alloc.attr.dentries.dir; | |
1310 | ||
1311 | fail_page_alloc.ignore_gfp_wait_file = | |
1312 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
1313 | &fail_page_alloc.ignore_gfp_wait); | |
1314 | ||
1315 | fail_page_alloc.ignore_gfp_highmem_file = | |
1316 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
1317 | &fail_page_alloc.ignore_gfp_highmem); | |
1318 | fail_page_alloc.min_order_file = | |
1319 | debugfs_create_u32("min-order", mode, dir, | |
1320 | &fail_page_alloc.min_order); | |
1321 | ||
1322 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
1323 | !fail_page_alloc.ignore_gfp_highmem_file || | |
1324 | !fail_page_alloc.min_order_file) { | |
1325 | err = -ENOMEM; | |
1326 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
1327 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
1328 | debugfs_remove(fail_page_alloc.min_order_file); | |
1329 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); | |
1330 | } | |
1331 | ||
1332 | return err; | |
1333 | } | |
1334 | ||
1335 | late_initcall(fail_page_alloc_debugfs); | |
1336 | ||
1337 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
1338 | ||
1339 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
1340 | ||
1341 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
1342 | { | |
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
1347 | ||
1348 | /* | |
1349 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
1350 | * of the allocation. | |
1351 | */ | |
1352 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
1353 | int classzone_idx, int alloc_flags) | |
1354 | { | |
1355 | /* free_pages my go negative - that's OK */ | |
1356 | long min = mark; | |
1357 | long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1; | |
1358 | int o; | |
1359 | ||
1360 | if (alloc_flags & ALLOC_HIGH) | |
1361 | min -= min / 2; | |
1362 | if (alloc_flags & ALLOC_HARDER) | |
1363 | min -= min / 4; | |
1364 | ||
1365 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1366 | return 0; | |
1367 | for (o = 0; o < order; o++) { | |
1368 | /* At the next order, this order's pages become unavailable */ | |
1369 | free_pages -= z->free_area[o].nr_free << o; | |
1370 | ||
1371 | /* Require fewer higher order pages to be free */ | |
1372 | min >>= 1; | |
1373 | ||
1374 | if (free_pages <= min) | |
1375 | return 0; | |
1376 | } | |
1377 | return 1; | |
1378 | } | |
1379 | ||
1380 | #ifdef CONFIG_NUMA | |
1381 | /* | |
1382 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1383 | * skip over zones that are not allowed by the cpuset, or that have | |
1384 | * been recently (in last second) found to be nearly full. See further | |
1385 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1386 | * that have to skip over a lot of full or unallowed zones. | |
1387 | * | |
1388 | * If the zonelist cache is present in the passed in zonelist, then | |
1389 | * returns a pointer to the allowed node mask (either the current | |
1390 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) | |
1391 | * | |
1392 | * If the zonelist cache is not available for this zonelist, does | |
1393 | * nothing and returns NULL. | |
1394 | * | |
1395 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1396 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1397 | * | |
1398 | * We hold off even calling zlc_setup, until after we've checked the | |
1399 | * first zone in the zonelist, on the theory that most allocations will | |
1400 | * be satisfied from that first zone, so best to examine that zone as | |
1401 | * quickly as we can. | |
1402 | */ | |
1403 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1404 | { | |
1405 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1406 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1407 | ||
1408 | zlc = zonelist->zlcache_ptr; | |
1409 | if (!zlc) | |
1410 | return NULL; | |
1411 | ||
1412 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { | |
1413 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1414 | zlc->last_full_zap = jiffies; | |
1415 | } | |
1416 | ||
1417 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1418 | &cpuset_current_mems_allowed : | |
1419 | &node_states[N_HIGH_MEMORY]; | |
1420 | return allowednodes; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1425 | * if it is worth looking at further for free memory: | |
1426 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1427 | * bit set in the zonelist_cache fullzones BITMAP). | |
1428 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1429 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1430 | * Return true (non-zero) if zone is worth looking at further, or | |
1431 | * else return false (zero) if it is not. | |
1432 | * | |
1433 | * This check -ignores- the distinction between various watermarks, | |
1434 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1435 | * found to be full for any variation of these watermarks, it will | |
1436 | * be considered full for up to one second by all requests, unless | |
1437 | * we are so low on memory on all allowed nodes that we are forced | |
1438 | * into the second scan of the zonelist. | |
1439 | * | |
1440 | * In the second scan we ignore this zonelist cache and exactly | |
1441 | * apply the watermarks to all zones, even it is slower to do so. | |
1442 | * We are low on memory in the second scan, and should leave no stone | |
1443 | * unturned looking for a free page. | |
1444 | */ | |
1445 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, | |
1446 | nodemask_t *allowednodes) | |
1447 | { | |
1448 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1449 | int i; /* index of *z in zonelist zones */ | |
1450 | int n; /* node that zone *z is on */ | |
1451 | ||
1452 | zlc = zonelist->zlcache_ptr; | |
1453 | if (!zlc) | |
1454 | return 1; | |
1455 | ||
1456 | i = z - zonelist->_zonerefs; | |
1457 | n = zlc->z_to_n[i]; | |
1458 | ||
1459 | /* This zone is worth trying if it is allowed but not full */ | |
1460 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1461 | } | |
1462 | ||
1463 | /* | |
1464 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1465 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1466 | * from that zone don't waste time re-examining it. | |
1467 | */ | |
1468 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) | |
1469 | { | |
1470 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1471 | int i; /* index of *z in zonelist zones */ | |
1472 | ||
1473 | zlc = zonelist->zlcache_ptr; | |
1474 | if (!zlc) | |
1475 | return; | |
1476 | ||
1477 | i = z - zonelist->_zonerefs; | |
1478 | ||
1479 | set_bit(i, zlc->fullzones); | |
1480 | } | |
1481 | ||
1482 | #else /* CONFIG_NUMA */ | |
1483 | ||
1484 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1485 | { | |
1486 | return NULL; | |
1487 | } | |
1488 | ||
1489 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, | |
1490 | nodemask_t *allowednodes) | |
1491 | { | |
1492 | return 1; | |
1493 | } | |
1494 | ||
1495 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) | |
1496 | { | |
1497 | } | |
1498 | #endif /* CONFIG_NUMA */ | |
1499 | ||
1500 | /* | |
1501 | * get_page_from_freelist goes through the zonelist trying to allocate | |
1502 | * a page. | |
1503 | */ | |
1504 | static struct page * | |
1505 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, | |
1506 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, | |
1507 | struct zone *preferred_zone, int migratetype) | |
1508 | { | |
1509 | struct zoneref *z; | |
1510 | struct page *page = NULL; | |
1511 | int classzone_idx; | |
1512 | struct zone *zone; | |
1513 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ | |
1514 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1515 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
1516 | ||
1517 | classzone_idx = zone_idx(preferred_zone); | |
1518 | zonelist_scan: | |
1519 | /* | |
1520 | * Scan zonelist, looking for a zone with enough free. | |
1521 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
1522 | */ | |
1523 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
1524 | high_zoneidx, nodemask) { | |
1525 | if (NUMA_BUILD && zlc_active && | |
1526 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1527 | continue; | |
1528 | if ((alloc_flags & ALLOC_CPUSET) && | |
1529 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
1530 | goto try_next_zone; | |
1531 | ||
1532 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); | |
1533 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
1534 | unsigned long mark; | |
1535 | int ret; | |
1536 | ||
1537 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; | |
1538 | if (zone_watermark_ok(zone, order, mark, | |
1539 | classzone_idx, alloc_flags)) | |
1540 | goto try_this_zone; | |
1541 | ||
1542 | if (zone_reclaim_mode == 0) | |
1543 | goto this_zone_full; | |
1544 | ||
1545 | ret = zone_reclaim(zone, gfp_mask, order); | |
1546 | switch (ret) { | |
1547 | case ZONE_RECLAIM_NOSCAN: | |
1548 | /* did not scan */ | |
1549 | goto try_next_zone; | |
1550 | case ZONE_RECLAIM_FULL: | |
1551 | /* scanned but unreclaimable */ | |
1552 | goto this_zone_full; | |
1553 | default: | |
1554 | /* did we reclaim enough */ | |
1555 | if (!zone_watermark_ok(zone, order, mark, | |
1556 | classzone_idx, alloc_flags)) | |
1557 | goto this_zone_full; | |
1558 | } | |
1559 | } | |
1560 | ||
1561 | try_this_zone: | |
1562 | page = buffered_rmqueue(preferred_zone, zone, order, | |
1563 | gfp_mask, migratetype); | |
1564 | if (page) | |
1565 | break; | |
1566 | this_zone_full: | |
1567 | if (NUMA_BUILD) | |
1568 | zlc_mark_zone_full(zonelist, z); | |
1569 | try_next_zone: | |
1570 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { | |
1571 | /* | |
1572 | * we do zlc_setup after the first zone is tried but only | |
1573 | * if there are multiple nodes make it worthwhile | |
1574 | */ | |
1575 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1576 | zlc_active = 1; | |
1577 | did_zlc_setup = 1; | |
1578 | } | |
1579 | } | |
1580 | ||
1581 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1582 | /* Disable zlc cache for second zonelist scan */ | |
1583 | zlc_active = 0; | |
1584 | goto zonelist_scan; | |
1585 | } | |
1586 | return page; | |
1587 | } | |
1588 | ||
1589 | static inline int | |
1590 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, | |
1591 | unsigned long pages_reclaimed) | |
1592 | { | |
1593 | /* Do not loop if specifically requested */ | |
1594 | if (gfp_mask & __GFP_NORETRY) | |
1595 | return 0; | |
1596 | ||
1597 | /* | |
1598 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER | |
1599 | * means __GFP_NOFAIL, but that may not be true in other | |
1600 | * implementations. | |
1601 | */ | |
1602 | if (order <= PAGE_ALLOC_COSTLY_ORDER) | |
1603 | return 1; | |
1604 | ||
1605 | /* | |
1606 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is | |
1607 | * specified, then we retry until we no longer reclaim any pages | |
1608 | * (above), or we've reclaimed an order of pages at least as | |
1609 | * large as the allocation's order. In both cases, if the | |
1610 | * allocation still fails, we stop retrying. | |
1611 | */ | |
1612 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) | |
1613 | return 1; | |
1614 | ||
1615 | /* | |
1616 | * Don't let big-order allocations loop unless the caller | |
1617 | * explicitly requests that. | |
1618 | */ | |
1619 | if (gfp_mask & __GFP_NOFAIL) | |
1620 | return 1; | |
1621 | ||
1622 | return 0; | |
1623 | } | |
1624 | ||
1625 | static inline struct page * | |
1626 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, | |
1627 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1628 | nodemask_t *nodemask, struct zone *preferred_zone, | |
1629 | int migratetype) | |
1630 | { | |
1631 | struct page *page; | |
1632 | ||
1633 | /* Acquire the OOM killer lock for the zones in zonelist */ | |
1634 | if (!try_set_zone_oom(zonelist, gfp_mask)) { | |
1635 | schedule_timeout_uninterruptible(1); | |
1636 | return NULL; | |
1637 | } | |
1638 | ||
1639 | /* | |
1640 | * Go through the zonelist yet one more time, keep very high watermark | |
1641 | * here, this is only to catch a parallel oom killing, we must fail if | |
1642 | * we're still under heavy pressure. | |
1643 | */ | |
1644 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, | |
1645 | order, zonelist, high_zoneidx, | |
1646 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, | |
1647 | preferred_zone, migratetype); | |
1648 | if (page) | |
1649 | goto out; | |
1650 | ||
1651 | if (!(gfp_mask & __GFP_NOFAIL)) { | |
1652 | /* The OOM killer will not help higher order allocs */ | |
1653 | if (order > PAGE_ALLOC_COSTLY_ORDER) | |
1654 | goto out; | |
1655 | /* | |
1656 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. | |
1657 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. | |
1658 | * The caller should handle page allocation failure by itself if | |
1659 | * it specifies __GFP_THISNODE. | |
1660 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. | |
1661 | */ | |
1662 | if (gfp_mask & __GFP_THISNODE) | |
1663 | goto out; | |
1664 | } | |
1665 | /* Exhausted what can be done so it's blamo time */ | |
1666 | out_of_memory(zonelist, gfp_mask, order, nodemask); | |
1667 | ||
1668 | out: | |
1669 | clear_zonelist_oom(zonelist, gfp_mask); | |
1670 | return page; | |
1671 | } | |
1672 | ||
1673 | /* The really slow allocator path where we enter direct reclaim */ | |
1674 | static inline struct page * | |
1675 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, | |
1676 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1677 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, | |
1678 | int migratetype, unsigned long *did_some_progress) | |
1679 | { | |
1680 | struct page *page = NULL; | |
1681 | struct reclaim_state reclaim_state; | |
1682 | struct task_struct *p = current; | |
1683 | ||
1684 | cond_resched(); | |
1685 | ||
1686 | /* We now go into synchronous reclaim */ | |
1687 | cpuset_memory_pressure_bump(); | |
1688 | p->flags |= PF_MEMALLOC; | |
1689 | lockdep_set_current_reclaim_state(gfp_mask); | |
1690 | reclaim_state.reclaimed_slab = 0; | |
1691 | p->reclaim_state = &reclaim_state; | |
1692 | ||
1693 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); | |
1694 | ||
1695 | p->reclaim_state = NULL; | |
1696 | lockdep_clear_current_reclaim_state(); | |
1697 | p->flags &= ~PF_MEMALLOC; | |
1698 | ||
1699 | cond_resched(); | |
1700 | ||
1701 | if (order != 0) | |
1702 | drain_all_pages(); | |
1703 | ||
1704 | if (likely(*did_some_progress)) | |
1705 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
1706 | zonelist, high_zoneidx, | |
1707 | alloc_flags, preferred_zone, | |
1708 | migratetype); | |
1709 | return page; | |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * This is called in the allocator slow-path if the allocation request is of | |
1714 | * sufficient urgency to ignore watermarks and take other desperate measures | |
1715 | */ | |
1716 | static inline struct page * | |
1717 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, | |
1718 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1719 | nodemask_t *nodemask, struct zone *preferred_zone, | |
1720 | int migratetype) | |
1721 | { | |
1722 | struct page *page; | |
1723 | ||
1724 | do { | |
1725 | page = get_page_from_freelist(gfp_mask, nodemask, order, | |
1726 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, | |
1727 | preferred_zone, migratetype); | |
1728 | ||
1729 | if (!page && gfp_mask & __GFP_NOFAIL) | |
1730 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
1731 | } while (!page && (gfp_mask & __GFP_NOFAIL)); | |
1732 | ||
1733 | return page; | |
1734 | } | |
1735 | ||
1736 | static inline | |
1737 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, | |
1738 | enum zone_type high_zoneidx) | |
1739 | { | |
1740 | struct zoneref *z; | |
1741 | struct zone *zone; | |
1742 | ||
1743 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) | |
1744 | wakeup_kswapd(zone, order); | |
1745 | } | |
1746 | ||
1747 | static inline int | |
1748 | gfp_to_alloc_flags(gfp_t gfp_mask) | |
1749 | { | |
1750 | struct task_struct *p = current; | |
1751 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; | |
1752 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1753 | ||
1754 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ | |
1755 | BUILD_BUG_ON(__GFP_HIGH != ALLOC_HIGH); | |
1756 | ||
1757 | /* | |
1758 | * The caller may dip into page reserves a bit more if the caller | |
1759 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1760 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1761 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1762 | */ | |
1763 | alloc_flags |= (gfp_mask & __GFP_HIGH); | |
1764 | ||
1765 | if (!wait) { | |
1766 | alloc_flags |= ALLOC_HARDER; | |
1767 | /* | |
1768 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
1769 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
1770 | */ | |
1771 | alloc_flags &= ~ALLOC_CPUSET; | |
1772 | } else if (unlikely(rt_task(p)) && !in_interrupt()) | |
1773 | alloc_flags |= ALLOC_HARDER; | |
1774 | ||
1775 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { | |
1776 | if (!in_interrupt() && | |
1777 | ((p->flags & PF_MEMALLOC) || | |
1778 | unlikely(test_thread_flag(TIF_MEMDIE)))) | |
1779 | alloc_flags |= ALLOC_NO_WATERMARKS; | |
1780 | } | |
1781 | ||
1782 | return alloc_flags; | |
1783 | } | |
1784 | ||
1785 | static inline struct page * | |
1786 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, | |
1787 | struct zonelist *zonelist, enum zone_type high_zoneidx, | |
1788 | nodemask_t *nodemask, struct zone *preferred_zone, | |
