]>
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/bootmem.h> | |
23 | #include <linux/compiler.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/suspend.h> | |
27 | #include <linux/pagevec.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/notifier.h> | |
31 | #include <linux/topology.h> | |
32 | #include <linux/sysctl.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
35 | #include <linux/memory_hotplug.h> | |
36 | #include <linux/nodemask.h> | |
37 | #include <linux/vmalloc.h> | |
38 | #include <linux/mempolicy.h> | |
39 | #include <linux/stop_machine.h> | |
40 | #include <linux/sort.h> | |
41 | #include <linux/pfn.h> | |
42 | #include <linux/backing-dev.h> | |
43 | #include <linux/fault-inject.h> | |
44 | ||
45 | #include <asm/tlbflush.h> | |
46 | #include <asm/div64.h> | |
47 | #include "internal.h" | |
48 | ||
49 | /* | |
50 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
51 | * initializer cleaner | |
52 | */ | |
53 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; | |
54 | EXPORT_SYMBOL(node_online_map); | |
55 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; | |
56 | EXPORT_SYMBOL(node_possible_map); | |
57 | unsigned long totalram_pages __read_mostly; | |
58 | unsigned long totalreserve_pages __read_mostly; | |
59 | long nr_swap_pages; | |
60 | int percpu_pagelist_fraction; | |
61 | ||
62 | static void __free_pages_ok(struct page *page, unsigned int order); | |
63 | ||
64 | /* | |
65 | * results with 256, 32 in the lowmem_reserve sysctl: | |
66 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
67 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
68 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
69 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
70 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
71 | * | |
72 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
73 | * don't need any ZONE_NORMAL reservation | |
74 | */ | |
75 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { | |
76 | 256, | |
77 | #ifdef CONFIG_ZONE_DMA32 | |
78 | 256, | |
79 | #endif | |
80 | #ifdef CONFIG_HIGHMEM | |
81 | 32 | |
82 | #endif | |
83 | }; | |
84 | ||
85 | EXPORT_SYMBOL(totalram_pages); | |
86 | ||
87 | static char * const zone_names[MAX_NR_ZONES] = { | |
88 | "DMA", | |
89 | #ifdef CONFIG_ZONE_DMA32 | |
90 | "DMA32", | |
91 | #endif | |
92 | "Normal", | |
93 | #ifdef CONFIG_HIGHMEM | |
94 | "HighMem" | |
95 | #endif | |
96 | }; | |
97 | ||
98 | int min_free_kbytes = 1024; | |
99 | ||
100 | unsigned long __meminitdata nr_kernel_pages; | |
101 | unsigned long __meminitdata nr_all_pages; | |
102 | static unsigned long __initdata dma_reserve; | |
103 | ||
104 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
105 | /* | |
106 | * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct | |
107 | * ranges of memory (RAM) that may be registered with add_active_range(). | |
108 | * Ranges passed to add_active_range() will be merged if possible | |
109 | * so the number of times add_active_range() can be called is | |
110 | * related to the number of nodes and the number of holes | |
111 | */ | |
112 | #ifdef CONFIG_MAX_ACTIVE_REGIONS | |
113 | /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */ | |
114 | #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS | |
115 | #else | |
116 | #if MAX_NUMNODES >= 32 | |
117 | /* If there can be many nodes, allow up to 50 holes per node */ | |
118 | #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50) | |
119 | #else | |
120 | /* By default, allow up to 256 distinct regions */ | |
121 | #define MAX_ACTIVE_REGIONS 256 | |
122 | #endif | |
123 | #endif | |
124 | ||
125 | struct node_active_region __initdata early_node_map[MAX_ACTIVE_REGIONS]; | |
126 | int __initdata nr_nodemap_entries; | |
127 | unsigned long __initdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; | |
128 | unsigned long __initdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; | |
129 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
130 | unsigned long __initdata node_boundary_start_pfn[MAX_NUMNODES]; | |
131 | unsigned long __initdata node_boundary_end_pfn[MAX_NUMNODES]; | |
132 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
133 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
134 | ||
135 | #ifdef CONFIG_DEBUG_VM | |
136 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) | |
137 | { | |
138 | int ret = 0; | |
139 | unsigned seq; | |
140 | unsigned long pfn = page_to_pfn(page); | |
141 | ||
142 | do { | |
143 | seq = zone_span_seqbegin(zone); | |
144 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
145 | ret = 1; | |
146 | else if (pfn < zone->zone_start_pfn) | |
147 | ret = 1; | |
148 | } while (zone_span_seqretry(zone, seq)); | |
149 | ||
150 | return ret; | |
151 | } | |
152 | ||
153 | static int page_is_consistent(struct zone *zone, struct page *page) | |
154 | { | |
155 | #ifdef CONFIG_HOLES_IN_ZONE | |
156 | if (!pfn_valid(page_to_pfn(page))) | |
157 | return 0; | |
158 | #endif | |
159 | if (zone != page_zone(page)) | |
160 | return 0; | |
161 | ||
162 | return 1; | |
163 | } | |
164 | /* | |
165 | * Temporary debugging check for pages not lying within a given zone. | |
166 | */ | |
167 | static int bad_range(struct zone *zone, struct page *page) | |
168 | { | |
169 | if (page_outside_zone_boundaries(zone, page)) | |
170 | return 1; | |
171 | if (!page_is_consistent(zone, page)) | |
172 | return 1; | |
173 | ||
174 | return 0; | |
175 | } | |
176 | #else | |
177 | static inline int bad_range(struct zone *zone, struct page *page) | |
178 | { | |
179 | return 0; | |
180 | } | |
181 | #endif | |
182 | ||
183 | static void bad_page(struct page *page) | |
184 | { | |
185 | printk(KERN_EMERG "Bad page state in process '%s'\n" | |
186 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" | |
187 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
188 | KERN_EMERG "Backtrace:\n", | |
189 | current->comm, page, (int)(2*sizeof(unsigned long)), | |
190 | (unsigned long)page->flags, page->mapping, | |
191 | page_mapcount(page), page_count(page)); | |
192 | dump_stack(); | |
193 | page->flags &= ~(1 << PG_lru | | |
194 | 1 << PG_private | | |
195 | 1 << PG_locked | | |
196 | 1 << PG_active | | |
197 | 1 << PG_dirty | | |
198 | 1 << PG_reclaim | | |
199 | 1 << PG_slab | | |
200 | 1 << PG_swapcache | | |
201 | 1 << PG_writeback | | |
202 | 1 << PG_buddy ); | |
203 | set_page_count(page, 0); | |
204 | reset_page_mapcount(page); | |
205 | page->mapping = NULL; | |
206 | add_taint(TAINT_BAD_PAGE); | |
207 | } | |
208 | ||
209 | /* | |
210 | * Higher-order pages are called "compound pages". They are structured thusly: | |
211 | * | |
212 | * The first PAGE_SIZE page is called the "head page". | |
213 | * | |
214 | * The remaining PAGE_SIZE pages are called "tail pages". | |
215 | * | |
216 | * All pages have PG_compound set. All pages have their ->private pointing at | |
217 | * the head page (even the head page has this). | |
218 | * | |
219 | * The first tail page's ->lru.next holds the address of the compound page's | |
220 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
221 | * This usage means that zero-order pages may not be compound. | |
222 | */ | |
223 | ||
224 | static void free_compound_page(struct page *page) | |
225 | { | |
226 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
227 | } | |
228 | ||
229 | static void prep_compound_page(struct page *page, unsigned long order) | |
230 | { | |
231 | int i; | |
232 | int nr_pages = 1 << order; | |
233 | ||
234 | set_compound_page_dtor(page, free_compound_page); | |
235 | page[1].lru.prev = (void *)order; | |
236 | for (i = 0; i < nr_pages; i++) { | |
237 | struct page *p = page + i; | |
238 | ||
239 | __SetPageCompound(p); | |
240 | set_page_private(p, (unsigned long)page); | |
241 | } | |
242 | } | |