1789 | int migratetype) | |
1790 | { | |
1791 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1792 | struct page *page = NULL; | |
1793 | int alloc_flags; | |
1794 | unsigned long pages_reclaimed = 0; | |
1795 | unsigned long did_some_progress; | |
1796 | struct task_struct *p = current; | |
1797 | ||
1798 | /* | |
1799 | * In the slowpath, we sanity check order to avoid ever trying to | |
1800 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may | |
1801 | * be using allocators in order of preference for an area that is | |
1802 | * too large. | |
1803 | */ | |
1804 | if (order >= MAX_ORDER) { | |
1805 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); | |
1806 | return NULL; | |
1807 | } | |
1808 | ||
1809 | /* | |
1810 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1811 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1812 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1813 | * using a larger set of nodes after it has established that the | |
1814 | * allowed per node queues are empty and that nodes are | |
1815 | * over allocated. | |
1816 | */ | |
1817 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1818 | goto nopage; | |
1819 | ||
1820 | restart: | |
1821 | wake_all_kswapd(order, zonelist, high_zoneidx); | |
1822 | ||
1823 | /* | |
1824 | * OK, we're below the kswapd watermark and have kicked background | |
1825 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1826 | * to how we want to proceed. | |
1827 | */ | |
1828 | alloc_flags = gfp_to_alloc_flags(gfp_mask); | |
1829 | ||
1830 | /* This is the last chance, in general, before the goto nopage. */ | |
1831 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, | |
1832 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, | |
1833 | preferred_zone, migratetype); | |
1834 | if (page) | |
1835 | goto got_pg; | |
1836 | ||
1837 | rebalance: | |
1838 | /* Allocate without watermarks if the context allows */ | |
1839 | if (alloc_flags & ALLOC_NO_WATERMARKS) { | |
1840 | page = __alloc_pages_high_priority(gfp_mask, order, | |
1841 | zonelist, high_zoneidx, nodemask, | |
1842 | preferred_zone, migratetype); | |
1843 | if (page) | |
1844 | goto got_pg; | |
1845 | } | |
1846 | ||
1847 | /* Atomic allocations - we can't balance anything */ | |
1848 | if (!wait) | |
1849 | goto nopage; | |
1850 | ||
1851 | /* Avoid recursion of direct reclaim */ | |
1852 | if (p->flags & PF_MEMALLOC) | |
1853 | goto nopage; | |
1854 | ||
1855 | /* Avoid allocations with no watermarks from looping endlessly */ | |
1856 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) | |
1857 | goto nopage; | |
1858 | ||
1859 | /* Try direct reclaim and then allocating */ | |
1860 | page = __alloc_pages_direct_reclaim(gfp_mask, order, | |
1861 | zonelist, high_zoneidx, | |
1862 | nodemask, | |
1863 | alloc_flags, preferred_zone, | |
1864 | migratetype, &did_some_progress); | |
1865 | if (page) | |
1866 | goto got_pg; | |
1867 | ||
1868 | /* | |
1869 | * If we failed to make any progress reclaiming, then we are | |
1870 | * running out of options and have to consider going OOM | |
1871 | */ | |
1872 | if (!did_some_progress) { | |
1873 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
1874 | if (oom_killer_disabled) | |
1875 | goto nopage; | |
1876 | page = __alloc_pages_may_oom(gfp_mask, order, | |
1877 | zonelist, high_zoneidx, | |
1878 | nodemask, preferred_zone, | |
1879 | migratetype); | |
1880 | if (page) | |
1881 | goto got_pg; | |
1882 | ||
1883 | /* | |
1884 | * The OOM killer does not trigger for high-order | |
1885 | * ~__GFP_NOFAIL allocations so if no progress is being | |
1886 | * made, there are no other options and retrying is | |
1887 | * unlikely to help. | |
1888 | */ | |
1889 | if (order > PAGE_ALLOC_COSTLY_ORDER && | |
1890 | !(gfp_mask & __GFP_NOFAIL)) | |
1891 | goto nopage; | |
1892 | ||
1893 | goto restart; | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | /* Check if we should retry the allocation */ | |
1898 | pages_reclaimed += did_some_progress; | |
1899 | if (should_alloc_retry(gfp_mask, order, pages_reclaimed)) { | |
1900 | /* Wait for some write requests to complete then retry */ | |
1901 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
1902 | goto rebalance; | |
1903 | } | |
1904 | ||
1905 | nopage: | |
1906 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1907 | printk(KERN_WARNING "%s: page allocation failure." | |
1908 | " order:%d, mode:0x%x\n", | |
1909 | p->comm, order, gfp_mask); | |
1910 | dump_stack(); | |
1911 | show_mem(); | |
1912 | } | |
1913 | return page; | |
1914 | got_pg: | |
1915 | if (kmemcheck_enabled) | |
1916 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); | |
1917 | return page; | |
1918 | ||
1919 | } | |
1920 | ||
1921 | /* | |
1922 | * This is the 'heart' of the zoned buddy allocator. | |
1923 | */ | |
1924 | struct page * | |
1925 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, | |
1926 | struct zonelist *zonelist, nodemask_t *nodemask) | |
1927 | { | |
1928 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
1929 | struct zone *preferred_zone; | |
1930 | struct page *page; | |
1931 | int migratetype = allocflags_to_migratetype(gfp_mask); | |
1932 | ||
1933 | gfp_mask &= gfp_allowed_mask; | |
1934 | ||
1935 | lockdep_trace_alloc(gfp_mask); | |
1936 | ||
1937 | might_sleep_if(gfp_mask & __GFP_WAIT); | |
1938 | ||
1939 | if (should_fail_alloc_page(gfp_mask, order)) | |
1940 | return NULL; | |
1941 | ||
1942 | /* | |
1943 | * Check the zones suitable for the gfp_mask contain at least one | |
1944 | * valid zone. It's possible to have an empty zonelist as a result | |
1945 | * of GFP_THISNODE and a memoryless node | |
1946 | */ | |
1947 | if (unlikely(!zonelist->_zonerefs->zone)) | |
1948 | return NULL; | |
1949 | ||
1950 | /* The preferred zone is used for statistics later */ | |
1951 | first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone); | |
1952 | if (!preferred_zone) | |
1953 | return NULL; | |
1954 | ||
1955 | /* First allocation attempt */ | |
1956 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, | |
1957 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, | |
1958 | preferred_zone, migratetype); | |
1959 | if (unlikely(!page)) | |
1960 | page = __alloc_pages_slowpath(gfp_mask, order, | |
1961 | zonelist, high_zoneidx, nodemask, | |
1962 | preferred_zone, migratetype); | |
1963 | ||
1964 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); | |
1965 | return page; | |
1966 | } | |
1967 | EXPORT_SYMBOL(__alloc_pages_nodemask); | |
1968 | ||
1969 | /* | |
1970 | * Common helper functions. | |
1971 | */ | |
1972 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) | |
1973 | { | |
1974 | struct page *page; | |
1975 | ||
1976 | /* | |
1977 | * __get_free_pages() returns a 32-bit address, which cannot represent | |
1978 | * a highmem page | |
1979 | */ | |
1980 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
1981 | ||
1982 | page = alloc_pages(gfp_mask, order); | |
1983 | if (!page) | |
1984 | return 0; | |
1985 | return (unsigned long) page_address(page); | |
1986 | } | |
1987 | EXPORT_SYMBOL(__get_free_pages); | |
1988 | ||
1989 | unsigned long get_zeroed_page(gfp_t gfp_mask) | |
1990 | { | |
1991 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); | |
1992 | } | |
1993 | EXPORT_SYMBOL(get_zeroed_page); | |
1994 | ||
1995 | void __pagevec_free(struct pagevec *pvec) | |
1996 | { | |
1997 | int i = pagevec_count(pvec); | |
1998 | ||
1999 | while (--i >= 0) { | |
2000 | trace_mm_pagevec_free(pvec->pages[i], pvec->cold); | |
2001 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
2002 | } | |
2003 | } | |
2004 | ||
2005 | void __free_pages(struct page *page, unsigned int order) | |
2006 | { | |
2007 | if (put_page_testzero(page)) { | |
2008 | if (order == 0) | |
2009 | free_hot_cold_page(page, 0); | |
2010 | else | |
2011 | __free_pages_ok(page, order); | |
2012 | } | |
2013 | } | |
2014 | ||
2015 | EXPORT_SYMBOL(__free_pages); | |
2016 | ||
2017 | void free_pages(unsigned long addr, unsigned int order) | |
2018 | { | |
2019 | if (addr != 0) { | |
2020 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
2021 | __free_pages(virt_to_page((void *)addr), order); | |
2022 | } | |
2023 | } | |
2024 | ||
2025 | EXPORT_SYMBOL(free_pages); | |
2026 | ||
2027 | /** | |
2028 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. | |
2029 | * @size: the number of bytes to allocate | |
2030 | * @gfp_mask: GFP flags for the allocation | |
2031 | * | |
2032 | * This function is similar to alloc_pages(), except that it allocates the | |
2033 | * minimum number of pages to satisfy the request. alloc_pages() can only | |
2034 | * allocate memory in power-of-two pages. | |
2035 | * | |
2036 | * This function is also limited by MAX_ORDER. | |
2037 | * | |
2038 | * Memory allocated by this function must be released by free_pages_exact(). | |
2039 | */ | |
2040 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) | |
2041 | { | |
2042 | unsigned int order = get_order(size); | |
2043 | unsigned long addr; | |
2044 | ||
2045 | addr = __get_free_pages(gfp_mask, order); | |
2046 | if (addr) { | |
2047 | unsigned long alloc_end = addr + (PAGE_SIZE << order); | |
2048 | unsigned long used = addr + PAGE_ALIGN(size); | |
2049 | ||
2050 | split_page(virt_to_page((void *)addr), order); | |
2051 | while (used < alloc_end) { | |
2052 | free_page(used); | |
2053 | used += PAGE_SIZE; | |
2054 | } | |
2055 | } | |
2056 | ||
2057 | return (void *)addr; | |
2058 | } | |
2059 | EXPORT_SYMBOL(alloc_pages_exact); | |
2060 | ||
2061 | /** | |
2062 | * free_pages_exact - release memory allocated via alloc_pages_exact() | |
2063 | * @virt: the value returned by alloc_pages_exact. | |
2064 | * @size: size of allocation, same value as passed to alloc_pages_exact(). | |
2065 | * | |
2066 | * Release the memory allocated by a previous call to alloc_pages_exact. | |
2067 | */ | |
2068 | void free_pages_exact(void *virt, size_t size) | |
2069 | { | |
2070 | unsigned long addr = (unsigned long)virt; | |
2071 | unsigned long end = addr + PAGE_ALIGN(size); | |
2072 | ||
2073 | while (addr < end) { | |
2074 | free_page(addr); | |
2075 | addr += PAGE_SIZE; | |
2076 | } | |
2077 | } | |
2078 | EXPORT_SYMBOL(free_pages_exact); | |
2079 | ||
2080 | static unsigned int nr_free_zone_pages(int offset) | |
2081 | { | |
2082 | struct zoneref *z; | |
2083 | struct zone *zone; | |
2084 | ||
2085 | /* Just pick one node, since fallback list is circular */ | |
2086 | unsigned int sum = 0; | |
2087 | ||
2088 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); | |
2089 | ||
2090 | for_each_zone_zonelist(zone, z, zonelist, offset) { | |
2091 | unsigned long size = zone->present_pages; | |
2092 | unsigned long high = high_wmark_pages(zone); | |
2093 | if (size > high) | |
2094 | sum += size - high; | |
2095 | } | |
2096 | ||
2097 | return sum; | |
2098 | } | |
2099 | ||
2100 | /* | |
2101 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
2102 | */ | |
2103 | unsigned int nr_free_buffer_pages(void) | |
2104 | { | |
2105 | return nr_free_zone_pages(gfp_zone(GFP_USER)); | |
2106 | } | |
2107 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); | |
2108 | ||
2109 | /* | |
2110 | * Amount of free RAM allocatable within all zones | |
2111 | */ | |
2112 | unsigned int nr_free_pagecache_pages(void) | |
2113 | { | |
2114 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); | |
2115 | } | |
2116 | ||
2117 | static inline void show_node(struct zone *zone) | |
2118 | { | |
2119 | if (NUMA_BUILD) | |
2120 | printk("Node %d ", zone_to_nid(zone)); | |
2121 | } | |
2122 | ||
2123 | void si_meminfo(struct sysinfo *val) | |
2124 | { | |
2125 | val->totalram = totalram_pages; | |
2126 | val->sharedram = 0; | |
2127 | val->freeram = global_page_state(NR_FREE_PAGES); | |
2128 | val->bufferram = nr_blockdev_pages(); | |
2129 | val->totalhigh = totalhigh_pages; | |
2130 | val->freehigh = nr_free_highpages(); | |
2131 | val->mem_unit = PAGE_SIZE; | |
2132 | } | |
2133 | ||
2134 | EXPORT_SYMBOL(si_meminfo); | |
2135 | ||
2136 | #ifdef CONFIG_NUMA | |
2137 | void si_meminfo_node(struct sysinfo *val, int nid) | |
2138 | { | |
2139 | pg_data_t *pgdat = NODE_DATA(nid); | |
2140 | ||
2141 | val->totalram = pgdat->node_present_pages; | |
2142 | val->freeram = node_page_state(nid, NR_FREE_PAGES); | |
2143 | #ifdef CONFIG_HIGHMEM | |
2144 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
2145 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], | |
2146 | NR_FREE_PAGES); | |
2147 | #else | |
2148 | val->totalhigh = 0; | |
2149 | val->freehigh = 0; | |
2150 | #endif | |
2151 | val->mem_unit = PAGE_SIZE; | |
2152 | } | |
2153 | #endif | |
2154 | ||
2155 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
2156 | ||
2157 | /* | |
2158 | * Show free area list (used inside shift_scroll-lock stuff) | |
2159 | * We also calculate the percentage fragmentation. We do this by counting the | |
2160 | * memory on each free list with the exception of the first item on the list. | |
2161 | */ | |
2162 | void show_free_areas(void) | |
2163 | { | |
2164 | int cpu; | |
2165 | struct zone *zone; | |
2166 | ||
2167 | for_each_populated_zone(zone) { | |
2168 | show_node(zone); | |
2169 | printk("%s per-cpu:\n", zone->name); | |
2170 | ||
2171 | for_each_online_cpu(cpu) { | |
2172 | struct per_cpu_pageset *pageset; | |
2173 | ||
2174 | pageset = per_cpu_ptr(zone->pageset, cpu); | |
2175 | ||
2176 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", | |
2177 | cpu, pageset->pcp.high, | |
2178 | pageset->pcp.batch, pageset->pcp.count); | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" | |
2183 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" | |
2184 | " unevictable:%lu" | |
2185 | " dirty:%lu writeback:%lu unstable:%lu\n" | |
2186 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" | |
2187 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", | |
2188 | global_page_state(NR_ACTIVE_ANON), | |
2189 | global_page_state(NR_INACTIVE_ANON), | |
2190 | global_page_state(NR_ISOLATED_ANON), | |
2191 | global_page_state(NR_ACTIVE_FILE), | |
2192 | global_page_state(NR_INACTIVE_FILE), | |
2193 | global_page_state(NR_ISOLATED_FILE), | |
2194 | global_page_state(NR_UNEVICTABLE), | |
2195 | global_page_state(NR_FILE_DIRTY), | |
2196 | global_page_state(NR_WRITEBACK), | |
2197 | global_page_state(NR_UNSTABLE_NFS), | |
2198 | global_page_state(NR_FREE_PAGES), | |
2199 | global_page_state(NR_SLAB_RECLAIMABLE), | |
2200 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
2201 | global_page_state(NR_FILE_MAPPED), | |
2202 | global_page_state(NR_SHMEM), | |
2203 | global_page_state(NR_PAGETABLE), | |
2204 | global_page_state(NR_BOUNCE)); | |
2205 | ||
2206 | for_each_populated_zone(zone) { | |
2207 | int i; | |
2208 | ||
2209 | show_node(zone); | |
2210 | printk("%s" | |
2211 | " free:%lukB" | |
2212 | " min:%lukB" | |
2213 | " low:%lukB" | |
2214 | " high:%lukB" | |
2215 | " active_anon:%lukB" | |
2216 | " inactive_anon:%lukB" | |
2217 | " active_file:%lukB" | |
2218 | " inactive_file:%lukB" | |
2219 | " unevictable:%lukB" | |
2220 | " isolated(anon):%lukB" | |
2221 | " isolated(file):%lukB" | |
2222 | " present:%lukB" | |
2223 | " mlocked:%lukB" | |
2224 | " dirty:%lukB" | |
2225 | " writeback:%lukB" | |
2226 | " mapped:%lukB" | |
2227 | " shmem:%lukB" | |
2228 | " slab_reclaimable:%lukB" | |
2229 | " slab_unreclaimable:%lukB" | |
2230 | " kernel_stack:%lukB" | |
2231 | " pagetables:%lukB" | |
2232 | " unstable:%lukB" | |
2233 | " bounce:%lukB" | |
2234 | " writeback_tmp:%lukB" | |
2235 | " pages_scanned:%lu" | |
2236 | " all_unreclaimable? %s" | |
2237 | "\n", | |
2238 | zone->name, | |
2239 | K(zone_page_state(zone, NR_FREE_PAGES)), | |
2240 | K(min_wmark_pages(zone)), | |
2241 | K(low_wmark_pages(zone)), | |
2242 | K(high_wmark_pages(zone)), | |
2243 | K(zone_page_state(zone, NR_ACTIVE_ANON)), | |
2244 | K(zone_page_state(zone, NR_INACTIVE_ANON)), | |
2245 | K(zone_page_state(zone, NR_ACTIVE_FILE)), | |
2246 | K(zone_page_state(zone, NR_INACTIVE_FILE)), | |
2247 | K(zone_page_state(zone, NR_UNEVICTABLE)), | |
2248 | K(zone_page_state(zone, NR_ISOLATED_ANON)), | |
2249 | K(zone_page_state(zone, NR_ISOLATED_FILE)), | |
2250 | K(zone->present_pages), | |
2251 | K(zone_page_state(zone, NR_MLOCK)), | |
2252 | K(zone_page_state(zone, NR_FILE_DIRTY)), | |
2253 | K(zone_page_state(zone, NR_WRITEBACK)), | |
2254 | K(zone_page_state(zone, NR_FILE_MAPPED)), | |
2255 | K(zone_page_state(zone, NR_SHMEM)), | |
2256 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), | |
2257 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), | |
2258 | zone_page_state(zone, NR_KERNEL_STACK) * | |
2259 | THREAD_SIZE / 1024, | |
2260 | K(zone_page_state(zone, NR_PAGETABLE)), | |
2261 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), | |
2262 | K(zone_page_state(zone, NR_BOUNCE)), | |
2263 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), | |
2264 | zone->pages_scanned, | |
2265 | (zone->all_unreclaimable ? "yes" : "no") | |
2266 | ); | |
2267 | printk("lowmem_reserve[]:"); | |
2268 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2269 | printk(" %lu", zone->lowmem_reserve[i]); | |
2270 | printk("\n"); | |
2271 | } | |
2272 | ||
2273 | for_each_populated_zone(zone) { | |
2274 | unsigned long nr[MAX_ORDER], flags, order, total = 0; | |
2275 | ||
2276 | show_node(zone); | |
2277 | printk("%s: ", zone->name); | |
2278 | ||
2279 | spin_lock_irqsave(&zone->lock, flags); | |
2280 | for (order = 0; order < MAX_ORDER; order++) { | |
2281 | nr[order] = zone->free_area[order].nr_free; | |
2282 | total += nr[order] << order; | |
2283 | } | |
2284 | spin_unlock_irqrestore(&zone->lock, flags); | |
2285 | for (order = 0; order < MAX_ORDER; order++) | |
2286 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
2287 | printk("= %lukB\n", K(total)); | |
2288 | } | |
2289 | ||
2290 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); | |
2291 | ||
2292 | show_swap_cache_info(); | |
2293 | } | |
2294 | ||
2295 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) | |
2296 | { | |
2297 | zoneref->zone = zone; | |
2298 | zoneref->zone_idx = zone_idx(zone); | |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * Builds allocation fallback zone lists. | |
2303 | * | |
2304 | * Add all populated zones of a node to the zonelist. | |
2305 | */ | |
2306 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, | |
2307 | int nr_zones, enum zone_type zone_type) | |
2308 | { | |
2309 | struct zone *zone; | |
2310 | ||
2311 | BUG_ON(zone_type >= MAX_NR_ZONES); | |
2312 | zone_type++; | |
2313 | ||
2314 | do { | |
2315 | zone_type--; | |
2316 | zone = pgdat->node_zones + zone_type; | |
2317 | if (populated_zone(zone)) { | |
2318 | zoneref_set_zone(zone, | |
2319 | &zonelist->_zonerefs[nr_zones++]); | |
2320 | check_highest_zone(zone_type); | |
2321 | } | |
2322 | ||
2323 | } while (zone_type); | |
2324 | return nr_zones; | |
2325 | } | |
2326 | ||
2327 | ||
2328 | /* | |
2329 | * zonelist_order: | |
2330 | * 0 = automatic detection of better ordering. | |
2331 | * 1 = order by ([node] distance, -zonetype) | |
2332 | * 2 = order by (-zonetype, [node] distance) | |
2333 | * | |
2334 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create | |
2335 | * the same zonelist. So only NUMA can configure this param. | |
2336 | */ | |
2337 | #define ZONELIST_ORDER_DEFAULT 0 | |
2338 | #define ZONELIST_ORDER_NODE 1 | |
2339 | #define ZONELIST_ORDER_ZONE 2 | |
2340 | ||
2341 | /* zonelist order in the kernel. | |
2342 | * set_zonelist_order() will set this to NODE or ZONE. | |
2343 | */ | |
2344 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2345 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; | |
2346 | ||
2347 | ||
2348 | #ifdef CONFIG_NUMA | |
2349 | /* The value user specified ....changed by config */ | |
2350 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2351 | /* string for sysctl */ | |
2352 | #define NUMA_ZONELIST_ORDER_LEN 16 | |
2353 | char numa_zonelist_order[16] = "default"; | |
2354 | ||
2355 | /* | |
2356 | * interface for configure zonelist ordering. | |
2357 | * command line option "numa_zonelist_order" | |
2358 | * = "[dD]efault - default, automatic configuration. | |
2359 | * = "[nN]ode - order by node locality, then by zone within node | |
2360 | * = "[zZ]one - order by zone, then by locality within zone | |
2361 | */ | |
2362 | ||
2363 | static int __parse_numa_zonelist_order(char *s) | |
2364 | { | |
2365 | if (*s == 'd' || *s == 'D') { | |
2366 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; | |
2367 | } else if (*s == 'n' || *s == 'N') { | |
2368 | user_zonelist_order = ZONELIST_ORDER_NODE; | |
2369 | } else if (*s == 'z' || *s == 'Z') { | |
2370 | user_zonelist_order = ZONELIST_ORDER_ZONE; | |
2371 | } else { | |
2372 | printk(KERN_WARNING | |
2373 | "Ignoring invalid numa_zonelist_order value: " | |
2374 | "%s\n", s); | |
2375 | return -EINVAL; | |
2376 | } | |
2377 | return 0; | |
2378 | } | |
2379 | ||
2380 | static __init int setup_numa_zonelist_order(char *s) | |
2381 | { | |
2382 | if (s) | |
2383 | return __parse_numa_zonelist_order(s); | |
2384 | return 0; | |
2385 | } | |
2386 | early_param("numa_zonelist_order", setup_numa_zonelist_order); | |
2387 | ||
2388 | /* | |
2389 | * sysctl handler for numa_zonelist_order | |
2390 | */ | |
2391 | int numa_zonelist_order_handler(ctl_table *table, int write, | |
2392 | void __user *buffer, size_t *length, | |
2393 | loff_t *ppos) | |
2394 | { | |
2395 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; | |
2396 | int ret; | |
2397 | static DEFINE_MUTEX(zl_order_mutex); | |
2398 | ||
2399 | mutex_lock(&zl_order_mutex); | |
2400 | if (write) | |
2401 | strcpy(saved_string, (char*)table->data); | |
2402 | ret = proc_dostring(table, write, buffer, length, ppos); | |
2403 | if (ret) | |
2404 | goto out; | |
2405 | if (write) { | |
2406 | int oldval = user_zonelist_order; | |
2407 | if (__parse_numa_zonelist_order((char*)table->data)) { | |
2408 | /* | |
2409 | * bogus value. restore saved string | |
2410 | */ | |
2411 | strncpy((char*)table->data, saved_string, | |
2412 | NUMA_ZONELIST_ORDER_LEN); | |
2413 | user_zonelist_order = oldval; | |
2414 | } else if (oldval != user_zonelist_order) | |
2415 | build_all_zonelists(); | |
2416 | } | |
2417 | out: | |
2418 | mutex_unlock(&zl_order_mutex); | |
2419 | return ret; | |
2420 | } | |
2421 | ||
2422 | ||
2423 | #define MAX_NODE_LOAD (nr_online_nodes) | |
2424 | static int node_load[MAX_NUMNODES]; | |
2425 | ||
2426 | /** | |
2427 | * find_next_best_node - find the next node that should appear in a given node's fallback list | |
2428 | * @node: node whose fallback list we're appending | |
2429 | * @used_node_mask: nodemask_t of already used nodes | |
2430 | * | |
2431 | * We use a number of factors to determine which is the next node that should | |
2432 | * appear on a given node's fallback list. The node should not have appeared | |
2433 | * already in @node's fallback list, and it should be the next closest node | |
2434 | * according to the distance array (which contains arbitrary distance values | |
2435 | * from each node to each node in the system), and should also prefer nodes | |
2436 | * with no CPUs, since presumably they'll have very little allocation pressure | |
2437 | * on them otherwise. | |
2438 | * It returns -1 if no node is found. | |
2439 | */ | |
2440 | static int find_next_best_node(int node, nodemask_t *used_node_mask) | |
2441 | { | |
2442 | int n, val; | |
2443 | int min_val = INT_MAX; | |
2444 | int best_node = -1; | |
2445 | const struct cpumask *tmp = cpumask_of_node(0); | |
2446 | ||
2447 | /* Use the local node if we haven't already */ | |
2448 | if (!node_isset(node, *used_node_mask)) { | |
2449 | node_set(node, *used_node_mask); | |
2450 | return node; | |
2451 | } | |
2452 | ||
2453 | for_each_node_state(n, N_HIGH_MEMORY) { | |
2454 | ||
2455 | /* Don't want a node to appear more than once */ | |
2456 | if (node_isset(n, *used_node_mask)) | |
2457 | continue; | |
2458 | ||
2459 | /* Use the distance array to find the distance */ | |
2460 | val = node_distance(node, n); | |
2461 | ||
2462 | /* Penalize nodes under us ("prefer the next node") */ | |
2463 | val += (n < node); | |
2464 | ||
2465 | /* Give preference to headless and unused nodes */ | |
2466 | tmp = cpumask_of_node(n); | |
2467 | if (!cpumask_empty(tmp)) | |
2468 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
2469 | ||
2470 | /* Slight preference for less loaded node */ | |
2471 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
2472 | val += node_load[n]; | |
2473 | ||
2474 | if (val < min_val) { | |
2475 | min_val = val; | |
2476 | best_node = n; | |
2477 | } | |
2478 | } | |
2479 | ||
2480 | if (best_node >= 0) | |
2481 | node_set(best_node, *used_node_mask); | |
2482 | ||
2483 | return best_node; | |
2484 | } | |
2485 | ||
2486 | ||
2487 | /* | |
2488 | * Build zonelists ordered by node and zones within node. | |
2489 | * This results in maximum locality--normal zone overflows into local | |
2490 | * DMA zone, if any--but risks exhausting DMA zone. | |
2491 | */ | |
2492 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) | |
2493 | { | |
2494 | int j; | |
2495 | struct zonelist *zonelist; | |
2496 | ||
2497 | zonelist = &pgdat->node_zonelists[0]; | |
2498 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) | |
2499 | ; | |
2500 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2501 | MAX_NR_ZONES - 1); | |
2502 | zonelist->_zonerefs[j].zone = NULL; | |
2503 | zonelist->_zonerefs[j].zone_idx = 0; | |
2504 | } | |
2505 | ||
2506 | /* | |
2507 | * Build gfp_thisnode zonelists | |
2508 | */ | |
2509 | static void build_thisnode_zonelists(pg_data_t *pgdat) | |
2510 | { | |
2511 | int j; | |
2512 | struct zonelist *zonelist; | |
2513 | ||
2514 | zonelist = &pgdat->node_zonelists[1]; | |
2515 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
2516 | zonelist->_zonerefs[j].zone = NULL; | |
2517 | zonelist->_zonerefs[j].zone_idx = 0; | |
2518 | } | |
2519 | ||
2520 | /* | |
2521 | * Build zonelists ordered by zone and nodes within zones. | |
2522 | * This results in conserving DMA zone[s] until all Normal memory is | |
2523 | * exhausted, but results in overflowing to remote node while memory | |
2524 | * may still exist in local DMA zone. | |
2525 | */ | |
2526 | static int node_order[MAX_NUMNODES]; | |
2527 | ||
2528 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) | |
2529 | { | |
2530 | int pos, j, node; | |
2531 | int zone_type; /* needs to be signed */ | |
2532 | struct zone *z; | |
2533 | struct zonelist *zonelist; | |
2534 | ||
2535 | zonelist = &pgdat->node_zonelists[0]; | |
2536 | pos = 0; | |
2537 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { | |
2538 | for (j = 0; j < nr_nodes; j++) { | |
2539 | node = node_order[j]; | |
2540 | z = &NODE_DATA(node)->node_zones[zone_type]; | |
2541 | if (populated_zone(z)) { | |
2542 | zoneref_set_zone(z, | |
2543 | &zonelist->_zonerefs[pos++]); | |
2544 | check_highest_zone(zone_type); | |
2545 | } | |
2546 | } | |
2547 | } | |
2548 | zonelist->_zonerefs[pos].zone = NULL; | |
2549 | zonelist->_zonerefs[pos].zone_idx = 0; | |
2550 | } | |
2551 | ||
2552 | static int default_zonelist_order(void) | |
2553 | { | |
2554 | int nid, zone_type; | |
2555 | unsigned long low_kmem_size,total_size; | |
2556 | struct zone *z; | |
2557 | int average_size; | |
2558 | /* | |
2559 | * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem. | |
2560 | * If they are really small and used heavily, the system can fall | |
2561 | * into OOM very easily. | |
2562 | * This function detect ZONE_DMA/DMA32 size and confgigures zone order. | |
2563 | */ | |
2564 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ | |
2565 | low_kmem_size = 0; | |
2566 | total_size = 0; | |
2567 | for_each_online_node(nid) { | |
2568 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2569 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2570 | if (populated_zone(z)) { | |
2571 | if (zone_type < ZONE_NORMAL) | |
2572 | low_kmem_size += z->present_pages; | |
2573 | total_size += z->present_pages; | |
2574 | } | |
2575 | } | |
2576 | } | |
2577 | if (!low_kmem_size || /* there are no DMA area. */ | |
2578 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ | |
2579 | return ZONELIST_ORDER_NODE; | |
2580 | /* | |
2581 | * look into each node's config. | |
2582 | * If there is a node whose DMA/DMA32 memory is very big area on | |
2583 | * local memory, NODE_ORDER may be suitable. | |
2584 | */ | |
2585 | average_size = total_size / | |
2586 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); | |
2587 | for_each_online_node(nid) { | |
2588 | low_kmem_size = 0; | |
2589 | total_size = 0; | |
2590 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { | |
2591 | z = &NODE_DATA(nid)->node_zones[zone_type]; | |
2592 | if (populated_zone(z)) { | |
2593 | if (zone_type < ZONE_NORMAL) | |
2594 | low_kmem_size += z->present_pages; | |
2595 | total_size += z->present_pages; | |
2596 | } | |
2597 | } | |
2598 | if (low_kmem_size && | |
2599 | total_size > average_size && /* ignore small node */ | |
2600 | low_kmem_size > total_size * 70/100) | |
2601 | return ZONELIST_ORDER_NODE; | |
2602 | } | |
2603 | return ZONELIST_ORDER_ZONE; | |
2604 | } | |
2605 | ||
2606 | static void set_zonelist_order(void) | |
2607 | { | |
2608 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) | |
2609 | current_zonelist_order = default_zonelist_order(); | |
2610 | else | |
2611 | current_zonelist_order = user_zonelist_order; | |
2612 | } | |
2613 | ||
2614 | static void build_zonelists(pg_data_t *pgdat) | |
2615 | { | |
2616 | int j, node, load; | |
2617 | enum zone_type i; | |
2618 | nodemask_t used_mask; | |
2619 | int local_node, prev_node; | |
2620 | struct zonelist *zonelist; | |
2621 | int order = current_zonelist_order; | |
2622 | ||
2623 | /* initialize zonelists */ | |
2624 | for (i = 0; i < MAX_ZONELISTS; i++) { | |
2625 | zonelist = pgdat->node_zonelists + i; | |
2626 | zonelist->_zonerefs[0].zone = NULL; | |
2627 | zonelist->_zonerefs[0].zone_idx = 0; | |
2628 | } | |
2629 | ||
2630 | /* NUMA-aware ordering of nodes */ | |
2631 | local_node = pgdat->node_id; | |
2632 | load = nr_online_nodes; | |
2633 | prev_node = local_node; | |
2634 | nodes_clear(used_mask); | |
2635 | ||
2636 | memset(node_order, 0, sizeof(node_order)); | |
2637 | j = 0; | |
2638 | ||
2639 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
2640 | int distance = node_distance(local_node, node); | |
2641 | ||
2642 | /* | |
2643 | * If another node is sufficiently far away then it is better | |
2644 | * to reclaim pages in a zone before going off node. | |
2645 | */ | |
2646 | if (distance > RECLAIM_DISTANCE) | |
2647 | zone_reclaim_mode = 1; | |
2648 | ||
2649 | /* | |
2650 | * We don't want to pressure a particular node. | |
2651 | * So adding penalty to the first node in same | |
2652 | * distance group to make it round-robin. | |
2653 | */ | |
2654 | if (distance != node_distance(local_node, prev_node)) | |
2655 | node_load[node] = load; | |
2656 | ||
2657 | prev_node = node; | |
2658 | load--; | |
2659 | if (order == ZONELIST_ORDER_NODE) | |
2660 | build_zonelists_in_node_order(pgdat, node); | |
2661 | else | |
2662 | node_order[j++] = node; /* remember order */ | |
2663 | } | |
2664 | ||
2665 | if (order == ZONELIST_ORDER_ZONE) { | |
2666 | /* calculate node order -- i.e., DMA last! */ | |
2667 | build_zonelists_in_zone_order(pgdat, j); | |
2668 | } | |
2669 | ||
2670 | build_thisnode_zonelists(pgdat); | |
2671 | } | |
2672 | ||
2673 | /* Construct the zonelist performance cache - see further mmzone.h */ | |
2674 | static void build_zonelist_cache(pg_data_t *pgdat) | |
2675 | { | |
2676 | struct zonelist *zonelist; | |
2677 | struct zonelist_cache *zlc; | |
2678 | struct zoneref *z; | |
2679 | ||
2680 | zonelist = &pgdat->node_zonelists[0]; | |
2681 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
2682 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
2683 | for (z = zonelist->_zonerefs; z->zone; z++) | |
2684 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); | |
2685 | } | |
2686 | ||
2687 | ||
2688 | #else /* CONFIG_NUMA */ | |
2689 | ||
2690 | static void set_zonelist_order(void) | |
2691 | { | |
2692 | current_zonelist_order = ZONELIST_ORDER_ZONE; | |
2693 | } | |
2694 | ||
2695 | static void build_zonelists(pg_data_t *pgdat) | |
2696 | { | |
2697 | int node, local_node; | |
2698 | enum zone_type j; | |
2699 | struct zonelist *zonelist; | |
2700 | ||
2701 | local_node = pgdat->node_id; | |
2702 | ||
2703 | zonelist = &pgdat->node_zonelists[0]; | |
2704 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); | |
2705 | ||
2706 | /* | |
2707 | * Now we build the zonelist so that it contains the zones | |
2708 | * of all the other nodes. | |
2709 | * We don't want to pressure a particular node, so when | |
2710 | * building the zones for node N, we make sure that the | |
2711 | * zones coming right after the local ones are those from | |
2712 | * node N+1 (modulo N) | |
2713 | */ | |
2714 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
2715 | if (!node_online(node)) | |
2716 | continue; | |
2717 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2718 | MAX_NR_ZONES - 1); | |
2719 | } | |
2720 | for (node = 0; node < local_node; node++) { | |
2721 | if (!node_online(node)) | |
2722 | continue; | |
2723 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, | |
2724 | MAX_NR_ZONES - 1); | |
2725 | } | |
2726 | ||
2727 | zonelist->_zonerefs[j].zone = NULL; | |
2728 | zonelist->_zonerefs[j].zone_idx = 0; | |
2729 | } | |
2730 | ||
2731 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ | |
2732 | static void build_zonelist_cache(pg_data_t *pgdat) | |
2733 | { | |
2734 | pgdat->node_zonelists[0].zlcache_ptr = NULL; | |
2735 | } | |
2736 | ||
2737 | #endif /* CONFIG_NUMA */ | |
2738 | ||
2739 | /* | |
2740 | * Boot pageset table. One per cpu which is going to be used for all | |
2741 | * zones and all nodes. The parameters will be set in such a way | |
2742 | * that an item put on a list will immediately be handed over to | |
2743 | * the buddy list. This is safe since pageset manipulation is done | |
2744 | * with interrupts disabled. | |
2745 | * | |
2746 | * The boot_pagesets must be kept even after bootup is complete for | |
2747 | * unused processors and/or zones. They do play a role for bootstrapping | |
2748 | * hotplugged processors. | |
2749 | * | |
2750 | * zoneinfo_show() and maybe other functions do | |
2751 | * not check if the processor is online before following the pageset pointer. | |
2752 | * Other parts of the kernel may not check if the zone is available. | |
2753 | */ | |
2754 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); | |
2755 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); | |
2756 | ||
2757 | /* return values int ....just for stop_machine() */ | |
2758 | static int __build_all_zonelists(void *dummy) | |
2759 | { | |
2760 | int nid; | |
2761 | int cpu; | |
2762 | ||
2763 | #ifdef CONFIG_NUMA | |
2764 | memset(node_load, 0, sizeof(node_load)); | |
2765 | #endif | |
2766 | for_each_online_node(nid) { | |
2767 | pg_data_t *pgdat = NODE_DATA(nid); | |
2768 | ||
2769 | build_zonelists(pgdat); | |