243 | ||
244 | static void destroy_compound_page(struct page *page, unsigned long order) | |
245 | { | |
246 | int i; | |
247 | int nr_pages = 1 << order; | |
248 | ||
249 | if (unlikely((unsigned long)page[1].lru.prev != order)) | |
250 | bad_page(page); | |
251 | ||
252 | for (i = 0; i < nr_pages; i++) { | |
253 | struct page *p = page + i; | |
254 | ||
255 | if (unlikely(!PageCompound(p) | | |
256 | (page_private(p) != (unsigned long)page))) | |
257 | bad_page(page); | |
258 | __ClearPageCompound(p); | |
259 | } | |
260 | } | |
261 | ||
262 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) | |
263 | { | |
264 | int i; | |
265 | ||
266 | VM_BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); | |
267 | /* | |
268 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
269 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
270 | */ | |
271 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); | |
272 | for (i = 0; i < (1 << order); i++) | |
273 | clear_highpage(page + i); | |
274 | } | |
275 | ||
276 | /* | |
277 | * function for dealing with page's order in buddy system. | |
278 | * zone->lock is already acquired when we use these. | |
279 | * So, we don't need atomic page->flags operations here. | |
280 | */ | |
281 | static inline unsigned long page_order(struct page *page) | |
282 | { | |
283 | return page_private(page); | |
284 | } | |
285 | ||
286 | static inline void set_page_order(struct page *page, int order) | |
287 | { | |
288 | set_page_private(page, order); | |
289 | __SetPageBuddy(page); | |
290 | } | |
291 | ||
292 | static inline void rmv_page_order(struct page *page) | |
293 | { | |
294 | __ClearPageBuddy(page); | |
295 | set_page_private(page, 0); | |
296 | } | |
297 | ||
298 | /* | |
299 | * Locate the struct page for both the matching buddy in our | |
300 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
301 | * | |
302 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
303 | * the following equation: | |
304 | * B2 = B1 ^ (1 << O) | |
305 | * For example, if the starting buddy (buddy2) is #8 its order | |
306 | * 1 buddy is #10: | |
307 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
308 | * | |
309 | * 2) Any buddy B will have an order O+1 parent P which | |
310 | * satisfies the following equation: | |
311 | * P = B & ~(1 << O) | |
312 | * | |
313 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER | |
314 | */ | |
315 | static inline struct page * | |
316 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
317 | { | |
318 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
319 | ||
320 | return page + (buddy_idx - page_idx); | |
321 | } | |
322 | ||
323 | static inline unsigned long | |
324 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
325 | { | |
326 | return (page_idx & ~(1 << order)); | |
327 | } | |
328 | ||
329 | /* | |
330 | * This function checks whether a page is free && is the buddy | |
331 | * we can do coalesce a page and its buddy if | |
332 | * (a) the buddy is not in a hole && | |
333 | * (b) the buddy is in the buddy system && | |
334 | * (c) a page and its buddy have the same order && | |
335 | * (d) a page and its buddy are in the same zone. | |
336 | * | |
337 | * For recording whether a page is in the buddy system, we use PG_buddy. | |
338 | * Setting, clearing, and testing PG_buddy is serialized by zone->lock. | |
339 | * | |
340 | * For recording page's order, we use page_private(page). | |
341 | */ | |
342 | static inline int page_is_buddy(struct page *page, struct page *buddy, | |
343 | int order) | |
344 | { | |
345 | #ifdef CONFIG_HOLES_IN_ZONE | |
346 | if (!pfn_valid(page_to_pfn(buddy))) | |
347 | return 0; | |
348 | #endif | |
349 | ||
350 | if (page_zone_id(page) != page_zone_id(buddy)) | |
351 | return 0; | |
352 | ||
353 | if (PageBuddy(buddy) && page_order(buddy) == order) { | |
354 | BUG_ON(page_count(buddy) != 0); | |
355 | return 1; | |
356 | } | |
357 | return 0; | |
358 | } | |
359 | ||
360 | /* | |
361 | * Freeing function for a buddy system allocator. | |
362 | * | |
363 | * The concept of a buddy system is to maintain direct-mapped table | |
364 | * (containing bit values) for memory blocks of various "orders". | |
365 | * The bottom level table contains the map for the smallest allocatable | |
366 | * units of memory (here, pages), and each level above it describes | |
367 | * pairs of units from the levels below, hence, "buddies". | |
368 | * At a high level, all that happens here is marking the table entry | |
369 | * at the bottom level available, and propagating the changes upward | |
370 | * as necessary, plus some accounting needed to play nicely with other | |
371 | * parts of the VM system. | |
372 | * At each level, we keep a list of pages, which are heads of continuous | |
373 | * free pages of length of (1 << order) and marked with PG_buddy. Page's | |
374 | * order is recorded in page_private(page) field. | |
375 | * So when we are allocating or freeing one, we can derive the state of the | |
376 | * other. That is, if we allocate a small block, and both were | |
377 | * free, the remainder of the region must be split into blocks. | |
378 | * If a block is freed, and its buddy is also free, then this | |
379 | * triggers coalescing into a block of larger size. | |
380 | * | |
381 | * -- wli | |
382 | */ | |
383 | ||
384 | static inline void __free_one_page(struct page *page, | |
385 | struct zone *zone, unsigned int order) | |
386 | { | |
387 | unsigned long page_idx; | |
388 | int order_size = 1 << order; | |
389 | ||
390 | if (unlikely(PageCompound(page))) | |
391 | destroy_compound_page(page, order); | |
392 | ||
393 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
394 | ||
395 | VM_BUG_ON(page_idx & (order_size - 1)); | |
396 | VM_BUG_ON(bad_range(zone, page)); | |
397 | ||
398 | zone->free_pages += order_size; | |
399 | while (order < MAX_ORDER-1) { | |
400 | unsigned long combined_idx; | |
401 | struct free_area *area; | |
402 | struct page *buddy; | |
403 | ||
404 | buddy = __page_find_buddy(page, page_idx, order); | |
405 | if (!page_is_buddy(page, buddy, order)) | |
406 | break; /* Move the buddy up one level. */ | |
407 | ||
408 | list_del(&buddy->lru); | |
409 | area = zone->free_area + order; | |
410 | area->nr_free--; | |
411 | rmv_page_order(buddy); | |
412 | combined_idx = __find_combined_index(page_idx, order); | |
413 | page = page + (combined_idx - page_idx); | |
414 | page_idx = combined_idx; | |
415 | order++; | |
416 | } | |
417 | set_page_order(page, order); | |
418 | list_add(&page->lru, &zone->free_area[order].free_list); | |
419 | zone->free_area[order].nr_free++; | |
420 | } | |
421 | ||
422 | static inline int free_pages_check(struct page *page) | |
423 | { | |
424 | if (unlikely(page_mapcount(page) | | |
425 | (page->mapping != NULL) | | |
426 | (page_count(page) != 0) | | |
427 | (page->flags & ( | |
428 | 1 << PG_lru | | |
429 | 1 << PG_private | | |
430 | 1 << PG_locked | | |
431 | 1 << PG_active | | |
432 | 1 << PG_reclaim | | |
433 | 1 << PG_slab | | |
434 | 1 << PG_swapcache | | |
435 | 1 << PG_writeback | | |
436 | 1 << PG_reserved | | |
437 | 1 << PG_buddy )))) | |
438 | bad_page(page); | |
439 | if (PageDirty(page)) | |
440 | __ClearPageDirty(page); | |
441 | /* | |
442 | * For now, we report if PG_reserved was found set, but do not | |
443 | * clear it, and do not free the page. But we shall soon need | |
444 | * to do more, for when the ZERO_PAGE count wraps negative. | |
445 | */ | |
446 | return PageReserved(page); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Frees a list of pages. | |
451 | * Assumes all pages on list are in same zone, and of same order. | |
452 | * count is the number of pages to free. | |
453 | * | |
454 | * If the zone was previously in an "all pages pinned" state then look to | |