2770 | build_zonelist_cache(pgdat); | |
2771 | } | |
2772 | ||
2773 | /* | |
2774 | * Initialize the boot_pagesets that are going to be used | |
2775 | * for bootstrapping processors. The real pagesets for | |
2776 | * each zone will be allocated later when the per cpu | |
2777 | * allocator is available. | |
2778 | * | |
2779 | * boot_pagesets are used also for bootstrapping offline | |
2780 | * cpus if the system is already booted because the pagesets | |
2781 | * are needed to initialize allocators on a specific cpu too. | |
2782 | * F.e. the percpu allocator needs the page allocator which | |
2783 | * needs the percpu allocator in order to allocate its pagesets | |
2784 | * (a chicken-egg dilemma). | |
2785 | */ | |
2786 | for_each_possible_cpu(cpu) | |
2787 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); | |
2788 | ||
2789 | return 0; | |
2790 | } | |
2791 | ||
2792 | void build_all_zonelists(void) | |
2793 | { | |
2794 | set_zonelist_order(); | |
2795 | ||
2796 | if (system_state == SYSTEM_BOOTING) { | |
2797 | __build_all_zonelists(NULL); | |
2798 | mminit_verify_zonelist(); | |
2799 | cpuset_init_current_mems_allowed(); | |
2800 | } else { | |
2801 | /* we have to stop all cpus to guarantee there is no user | |
2802 | of zonelist */ | |
2803 | stop_machine(__build_all_zonelists, NULL, NULL); | |
2804 | /* cpuset refresh routine should be here */ | |
2805 | } | |
2806 | vm_total_pages = nr_free_pagecache_pages(); | |
2807 | /* | |
2808 | * Disable grouping by mobility if the number of pages in the | |
2809 | * system is too low to allow the mechanism to work. It would be | |
2810 | * more accurate, but expensive to check per-zone. This check is | |
2811 | * made on memory-hotadd so a system can start with mobility | |
2812 | * disabled and enable it later | |
2813 | */ | |
2814 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) | |
2815 | page_group_by_mobility_disabled = 1; | |
2816 | else | |
2817 | page_group_by_mobility_disabled = 0; | |
2818 | ||
2819 | printk("Built %i zonelists in %s order, mobility grouping %s. " | |
2820 | "Total pages: %ld\n", | |
2821 | nr_online_nodes, | |
2822 | zonelist_order_name[current_zonelist_order], | |
2823 | page_group_by_mobility_disabled ? "off" : "on", | |
2824 | vm_total_pages); | |
2825 | #ifdef CONFIG_NUMA | |
2826 | printk("Policy zone: %s\n", zone_names[policy_zone]); | |
2827 | #endif | |
2828 | } | |
2829 | ||
2830 | /* | |
2831 | * Helper functions to size the waitqueue hash table. | |
2832 | * Essentially these want to choose hash table sizes sufficiently | |
2833 | * large so that collisions trying to wait on pages are rare. | |
2834 | * But in fact, the number of active page waitqueues on typical | |
2835 | * systems is ridiculously low, less than 200. So this is even | |
2836 | * conservative, even though it seems large. | |
2837 | * | |
2838 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
2839 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
2840 | */ | |
2841 | #define PAGES_PER_WAITQUEUE 256 | |
2842 | ||
2843 | #ifndef CONFIG_MEMORY_HOTPLUG | |
2844 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2845 | { | |
2846 | unsigned long size = 1; | |
2847 | ||
2848 | pages /= PAGES_PER_WAITQUEUE; | |
2849 | ||
2850 | while (size < pages) | |
2851 | size <<= 1; | |
2852 | ||
2853 | /* | |
2854 | * Once we have dozens or even hundreds of threads sleeping | |
2855 | * on IO we've got bigger problems than wait queue collision. | |
2856 | * Limit the size of the wait table to a reasonable size. | |
2857 | */ | |
2858 | size = min(size, 4096UL); | |
2859 | ||
2860 | return max(size, 4UL); | |
2861 | } | |
2862 | #else | |
2863 | /* | |
2864 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
2865 | * a suitable size for its wait_table. So we use the maximum size now. | |
2866 | * | |
2867 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
2868 | * | |
2869 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
2870 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
2871 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
2872 | * | |
2873 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
2874 | * or more by the traditional way. (See above). It equals: | |
2875 | * | |
2876 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
2877 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
2878 | * powerpc (64K page size) : = (32G +16M)byte. | |
2879 | */ | |
2880 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
2881 | { | |
2882 | return 4096UL; | |
2883 | } | |
2884 | #endif | |
2885 | ||
2886 | /* | |
2887 | * This is an integer logarithm so that shifts can be used later | |
2888 | * to extract the more random high bits from the multiplicative | |
2889 | * hash function before the remainder is taken. | |
2890 | */ | |
2891 | static inline unsigned long wait_table_bits(unsigned long size) | |
2892 | { | |
2893 | return ffz(~size); | |
2894 | } | |
2895 | ||
2896 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
2897 | ||
2898 | /* | |
2899 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number | |
2900 | * of blocks reserved is based on min_wmark_pages(zone). The memory within | |
2901 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes | |
2902 | * higher will lead to a bigger reserve which will get freed as contiguous | |
2903 | * blocks as reclaim kicks in | |
2904 | */ | |
2905 | static void setup_zone_migrate_reserve(struct zone *zone) | |
2906 | { | |
2907 | unsigned long start_pfn, pfn, end_pfn; | |
2908 | struct page *page; | |
2909 | unsigned long block_migratetype; | |
2910 | int reserve; | |
2911 | ||
2912 | /* Get the start pfn, end pfn and the number of blocks to reserve */ | |
2913 | start_pfn = zone->zone_start_pfn; | |
2914 | end_pfn = start_pfn + zone->spanned_pages; | |
2915 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> | |
2916 | pageblock_order; | |
2917 | ||
2918 | /* | |
2919 | * Reserve blocks are generally in place to help high-order atomic | |
2920 | * allocations that are short-lived. A min_free_kbytes value that | |
2921 | * would result in more than 2 reserve blocks for atomic allocations | |
2922 | * is assumed to be in place to help anti-fragmentation for the | |
2923 | * future allocation of hugepages at runtime. | |
2924 | */ | |
2925 | reserve = min(2, reserve); | |
2926 | ||
2927 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { | |
2928 | if (!pfn_valid(pfn)) | |
2929 | continue; | |
2930 | page = pfn_to_page(pfn); | |
2931 | ||
2932 | /* Watch out for overlapping nodes */ | |
2933 | if (page_to_nid(page) != zone_to_nid(zone)) | |
2934 | continue; | |
2935 | ||
2936 | /* Blocks with reserved pages will never free, skip them. */ | |
2937 | if (PageReserved(page)) | |
2938 | continue; | |
2939 | ||
2940 | block_migratetype = get_pageblock_migratetype(page); | |
2941 | ||
2942 | /* If this block is reserved, account for it */ | |
2943 | if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) { | |
2944 | reserve--; | |
2945 | continue; | |
2946 | } | |
2947 | ||
2948 | /* Suitable for reserving if this block is movable */ | |
2949 | if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) { | |
2950 | set_pageblock_migratetype(page, MIGRATE_RESERVE); | |
2951 | move_freepages_block(zone, page, MIGRATE_RESERVE); | |
2952 | reserve--; | |
2953 | continue; | |
2954 | } | |
2955 | ||
2956 | /* | |
2957 | * If the reserve is met and this is a previous reserved block, | |
2958 | * take it back | |
2959 | */ | |
2960 | if (block_migratetype == MIGRATE_RESERVE) { | |
2961 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
2962 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
2963 | } | |
2964 | } | |
2965 | } | |
2966 | ||
2967 | /* | |
2968 | * Initially all pages are reserved - free ones are freed | |
2969 | * up by free_all_bootmem() once the early boot process is | |
2970 | * done. Non-atomic initialization, single-pass. | |
2971 | */ | |
2972 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, | |
2973 | unsigned long start_pfn, enum memmap_context context) | |
2974 | { | |
2975 | struct page *page; | |
2976 | unsigned long end_pfn = start_pfn + size; | |
2977 | unsigned long pfn; | |
2978 | struct zone *z; | |
2979 | ||
2980 | if (highest_memmap_pfn < end_pfn - 1) | |
2981 | highest_memmap_pfn = end_pfn - 1; | |
2982 | ||
2983 | z = &NODE_DATA(nid)->node_zones[zone]; | |
2984 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
2985 | /* | |
2986 | * There can be holes in boot-time mem_map[]s | |
2987 | * handed to this function. They do not | |
2988 | * exist on hotplugged memory. | |
2989 | */ | |
2990 | if (context == MEMMAP_EARLY) { | |
2991 | if (!early_pfn_valid(pfn)) | |
2992 | continue; | |
2993 | if (!early_pfn_in_nid(pfn, nid)) | |
2994 | continue; | |
2995 | } | |
2996 | page = pfn_to_page(pfn); | |
2997 | set_page_links(page, zone, nid, pfn); | |
2998 | mminit_verify_page_links(page, zone, nid, pfn); | |
2999 | init_page_count(page); | |
3000 | reset_page_mapcount(page); | |
3001 | SetPageReserved(page); | |
3002 | /* | |
3003 | * Mark the block movable so that blocks are reserved for | |
3004 | * movable at startup. This will force kernel allocations | |
3005 | * to reserve their blocks rather than leaking throughout | |
3006 | * the address space during boot when many long-lived | |
3007 | * kernel allocations are made. Later some blocks near | |
3008 | * the start are marked MIGRATE_RESERVE by | |
3009 | * setup_zone_migrate_reserve() | |
3010 | * | |
3011 | * bitmap is created for zone's valid pfn range. but memmap | |
3012 | * can be created for invalid pages (for alignment) | |
3013 | * check here not to call set_pageblock_migratetype() against | |
3014 | * pfn out of zone. | |
3015 | */ | |
3016 | if ((z->zone_start_pfn <= pfn) | |
3017 | && (pfn < z->zone_start_pfn + z->spanned_pages) | |
3018 | && !(pfn & (pageblock_nr_pages - 1))) | |
3019 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
3020 | ||
3021 | INIT_LIST_HEAD(&page->lru); | |
3022 | #ifdef WANT_PAGE_VIRTUAL | |
3023 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
3024 | if (!is_highmem_idx(zone)) | |
3025 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
3026 | #endif | |
3027 | } | |
3028 | } | |
3029 | ||
3030 | static void __meminit zone_init_free_lists(struct zone *zone) | |
3031 | { | |
3032 | int order, t; | |
3033 | for_each_migratetype_order(order, t) { | |
3034 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); | |
3035 | zone->free_area[order].nr_free = 0; | |
3036 | } | |
3037 | } | |
3038 | ||
3039 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
3040 | #define memmap_init(size, nid, zone, start_pfn) \ | |
3041 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) | |
3042 | #endif | |
3043 | ||
3044 | static int zone_batchsize(struct zone *zone) | |
3045 | { | |
3046 | #ifdef CONFIG_MMU | |
3047 | int batch; | |
3048 | ||
3049 | /* | |
3050 | * The per-cpu-pages pools are set to around 1000th of the | |
3051 | * size of the zone. But no more than 1/2 of a meg. | |
3052 | * | |
3053 | * OK, so we don't know how big the cache is. So guess. | |
3054 | */ | |
3055 | batch = zone->present_pages / 1024; | |
3056 | if (batch * PAGE_SIZE > 512 * 1024) | |
3057 | batch = (512 * 1024) / PAGE_SIZE; | |
3058 | batch /= 4; /* We effectively *= 4 below */ | |
3059 | if (batch < 1) | |
3060 | batch = 1; | |
3061 | ||
3062 | /* | |
3063 | * Clamp the batch to a 2^n - 1 value. Having a power | |
3064 | * of 2 value was found to be more likely to have | |
3065 | * suboptimal cache aliasing properties in some cases. | |
3066 | * | |
3067 | * For example if 2 tasks are alternately allocating | |
3068 | * batches of pages, one task can end up with a lot | |
3069 | * of pages of one half of the possible page colors | |
3070 | * and the other with pages of the other colors. | |
3071 | */ | |
3072 | batch = rounddown_pow_of_two(batch + batch/2) - 1; | |
3073 | ||
3074 | return batch; | |
3075 | ||
3076 | #else | |
3077 | /* The deferral and batching of frees should be suppressed under NOMMU | |
3078 | * conditions. | |
3079 | * | |
3080 | * The problem is that NOMMU needs to be able to allocate large chunks | |
3081 | * of contiguous memory as there's no hardware page translation to | |
3082 | * assemble apparent contiguous memory from discontiguous pages. | |
3083 | * | |
3084 | * Queueing large contiguous runs of pages for batching, however, | |
3085 | * causes the pages to actually be freed in smaller chunks. As there | |
3086 | * can be a significant delay between the individual batches being | |
3087 | * recycled, this leads to the once large chunks of space being | |
3088 | * fragmented and becoming unavailable for high-order allocations. | |
3089 | */ | |
3090 | return 0; | |
3091 | #endif | |
3092 | } | |
3093 | ||
3094 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) | |
3095 | { | |
3096 | struct per_cpu_pages *pcp; | |
3097 | int migratetype; | |
3098 | ||
3099 | memset(p, 0, sizeof(*p)); | |
3100 | ||
3101 | pcp = &p->pcp; | |
3102 | pcp->count = 0; | |
3103 | pcp->high = 6 * batch; | |
3104 | pcp->batch = max(1UL, 1 * batch); | |
3105 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) | |
3106 | INIT_LIST_HEAD(&pcp->lists[migratetype]); | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
3111 | * to the value high for the pageset p. | |
3112 | */ | |
3113 | ||
3114 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
3115 | unsigned long high) | |
3116 | { | |
3117 | struct per_cpu_pages *pcp; | |
3118 | ||
3119 | pcp = &p->pcp; | |
3120 | pcp->high = high; | |
3121 | pcp->batch = max(1UL, high/4); | |
3122 | if ((high/4) > (PAGE_SHIFT * 8)) | |
3123 | pcp->batch = PAGE_SHIFT * 8; | |
3124 | } | |
3125 | ||
3126 | /* | |
3127 | * Allocate per cpu pagesets and initialize them. | |
3128 | * Before this call only boot pagesets were available. | |
3129 | * Boot pagesets will no longer be used by this processorr | |
3130 | * after setup_per_cpu_pageset(). | |
3131 | */ | |
3132 | void __init setup_per_cpu_pageset(void) | |
3133 | { | |
3134 | struct zone *zone; | |
3135 | int cpu; | |
3136 | ||
3137 | for_each_populated_zone(zone) { | |
3138 | zone->pageset = alloc_percpu(struct per_cpu_pageset); | |
3139 | ||
3140 | for_each_possible_cpu(cpu) { | |
3141 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); | |
3142 | ||
3143 | setup_pageset(pcp, zone_batchsize(zone)); | |
3144 | ||
3145 | if (percpu_pagelist_fraction) | |
3146 | setup_pagelist_highmark(pcp, | |
3147 | (zone->present_pages / | |
3148 | percpu_pagelist_fraction)); | |
3149 | } | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | static noinline __init_refok | |
3154 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) | |
3155 | { | |
3156 | int i; | |
3157 | struct pglist_data *pgdat = zone->zone_pgdat; | |
3158 | size_t alloc_size; | |
3159 | ||
3160 | /* | |
3161 | * The per-page waitqueue mechanism uses hashed waitqueues | |
3162 | * per zone. | |
3163 | */ | |
3164 | zone->wait_table_hash_nr_entries = | |
3165 | wait_table_hash_nr_entries(zone_size_pages); | |
3166 | zone->wait_table_bits = | |
3167 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
3168 | alloc_size = zone->wait_table_hash_nr_entries | |
3169 | * sizeof(wait_queue_head_t); | |
3170 | ||
3171 | if (!slab_is_available()) { | |
3172 | zone->wait_table = (wait_queue_head_t *) | |
3173 | alloc_bootmem_node(pgdat, alloc_size); | |
3174 | } else { | |
3175 | /* | |
3176 | * This case means that a zone whose size was 0 gets new memory | |
3177 | * via memory hot-add. | |
3178 | * But it may be the case that a new node was hot-added. In | |
3179 | * this case vmalloc() will not be able to use this new node's | |
3180 | * memory - this wait_table must be initialized to use this new | |
3181 | * node itself as well. | |
3182 | * To use this new node's memory, further consideration will be | |
3183 | * necessary. | |
3184 | */ | |
3185 | zone->wait_table = vmalloc(alloc_size); | |
3186 | } | |
3187 | if (!zone->wait_table) | |
3188 | return -ENOMEM; | |
3189 | ||
3190 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) | |
3191 | init_waitqueue_head(zone->wait_table + i); | |
3192 | ||
3193 | return 0; | |
3194 | } | |
3195 | ||
3196 | static int __zone_pcp_update(void *data) | |
3197 | { | |
3198 | struct zone *zone = data; | |
3199 | int cpu; | |
3200 | unsigned long batch = zone_batchsize(zone), flags; | |
3201 | ||
3202 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
3203 | struct per_cpu_pageset *pset; | |
3204 | struct per_cpu_pages *pcp; | |
3205 | ||
3206 | pset = per_cpu_ptr(zone->pageset, cpu); | |
3207 | pcp = &pset->pcp; | |
3208 | ||
3209 | local_irq_save(flags); | |
3210 | free_pcppages_bulk(zone, pcp->count, pcp); | |
3211 | setup_pageset(pset, batch); | |
3212 | local_irq_restore(flags); | |
3213 | } | |
3214 | return 0; | |
3215 | } | |
3216 | ||
3217 | void zone_pcp_update(struct zone *zone) | |
3218 | { | |
3219 | stop_machine(__zone_pcp_update, zone, NULL); | |
3220 | } | |
3221 | ||
3222 | static __meminit void zone_pcp_init(struct zone *zone) | |
3223 | { | |
3224 | /* | |
3225 | * per cpu subsystem is not up at this point. The following code | |
3226 | * relies on the ability of the linker to provide the | |
3227 | * offset of a (static) per cpu variable into the per cpu area. | |
3228 | */ | |
3229 | zone->pageset = &boot_pageset; | |
3230 | ||
3231 | if (zone->present_pages) | |
3232 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", | |
3233 | zone->name, zone->present_pages, | |
3234 | zone_batchsize(zone)); | |
3235 | } | |
3236 | ||
3237 | __meminit int init_currently_empty_zone(struct zone *zone, | |
3238 | unsigned long zone_start_pfn, | |
3239 | unsigned long size, | |
3240 | enum memmap_context context) | |
3241 | { | |
3242 | struct pglist_data *pgdat = zone->zone_pgdat; | |
3243 | int ret; | |
3244 | ret = zone_wait_table_init(zone, size); | |
3245 | if (ret) | |
3246 | return ret; | |
3247 | pgdat->nr_zones = zone_idx(zone) + 1; | |
3248 | ||
3249 | zone->zone_start_pfn = zone_start_pfn; | |
3250 | ||
3251 | mminit_dprintk(MMINIT_TRACE, "memmap_init", | |
3252 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", | |
3253 | pgdat->node_id, | |
3254 | (unsigned long)zone_idx(zone), | |
3255 | zone_start_pfn, (zone_start_pfn + size)); | |
3256 | ||
3257 | zone_init_free_lists(zone); | |
3258 | ||
3259 | return 0; | |
3260 | } | |
3261 | ||
3262 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
3263 | /* | |
3264 | * Basic iterator support. Return the first range of PFNs for a node | |
3265 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
3266 | */ | |
3267 | static int __meminit first_active_region_index_in_nid(int nid) | |
3268 | { | |
3269 | int i; | |
3270 | ||
3271 | for (i = 0; i < nr_nodemap_entries; i++) | |
3272 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
3273 | return i; | |
3274 | ||
3275 | return -1; | |
3276 | } | |
3277 | ||
3278 | /* | |
3279 | * Basic iterator support. Return the next active range of PFNs for a node | |
3280 | * Note: nid == MAX_NUMNODES returns next region regardless of node | |
3281 | */ | |
3282 | static int __meminit next_active_region_index_in_nid(int index, int nid) | |
3283 | { | |
3284 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
3285 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
3286 | return index; | |
3287 | ||
3288 | return -1; | |
3289 | } | |
3290 | ||
3291 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
3292 | /* | |
3293 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
3294 | * Architectures may implement their own version but if add_active_range() | |
3295 | * was used and there are no special requirements, this is a convenient | |
3296 | * alternative | |
3297 | */ | |
3298 | int __meminit __early_pfn_to_nid(unsigned long pfn) | |
3299 | { | |
3300 | int i; | |
3301 | ||
3302 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3303 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
3304 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3305 | ||
3306 | if (start_pfn <= pfn && pfn < end_pfn) | |
3307 | return early_node_map[i].nid; | |
3308 | } | |
3309 | /* This is a memory hole */ | |
3310 | return -1; | |
3311 | } | |
3312 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
3313 | ||
3314 | int __meminit early_pfn_to_nid(unsigned long pfn) | |
3315 | { | |
3316 | int nid; | |
3317 | ||
3318 | nid = __early_pfn_to_nid(pfn); | |
3319 | if (nid >= 0) | |
3320 | return nid; | |
3321 | /* just returns 0 */ | |
3322 | return 0; | |
3323 | } | |
3324 | ||
3325 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES | |
3326 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) | |
3327 | { | |
3328 | int nid; | |
3329 | ||
3330 | nid = __early_pfn_to_nid(pfn); | |
3331 | if (nid >= 0 && nid != node) | |
3332 | return false; | |
3333 | return true; | |
3334 | } | |
3335 | #endif | |
3336 | ||
3337 | /* Basic iterator support to walk early_node_map[] */ | |
3338 | #define for_each_active_range_index_in_nid(i, nid) \ | |
3339 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
3340 | i = next_active_region_index_in_nid(i, nid)) | |
3341 | ||
3342 | /** | |
3343 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
3344 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. | |
3345 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
3346 | * | |
3347 | * If an architecture guarantees that all ranges registered with | |
3348 | * add_active_ranges() contain no holes and may be freed, this | |
3349 | * this function may be used instead of calling free_bootmem() manually. | |
3350 | */ | |
3351 | void __init free_bootmem_with_active_regions(int nid, | |
3352 | unsigned long max_low_pfn) | |
3353 | { | |
3354 | int i; | |
3355 | ||
3356 | for_each_active_range_index_in_nid(i, nid) { | |
3357 | unsigned long size_pages = 0; | |
3358 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
3359 | ||
3360 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
3361 | continue; | |
3362 | ||
3363 | if (end_pfn > max_low_pfn) | |
3364 | end_pfn = max_low_pfn; | |
3365 | ||
3366 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
3367 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
3368 | PFN_PHYS(early_node_map[i].start_pfn), | |
3369 | size_pages << PAGE_SHIFT); | |
3370 | } | |
3371 | } | |
3372 | ||
3373 | int __init add_from_early_node_map(struct range *range, int az, | |
3374 | int nr_range, int nid) | |
3375 | { | |
3376 | int i; | |
3377 | u64 start, end; | |
3378 | ||
3379 | /* need to go over early_node_map to find out good range for node */ | |
3380 | for_each_active_range_index_in_nid(i, nid) { | |
3381 | start = early_node_map[i].start_pfn; | |
3382 | end = early_node_map[i].end_pfn; | |
3383 | nr_range = add_range(range, az, nr_range, start, end); | |
3384 | } | |
3385 | return nr_range; | |
3386 | } | |
3387 | ||
3388 | #ifdef CONFIG_NO_BOOTMEM | |
3389 | void * __init __alloc_memory_core_early(int nid, u64 size, u64 align, | |
3390 | u64 goal, u64 limit) | |
3391 | { | |
3392 | int i; | |
3393 | void *ptr; | |
3394 | ||
3395 | /* need to go over early_node_map to find out good range for node */ | |
3396 | for_each_active_range_index_in_nid(i, nid) { | |
3397 | u64 addr; | |
3398 | u64 ei_start, ei_last; | |
3399 | ||
3400 | ei_last = early_node_map[i].end_pfn; | |
3401 | ei_last <<= PAGE_SHIFT; | |
3402 | ei_start = early_node_map[i].start_pfn; | |
3403 | ei_start <<= PAGE_SHIFT; | |
3404 | addr = find_early_area(ei_start, ei_last, | |
3405 | goal, limit, size, align); | |
3406 | ||
3407 | if (addr == -1ULL) | |
3408 | continue; | |
3409 | ||
3410 | #if 0 | |
3411 | printk(KERN_DEBUG "alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n", | |
3412 | nid, | |
3413 | ei_start, ei_last, goal, limit, size, | |
3414 | align, addr); | |
3415 | #endif | |
3416 | ||
3417 | ptr = phys_to_virt(addr); | |
3418 | memset(ptr, 0, size); | |
3419 | reserve_early_without_check(addr, addr + size, "BOOTMEM"); | |
3420 | return ptr; | |
3421 | } | |
3422 | ||
3423 | return NULL; | |
3424 | } | |
3425 | #endif | |
3426 | ||
3427 | ||
3428 | void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) | |
3429 | { | |
3430 | int i; | |
3431 | int ret; | |
3432 | ||
3433 | for_each_active_range_index_in_nid(i, nid) { | |
3434 | ret = work_fn(early_node_map[i].start_pfn, | |
3435 | early_node_map[i].end_pfn, data); | |
3436 | if (ret) | |
3437 | break; | |
3438 | } | |
3439 | } | |
3440 | /** | |
3441 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
3442 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. | |
3443 | * | |
3444 | * If an architecture guarantees that all ranges registered with | |
3445 | * add_active_ranges() contain no holes and may be freed, this | |
3446 | * function may be used instead of calling memory_present() manually. | |
3447 | */ | |
3448 | void __init sparse_memory_present_with_active_regions(int nid) | |
3449 | { | |
3450 | int i; | |
3451 | ||
3452 | for_each_active_range_index_in_nid(i, nid) | |
3453 | memory_present(early_node_map[i].nid, | |
3454 | early_node_map[i].start_pfn, | |
3455 | early_node_map[i].end_pfn); | |
3456 | } | |
3457 | ||
3458 | /** | |
3459 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
3460 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. | |
3461 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
3462 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
3463 | * | |
3464 | * It returns the start and end page frame of a node based on information | |
3465 | * provided by an arch calling add_active_range(). If called for a node | |
3466 | * with no available memory, a warning is printed and the start and end | |
3467 | * PFNs will be 0. | |
3468 | */ | |
3469 | void __meminit get_pfn_range_for_nid(unsigned int nid, | |
3470 | unsigned long *start_pfn, unsigned long *end_pfn) | |
3471 | { | |
3472 | int i; | |
3473 | *start_pfn = -1UL; | |
3474 | *end_pfn = 0; | |
3475 | ||
3476 | for_each_active_range_index_in_nid(i, nid) { | |
3477 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
3478 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
3479 | } | |
3480 | ||
3481 | if (*start_pfn == -1UL) | |
3482 | *start_pfn = 0; | |
3483 | } | |
3484 | ||
3485 | /* | |
3486 | * This finds a zone that can be used for ZONE_MOVABLE pages. The | |
3487 | * assumption is made that zones within a node are ordered in monotonic | |
3488 | * increasing memory addresses so that the "highest" populated zone is used | |
3489 | */ | |
3490 | static void __init find_usable_zone_for_movable(void) | |
3491 | { | |
3492 | int zone_index; | |
3493 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { | |
3494 | if (zone_index == ZONE_MOVABLE) | |
3495 | continue; | |
3496 | ||
3497 | if (arch_zone_highest_possible_pfn[zone_index] > | |
3498 | arch_zone_lowest_possible_pfn[zone_index]) | |
3499 | break; | |
3500 | } | |
3501 | ||
3502 | VM_BUG_ON(zone_index == -1); | |
3503 | movable_zone = zone_index; | |
3504 | } | |
3505 | ||
3506 | /* | |
3507 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE | |
3508 | * because it is sized independant of architecture. Unlike the other zones, | |
3509 | * the starting point for ZONE_MOVABLE is not fixed. It may be different | |
3510 | * in each node depending on the size of each node and how evenly kernelcore | |
3511 | * is distributed. This helper function adjusts the zone ranges | |
3512 | * provided by the architecture for a given node by using the end of the | |
3513 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that | |
3514 | * zones within a node are in order of monotonic increases memory addresses | |
3515 | */ | |
3516 | static void __meminit adjust_zone_range_for_zone_movable(int nid, | |
3517 | unsigned long zone_type, | |
3518 | unsigned long node_start_pfn, | |
3519 | unsigned long node_end_pfn, | |
3520 | unsigned long *zone_start_pfn, | |
3521 | unsigned long *zone_end_pfn) | |
3522 | { | |
3523 | /* Only adjust if ZONE_MOVABLE is on this node */ | |
3524 | if (zone_movable_pfn[nid]) { | |
3525 | /* Size ZONE_MOVABLE */ | |
3526 | if (zone_type == ZONE_MOVABLE) { | |
3527 | *zone_start_pfn = zone_movable_pfn[nid]; | |
3528 | *zone_end_pfn = min(node_end_pfn, | |
3529 | arch_zone_highest_possible_pfn[movable_zone]); | |
3530 | ||
3531 | /* Adjust for ZONE_MOVABLE starting within this range */ | |
3532 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && | |
3533 | *zone_end_pfn > zone_movable_pfn[nid]) { | |
3534 | *zone_end_pfn = zone_movable_pfn[nid]; | |
3535 | ||
3536 | /* Check if this whole range is within ZONE_MOVABLE */ | |
3537 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) | |
3538 | *zone_start_pfn = *zone_end_pfn; | |
3539 | } | |
3540 | } | |
3541 | ||
3542 | /* | |
3543 | * Return the number of pages a zone spans in a node, including holes | |
3544 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
3545 | */ | |
3546 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, | |
3547 | unsigned long zone_type, | |
3548 | unsigned long *ignored) | |
3549 | { | |
3550 | unsigned long node_start_pfn, node_end_pfn; | |
3551 | unsigned long zone_start_pfn, zone_end_pfn; | |
3552 | ||
3553 | /* Get the start and end of the node and zone */ | |
3554 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3555 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
3556 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
3557 | adjust_zone_range_for_zone_movable(nid, zone_type, | |
3558 | node_start_pfn, node_end_pfn, | |
3559 | &zone_start_pfn, &zone_end_pfn); | |
3560 | ||
3561 | /* Check that this node has pages within the zone's required range */ | |
3562 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
3563 | return 0; | |
3564 | ||
3565 | /* Move the zone boundaries inside the node if necessary */ | |
3566 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
3567 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
3568 | ||
3569 | /* Return the spanned pages */ | |
3570 | return zone_end_pfn - zone_start_pfn; | |
3571 | } | |
3572 | ||
3573 | /* | |
3574 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
3575 | * then all holes in the requested range will be accounted for. | |
3576 | */ | |
3577 | unsigned long __meminit __absent_pages_in_range(int nid, | |
3578 | unsigned long range_start_pfn, | |
3579 | unsigned long range_end_pfn) | |
3580 | { | |
3581 | int i = 0; | |
3582 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
3583 | unsigned long start_pfn; | |
3584 | ||
3585 | /* Find the end_pfn of the first active range of pfns in the node */ | |
3586 | i = first_active_region_index_in_nid(nid); | |
3587 | if (i == -1) | |
3588 | return 0; | |
3589 | ||
3590 | prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3591 | ||
3592 | /* Account for ranges before physical memory on this node */ | |
3593 | if (early_node_map[i].start_pfn > range_start_pfn) | |
3594 | hole_pages = prev_end_pfn - range_start_pfn; | |
3595 | ||
3596 | /* Find all holes for the zone within the node */ | |
3597 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
3598 | ||
3599 | /* No need to continue if prev_end_pfn is outside the zone */ | |
3600 | if (prev_end_pfn >= range_end_pfn) | |
3601 | break; | |
3602 | ||
3603 | /* Make sure the end of the zone is not within the hole */ | |
3604 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
3605 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
3606 | ||
3607 | /* Update the hole size cound and move on */ | |
3608 | if (start_pfn > range_start_pfn) { | |
3609 | BUG_ON(prev_end_pfn > start_pfn); | |
3610 | hole_pages += start_pfn - prev_end_pfn; | |
3611 | } | |
3612 | prev_end_pfn = early_node_map[i].end_pfn; | |
3613 | } | |
3614 | ||
3615 | /* Account for ranges past physical memory on this node */ | |
3616 | if (range_end_pfn > prev_end_pfn) | |
3617 | hole_pages += range_end_pfn - | |
3618 | max(range_start_pfn, prev_end_pfn); | |
3619 | ||
3620 | return hole_pages; | |
3621 | } | |
3622 | ||
3623 | /** | |
3624 | * absent_pages_in_range - Return number of page frames in holes within a range | |
3625 | * @start_pfn: The start PFN to start searching for holes | |
3626 | * @end_pfn: The end PFN to stop searching for holes | |
3627 | * | |
3628 | * It returns the number of pages frames in memory holes within a range. | |
3629 | */ | |
3630 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
3631 | unsigned long end_pfn) | |
3632 | { | |
3633 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
3634 | } | |
3635 | ||
3636 | /* Return the number of page frames in holes in a zone on a node */ | |
3637 | static unsigned long __meminit zone_absent_pages_in_node(int nid, | |
3638 | unsigned long zone_type, | |
3639 | unsigned long *ignored) | |
3640 | { | |
3641 | unsigned long node_start_pfn, node_end_pfn; | |
3642 | unsigned long zone_start_pfn, zone_end_pfn; | |
3643 | ||
3644 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
3645 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
3646 | node_start_pfn); | |
3647 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
3648 | node_end_pfn); | |
3649 | ||
3650 | adjust_zone_range_for_zone_movable(nid, zone_type, | |
3651 | node_start_pfn, node_end_pfn, | |
3652 | &zone_start_pfn, &zone_end_pfn); | |
3653 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); | |
3654 | } | |
3655 | ||
3656 | #else | |
3657 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, | |
3658 | unsigned long zone_type, | |
3659 | unsigned long *zones_size) | |
3660 | { | |
3661 | return zones_size[zone_type]; | |
3662 | } | |
3663 | ||
3664 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, | |
3665 | unsigned long zone_type, | |
3666 | unsigned long *zholes_size) | |
3667 | { | |
3668 | if (!zholes_size) | |
3669 | return 0; | |
3670 | ||
3671 | return zholes_size[zone_type]; | |
3672 | } | |
3673 | ||
3674 | #endif | |
3675 | ||
3676 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, | |
3677 | unsigned long *zones_size, unsigned long *zholes_size) | |
3678 | { | |
3679 | unsigned long realtotalpages, totalpages = 0; | |
3680 | enum zone_type i; | |
3681 | ||
3682 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3683 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
3684 | zones_size); | |
3685 | pgdat->node_spanned_pages = totalpages; | |
3686 | ||
3687 | realtotalpages = totalpages; | |
3688 | for (i = 0; i < MAX_NR_ZONES; i++) | |
3689 | realtotalpages -= | |
3690 | zone_absent_pages_in_node(pgdat->node_id, i, | |
3691 | zholes_size); | |
3692 | pgdat->node_present_pages = realtotalpages; | |
3693 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
3694 | realtotalpages); | |
3695 | } | |
3696 | ||
3697 | #ifndef CONFIG_SPARSEMEM | |
3698 | /* | |
3699 | * Calculate the size of the zone->blockflags rounded to an unsigned long | |
3700 | * Start by making sure zonesize is a multiple of pageblock_order by rounding | |
3701 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally | |
3702 | * round what is now in bits to nearest long in bits, then return it in | |
3703 | * bytes. | |
3704 | */ | |
3705 | static unsigned long __init usemap_size(unsigned long zonesize) | |
3706 | { | |
3707 | unsigned long usemapsize; | |
3708 | ||