455 | * see if this freeing clears that state. | |
456 | * | |
457 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
458 | * pinned" detection logic. | |
459 | */ | |
460 | static void free_pages_bulk(struct zone *zone, int count, | |
461 | struct list_head *list, int order) | |
462 | { | |
463 | spin_lock(&zone->lock); | |
464 | zone->all_unreclaimable = 0; | |
465 | zone->pages_scanned = 0; | |
466 | while (count--) { | |
467 | struct page *page; | |
468 | ||
469 | VM_BUG_ON(list_empty(list)); | |
470 | page = list_entry(list->prev, struct page, lru); | |
471 | /* have to delete it as __free_one_page list manipulates */ | |
472 | list_del(&page->lru); | |
473 | __free_one_page(page, zone, order); | |
474 | } | |
475 | spin_unlock(&zone->lock); | |
476 | } | |
477 | ||
478 | static void free_one_page(struct zone *zone, struct page *page, int order) | |
479 | { | |
480 | spin_lock(&zone->lock); | |
481 | zone->all_unreclaimable = 0; | |
482 | zone->pages_scanned = 0; | |
483 | __free_one_page(page, zone, order); | |
484 | spin_unlock(&zone->lock); | |
485 | } | |
486 | ||
487 | static void __free_pages_ok(struct page *page, unsigned int order) | |
488 | { | |
489 | unsigned long flags; | |
490 | int i; | |
491 | int reserved = 0; | |
492 | ||
493 | for (i = 0 ; i < (1 << order) ; ++i) | |
494 | reserved += free_pages_check(page + i); | |
495 | if (reserved) | |
496 | return; | |
497 | ||
498 | if (!PageHighMem(page)) | |
499 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); | |
500 | arch_free_page(page, order); | |
501 | kernel_map_pages(page, 1 << order, 0); | |
502 | ||
503 | local_irq_save(flags); | |
504 | __count_vm_events(PGFREE, 1 << order); | |
505 | free_one_page(page_zone(page), page, order); | |
506 | local_irq_restore(flags); | |
507 | } | |
508 | ||
509 | /* | |
510 | * permit the bootmem allocator to evade page validation on high-order frees | |
511 | */ | |
512 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
513 | { | |
514 | if (order == 0) { | |
515 | __ClearPageReserved(page); | |
516 | set_page_count(page, 0); | |
517 | set_page_refcounted(page); | |
518 | __free_page(page); | |
519 | } else { | |
520 | int loop; | |
521 | ||
522 | prefetchw(page); | |
523 | for (loop = 0; loop < BITS_PER_LONG; loop++) { | |
524 | struct page *p = &page[loop]; | |
525 | ||
526 | if (loop + 1 < BITS_PER_LONG) | |
527 | prefetchw(p + 1); | |
528 | __ClearPageReserved(p); | |
529 | set_page_count(p, 0); | |
530 | } | |
531 | ||
532 | set_page_refcounted(page); | |
533 | __free_pages(page, order); | |
534 | } | |
535 | } | |
536 | ||
537 | ||
538 | /* | |
539 | * The order of subdivision here is critical for the IO subsystem. | |
540 | * Please do not alter this order without good reasons and regression | |
541 | * testing. Specifically, as large blocks of memory are subdivided, | |
542 | * the order in which smaller blocks are delivered depends on the order | |
543 | * they're subdivided in this function. This is the primary factor | |
544 | * influencing the order in which pages are delivered to the IO | |
545 | * subsystem according to empirical testing, and this is also justified | |
546 | * by considering the behavior of a buddy system containing a single | |
547 | * large block of memory acted on by a series of small allocations. | |
548 | * This behavior is a critical factor in sglist merging's success. | |
549 | * | |
550 | * -- wli | |
551 | */ | |
552 | static inline void expand(struct zone *zone, struct page *page, | |
553 | int low, int high, struct free_area *area) | |
554 | { | |
555 | unsigned long size = 1 << high; | |
556 | ||
557 | while (high > low) { | |
558 | area--; | |
559 | high--; | |
560 | size >>= 1; | |
561 | VM_BUG_ON(bad_range(zone, &page[size])); | |
562 | list_add(&page[size].lru, &area->free_list); | |
563 | area->nr_free++; | |
564 | set_page_order(&page[size], high); | |
565 | } | |
566 | } | |
567 | ||
568 | /* | |
569 | * This page is about to be returned from the page allocator | |
570 | */ | |
571 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) | |
572 | { | |
573 | if (unlikely(page_mapcount(page) | | |
574 | (page->mapping != NULL) | | |
575 | (page_count(page) != 0) | | |
576 | (page->flags & ( | |
577 | 1 << PG_lru | | |
578 | 1 << PG_private | | |
579 | 1 << PG_locked | | |
580 | 1 << PG_active | | |
581 | 1 << PG_dirty | | |
582 | 1 << PG_reclaim | | |
583 | 1 << PG_slab | | |
584 | 1 << PG_swapcache | | |
585 | 1 << PG_writeback | | |
586 | 1 << PG_reserved | | |
587 | 1 << PG_buddy )))) | |
588 | bad_page(page); | |
589 | ||
590 | /* | |
591 | * For now, we report if PG_reserved was found set, but do not | |
592 | * clear it, and do not allocate the page: as a safety net. | |
593 | */ | |
594 | if (PageReserved(page)) | |
595 | return 1; | |
596 | ||
597 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | | |
598 | 1 << PG_referenced | 1 << PG_arch_1 | | |
599 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
600 | set_page_private(page, 0); | |
601 | set_page_refcounted(page); | |
602 | ||
603 | arch_alloc_page(page, order); | |
604 | kernel_map_pages(page, 1 << order, 1); | |
605 | ||
606 | if (gfp_flags & __GFP_ZERO) | |
607 | prep_zero_page(page, order, gfp_flags); | |
608 | ||
609 | if (order && (gfp_flags & __GFP_COMP)) | |
610 | prep_compound_page(page, order); | |
611 | ||
612 | return 0; | |
613 | } | |
614 | ||
615 | /* | |
616 | * Do the hard work of removing an element from the buddy allocator. | |
617 | * Call me with the zone->lock already held. | |
618 | */ | |
619 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
620 | { | |
621 | struct free_area * area; | |
622 | unsigned int current_order; | |
623 | struct page *page; | |
624 | ||
625 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
626 | area = zone->free_area + current_order; | |
627 | if (list_empty(&area->free_list)) | |
628 | continue; | |
629 | ||
630 | page = list_entry(area->free_list.next, struct page, lru); | |
631 | list_del(&page->lru); | |
632 | rmv_page_order(page); | |
633 | area->nr_free--; | |
634 | zone->free_pages -= 1UL << order; | |
635 | expand(zone, page, order, current_order, area); | |
636 | return page; | |
637 | } | |
638 | ||
639 | return NULL; | |
640 | } | |
641 | ||
642 | /* | |
643 | * Obtain a specified number of elements from the buddy allocator, all under | |
644 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
645 | * Returns the number of new pages which were placed at *list. | |
646 | */ | |
647 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
648 | unsigned long count, struct list_head *list) | |
649 | { | |
650 | int i; | |
651 | ||
652 | spin_lock(&zone->lock); | |
653 | for (i = 0; i < count; ++i) { | |
654 | struct page *page = __rmqueue(zone, order); | |
655 | if (unlikely(page == NULL)) | |
656 | break; | |
657 | list_add_tail(&page->lru, list); | |
658 | } | |
659 | spin_unlock(&zone->lock); | |
660 | return i; | |
661 | } | |
662 | ||
663 | #ifdef CONFIG_NUMA | |
664 | /* | |
665 | * Called from the slab reaper to drain pagesets on a particular node that | |
666 | * belongs to the currently executing processor. | |
667 | * Note that this function must be called with the thread pinned to | |
668 | * a single processor. | |
669 | */ | |
670 | void drain_node_pages(int nodeid) | |
671 | { | |
672 | int i; | |
673 | enum zone_type z; | |
674 | unsigned long flags; | |
675 | ||
676 | for (z = 0; z < MAX_NR_ZONES; z++) { | |
677 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
678 | struct per_cpu_pageset *pset; | |
679 | ||
680 | if (!populated_zone(zone)) | |
681 | continue; | |
682 | ||
683 | pset = zone_pcp(zone, smp_processor_id()); | |
684 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { | |
685 | struct per_cpu_pages *pcp; | |
686 | ||