3709 | usemapsize = roundup(zonesize, pageblock_nr_pages); | |
3710 | usemapsize = usemapsize >> pageblock_order; | |
3711 | usemapsize *= NR_PAGEBLOCK_BITS; | |
3712 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); | |
3713 | ||
3714 | return usemapsize / 8; | |
3715 | } | |
3716 | ||
3717 | static void __init setup_usemap(struct pglist_data *pgdat, | |
3718 | struct zone *zone, unsigned long zonesize) | |
3719 | { | |
3720 | unsigned long usemapsize = usemap_size(zonesize); | |
3721 | zone->pageblock_flags = NULL; | |
3722 | if (usemapsize) | |
3723 | zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize); | |
3724 | } | |
3725 | #else | |
3726 | static void inline setup_usemap(struct pglist_data *pgdat, | |
3727 | struct zone *zone, unsigned long zonesize) {} | |
3728 | #endif /* CONFIG_SPARSEMEM */ | |
3729 | ||
3730 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE | |
3731 | ||
3732 | /* Return a sensible default order for the pageblock size. */ | |
3733 | static inline int pageblock_default_order(void) | |
3734 | { | |
3735 | if (HPAGE_SHIFT > PAGE_SHIFT) | |
3736 | return HUGETLB_PAGE_ORDER; | |
3737 | ||
3738 | return MAX_ORDER-1; | |
3739 | } | |
3740 | ||
3741 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ | |
3742 | static inline void __init set_pageblock_order(unsigned int order) | |
3743 | { | |
3744 | /* Check that pageblock_nr_pages has not already been setup */ | |
3745 | if (pageblock_order) | |
3746 | return; | |
3747 | ||
3748 | /* | |
3749 | * Assume the largest contiguous order of interest is a huge page. | |
3750 | * This value may be variable depending on boot parameters on IA64 | |
3751 | */ | |
3752 | pageblock_order = order; | |
3753 | } | |
3754 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3755 | ||
3756 | /* | |
3757 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() | |
3758 | * and pageblock_default_order() are unused as pageblock_order is set | |
3759 | * at compile-time. See include/linux/pageblock-flags.h for the values of | |
3760 | * pageblock_order based on the kernel config | |
3761 | */ | |
3762 | static inline int pageblock_default_order(unsigned int order) | |
3763 | { | |
3764 | return MAX_ORDER-1; | |
3765 | } | |
3766 | #define set_pageblock_order(x) do {} while (0) | |
3767 | ||
3768 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ | |
3769 | ||
3770 | /* | |
3771 | * Set up the zone data structures: | |
3772 | * - mark all pages reserved | |
3773 | * - mark all memory queues empty | |
3774 | * - clear the memory bitmaps | |
3775 | */ | |
3776 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, | |
3777 | unsigned long *zones_size, unsigned long *zholes_size) | |
3778 | { | |
3779 | enum zone_type j; | |
3780 | int nid = pgdat->node_id; | |
3781 | unsigned long zone_start_pfn = pgdat->node_start_pfn; | |
3782 | int ret; | |
3783 | ||
3784 | pgdat_resize_init(pgdat); | |
3785 | pgdat->nr_zones = 0; | |
3786 | init_waitqueue_head(&pgdat->kswapd_wait); | |
3787 | pgdat->kswapd_max_order = 0; | |
3788 | pgdat_page_cgroup_init(pgdat); | |
3789 | ||
3790 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3791 | struct zone *zone = pgdat->node_zones + j; | |
3792 | unsigned long size, realsize, memmap_pages; | |
3793 | enum lru_list l; | |
3794 | ||
3795 | size = zone_spanned_pages_in_node(nid, j, zones_size); | |
3796 | realsize = size - zone_absent_pages_in_node(nid, j, | |
3797 | zholes_size); | |
3798 | ||
3799 | /* | |
3800 | * Adjust realsize so that it accounts for how much memory | |
3801 | * is used by this zone for memmap. This affects the watermark | |
3802 | * and per-cpu initialisations | |
3803 | */ | |
3804 | memmap_pages = | |
3805 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; | |
3806 | if (realsize >= memmap_pages) { | |
3807 | realsize -= memmap_pages; | |
3808 | if (memmap_pages) | |
3809 | printk(KERN_DEBUG | |
3810 | " %s zone: %lu pages used for memmap\n", | |
3811 | zone_names[j], memmap_pages); | |
3812 | } else | |
3813 | printk(KERN_WARNING | |
3814 | " %s zone: %lu pages exceeds realsize %lu\n", | |
3815 | zone_names[j], memmap_pages, realsize); | |
3816 | ||
3817 | /* Account for reserved pages */ | |
3818 | if (j == 0 && realsize > dma_reserve) { | |
3819 | realsize -= dma_reserve; | |
3820 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", | |
3821 | zone_names[0], dma_reserve); | |
3822 | } | |
3823 | ||
3824 | if (!is_highmem_idx(j)) | |
3825 | nr_kernel_pages += realsize; | |
3826 | nr_all_pages += realsize; | |
3827 | ||
3828 | zone->spanned_pages = size; | |
3829 | zone->present_pages = realsize; | |
3830 | #ifdef CONFIG_NUMA | |
3831 | zone->node = nid; | |
3832 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) | |
3833 | / 100; | |
3834 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; | |
3835 | #endif | |
3836 | zone->name = zone_names[j]; | |
3837 | spin_lock_init(&zone->lock); | |
3838 | spin_lock_init(&zone->lru_lock); | |
3839 | zone_seqlock_init(zone); | |
3840 | zone->zone_pgdat = pgdat; | |
3841 | ||
3842 | zone->prev_priority = DEF_PRIORITY; | |
3843 | ||
3844 | zone_pcp_init(zone); | |
3845 | for_each_lru(l) { | |
3846 | INIT_LIST_HEAD(&zone->lru[l].list); | |
3847 | zone->reclaim_stat.nr_saved_scan[l] = 0; | |
3848 | } | |
3849 | zone->reclaim_stat.recent_rotated[0] = 0; | |
3850 | zone->reclaim_stat.recent_rotated[1] = 0; | |
3851 | zone->reclaim_stat.recent_scanned[0] = 0; | |
3852 | zone->reclaim_stat.recent_scanned[1] = 0; | |
3853 | zap_zone_vm_stats(zone); | |
3854 | zone->flags = 0; | |
3855 | if (!size) | |
3856 | continue; | |
3857 | ||
3858 | set_pageblock_order(pageblock_default_order()); | |
3859 | setup_usemap(pgdat, zone, size); | |
3860 | ret = init_currently_empty_zone(zone, zone_start_pfn, | |
3861 | size, MEMMAP_EARLY); | |
3862 | BUG_ON(ret); | |
3863 | memmap_init(size, nid, j, zone_start_pfn); | |
3864 | zone_start_pfn += size; | |
3865 | } | |
3866 | } | |
3867 | ||
3868 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) | |
3869 | { | |
3870 | /* Skip empty nodes */ | |
3871 | if (!pgdat->node_spanned_pages) | |
3872 | return; | |
3873 | ||
3874 | #ifdef CONFIG_FLAT_NODE_MEM_MAP | |
3875 | /* ia64 gets its own node_mem_map, before this, without bootmem */ | |
3876 | if (!pgdat->node_mem_map) { | |
3877 | unsigned long size, start, end; | |
3878 | struct page *map; | |
3879 | ||
3880 | /* | |
3881 | * The zone's endpoints aren't required to be MAX_ORDER | |
3882 | * aligned but the node_mem_map endpoints must be in order | |
3883 | * for the buddy allocator to function correctly. | |
3884 | */ | |
3885 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
3886 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
3887 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
3888 | size = (end - start) * sizeof(struct page); | |
3889 | map = alloc_remap(pgdat->node_id, size); | |
3890 | if (!map) | |
3891 | map = alloc_bootmem_node(pgdat, size); | |
3892 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); | |
3893 | } | |
3894 | #ifndef CONFIG_NEED_MULTIPLE_NODES | |
3895 | /* | |
3896 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
3897 | */ | |
3898 | if (pgdat == NODE_DATA(0)) { | |
3899 | mem_map = NODE_DATA(0)->node_mem_map; | |
3900 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
3901 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
3902 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); | |
3903 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
3904 | } | |
3905 | #endif | |
3906 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
3907 | } | |
3908 | ||
3909 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, | |
3910 | unsigned long node_start_pfn, unsigned long *zholes_size) | |
3911 | { | |
3912 | pg_data_t *pgdat = NODE_DATA(nid); | |
3913 | ||
3914 | pgdat->node_id = nid; | |
3915 | pgdat->node_start_pfn = node_start_pfn; | |
3916 | calculate_node_totalpages(pgdat, zones_size, zholes_size); | |
3917 | ||
3918 | alloc_node_mem_map(pgdat); | |
3919 | #ifdef CONFIG_FLAT_NODE_MEM_MAP | |
3920 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", | |
3921 | nid, (unsigned long)pgdat, | |
3922 | (unsigned long)pgdat->node_mem_map); | |
3923 | #endif | |
3924 | ||
3925 | free_area_init_core(pgdat, zones_size, zholes_size); | |
3926 | } | |
3927 | ||
3928 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
3929 | ||
3930 | #if MAX_NUMNODES > 1 | |
3931 | /* | |
3932 | * Figure out the number of possible node ids. | |
3933 | */ | |
3934 | static void __init setup_nr_node_ids(void) | |
3935 | { | |
3936 | unsigned int node; | |
3937 | unsigned int highest = 0; | |
3938 | ||
3939 | for_each_node_mask(node, node_possible_map) | |
3940 | highest = node; | |
3941 | nr_node_ids = highest + 1; | |
3942 | } | |
3943 | #else | |
3944 | static inline void setup_nr_node_ids(void) | |
3945 | { | |
3946 | } | |
3947 | #endif | |
3948 | ||
3949 | /** | |
3950 | * add_active_range - Register a range of PFNs backed by physical memory | |
3951 | * @nid: The node ID the range resides on | |
3952 | * @start_pfn: The start PFN of the available physical memory | |
3953 | * @end_pfn: The end PFN of the available physical memory | |
3954 | * | |
3955 | * These ranges are stored in an early_node_map[] and later used by | |
3956 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
3957 | * range spans a memory hole, it is up to the architecture to ensure | |
3958 | * the memory is not freed by the bootmem allocator. If possible | |
3959 | * the range being registered will be merged with existing ranges. | |
3960 | */ | |
3961 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
3962 | unsigned long end_pfn) | |
3963 | { | |
3964 | int i; | |
3965 | ||
3966 | mminit_dprintk(MMINIT_TRACE, "memory_register", | |
3967 | "Entering add_active_range(%d, %#lx, %#lx) " | |
3968 | "%d entries of %d used\n", | |
3969 | nid, start_pfn, end_pfn, | |
3970 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
3971 | ||
3972 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); | |
3973 | ||
3974 | /* Merge with existing active regions if possible */ | |
3975 | for (i = 0; i < nr_nodemap_entries; i++) { | |
3976 | if (early_node_map[i].nid != nid) | |
3977 | continue; | |
3978 | ||
3979 | /* Skip if an existing region covers this new one */ | |
3980 | if (start_pfn >= early_node_map[i].start_pfn && | |
3981 | end_pfn <= early_node_map[i].end_pfn) | |
3982 | return; | |
3983 | ||
3984 | /* Merge forward if suitable */ | |
3985 | if (start_pfn <= early_node_map[i].end_pfn && | |
3986 | end_pfn > early_node_map[i].end_pfn) { | |
3987 | early_node_map[i].end_pfn = end_pfn; | |
3988 | return; | |
3989 | } | |
3990 | ||
3991 | /* Merge backward if suitable */ | |
3992 | if (start_pfn < early_node_map[i].start_pfn && | |
3993 | end_pfn >= early_node_map[i].start_pfn) { | |
3994 | early_node_map[i].start_pfn = start_pfn; | |
3995 | return; | |
3996 | } | |
3997 | } | |
3998 | ||
3999 | /* Check that early_node_map is large enough */ | |
4000 | if (i >= MAX_ACTIVE_REGIONS) { | |
4001 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
4002 | MAX_ACTIVE_REGIONS); | |
4003 | return; | |
4004 | } | |
4005 | ||
4006 | early_node_map[i].nid = nid; | |
4007 | early_node_map[i].start_pfn = start_pfn; | |
4008 | early_node_map[i].end_pfn = end_pfn; | |
4009 | nr_nodemap_entries = i + 1; | |
4010 | } | |
4011 | ||
4012 | /** | |
4013 | * remove_active_range - Shrink an existing registered range of PFNs | |
4014 | * @nid: The node id the range is on that should be shrunk | |
4015 | * @start_pfn: The new PFN of the range | |
4016 | * @end_pfn: The new PFN of the range | |
4017 | * | |
4018 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
4019 | * The map is kept near the end physical page range that has already been | |
4020 | * registered. This function allows an arch to shrink an existing registered | |
4021 | * range. | |
4022 | */ | |
4023 | void __init remove_active_range(unsigned int nid, unsigned long start_pfn, | |
4024 | unsigned long end_pfn) | |
4025 | { | |
4026 | int i, j; | |
4027 | int removed = 0; | |
4028 | ||
4029 | printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n", | |
4030 | nid, start_pfn, end_pfn); | |
4031 | ||
4032 | /* Find the old active region end and shrink */ | |
4033 | for_each_active_range_index_in_nid(i, nid) { | |
4034 | if (early_node_map[i].start_pfn >= start_pfn && | |
4035 | early_node_map[i].end_pfn <= end_pfn) { | |
4036 | /* clear it */ | |
4037 | early_node_map[i].start_pfn = 0; | |
4038 | early_node_map[i].end_pfn = 0; | |
4039 | removed = 1; | |
4040 | continue; | |
4041 | } | |
4042 | if (early_node_map[i].start_pfn < start_pfn && | |
4043 | early_node_map[i].end_pfn > start_pfn) { | |
4044 | unsigned long temp_end_pfn = early_node_map[i].end_pfn; | |
4045 | early_node_map[i].end_pfn = start_pfn; | |
4046 | if (temp_end_pfn > end_pfn) | |
4047 | add_active_range(nid, end_pfn, temp_end_pfn); | |
4048 | continue; | |
4049 | } | |
4050 | if (early_node_map[i].start_pfn >= start_pfn && | |
4051 | early_node_map[i].end_pfn > end_pfn && | |
4052 | early_node_map[i].start_pfn < end_pfn) { | |
4053 | early_node_map[i].start_pfn = end_pfn; | |
4054 | continue; | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | if (!removed) | |
4059 | return; | |
4060 | ||
4061 | /* remove the blank ones */ | |
4062 | for (i = nr_nodemap_entries - 1; i > 0; i--) { | |
4063 | if (early_node_map[i].nid != nid) | |
4064 | continue; | |
4065 | if (early_node_map[i].end_pfn) | |
4066 | continue; | |
4067 | /* we found it, get rid of it */ | |
4068 | for (j = i; j < nr_nodemap_entries - 1; j++) | |
4069 | memcpy(&early_node_map[j], &early_node_map[j+1], | |
4070 | sizeof(early_node_map[j])); | |
4071 | j = nr_nodemap_entries - 1; | |
4072 | memset(&early_node_map[j], 0, sizeof(early_node_map[j])); | |
4073 | nr_nodemap_entries--; | |
4074 | } | |
4075 | } | |
4076 | ||
4077 | /** | |
4078 | * remove_all_active_ranges - Remove all currently registered regions | |
4079 | * | |
4080 | * During discovery, it may be found that a table like SRAT is invalid | |
4081 | * and an alternative discovery method must be used. This function removes | |
4082 | * all currently registered regions. | |
4083 | */ | |
4084 | void __init remove_all_active_ranges(void) | |
4085 | { | |
4086 | memset(early_node_map, 0, sizeof(early_node_map)); | |
4087 | nr_nodemap_entries = 0; | |
4088 | } | |
4089 | ||
4090 | /* Compare two active node_active_regions */ | |
4091 | static int __init cmp_node_active_region(const void *a, const void *b) | |
4092 | { | |
4093 | struct node_active_region *arange = (struct node_active_region *)a; | |
4094 | struct node_active_region *brange = (struct node_active_region *)b; | |
4095 | ||
4096 | /* Done this way to avoid overflows */ | |
4097 | if (arange->start_pfn > brange->start_pfn) | |
4098 | return 1; | |
4099 | if (arange->start_pfn < brange->start_pfn) | |
4100 | return -1; | |
4101 | ||
4102 | return 0; | |
4103 | } | |
4104 | ||
4105 | /* sort the node_map by start_pfn */ | |
4106 | void __init sort_node_map(void) | |
4107 | { | |
4108 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
4109 | sizeof(struct node_active_region), | |
4110 | cmp_node_active_region, NULL); | |
4111 | } | |
4112 | ||
4113 | /* Find the lowest pfn for a node */ | |
4114 | static unsigned long __init find_min_pfn_for_node(int nid) | |
4115 | { | |
4116 | int i; | |
4117 | unsigned long min_pfn = ULONG_MAX; | |
4118 | ||
4119 | /* Assuming a sorted map, the first range found has the starting pfn */ | |
4120 | for_each_active_range_index_in_nid(i, nid) | |
4121 | min_pfn = min(min_pfn, early_node_map[i].start_pfn); | |
4122 | ||
4123 | if (min_pfn == ULONG_MAX) { | |
4124 | printk(KERN_WARNING | |
4125 | "Could not find start_pfn for node %d\n", nid); | |
4126 | return 0; | |
4127 | } | |
4128 | ||
4129 | return min_pfn; | |
4130 | } | |
4131 | ||
4132 | /** | |
4133 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
4134 | * | |
4135 | * It returns the minimum PFN based on information provided via | |
4136 | * add_active_range(). | |
4137 | */ | |
4138 | unsigned long __init find_min_pfn_with_active_regions(void) | |
4139 | { | |
4140 | return find_min_pfn_for_node(MAX_NUMNODES); | |
4141 | } | |
4142 | ||
4143 | /* | |
4144 | * early_calculate_totalpages() | |
4145 | * Sum pages in active regions for movable zone. | |
4146 | * Populate N_HIGH_MEMORY for calculating usable_nodes. | |
4147 | */ | |
4148 | static unsigned long __init early_calculate_totalpages(void) | |
4149 | { | |
4150 | int i; | |
4151 | unsigned long totalpages = 0; | |
4152 | ||
4153 | for (i = 0; i < nr_nodemap_entries; i++) { | |
4154 | unsigned long pages = early_node_map[i].end_pfn - | |
4155 | early_node_map[i].start_pfn; | |
4156 | totalpages += pages; | |
4157 | if (pages) | |
4158 | node_set_state(early_node_map[i].nid, N_HIGH_MEMORY); | |
4159 | } | |
4160 | return totalpages; | |
4161 | } | |
4162 | ||
4163 | /* | |
4164 | * Find the PFN the Movable zone begins in each node. Kernel memory | |
4165 | * is spread evenly between nodes as long as the nodes have enough | |
4166 | * memory. When they don't, some nodes will have more kernelcore than | |
4167 | * others | |
4168 | */ | |
4169 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) | |
4170 | { | |
4171 | int i, nid; | |
4172 | unsigned long usable_startpfn; | |
4173 | unsigned long kernelcore_node, kernelcore_remaining; | |
4174 | /* save the state before borrow the nodemask */ | |
4175 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; | |
4176 | unsigned long totalpages = early_calculate_totalpages(); | |