687 | pcp = &pset->pcp[i]; | |
688 | if (pcp->count) { | |
689 | int to_drain; | |
690 | ||
691 | local_irq_save(flags); | |
692 | if (pcp->count >= pcp->batch) | |
693 | to_drain = pcp->batch; | |
694 | else | |
695 | to_drain = pcp->count; | |
696 | free_pages_bulk(zone, to_drain, &pcp->list, 0); | |
697 | pcp->count -= to_drain; | |
698 | local_irq_restore(flags); | |
699 | } | |
700 | } | |
701 | } | |
702 | } | |
703 | #endif | |
704 | ||
705 | static void __drain_pages(unsigned int cpu) | |
706 | { | |
707 | unsigned long flags; | |
708 | struct zone *zone; | |
709 | int i; | |
710 | ||
711 | for_each_zone(zone) { | |
712 | struct per_cpu_pageset *pset; | |
713 | ||
714 | if (!populated_zone(zone)) | |
715 | continue; | |
716 | ||
717 | pset = zone_pcp(zone, cpu); | |
718 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { | |
719 | struct per_cpu_pages *pcp; | |
720 | ||
721 | pcp = &pset->pcp[i]; | |
722 | local_irq_save(flags); | |
723 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
724 | pcp->count = 0; | |
725 | local_irq_restore(flags); | |
726 | } | |
727 | } | |
728 | } | |
729 | ||
730 | #ifdef CONFIG_PM | |
731 | ||
732 | void mark_free_pages(struct zone *zone) | |
733 | { | |
734 | unsigned long pfn, max_zone_pfn; | |
735 | unsigned long flags; | |
736 | int order; | |
737 | struct list_head *curr; | |
738 | ||
739 | if (!zone->spanned_pages) | |
740 | return; | |
741 | ||
742 | spin_lock_irqsave(&zone->lock, flags); | |
743 | ||
744 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
745 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
746 | if (pfn_valid(pfn)) { | |
747 | struct page *page = pfn_to_page(pfn); | |
748 | ||
749 | if (!PageNosave(page)) | |
750 | ClearPageNosaveFree(page); | |
751 | } | |
752 | ||
753 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
754 | list_for_each(curr, &zone->free_area[order].free_list) { | |
755 | unsigned long i; | |
756 | ||
757 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
758 | for (i = 0; i < (1UL << order); i++) | |
759 | SetPageNosaveFree(pfn_to_page(pfn + i)); | |
760 | } | |
761 | ||
762 | spin_unlock_irqrestore(&zone->lock, flags); | |
763 | } | |
764 | ||
765 | /* | |
766 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
767 | */ | |
768 | void drain_local_pages(void) | |
769 | { | |
770 | unsigned long flags; | |
771 | ||
772 | local_irq_save(flags); | |
773 | __drain_pages(smp_processor_id()); | |
774 | local_irq_restore(flags); | |
775 | } | |
776 | #endif /* CONFIG_PM */ | |
777 | ||
778 | /* | |
779 | * Free a 0-order page | |
780 | */ | |
781 | static void fastcall free_hot_cold_page(struct page *page, int cold) | |
782 | { | |
783 | struct zone *zone = page_zone(page); | |
784 | struct per_cpu_pages *pcp; | |
785 | unsigned long flags; | |
786 | ||
787 | if (PageAnon(page)) | |
788 | page->mapping = NULL; | |
789 | if (free_pages_check(page)) | |
790 | return; | |
791 | ||
792 | if (!PageHighMem(page)) | |
793 | debug_check_no_locks_freed(page_address(page), PAGE_SIZE); | |
794 | arch_free_page(page, 0); | |
795 | kernel_map_pages(page, 1, 0); | |
796 | ||
797 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; | |
798 | local_irq_save(flags); | |
799 | __count_vm_event(PGFREE); | |
800 | list_add(&page->lru, &pcp->list); | |
801 | pcp->count++; | |
802 | if (pcp->count >= pcp->high) { | |
803 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
804 | pcp->count -= pcp->batch; | |
805 | } | |
806 | local_irq_restore(flags); | |
807 | put_cpu(); | |
808 | } | |
809 | ||
810 | void fastcall free_hot_page(struct page *page) | |
811 | { | |
812 | free_hot_cold_page(page, 0); | |
813 | } | |
814 | ||
815 | void fastcall free_cold_page(struct page *page) | |
816 | { | |
817 | free_hot_cold_page(page, 1); | |
818 | } | |
819 | ||
820 | /* | |
821 | * split_page takes a non-compound higher-order page, and splits it into | |
822 | * n (1<<order) sub-pages: page[0..n] | |
823 | * Each sub-page must be freed individually. | |
824 | * | |
825 | * Note: this is probably too low level an operation for use in drivers. | |
826 | * Please consult with lkml before using this in your driver. | |
827 | */ | |
828 | void split_page(struct page *page, unsigned int order) | |
829 | { | |
830 | int i; | |
831 | ||
832 | VM_BUG_ON(PageCompound(page)); | |
833 | VM_BUG_ON(!page_count(page)); | |
834 | for (i = 1; i < (1 << order); i++) | |
835 | set_page_refcounted(page + i); | |
836 | } | |
837 | ||
838 | /* | |
839 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
840 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
841 | * or two. | |
842 | */ | |
843 | static struct page *buffered_rmqueue(struct zonelist *zonelist, | |
844 | struct zone *zone, int order, gfp_t gfp_flags) | |
845 | { | |
846 | unsigned long flags; | |
847 | struct page *page; | |
848 | int cold = !!(gfp_flags & __GFP_COLD); | |
849 | int cpu; | |
850 | ||
851 | again: | |
852 | cpu = get_cpu(); | |
853 | if (likely(order == 0)) { | |
854 | struct per_cpu_pages *pcp; | |
855 | ||
856 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; | |
857 | local_irq_save(flags); | |
858 | if (!pcp->count) { | |
859 | pcp->count = rmqueue_bulk(zone, 0, | |
860 | pcp->batch, &pcp->list); | |
861 | if (unlikely(!pcp->count)) | |
862 | goto failed; | |
863 | } | |
864 | page = list_entry(pcp->list.next, struct page, lru); | |
865 | list_del(&page->lru); | |
866 | pcp->count--; | |
867 | } else { | |
868 | spin_lock_irqsave(&zone->lock, flags); | |
869 | page = __rmqueue(zone, order); | |
870 | spin_unlock(&zone->lock); | |
871 | if (!page) | |
872 | goto failed; | |
873 | } | |
874 | ||
875 | __count_zone_vm_events(PGALLOC, zone, 1 << order); | |
876 | zone_statistics(zonelist, zone); | |
877 | local_irq_restore(flags); | |
878 | put_cpu(); | |
879 | ||
880 | VM_BUG_ON(bad_range(zone, page)); | |
881 | if (prep_new_page(page, order, gfp_flags)) | |
882 | goto again; | |
883 | return page; | |
884 | ||
885 | failed: | |
886 | local_irq_restore(flags); | |
887 | put_cpu(); | |
888 | return NULL; | |
889 | } | |
890 | ||
891 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ | |
892 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ | |
893 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
894 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
895 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
896 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
897 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
898 | ||
899 | #ifdef CONFIG_FAIL_PAGE_ALLOC | |
900 | ||
901 | static struct fail_page_alloc_attr { | |
902 | struct fault_attr attr; | |
903 | ||
904 | u32 ignore_gfp_highmem; | |
905 | u32 ignore_gfp_wait; | |
906 | ||
907 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
908 | ||
909 | struct dentry *ignore_gfp_highmem_file; | |
910 | struct dentry *ignore_gfp_wait_file; | |
911 | ||
912 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
913 | ||
914 | } fail_page_alloc = { | |
915 | .attr = FAULT_ATTR_INITIALIZER, | |
916 | .ignore_gfp_wait = 1, | |
917 | .ignore_gfp_highmem = 1, | |
918 | }; | |
919 | ||
920 | static int __init setup_fail_page_alloc(char *str) | |
921 | { | |
922 | return setup_fault_attr(&fail_page_alloc.attr, str); | |
923 | } | |
924 | __setup("fail_page_alloc=", setup_fail_page_alloc); | |
925 | ||
926 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
927 | { | |
928 | if (gfp_mask & __GFP_NOFAIL) | |
929 | return 0; | |
930 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) | |
931 | return 0; | |
932 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) | |
933 | return 0; | |
934 | ||
935 | return should_fail(&fail_page_alloc.attr, 1 << order); | |
936 | } | |
937 | ||
938 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
939 | ||