4177 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); | |
4178 | ||
4179 | /* | |
4180 | * If movablecore was specified, calculate what size of | |
4181 | * kernelcore that corresponds so that memory usable for | |
4182 | * any allocation type is evenly spread. If both kernelcore | |
4183 | * and movablecore are specified, then the value of kernelcore | |
4184 | * will be used for required_kernelcore if it's greater than | |
4185 | * what movablecore would have allowed. | |
4186 | */ | |
4187 | if (required_movablecore) { | |
4188 | unsigned long corepages; | |
4189 | ||
4190 | /* | |
4191 | * Round-up so that ZONE_MOVABLE is at least as large as what | |
4192 | * was requested by the user | |
4193 | */ | |
4194 | required_movablecore = | |
4195 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); | |
4196 | corepages = totalpages - required_movablecore; | |
4197 | ||
4198 | required_kernelcore = max(required_kernelcore, corepages); | |
4199 | } | |
4200 | ||
4201 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ | |
4202 | if (!required_kernelcore) | |
4203 | goto out; | |
4204 | ||
4205 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ | |
4206 | find_usable_zone_for_movable(); | |
4207 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; | |
4208 | ||
4209 | restart: | |
4210 | /* Spread kernelcore memory as evenly as possible throughout nodes */ | |
4211 | kernelcore_node = required_kernelcore / usable_nodes; | |
4212 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4213 | /* | |
4214 | * Recalculate kernelcore_node if the division per node | |
4215 | * now exceeds what is necessary to satisfy the requested | |
4216 | * amount of memory for the kernel | |
4217 | */ | |
4218 | if (required_kernelcore < kernelcore_node) | |
4219 | kernelcore_node = required_kernelcore / usable_nodes; | |
4220 | ||
4221 | /* | |
4222 | * As the map is walked, we track how much memory is usable | |
4223 | * by the kernel using kernelcore_remaining. When it is | |
4224 | * 0, the rest of the node is usable by ZONE_MOVABLE | |
4225 | */ | |
4226 | kernelcore_remaining = kernelcore_node; | |
4227 | ||
4228 | /* Go through each range of PFNs within this node */ | |
4229 | for_each_active_range_index_in_nid(i, nid) { | |
4230 | unsigned long start_pfn, end_pfn; | |
4231 | unsigned long size_pages; | |
4232 | ||
4233 | start_pfn = max(early_node_map[i].start_pfn, | |
4234 | zone_movable_pfn[nid]); | |
4235 | end_pfn = early_node_map[i].end_pfn; | |
4236 | if (start_pfn >= end_pfn) | |
4237 | continue; | |
4238 | ||
4239 | /* Account for what is only usable for kernelcore */ | |
4240 | if (start_pfn < usable_startpfn) { | |
4241 | unsigned long kernel_pages; | |
4242 | kernel_pages = min(end_pfn, usable_startpfn) | |
4243 | - start_pfn; | |
4244 | ||
4245 | kernelcore_remaining -= min(kernel_pages, | |
4246 | kernelcore_remaining); | |
4247 | required_kernelcore -= min(kernel_pages, | |
4248 | required_kernelcore); | |
4249 | ||
4250 | /* Continue if range is now fully accounted */ | |
4251 | if (end_pfn <= usable_startpfn) { | |
4252 | ||
4253 | /* | |
4254 | * Push zone_movable_pfn to the end so | |
4255 | * that if we have to rebalance | |
4256 | * kernelcore across nodes, we will | |
4257 | * not double account here | |
4258 | */ | |
4259 | zone_movable_pfn[nid] = end_pfn; | |
4260 | continue; | |
4261 | } | |
4262 | start_pfn = usable_startpfn; | |
4263 | } | |
4264 | ||
4265 | /* | |
4266 | * The usable PFN range for ZONE_MOVABLE is from | |
4267 | * start_pfn->end_pfn. Calculate size_pages as the | |
4268 | * number of pages used as kernelcore | |
4269 | */ | |
4270 | size_pages = end_pfn - start_pfn; | |
4271 | if (size_pages > kernelcore_remaining) | |
4272 | size_pages = kernelcore_remaining; | |
4273 | zone_movable_pfn[nid] = start_pfn + size_pages; | |
4274 | ||
4275 | /* | |
4276 | * Some kernelcore has been met, update counts and | |
4277 | * break if the kernelcore for this node has been | |
4278 | * satisified | |
4279 | */ | |
4280 | required_kernelcore -= min(required_kernelcore, | |
4281 | size_pages); | |
4282 | kernelcore_remaining -= size_pages; | |
4283 | if (!kernelcore_remaining) | |
4284 | break; | |
4285 | } | |
4286 | } | |
4287 | ||
4288 | /* | |
4289 | * If there is still required_kernelcore, we do another pass with one | |
4290 | * less node in the count. This will push zone_movable_pfn[nid] further | |
4291 | * along on the nodes that still have memory until kernelcore is | |
4292 | * satisified | |
4293 | */ | |
4294 | usable_nodes--; | |
4295 | if (usable_nodes && required_kernelcore > usable_nodes) | |
4296 | goto restart; | |
4297 | ||
4298 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ | |
4299 | for (nid = 0; nid < MAX_NUMNODES; nid++) | |
4300 | zone_movable_pfn[nid] = | |
4301 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); | |
4302 | ||
4303 | out: | |
4304 | /* restore the node_state */ | |
4305 | node_states[N_HIGH_MEMORY] = saved_node_state; | |
4306 | } | |
4307 | ||
4308 | /* Any regular memory on that node ? */ | |
4309 | static void check_for_regular_memory(pg_data_t *pgdat) | |
4310 | { | |
4311 | #ifdef CONFIG_HIGHMEM | |
4312 | enum zone_type zone_type; | |
4313 | ||
4314 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { | |
4315 | struct zone *zone = &pgdat->node_zones[zone_type]; | |
4316 | if (zone->present_pages) | |
4317 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); | |
4318 | } | |
4319 | #endif | |
4320 | } | |
4321 | ||
4322 | /** | |
4323 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
4324 | * @max_zone_pfn: an array of max PFNs for each zone | |
4325 | * | |
4326 | * This will call free_area_init_node() for each active node in the system. | |
4327 | * Using the page ranges provided by add_active_range(), the size of each | |
4328 | * zone in each node and their holes is calculated. If the maximum PFN | |
4329 | * between two adjacent zones match, it is assumed that the zone is empty. | |
4330 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
4331 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
4332 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
4333 | * at arch_max_dma_pfn. | |
4334 | */ | |
4335 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
4336 | { | |
4337 | unsigned long nid; | |
4338 | int i; | |
4339 | ||
4340 | /* Sort early_node_map as initialisation assumes it is sorted */ | |
4341 | sort_node_map(); | |
4342 | ||
4343 | /* Record where the zone boundaries are */ | |
4344 | memset(arch_zone_lowest_possible_pfn, 0, | |
4345 | sizeof(arch_zone_lowest_possible_pfn)); | |
4346 | memset(arch_zone_highest_possible_pfn, 0, | |
4347 | sizeof(arch_zone_highest_possible_pfn)); | |
4348 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
4349 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
4350 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
4351 | if (i == ZONE_MOVABLE) | |
4352 | continue; | |
4353 | arch_zone_lowest_possible_pfn[i] = | |
4354 | arch_zone_highest_possible_pfn[i-1]; | |
4355 | arch_zone_highest_possible_pfn[i] = | |
4356 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
4357 | } | |
4358 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; | |
4359 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; | |
4360 | ||
4361 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ | |
4362 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); | |
4363 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); | |
4364 | ||
4365 | /* Print out the zone ranges */ | |
4366 | printk("Zone PFN ranges:\n"); | |
4367 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4368 | if (i == ZONE_MOVABLE) | |
4369 | continue; | |
4370 | printk(" %-8s %0#10lx -> %0#10lx\n", | |
4371 | zone_names[i], | |
4372 | arch_zone_lowest_possible_pfn[i], | |
4373 | arch_zone_highest_possible_pfn[i]); | |
4374 | } | |
4375 | ||
4376 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ | |
4377 | printk("Movable zone start PFN for each node\n"); | |
4378 | for (i = 0; i < MAX_NUMNODES; i++) { | |
4379 | if (zone_movable_pfn[i]) | |
4380 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); | |
4381 | } | |
4382 | ||
4383 | /* Print out the early_node_map[] */ | |
4384 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
4385 | for (i = 0; i < nr_nodemap_entries; i++) | |
4386 | printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid, | |
4387 | early_node_map[i].start_pfn, | |
4388 | early_node_map[i].end_pfn); | |
4389 | ||
4390 | /* Initialise every node */ | |
4391 | mminit_verify_pageflags_layout(); | |
4392 | setup_nr_node_ids(); | |
4393 | for_each_online_node(nid) { | |
4394 | pg_data_t *pgdat = NODE_DATA(nid); | |
4395 | free_area_init_node(nid, NULL, | |
4396 | find_min_pfn_for_node(nid), NULL); | |
4397 | ||
4398 | /* Any memory on that node */ | |
4399 | if (pgdat->node_present_pages) | |
4400 | node_set_state(nid, N_HIGH_MEMORY); | |
4401 | check_for_regular_memory(pgdat); | |
4402 | } | |
4403 | } | |
4404 | ||
4405 | static int __init cmdline_parse_core(char *p, unsigned long *core) | |
4406 | { | |
4407 | unsigned long long coremem; | |
4408 | if (!p) | |
4409 | return -EINVAL; | |
4410 | ||
4411 | coremem = memparse(p, &p); | |
4412 | *core = coremem >> PAGE_SHIFT; | |
4413 | ||
4414 | /* Paranoid check that UL is enough for the coremem value */ | |
4415 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); | |
4416 | ||
4417 | return 0; | |
4418 | } | |
4419 | ||
4420 | /* | |
4421 | * kernelcore=size sets the amount of memory for use for allocations that | |
4422 | * cannot be reclaimed or migrated. | |
4423 | */ | |
4424 | static int __init cmdline_parse_kernelcore(char *p) | |
4425 | { | |
4426 | return cmdline_parse_core(p, &required_kernelcore); | |
4427 | } | |
4428 | ||
4429 | /* | |
4430 | * movablecore=size sets the amount of memory for use for allocations that | |
4431 | * can be reclaimed or migrated. | |
4432 | */ | |
4433 | static int __init cmdline_parse_movablecore(char *p) | |
4434 | { | |
4435 | return cmdline_parse_core(p, &required_movablecore); | |
4436 | } | |
4437 | ||
4438 | early_param("kernelcore", cmdline_parse_kernelcore); | |
4439 | early_param("movablecore", cmdline_parse_movablecore); | |
4440 | ||
4441 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
4442 | ||
4443 | /** | |
4444 | * set_dma_reserve - set the specified number of pages reserved in the first zone | |
4445 | * @new_dma_reserve: The number of pages to mark reserved | |
4446 | * | |
4447 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
4448 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
4449 | * and other unfreeable allocations which can skew the watermarks badly. This | |
4450 | * function may optionally be used to account for unfreeable pages in the | |
4451 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
4452 | * smaller per-cpu batchsize. | |
4453 | */ | |
4454 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
4455 | { | |
4456 | dma_reserve = new_dma_reserve; | |
4457 | } | |
4458 | ||
4459 | #ifndef CONFIG_NEED_MULTIPLE_NODES | |
4460 | struct pglist_data __refdata contig_page_data = { | |
4461 | #ifndef CONFIG_NO_BOOTMEM | |
4462 | .bdata = &bootmem_node_data[0] | |
4463 | #endif | |
4464 | }; | |
4465 | EXPORT_SYMBOL(contig_page_data); | |
4466 | #endif | |
4467 | ||
4468 | void __init free_area_init(unsigned long *zones_size) | |
4469 | { | |
4470 | free_area_init_node(0, zones_size, | |
4471 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); | |
4472 | } | |
4473 | ||
4474 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
4475 | unsigned long action, void *hcpu) | |
4476 | { | |
4477 | int cpu = (unsigned long)hcpu; | |
4478 | ||
4479 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { | |
4480 | drain_pages(cpu); | |
4481 | ||
4482 | /* | |
4483 | * Spill the event counters of the dead processor | |
4484 | * into the current processors event counters. | |
4485 | * This artificially elevates the count of the current | |
4486 | * processor. | |
4487 | */ | |
4488 | vm_events_fold_cpu(cpu); | |
4489 | ||
4490 | /* | |
4491 | * Zero the differential counters of the dead processor | |
4492 | * so that the vm statistics are consistent. | |
4493 | * | |
4494 | * This is only okay since the processor is dead and cannot | |
4495 | * race with what we are doing. | |
4496 | */ | |
4497 | refresh_cpu_vm_stats(cpu); | |
4498 | } | |
4499 | return NOTIFY_OK; | |
4500 | } | |
4501 | ||
4502 | void __init page_alloc_init(void) | |
4503 | { | |
4504 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
4505 | } | |
4506 | ||
4507 | /* | |
4508 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
4509 | * or min_free_kbytes changes. | |
4510 | */ | |
4511 | static void calculate_totalreserve_pages(void) | |
4512 | { | |
4513 | struct pglist_data *pgdat; | |
4514 | unsigned long reserve_pages = 0; | |
4515 | enum zone_type i, j; | |
4516 | ||
4517 | for_each_online_pgdat(pgdat) { | |
4518 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
4519 | struct zone *zone = pgdat->node_zones + i; | |
4520 | unsigned long max = 0; | |
4521 | ||
4522 | /* Find valid and maximum lowmem_reserve in the zone */ | |
4523 | for (j = i; j < MAX_NR_ZONES; j++) { | |
4524 | if (zone->lowmem_reserve[j] > max) | |
4525 | max = zone->lowmem_reserve[j]; | |
4526 | } | |
4527 | ||
4528 | /* we treat the high watermark as reserved pages. */ | |
4529 | max += high_wmark_pages(zone); | |
4530 | ||
4531 | if (max > zone->present_pages) | |
4532 | max = zone->present_pages; | |
4533 | reserve_pages += max; | |
4534 | } | |
4535 | } | |
4536 | totalreserve_pages = reserve_pages; | |
4537 | } | |
4538 | ||
4539 | /* | |
4540 | * setup_per_zone_lowmem_reserve - called whenever | |
4541 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
4542 | * has a correct pages reserved value, so an adequate number of | |
4543 | * pages are left in the zone after a successful __alloc_pages(). | |
4544 | */ | |
4545 | static void setup_per_zone_lowmem_reserve(void) | |
4546 | { | |
4547 | struct pglist_data *pgdat; | |
4548 | enum zone_type j, idx; | |
4549 | ||
4550 | for_each_online_pgdat(pgdat) { | |
4551 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
4552 | struct zone *zone = pgdat->node_zones + j; | |
4553 | unsigned long present_pages = zone->present_pages; | |
4554 | ||
4555 | zone->lowmem_reserve[j] = 0; | |
4556 | ||
4557 | idx = j; | |
4558 | while (idx) { | |
4559 | struct zone *lower_zone; | |
4560 | ||
4561 | idx--; | |
4562 | ||
4563 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
4564 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
4565 | ||
4566 | lower_zone = pgdat->node_zones + idx; | |
4567 | lower_zone->lowmem_reserve[j] = present_pages / | |
4568 | sysctl_lowmem_reserve_ratio[idx]; | |
4569 | present_pages += lower_zone->present_pages; | |
4570 | } | |
4571 | } | |
4572 | } | |
4573 | ||
4574 | /* update totalreserve_pages */ | |
4575 | calculate_totalreserve_pages(); | |
4576 | } | |
4577 | ||
4578 | /** | |
4579 | * setup_per_zone_wmarks - called when min_free_kbytes changes | |
4580 | * or when memory is hot-{added|removed} | |
4581 | * | |
4582 | * Ensures that the watermark[min,low,high] values for each zone are set | |
4583 | * correctly with respect to min_free_kbytes. | |
4584 | */ | |
4585 | void setup_per_zone_wmarks(void) | |
4586 | { | |
4587 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
4588 | unsigned long lowmem_pages = 0; | |
4589 | struct zone *zone; | |
4590 | unsigned long flags; | |
4591 | ||
4592 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
4593 | for_each_zone(zone) { | |
4594 | if (!is_highmem(zone)) | |
4595 | lowmem_pages += zone->present_pages; | |
4596 | } | |
4597 | ||
4598 | for_each_zone(zone) { | |
4599 | u64 tmp; | |
4600 | ||
4601 | spin_lock_irqsave(&zone->lock, flags); | |
4602 | tmp = (u64)pages_min * zone->present_pages; | |
4603 | do_div(tmp, lowmem_pages); | |
4604 | if (is_highmem(zone)) { | |
4605 | /* | |
4606 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't | |
4607 | * need highmem pages, so cap pages_min to a small | |
4608 | * value here. | |
4609 | * | |
4610 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) | |
4611 | * deltas controls asynch page reclaim, and so should | |
4612 | * not be capped for highmem. | |
4613 | */ | |
4614 | int min_pages; | |
4615 | ||
4616 | min_pages = zone->present_pages / 1024; | |
4617 | if (min_pages < SWAP_CLUSTER_MAX) | |
4618 | min_pages = SWAP_CLUSTER_MAX; | |
4619 | if (min_pages > 128) | |
4620 | min_pages = 128; | |
4621 | zone->watermark[WMARK_MIN] = min_pages; | |
4622 | } else { | |
4623 | /* | |
4624 | * If it's a lowmem zone, reserve a number of pages | |
4625 | * proportionate to the zone's size. | |
4626 | */ | |
4627 | zone->watermark[WMARK_MIN] = tmp; | |
4628 | } | |
4629 | ||
4630 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); | |
4631 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); | |
4632 | setup_zone_migrate_reserve(zone); | |
4633 | spin_unlock_irqrestore(&zone->lock, flags); | |
4634 | } | |
4635 | ||
4636 | /* update totalreserve_pages */ | |
4637 | calculate_totalreserve_pages(); | |
4638 | } | |
4639 | ||
4640 | /* | |
4641 | * The inactive anon list should be small enough that the VM never has to | |
4642 | * do too much work, but large enough that each inactive page has a chance | |
4643 | * to be referenced again before it is swapped out. | |
4644 | * | |
4645 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to | |
4646 | * INACTIVE_ANON pages on this zone's LRU, maintained by the | |