940 | static int __init fail_page_alloc_debugfs(void) | |
941 | { | |
942 | mode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
943 | struct dentry *dir; | |
944 | int err; | |
945 | ||
946 | err = init_fault_attr_dentries(&fail_page_alloc.attr, | |
947 | "fail_page_alloc"); | |
948 | if (err) | |
949 | return err; | |
950 | dir = fail_page_alloc.attr.dentries.dir; | |
951 | ||
952 | fail_page_alloc.ignore_gfp_wait_file = | |
953 | debugfs_create_bool("ignore-gfp-wait", mode, dir, | |
954 | &fail_page_alloc.ignore_gfp_wait); | |
955 | ||
956 | fail_page_alloc.ignore_gfp_highmem_file = | |
957 | debugfs_create_bool("ignore-gfp-highmem", mode, dir, | |
958 | &fail_page_alloc.ignore_gfp_highmem); | |
959 | ||
960 | if (!fail_page_alloc.ignore_gfp_wait_file || | |
961 | !fail_page_alloc.ignore_gfp_highmem_file) { | |
962 | err = -ENOMEM; | |
963 | debugfs_remove(fail_page_alloc.ignore_gfp_wait_file); | |
964 | debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file); | |
965 | cleanup_fault_attr_dentries(&fail_page_alloc.attr); | |
966 | } | |
967 | ||
968 | return err; | |
969 | } | |
970 | ||
971 | late_initcall(fail_page_alloc_debugfs); | |
972 | ||
973 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
974 | ||
975 | #else /* CONFIG_FAIL_PAGE_ALLOC */ | |
976 | ||
977 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) | |
978 | { | |
979 | return 0; | |
980 | } | |
981 | ||
982 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ | |
983 | ||
984 | /* | |
985 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
986 | * of the allocation. | |
987 | */ | |
988 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
989 | int classzone_idx, int alloc_flags) | |
990 | { | |
991 | /* free_pages my go negative - that's OK */ | |
992 | unsigned long min = mark; | |
993 | long free_pages = z->free_pages - (1 << order) + 1; | |
994 | int o; | |
995 | ||
996 | if (alloc_flags & ALLOC_HIGH) | |
997 | min -= min / 2; | |
998 | if (alloc_flags & ALLOC_HARDER) | |
999 | min -= min / 4; | |
1000 | ||
1001 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
1002 | return 0; | |
1003 | for (o = 0; o < order; o++) { | |
1004 | /* At the next order, this order's pages become unavailable */ | |
1005 | free_pages -= z->free_area[o].nr_free << o; | |
1006 | ||
1007 | /* Require fewer higher order pages to be free */ | |
1008 | min >>= 1; | |
1009 | ||
1010 | if (free_pages <= min) | |
1011 | return 0; | |
1012 | } | |
1013 | return 1; | |
1014 | } | |
1015 | ||
1016 | #ifdef CONFIG_NUMA | |
1017 | /* | |
1018 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to | |
1019 | * skip over zones that are not allowed by the cpuset, or that have | |
1020 | * been recently (in last second) found to be nearly full. See further | |
1021 | * comments in mmzone.h. Reduces cache footprint of zonelist scans | |
1022 | * that have to skip over alot of full or unallowed zones. | |
1023 | * | |
1024 | * If the zonelist cache is present in the passed in zonelist, then | |
1025 | * returns a pointer to the allowed node mask (either the current | |
1026 | * tasks mems_allowed, or node_online_map.) | |
1027 | * | |
1028 | * If the zonelist cache is not available for this zonelist, does | |
1029 | * nothing and returns NULL. | |
1030 | * | |
1031 | * If the fullzones BITMAP in the zonelist cache is stale (more than | |
1032 | * a second since last zap'd) then we zap it out (clear its bits.) | |
1033 | * | |
1034 | * We hold off even calling zlc_setup, until after we've checked the | |
1035 | * first zone in the zonelist, on the theory that most allocations will | |
1036 | * be satisfied from that first zone, so best to examine that zone as | |
1037 | * quickly as we can. | |
1038 | */ | |
1039 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1040 | { | |
1041 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1042 | nodemask_t *allowednodes; /* zonelist_cache approximation */ | |
1043 | ||
1044 | zlc = zonelist->zlcache_ptr; | |
1045 | if (!zlc) | |
1046 | return NULL; | |
1047 | ||
1048 | if (jiffies - zlc->last_full_zap > 1 * HZ) { | |
1049 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1050 | zlc->last_full_zap = jiffies; | |
1051 | } | |
1052 | ||
1053 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? | |
1054 | &cpuset_current_mems_allowed : | |
1055 | &node_online_map; | |
1056 | return allowednodes; | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * Given 'z' scanning a zonelist, run a couple of quick checks to see | |
1061 | * if it is worth looking at further for free memory: | |
1062 | * 1) Check that the zone isn't thought to be full (doesn't have its | |
1063 | * bit set in the zonelist_cache fullzones BITMAP). | |
1064 | * 2) Check that the zones node (obtained from the zonelist_cache | |
1065 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. | |
1066 | * Return true (non-zero) if zone is worth looking at further, or | |
1067 | * else return false (zero) if it is not. | |
1068 | * | |
1069 | * This check -ignores- the distinction between various watermarks, | |
1070 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is | |
1071 | * found to be full for any variation of these watermarks, it will | |
1072 | * be considered full for up to one second by all requests, unless | |
1073 | * we are so low on memory on all allowed nodes that we are forced | |
1074 | * into the second scan of the zonelist. | |
1075 | * | |
1076 | * In the second scan we ignore this zonelist cache and exactly | |
1077 | * apply the watermarks to all zones, even it is slower to do so. | |
1078 | * We are low on memory in the second scan, and should leave no stone | |
1079 | * unturned looking for a free page. | |
1080 | */ | |
1081 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1082 | nodemask_t *allowednodes) | |
1083 | { | |
1084 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1085 | int i; /* index of *z in zonelist zones */ | |
1086 | int n; /* node that zone *z is on */ | |
1087 | ||
1088 | zlc = zonelist->zlcache_ptr; | |
1089 | if (!zlc) | |
1090 | return 1; | |
1091 | ||
1092 | i = z - zonelist->zones; | |
1093 | n = zlc->z_to_n[i]; | |
1094 | ||
1095 | /* This zone is worth trying if it is allowed but not full */ | |
1096 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); | |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * Given 'z' scanning a zonelist, set the corresponding bit in | |
1101 | * zlc->fullzones, so that subsequent attempts to allocate a page | |
1102 | * from that zone don't waste time re-examining it. | |
1103 | */ | |
1104 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1105 | { | |
1106 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ | |
1107 | int i; /* index of *z in zonelist zones */ | |
1108 | ||
1109 | zlc = zonelist->zlcache_ptr; | |
1110 | if (!zlc) | |
1111 | return; | |
1112 | ||
1113 | i = z - zonelist->zones; | |
1114 | ||
1115 | set_bit(i, zlc->fullzones); | |
1116 | } | |
1117 | ||
1118 | #else /* CONFIG_NUMA */ | |
1119 | ||
1120 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) | |
1121 | { | |
1122 | return NULL; | |
1123 | } | |
1124 | ||
1125 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zone **z, | |
1126 | nodemask_t *allowednodes) | |
1127 | { | |
1128 | return 1; | |
1129 | } | |
1130 | ||
1131 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zone **z) | |
1132 | { | |
1133 | } | |
1134 | #endif /* CONFIG_NUMA */ | |
1135 | ||
1136 | /* | |
1137 | * get_page_from_freelist goes through the zonelist trying to allocate | |
1138 | * a page. | |
1139 | */ | |
1140 | static struct page * | |
1141 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
1142 | struct zonelist *zonelist, int alloc_flags) | |
1143 | { | |
1144 | struct zone **z; | |
1145 | struct page *page = NULL; | |
1146 | int classzone_idx = zone_idx(zonelist->zones[0]); | |