4647 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of | |
4648 | * the anonymous pages are kept on the inactive list. | |
4649 | * | |
4650 | * total target max | |
4651 | * memory ratio inactive anon | |
4652 | * ------------------------------------- | |
4653 | * 10MB 1 5MB | |
4654 | * 100MB 1 50MB | |
4655 | * 1GB 3 250MB | |
4656 | * 10GB 10 0.9GB | |
4657 | * 100GB 31 3GB | |
4658 | * 1TB 101 10GB | |
4659 | * 10TB 320 32GB | |
4660 | */ | |
4661 | void calculate_zone_inactive_ratio(struct zone *zone) | |
4662 | { | |
4663 | unsigned int gb, ratio; | |
4664 | ||
4665 | /* Zone size in gigabytes */ | |
4666 | gb = zone->present_pages >> (30 - PAGE_SHIFT); | |
4667 | if (gb) | |
4668 | ratio = int_sqrt(10 * gb); | |
4669 | else | |
4670 | ratio = 1; | |
4671 | ||
4672 | zone->inactive_ratio = ratio; | |
4673 | } | |
4674 | ||
4675 | static void __init setup_per_zone_inactive_ratio(void) | |
4676 | { | |
4677 | struct zone *zone; | |
4678 | ||
4679 | for_each_zone(zone) | |
4680 | calculate_zone_inactive_ratio(zone); | |
4681 | } | |
4682 | ||
4683 | /* | |
4684 | * Initialise min_free_kbytes. | |
4685 | * | |
4686 | * For small machines we want it small (128k min). For large machines | |
4687 | * we want it large (64MB max). But it is not linear, because network | |
4688 | * bandwidth does not increase linearly with machine size. We use | |
4689 | * | |
4690 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
4691 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
4692 | * | |
4693 | * which yields | |
4694 | * | |
4695 | * 16MB: 512k | |
4696 | * 32MB: 724k | |
4697 | * 64MB: 1024k | |
4698 | * 128MB: 1448k | |
4699 | * 256MB: 2048k | |
4700 | * 512MB: 2896k | |
4701 | * 1024MB: 4096k | |
4702 | * 2048MB: 5792k | |
4703 | * 4096MB: 8192k | |
4704 | * 8192MB: 11584k | |
4705 | * 16384MB: 16384k | |
4706 | */ | |
4707 | static int __init init_per_zone_wmark_min(void) | |
4708 | { | |
4709 | unsigned long lowmem_kbytes; | |
4710 | ||
4711 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
4712 | ||
4713 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
4714 | if (min_free_kbytes < 128) | |
4715 | min_free_kbytes = 128; | |
4716 | if (min_free_kbytes > 65536) | |
4717 | min_free_kbytes = 65536; | |
4718 | setup_per_zone_wmarks(); | |
4719 | setup_per_zone_lowmem_reserve(); | |
4720 | setup_per_zone_inactive_ratio(); | |
4721 | return 0; | |
4722 | } | |
4723 | module_init(init_per_zone_wmark_min) | |
4724 | ||
4725 | /* | |
4726 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
4727 | * that we can call two helper functions whenever min_free_kbytes | |
4728 | * changes. | |
4729 | */ | |
4730 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
4731 | void __user *buffer, size_t *length, loff_t *ppos) | |
4732 | { | |
4733 | proc_dointvec(table, write, buffer, length, ppos); | |
4734 | if (write) | |
4735 | setup_per_zone_wmarks(); | |
4736 | return 0; | |
4737 | } | |
4738 | ||
4739 | #ifdef CONFIG_NUMA | |
4740 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
4741 | void __user *buffer, size_t *length, loff_t *ppos) | |
4742 | { | |
4743 | struct zone *zone; | |
4744 | int rc; | |
4745 | ||
4746 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
4747 | if (rc) | |
4748 | return rc; | |
4749 | ||
4750 | for_each_zone(zone) | |
4751 | zone->min_unmapped_pages = (zone->present_pages * | |
4752 | sysctl_min_unmapped_ratio) / 100; | |
4753 | return 0; | |
4754 | } | |
4755 | ||
4756 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
4757 | void __user *buffer, size_t *length, loff_t *ppos) | |
4758 | { | |
4759 | struct zone *zone; | |
4760 | int rc; | |
4761 | ||
4762 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
4763 | if (rc) | |
4764 | return rc; | |
4765 | ||
4766 | for_each_zone(zone) | |
4767 | zone->min_slab_pages = (zone->present_pages * | |
4768 | sysctl_min_slab_ratio) / 100; | |
4769 | return 0; | |
4770 | } | |
4771 | #endif | |
4772 | ||
4773 | /* | |
4774 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
4775 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
4776 | * whenever sysctl_lowmem_reserve_ratio changes. | |
4777 | * | |
4778 | * The reserve ratio obviously has absolutely no relation with the | |
4779 | * minimum watermarks. The lowmem reserve ratio can only make sense | |
4780 | * if in function of the boot time zone sizes. | |
4781 | */ | |
4782 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
4783 | void __user *buffer, size_t *length, loff_t *ppos) | |
4784 | { | |
4785 | proc_dointvec_minmax(table, write, buffer, length, ppos); | |
4786 | setup_per_zone_lowmem_reserve(); | |
4787 | return 0; | |
4788 | } | |
4789 | ||
4790 | /* | |
4791 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
4792 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
4793 | * can have before it gets flushed back to buddy allocator. | |
4794 | */ | |
4795 | ||
4796 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
4797 | void __user *buffer, size_t *length, loff_t *ppos) | |
4798 | { | |
4799 | struct zone *zone; | |
4800 | unsigned int cpu; | |
4801 | int ret; | |
4802 | ||
4803 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); | |
4804 | if (!write || (ret == -EINVAL)) | |
4805 | return ret; | |
4806 | for_each_populated_zone(zone) { | |
4807 | for_each_possible_cpu(cpu) { | |
4808 | unsigned long high; | |
4809 | high = zone->present_pages / percpu_pagelist_fraction; | |
4810 | setup_pagelist_highmark( | |
4811 | per_cpu_ptr(zone->pageset, cpu), high); | |
4812 | } | |
4813 | } | |
4814 | return 0; | |
4815 | } | |
4816 | ||
4817 | int hashdist = HASHDIST_DEFAULT; | |
4818 | ||
4819 | #ifdef CONFIG_NUMA | |
4820 | static int __init set_hashdist(char *str) | |
4821 | { | |
4822 | if (!str) | |
4823 | return 0; | |
4824 | hashdist = simple_strtoul(str, &str, 0); | |
4825 | return 1; | |
4826 | } | |
4827 | __setup("hashdist=", set_hashdist); | |
4828 | #endif | |
4829 | ||
4830 | /* | |
4831 | * allocate a large system hash table from bootmem | |
4832 | * - it is assumed that the hash table must contain an exact power-of-2 | |
4833 | * quantity of entries | |
4834 | * - limit is the number of hash buckets, not the total allocation size | |
4835 | */ | |
4836 | void *__init alloc_large_system_hash(const char *tablename, | |
4837 | unsigned long bucketsize, | |
4838 | unsigned long numentries, | |
4839 | int scale, | |
4840 | int flags, | |
4841 | unsigned int *_hash_shift, | |
4842 | unsigned int *_hash_mask, | |
4843 | unsigned long limit) | |
4844 | { | |
4845 | unsigned long long max = limit; | |
4846 | unsigned long log2qty, size; | |
4847 | void *table = NULL; | |
4848 | ||
4849 | /* allow the kernel cmdline to have a say */ | |
4850 | if (!numentries) { | |
4851 | /* round applicable memory size up to nearest megabyte */ | |
4852 | numentries = nr_kernel_pages; | |
4853 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
4854 | numentries >>= 20 - PAGE_SHIFT; | |
4855 | numentries <<= 20 - PAGE_SHIFT; | |
4856 | ||
4857 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
4858 | if (scale > PAGE_SHIFT) | |
4859 | numentries >>= (scale - PAGE_SHIFT); | |
4860 | else | |
4861 | numentries <<= (PAGE_SHIFT - scale); | |
4862 | ||
4863 | /* Make sure we've got at least a 0-order allocation.. */ | |
4864 | if (unlikely(flags & HASH_SMALL)) { | |
4865 | /* Makes no sense without HASH_EARLY */ | |
4866 | WARN_ON(!(flags & HASH_EARLY)); | |
4867 | if (!(numentries >> *_hash_shift)) { | |
4868 | numentries = 1UL << *_hash_shift; | |
4869 | BUG_ON(!numentries); | |
4870 | } | |
4871 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
4872 | numentries = PAGE_SIZE / bucketsize; | |
4873 | } | |
4874 | numentries = roundup_pow_of_two(numentries); | |
4875 | ||
4876 | /* limit allocation size to 1/16 total memory by default */ | |
4877 | if (max == 0) { | |
4878 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
4879 | do_div(max, bucketsize); | |
4880 | } | |
4881 | ||
4882 | if (numentries > max) | |
4883 | numentries = max; | |
4884 | ||
4885 | log2qty = ilog2(numentries); | |
4886 | ||
4887 | do { | |
4888 | size = bucketsize << log2qty; | |
4889 | if (flags & HASH_EARLY) | |
4890 | table = alloc_bootmem_nopanic(size); | |
4891 | else if (hashdist) | |
4892 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
4893 | else { | |
4894 | /* | |
4895 | * If bucketsize is not a power-of-two, we may free | |
4896 | * some pages at the end of hash table which | |
4897 | * alloc_pages_exact() automatically does | |
4898 | */ | |
4899 | if (get_order(size) < MAX_ORDER) { | |
4900 | table = alloc_pages_exact(size, GFP_ATOMIC); | |
4901 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); | |
4902 | } | |
4903 | } | |
4904 | } while (!table && size > PAGE_SIZE && --log2qty); | |
4905 | ||
4906 | if (!table) | |
4907 | panic("Failed to allocate %s hash table\n", tablename); | |
4908 | ||
4909 | printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n", | |
4910 | tablename, | |
4911 | (1U << log2qty), | |
4912 | ilog2(size) - PAGE_SHIFT, | |
4913 | size); | |
4914 | ||
4915 | if (_hash_shift) | |
4916 | *_hash_shift = log2qty; | |
4917 | if (_hash_mask) | |
4918 | *_hash_mask = (1 << log2qty) - 1; | |
4919 | ||
4920 | return table; | |
4921 | } | |
4922 | ||
4923 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ | |
4924 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, | |
4925 | unsigned long pfn) | |
4926 | { | |
4927 | #ifdef CONFIG_SPARSEMEM | |
4928 | return __pfn_to_section(pfn)->pageblock_flags; | |
4929 | #else | |
4930 | return zone->pageblock_flags; | |
4931 | #endif /* CONFIG_SPARSEMEM */ | |
4932 | } | |
4933 | ||
4934 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) | |
4935 | { | |
4936 | #ifdef CONFIG_SPARSEMEM | |
4937 | pfn &= (PAGES_PER_SECTION-1); | |
4938 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
4939 | #else | |
4940 | pfn = pfn - zone->zone_start_pfn; | |
4941 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; | |
4942 | #endif /* CONFIG_SPARSEMEM */ | |
4943 | } | |
4944 | ||
4945 | /** | |
4946 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages | |
4947 | * @page: The page within the block of interest | |
4948 | * @start_bitidx: The first bit of interest to retrieve | |
4949 | * @end_bitidx: The last bit of interest | |
4950 | * returns pageblock_bits flags | |
4951 | */ | |
4952 | unsigned long get_pageblock_flags_group(struct page *page, | |
4953 | int start_bitidx, int end_bitidx) | |
4954 | { | |
4955 | struct zone *zone; | |
4956 | unsigned long *bitmap; | |
4957 | unsigned long pfn, bitidx; | |
4958 | unsigned long flags = 0; | |
4959 | unsigned long value = 1; | |
4960 | ||
4961 | zone = page_zone(page); | |
4962 | pfn = page_to_pfn(page); | |
4963 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4964 | bitidx = pfn_to_bitidx(zone, pfn); | |
4965 | ||
4966 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4967 | if (test_bit(bitidx + start_bitidx, bitmap)) | |
4968 | flags |= value; | |
4969 | ||
4970 | return flags; | |
4971 | } | |
4972 | ||
4973 | /** | |
4974 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages | |
4975 | * @page: The page within the block of interest | |
4976 | * @start_bitidx: The first bit of interest | |
4977 | * @end_bitidx: The last bit of interest | |
4978 | * @flags: The flags to set | |
4979 | */ | |
4980 | void set_pageblock_flags_group(struct page *page, unsigned long flags, | |
4981 | int start_bitidx, int end_bitidx) | |
4982 | { | |
4983 | struct zone *zone; | |
4984 | unsigned long *bitmap; | |
4985 | unsigned long pfn, bitidx; | |
4986 | unsigned long value = 1; | |
4987 | ||
4988 | zone = page_zone(page); | |
4989 | pfn = page_to_pfn(page); | |
4990 | bitmap = get_pageblock_bitmap(zone, pfn); | |
4991 | bitidx = pfn_to_bitidx(zone, pfn); | |
4992 | VM_BUG_ON(pfn < zone->zone_start_pfn); | |
4993 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); | |
4994 | ||
4995 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) | |
4996 | if (flags & value) | |
4997 | __set_bit(bitidx + start_bitidx, bitmap); | |
4998 | else | |
4999 | __clear_bit(bitidx + start_bitidx, bitmap); | |
5000 | } | |
5001 | ||
5002 | /* | |
5003 | * This is designed as sub function...plz see page_isolation.c also. | |
5004 | * set/clear page block's type to be ISOLATE. | |
5005 | * page allocater never alloc memory from ISOLATE block. | |
5006 | */ | |
5007 | ||
5008 | int set_migratetype_isolate(struct page *page) | |
5009 | { | |
5010 | struct zone *zone; | |
5011 | struct page *curr_page; | |
5012 | unsigned long flags, pfn, iter; | |
5013 | unsigned long immobile = 0; | |
5014 | struct memory_isolate_notify arg; | |
5015 | int notifier_ret; | |
5016 | int ret = -EBUSY; | |
5017 | int zone_idx; | |
5018 | ||
5019 | zone = page_zone(page); | |
5020 | zone_idx = zone_idx(zone); | |
5021 | ||
5022 | spin_lock_irqsave(&zone->lock, flags); | |
5023 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE || | |
5024 | zone_idx == ZONE_MOVABLE) { | |
5025 | ret = 0; | |
5026 | goto out; | |
5027 | } | |
5028 | ||
5029 | pfn = page_to_pfn(page); | |
5030 | arg.start_pfn = pfn; | |
5031 | arg.nr_pages = pageblock_nr_pages; | |
5032 | arg.pages_found = 0; | |
5033 | ||
5034 | /* | |
5035 | * It may be possible to isolate a pageblock even if the | |
5036 | * migratetype is not MIGRATE_MOVABLE. The memory isolation | |
5037 | * notifier chain is used by balloon drivers to return the | |
5038 | * number of pages in a range that are held by the balloon | |
5039 | * driver to shrink memory. If all the pages are accounted for | |
5040 | * by balloons, are free, or on the LRU, isolation can continue. | |
5041 | * Later, for example, when memory hotplug notifier runs, these | |
5042 | * pages reported as "can be isolated" should be isolated(freed) | |
5043 | * by the balloon driver through the memory notifier chain. | |
5044 | */ | |
5045 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); | |
5046 | notifier_ret = notifier_to_errno(notifier_ret); | |
5047 | if (notifier_ret || !arg.pages_found) | |
5048 | goto out; | |
5049 | ||
5050 | for (iter = pfn; iter < (pfn + pageblock_nr_pages); iter++) { | |
5051 | if (!pfn_valid_within(pfn)) | |
5052 | continue; | |
5053 | ||
5054 | curr_page = pfn_to_page(iter); | |
5055 | if (!page_count(curr_page) || PageLRU(curr_page)) | |
5056 | continue; | |
5057 | ||
5058 | immobile++; | |
5059 | } | |
5060 | ||
5061 | if (arg.pages_found == immobile) | |
5062 | ret = 0; | |
5063 | ||
5064 | out: | |
5065 | if (!ret) { | |
5066 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); | |
5067 | move_freepages_block(zone, page, MIGRATE_ISOLATE); | |
5068 | } | |
5069 | ||
5070 | spin_unlock_irqrestore(&zone->lock, flags); | |
5071 | if (!ret) | |
5072 | drain_all_pages(); | |
5073 | return ret; | |
5074 | } | |
5075 | ||
5076 | void unset_migratetype_isolate(struct page *page) | |
5077 | { | |
5078 | struct zone *zone; | |
5079 | unsigned long flags; | |
5080 | zone = page_zone(page); | |
5081 | spin_lock_irqsave(&zone->lock, flags); | |
5082 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) | |
5083 | goto out; | |
5084 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); | |
5085 | move_freepages_block(zone, page, MIGRATE_MOVABLE); | |
5086 | out: | |
5087 | spin_unlock_irqrestore(&zone->lock, flags); | |
5088 | } | |
5089 | ||
5090 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
5091 | /* | |
5092 | * All pages in the range must be isolated before calling this. | |
5093 | */ | |
5094 | void | |
5095 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) | |
5096 | { | |
5097 | struct page *page; | |
5098 | struct zone *zone; | |
5099 | int order, i; | |
5100 | unsigned long pfn; | |
5101 | unsigned long flags; | |
5102 | /* find the first valid pfn */ | |
5103 | for (pfn = start_pfn; pfn < end_pfn; pfn++) | |
5104 | if (pfn_valid(pfn)) | |
5105 | break; | |
5106 | if (pfn == end_pfn) | |
5107 | return; | |
5108 | zone = page_zone(pfn_to_page(pfn)); | |
5109 | spin_lock_irqsave(&zone->lock, flags); | |
5110 | pfn = start_pfn; | |
5111 | while (pfn < end_pfn) { | |
5112 | if (!pfn_valid(pfn)) { | |
5113 | pfn++; | |
5114 | continue; | |
5115 | } | |
5116 | page = pfn_to_page(pfn); | |
5117 | BUG_ON(page_count(page)); | |
5118 | BUG_ON(!PageBuddy(page)); | |
5119 | order = page_order(page); | |
5120 | #ifdef CONFIG_DEBUG_VM | |
5121 | printk(KERN_INFO "remove from free list %lx %d %lx\n", | |
5122 | pfn, 1 << order, end_pfn); | |
5123 | #endif | |
5124 | list_del(&page->lru); | |
5125 | rmv_page_order(page); | |
5126 | zone->free_area[order].nr_free--; | |
5127 | __mod_zone_page_state(zone, NR_FREE_PAGES, | |
5128 | - (1UL << order)); | |
5129 | for (i = 0; i < (1 << order); i++) | |
5130 | SetPageReserved((page+i)); | |
5131 | pfn += (1 << order); | |
5132 | } | |
5133 | spin_unlock_irqrestore(&zone->lock, flags); | |
5134 | } | |
5135 | #endif | |
5136 | ||
5137 | #ifdef CONFIG_MEMORY_FAILURE | |
5138 | bool is_free_buddy_page(struct page *page) | |
5139 | { | |
5140 | struct zone *zone = page_zone(page); | |
5141 | unsigned long pfn = page_to_pfn(page); | |
5142 | unsigned long flags; | |
5143 | int order; | |
5144 | ||
5145 | spin_lock_irqsave(&zone->lock, flags); | |
5146 | for (order = 0; order < MAX_ORDER; order++) { | |
5147 | struct page *page_head = page - (pfn & ((1 << order) - 1)); | |
5148 | ||
5149 | if (PageBuddy(page_head) && page_order(page_head) >= order) | |
5150 | break; | |
5151 | } | |
5152 | spin_unlock_irqrestore(&zone->lock, flags); | |
5153 | ||
5154 | return order < MAX_ORDER; | |
5155 | } | |
5156 | #endif |