1147 | struct zone *zone; | |
1148 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ | |
1149 | int zlc_active = 0; /* set if using zonelist_cache */ | |
1150 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ | |
1151 | ||
1152 | zonelist_scan: | |
1153 | /* | |
1154 | * Scan zonelist, looking for a zone with enough free. | |
1155 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
1156 | */ | |
1157 | z = zonelist->zones; | |
1158 | ||
1159 | do { | |
1160 | if (NUMA_BUILD && zlc_active && | |
1161 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) | |
1162 | continue; | |
1163 | zone = *z; | |
1164 | if (unlikely(NUMA_BUILD && (gfp_mask & __GFP_THISNODE) && | |
1165 | zone->zone_pgdat != zonelist->zones[0]->zone_pgdat)) | |
1166 | break; | |
1167 | if ((alloc_flags & ALLOC_CPUSET) && | |
1168 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
1169 | goto try_next_zone; | |
1170 | ||
1171 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
1172 | unsigned long mark; | |
1173 | if (alloc_flags & ALLOC_WMARK_MIN) | |
1174 | mark = zone->pages_min; | |
1175 | else if (alloc_flags & ALLOC_WMARK_LOW) | |
1176 | mark = zone->pages_low; | |
1177 | else | |
1178 | mark = zone->pages_high; | |
1179 | if (!zone_watermark_ok(zone, order, mark, | |
1180 | classzone_idx, alloc_flags)) { | |
1181 | if (!zone_reclaim_mode || | |
1182 | !zone_reclaim(zone, gfp_mask, order)) | |
1183 | goto this_zone_full; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | page = buffered_rmqueue(zonelist, zone, order, gfp_mask); | |
1188 | if (page) | |
1189 | break; | |
1190 | this_zone_full: | |
1191 | if (NUMA_BUILD) | |
1192 | zlc_mark_zone_full(zonelist, z); | |
1193 | try_next_zone: | |
1194 | if (NUMA_BUILD && !did_zlc_setup) { | |
1195 | /* we do zlc_setup after the first zone is tried */ | |
1196 | allowednodes = zlc_setup(zonelist, alloc_flags); | |
1197 | zlc_active = 1; | |
1198 | did_zlc_setup = 1; | |
1199 | } | |
1200 | } while (*(++z) != NULL); | |
1201 | ||
1202 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { | |
1203 | /* Disable zlc cache for second zonelist scan */ | |
1204 | zlc_active = 0; | |
1205 | goto zonelist_scan; | |
1206 | } | |
1207 | return page; | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * This is the 'heart' of the zoned buddy allocator. | |
1212 | */ | |
1213 | struct page * fastcall | |
1214 | __alloc_pages(gfp_t gfp_mask, unsigned int order, | |
1215 | struct zonelist *zonelist) | |
1216 | { | |
1217 | const gfp_t wait = gfp_mask & __GFP_WAIT; | |
1218 | struct zone **z; | |
1219 | struct page *page; | |
1220 | struct reclaim_state reclaim_state; | |
1221 | struct task_struct *p = current; | |
1222 | int do_retry; | |
1223 | int alloc_flags; | |
1224 | int did_some_progress; | |
1225 | ||
1226 | might_sleep_if(wait); | |
1227 | ||
1228 | if (should_fail_alloc_page(gfp_mask, order)) | |
1229 | return NULL; | |
1230 | ||
1231 | restart: | |
1232 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ | |
1233 | ||
1234 | if (unlikely(*z == NULL)) { | |
1235 | /* Should this ever happen?? */ | |
1236 | return NULL; | |
1237 | } | |
1238 | ||
1239 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, | |
1240 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); | |
1241 | if (page) | |
1242 | goto got_pg; | |
1243 | ||
1244 | /* | |
1245 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and | |
1246 | * __GFP_NOWARN set) should not cause reclaim since the subsystem | |
1247 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim | |
1248 | * using a larger set of nodes after it has established that the | |
1249 | * allowed per node queues are empty and that nodes are | |
1250 | * over allocated. | |
1251 | */ | |
1252 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) | |
1253 | goto nopage; | |
1254 | ||
1255 | for (z = zonelist->zones; *z; z++) | |
1256 | wakeup_kswapd(*z, order); | |
1257 | ||
1258 | /* | |
1259 | * OK, we're below the kswapd watermark and have kicked background | |
1260 | * reclaim. Now things get more complex, so set up alloc_flags according | |
1261 | * to how we want to proceed. | |
1262 | * | |
1263 | * The caller may dip into page reserves a bit more if the caller | |
1264 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
1265 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will | |
1266 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
1267 | */ | |
1268 | alloc_flags = ALLOC_WMARK_MIN; | |
1269 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) | |
1270 | alloc_flags |= ALLOC_HARDER; | |
1271 | if (gfp_mask & __GFP_HIGH) | |
1272 | alloc_flags |= ALLOC_HIGH; | |
1273 | if (wait) | |
1274 | alloc_flags |= ALLOC_CPUSET; | |
1275 | ||
1276 | /* | |
1277 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
1278 | * coming from realtime tasks go deeper into reserves. | |
1279 | * | |
1280 | * This is the last chance, in general, before the goto nopage. | |
1281 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
1282 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
1283 | */ | |
1284 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); | |
1285 | if (page) | |
1286 | goto got_pg; | |
1287 | ||
1288 | /* This allocation should allow future memory freeing. */ | |
1289 | ||
1290 | rebalance: | |
1291 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
1292 | && !in_interrupt()) { | |
1293 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
1294 | nofail_alloc: | |
1295 | /* go through the zonelist yet again, ignoring mins */ | |
1296 | page = get_page_from_freelist(gfp_mask, order, | |
1297 | zonelist, ALLOC_NO_WATERMARKS); | |
1298 | if (page) | |
1299 | goto got_pg; | |
1300 | if (gfp_mask & __GFP_NOFAIL) { | |
1301 | congestion_wait(WRITE, HZ/50); | |
1302 | goto nofail_alloc; | |
1303 | } | |
1304 | } | |
1305 | goto nopage; | |
1306 | } | |
1307 | ||
1308 | /* Atomic allocations - we can't balance anything */ | |
1309 | if (!wait) | |
1310 | goto nopage; | |
1311 | ||
1312 | cond_resched(); | |
1313 | ||
1314 | /* We now go into synchronous reclaim */ | |
1315 | cpuset_memory_pressure_bump(); | |
1316 | p->flags |= PF_MEMALLOC; | |
1317 | reclaim_state.reclaimed_slab = 0; | |
1318 | p->reclaim_state = &reclaim_state; | |
1319 | ||
1320 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); | |
1321 | ||
1322 | p->reclaim_state = NULL; | |
1323 | p->flags &= ~PF_MEMALLOC; | |
1324 | ||
1325 | cond_resched(); | |
1326 | ||
1327 | if (likely(did_some_progress)) { | |
1328 | page = get_page_from_freelist(gfp_mask, order, | |
1329 | zonelist, alloc_flags); | |
1330 | if (page) | |
1331 | goto got_pg; | |
1332 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { | |
1333 | /* | |
1334 | * Go through the zonelist yet one more time, keep | |
1335 | * very high watermark here, this is only to catch | |
1336 | * a parallel oom killing, we must fail if we're still | |
1337 | * under heavy pressure. | |
1338 | */ | |
1339 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, | |
1340 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); | |
1341 | if (page) | |
1342 | goto got_pg; | |
1343 | ||
1344 | out_of_memory(zonelist, gfp_mask, order); | |
1345 | goto restart; | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * Don't let big-order allocations loop unless the caller explicitly | |
1350 | * requests that. Wait for some write requests to complete then retry. | |
1351 | * | |
1352 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1353 | * <= 3, but that may not be true in other implementations. | |
1354 | */ | |
1355 | do_retry = 0; | |
1356 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1357 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1358 | do_retry = 1; | |
1359 | if (gfp_mask & __GFP_NOFAIL) | |
1360 | do_retry = 1; | |
1361 | } | |
1362 | if (do_retry) { | |
1363 | congestion_wait(WRITE, HZ/50); | |
1364 | goto rebalance; | |
1365 | } | |
1366 | ||
1367 | nopage: | |
1368 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1369 | printk(KERN_WARNING "%s: page allocation failure." | |
1370 | " order:%d, mode:0x%x\n", | |
1371 | p->comm, order, gfp_mask); | |
1372 | dump_stack(); | |
1373 | show_mem(); | |
1374 | } | |
1375 | got_pg: | |
1376 | return page; | |
1377 | } | |
1378 | ||
1379 | EXPORT_SYMBOL(__alloc_pages); | |
1380 | ||
1381 | /* | |
1382 | * Common helper functions. | |
1383 | */ | |
1384 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) | |
1385 | { | |
1386 | struct page * page; | |
1387 | page = alloc_pages(gfp_mask, order); | |
1388 | if (!page) | |
1389 | return 0; | |
1390 | return (unsigned long) page_address(page); | |
1391 | } | |
1392 | ||
1393 | EXPORT_SYMBOL(__get_free_pages); | |
1394 | ||
1395 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) | |
1396 | { | |
1397 | struct page * page; | |
1398 | ||
1399 | /* | |
1400 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1401 | * a highmem page | |
1402 | */ | |
1403 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); | |
1404 | ||
1405 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1406 | if (page) | |
1407 | return (unsigned long) page_address(page); | |
1408 | return 0; | |
1409 | } | |
1410 | ||
1411 | EXPORT_SYMBOL(get_zeroed_page); | |
1412 | ||
1413 | void __pagevec_free(struct pagevec *pvec) | |
1414 | { | |
1415 | int i = pagevec_count(pvec); | |
1416 | ||
1417 | while (--i >= 0) | |
1418 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1419 | } | |
1420 | ||
1421 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1422 | { | |
1423 | if (put_page_testzero(page)) { | |
1424 | if (order == 0) | |
1425 | free_hot_page(page); | |
1426 | else | |
1427 | __free_pages_ok(page, order); | |
1428 | } | |
1429 | } | |
1430 | ||
1431 | EXPORT_SYMBOL(__free_pages); | |
1432 | ||
1433 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1434 | { | |
1435 | if (addr != 0) { | |
1436 | VM_BUG_ON(!virt_addr_valid((void *)addr)); | |
1437 | __free_pages(virt_to_page((void *)addr), order); | |
1438 | } | |
1439 | } | |
1440 | ||
1441 | EXPORT_SYMBOL(free_pages); | |
1442 | ||
1443 | /* | |
1444 | * Total amount of free (allocatable) RAM: | |
1445 | */ | |
1446 | unsigned int nr_free_pages(void) | |
1447 | { | |
1448 | unsigned int sum = 0; | |
1449 | struct zone *zone; | |
1450 | ||
1451 | for_each_zone(zone) | |
1452 | sum += zone->free_pages; | |
1453 | ||
1454 | return sum; | |
1455 | } | |
1456 | ||
1457 | EXPORT_SYMBOL(nr_free_pages); | |
1458 | ||
1459 | #ifdef CONFIG_NUMA | |
1460 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1461 | { | |
1462 | unsigned int sum = 0; | |
1463 | enum zone_type i; | |
1464 | ||
1465 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1466 | sum += pgdat->node_zones[i].free_pages; | |
1467 | ||
1468 | return sum; | |
1469 | } | |
1470 | #endif | |
1471 | ||
1472 | static unsigned int nr_free_zone_pages(int offset) | |
1473 | { | |
1474 | /* Just pick one node, since fallback list is circular */ | |
1475 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1476 | unsigned int sum = 0; | |
1477 | ||
1478 | struct zonelist *zonelist = pgdat->node_zonelists + offset; | |
1479 | struct zone **zonep = zonelist->zones; | |
1480 | struct zone *zone; | |
1481 | ||
1482 | for (zone = *zonep++; zone; zone = *zonep++) { | |
1483 | unsigned long size = zone->present_pages; | |
1484 | unsigned long high = zone->pages_high; | |
1485 | if (size > high) | |
1486 | sum += size - high; | |
1487 | } | |
1488 | ||
1489 | return sum; | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1494 | */ | |
1495 | unsigned int nr_free_buffer_pages(void) | |
1496 | { | |
1497 | return nr_free_zone_pages(gfp_zone(GFP_USER)); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Amount of free RAM allocatable within all zones | |
1502 | */ | |
1503 | unsigned int nr_free_pagecache_pages(void) | |
1504 | { | |
1505 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); | |
1506 | } | |
1507 | ||
1508 | static inline void show_node(struct zone *zone) | |
1509 | { | |
1510 | if (NUMA_BUILD) | |
1511 | printk("Node %d ", zone_to_nid(zone)); | |
1512 | } | |
1513 | ||
1514 | void si_meminfo(struct sysinfo *val) | |
1515 | { | |
1516 | val->totalram = totalram_pages; | |
1517 | val->sharedram = 0; | |
1518 | val->freeram = nr_free_pages(); | |
1519 | val->bufferram = nr_blockdev_pages(); | |
1520 | val->totalhigh = totalhigh_pages; | |
1521 | val->freehigh = nr_free_highpages(); | |
1522 | val->mem_unit = PAGE_SIZE; | |
1523 | } | |
1524 | ||
1525 | EXPORT_SYMBOL(si_meminfo); | |
1526 | ||
1527 | #ifdef CONFIG_NUMA | |
1528 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1529 | { | |
1530 | pg_data_t *pgdat = NODE_DATA(nid); | |
1531 | ||
1532 | val->totalram = pgdat->node_present_pages; | |
1533 | val->freeram = nr_free_pages_pgdat(pgdat); | |
1534 | #ifdef CONFIG_HIGHMEM | |
1535 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
1536 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1537 | #else | |
1538 | val->totalhigh = 0; | |
1539 | val->freehigh = 0; | |
1540 | #endif | |
1541 | val->mem_unit = PAGE_SIZE; | |
1542 | } | |
1543 | #endif | |
1544 | ||
1545 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1546 | ||
1547 | /* | |
1548 | * Show free area list (used inside shift_scroll-lock stuff) | |
1549 | * We also calculate the percentage fragmentation. We do this by counting the | |
1550 | * memory on each free list with the exception of the first item on the list. | |
1551 | */ | |
1552 | void show_free_areas(void) | |
1553 | { | |
1554 | int cpu; | |
1555 | unsigned long active; | |
1556 | unsigned long inactive; | |
1557 | unsigned long free; | |
1558 | struct zone *zone; | |
1559 | ||
1560 | for_each_zone(zone) { | |
1561 | if (!populated_zone(zone)) | |
1562 | continue; | |
1563 | ||
1564 | show_node(zone); | |
1565 | printk("%s per-cpu:\n", zone->name); | |
1566 | ||
1567 | for_each_online_cpu(cpu) { | |
1568 | struct per_cpu_pageset *pageset; | |
1569 | ||
1570 | pageset = zone_pcp(zone, cpu); | |
1571 | ||
1572 | printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d " | |
1573 | "Cold: hi:%5d, btch:%4d usd:%4d\n", | |
1574 | cpu, pageset->pcp[0].high, | |
1575 | pageset->pcp[0].batch, pageset->pcp[0].count, | |
1576 | pageset->pcp[1].high, pageset->pcp[1].batch, | |
1577 | pageset->pcp[1].count); | |
1578 | } | |
1579 | } | |
1580 | ||
1581 | get_zone_counts(&active, &inactive, &free); | |
1582 | ||
1583 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " | |
1584 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1585 | active, | |
1586 | inactive, | |
1587 | global_page_state(NR_FILE_DIRTY), | |
1588 | global_page_state(NR_WRITEBACK), | |
1589 | global_page_state(NR_UNSTABLE_NFS), | |
1590 | nr_free_pages(), | |
1591 | global_page_state(NR_SLAB_RECLAIMABLE) + | |
1592 | global_page_state(NR_SLAB_UNRECLAIMABLE), | |
1593 | global_page_state(NR_FILE_MAPPED), | |
1594 | global_page_state(NR_PAGETABLE)); | |
1595 | ||
1596 | for_each_zone(zone) { | |
1597 | int i; | |
1598 | ||
1599 | if (!populated_zone(zone)) | |
1600 | continue; | |
1601 | ||
1602 | show_node(zone); | |
1603 | printk("%s" | |
1604 | " free:%lukB" | |
1605 | " min:%lukB" | |
1606 | " low:%lukB" | |
1607 | " high:%lukB" | |
1608 | " active:%lukB" | |
1609 | " inactive:%lukB" | |
1610 | " present:%lukB" | |
1611 | " pages_scanned:%lu" | |
1612 | " all_unreclaimable? %s" | |
1613 | "\n", | |
1614 | zone->name, | |
1615 | K(zone->free_pages), | |
1616 | K(zone->pages_min), | |
1617 | K(zone->pages_low), | |
1618 | K(zone->pages_high), | |
1619 | K(zone->nr_active), | |
1620 | K(zone->nr_inactive), | |
1621 | K(zone->present_pages), | |
1622 | zone->pages_scanned, | |
1623 | (zone->all_unreclaimable ? "yes" : "no") | |
1624 | ); | |
1625 | printk("lowmem_reserve[]:"); | |
1626 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1627 | printk(" %lu", zone->lowmem_reserve[i]); | |
1628 | printk("\n"); | |
1629 | } | |
1630 | ||
1631 | for_each_zone(zone) { | |
1632 | unsigned long nr[MAX_ORDER], flags, order, total = 0; | |
1633 | ||
1634 | if (!populated_zone(zone)) | |
1635 | continue; | |
1636 | ||
1637 | show_node(zone); | |
1638 | printk("%s: ", zone->name); | |
1639 | ||
1640 | spin_lock_irqsave(&zone->lock, flags); | |
1641 | for (order = 0; order < MAX_ORDER; order++) { | |
1642 | nr[order] = zone->free_area[order].nr_free; | |
1643 | total += nr[order] << order; | |
1644 | } | |
1645 | spin_unlock_irqrestore(&zone->lock, flags); | |
1646 | for (order = 0; order < MAX_ORDER; order++) | |
1647 | printk("%lu*%lukB ", nr[order], K(1UL) << order); | |
1648 | printk("= %lukB\n", K(total)); | |
1649 | } | |
1650 | ||
1651 | show_swap_cache_info(); | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * Builds allocation fallback zone lists. | |
1656 | * | |
1657 | * Add all populated zones of a node to the zonelist. | |
1658 | */ | |
1659 | static int __meminit build_zonelists_node(pg_data_t *pgdat, | |
1660 | struct zonelist *zonelist, int nr_zones, enum zone_type zone_type) | |
1661 | { | |
1662 | struct zone *zone; | |
1663 | ||
1664 | BUG_ON(zone_type >= MAX_NR_ZONES); | |
1665 | zone_type++; | |
1666 | ||
1667 | do { | |
1668 | zone_type--; | |
1669 | zone = pgdat->node_zones + zone_type; | |
1670 | if (populated_zone(zone)) { | |
1671 | zonelist->zones[nr_zones++] = zone; | |
1672 | check_highest_zone(zone_type); | |
1673 | } | |
1674 | ||
1675 | } while (zone_type); | |
1676 | return nr_zones; | |
1677 | } | |
1678 | ||
1679 | #ifdef CONFIG_NUMA | |
1680 | #define MAX_NODE_LOAD (num_online_nodes()) | |
1681 | static int __meminitdata node_load[MAX_NUMNODES]; | |
1682 | /** | |
1683 | * find_next_best_node - find the next node that should appear in a given node's fallback list | |
1684 | * @node: node whose fallback list we're appending | |
1685 | * @used_node_mask: nodemask_t of already used nodes | |
1686 | * | |
1687 | * We use a number of factors to determine which is the next node that should | |
1688 | * appear on a given node's fallback list. The node should not have appeared | |
1689 | * already in @node's fallback list, and it should be the next closest node | |
1690 | * according to the distance array (which contains arbitrary distance values | |
1691 | * from each node to each node in the system), and should also prefer nodes | |
1692 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1693 | * on them otherwise. | |
1694 | * It returns -1 if no node is found. | |
1695 | */ | |
1696 | static int __meminit find_next_best_node(int node, nodemask_t *used_node_mask) | |
1697 | { | |
1698 | int n, val; | |
1699 | int min_val = INT_MAX; | |
1700 | int best_node = -1; | |
1701 | ||
1702 | /* Use the local node if we haven't already */ | |
1703 | if (!node_isset(node, *used_node_mask)) { | |
1704 | node_set(node, *used_node_mask); | |
1705 | return node; | |
1706 | } | |
1707 | ||
1708 | for_each_online_node(n) { | |
1709 | cpumask_t tmp; | |
1710 | ||
1711 | /* Don't want a node to appear more than once */ | |
1712 | if (node_isset(n, *used_node_mask)) | |
1713 | continue; | |
1714 | ||
1715 | /* Use the distance array to find the distance */ | |
1716 | val = node_distance(node, n); | |
1717 | ||
1718 | /* Penalize nodes under us ("prefer the next node") */ | |
1719 | val += (n < node); | |
1720 | ||
1721 | /* Give preference to headless and unused nodes */ | |
1722 | tmp = node_to_cpumask(n); | |
1723 | if (!cpus_empty(tmp)) | |
1724 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1725 | ||
1726 | /* Slight preference for less loaded node */ | |
1727 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1728 | val += node_load[n]; | |
1729 | ||
1730 | if (val < min_val) { | |
1731 | min_val = val; | |
1732 | best_node = n; | |
1733 | } | |
1734 | } | |
1735 | ||
1736 | if (best_node >= 0) | |
1737 | node_set(best_node, *used_node_mask); | |
1738 | ||
1739 | return best_node; | |
1740 | } | |
1741 | ||
1742 | static void __meminit build_zonelists(pg_data_t *pgdat) | |
1743 | { | |
1744 | int j, node, local_node; | |
1745 | enum zone_type i; | |
1746 | int prev_node, load; | |
1747 | struct zonelist *zonelist; | |
1748 | nodemask_t used_mask; | |
1749 | ||
1750 | /* initialize zonelists */ | |
1751 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1752 | zonelist = pgdat->node_zonelists + i; | |
1753 | zonelist->zones[0] = NULL; | |
1754 | } | |
1755 | ||
1756 | /* NUMA-aware ordering of nodes */ | |
1757 | local_node = pgdat->node_id; | |
1758 | load = num_online_nodes(); | |
1759 | prev_node = local_node; | |
1760 | nodes_clear(used_mask); | |
1761 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
1762 | int distance = node_distance(local_node, node); | |
1763 | ||
1764 | /* | |
1765 | * If another node is sufficiently far away then it is better | |
1766 | * to reclaim pages in a zone before going off node. | |
1767 | */ | |
1768 | if (distance > RECLAIM_DISTANCE) | |
1769 | zone_reclaim_mode = 1; | |
1770 | ||
1771 | /* | |
1772 | * We don't want to pressure a particular node. | |
1773 | * So adding penalty to the first node in same | |
1774 | * distance group to make it round-robin. | |
1775 | */ | |
1776 | ||
1777 | if (distance != node_distance(local_node, prev_node)) | |
1778 | node_load[node] += load; | |
1779 | prev_node = node; | |
1780 | load--; | |
1781 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1782 | zonelist = pgdat->node_zonelists + i; | |
1783 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1784 | ||
1785 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
1786 | zonelist->zones[j] = NULL; | |
1787 | } | |
1788 | } | |
1789 | } | |
1790 | ||
1791 | /* Construct the zonelist performance cache - see further mmzone.h */ | |
1792 | static void __meminit build_zonelist_cache(pg_data_t *pgdat) | |
1793 | { | |
1794 | int i; | |
1795 | ||
1796 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1797 | struct zonelist *zonelist; | |
1798 | struct zonelist_cache *zlc; | |
1799 | struct zone **z; | |
1800 | ||
1801 | zonelist = pgdat->node_zonelists + i; | |
1802 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; | |
1803 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); | |
1804 | for (z = zonelist->zones; *z; z++) | |
1805 | zlc->z_to_n[z - zonelist->zones] = zone_to_nid(*z); | |
1806 | } | |
1807 | } | |
1808 | ||
1809 | #else /* CONFIG_NUMA */ | |
1810 | ||
1811 | static void __meminit build_zonelists(pg_data_t *pgdat) | |
1812 | { | |
1813 | int node, local_node; | |
1814 | enum zone_type i,j; | |
1815 | ||
1816 | local_node = pgdat->node_id; | |
1817 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1818 | struct zonelist *zonelist; | |
1819 | ||
1820 | zonelist = pgdat->node_zonelists + i; | |
1821 | ||
1822 | j = build_zonelists_node(pgdat, zonelist, 0, i); | |
1823 | /* | |
1824 | * Now we build the zonelist so that it contains the zones | |
1825 | * of all the other nodes. | |
1826 | * We don't want to pressure a particular node, so when | |
1827 | * building the zones for node N, we make sure that the | |
1828 | * zones coming right after the local ones are those from | |
1829 | * node N+1 (modulo N) | |
1830 | */ | |
1831 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1832 | if (!node_online(node)) | |
1833 | continue; | |
1834 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
1835 | } | |
1836 | for (node = 0; node < local_node; node++) { | |
1837 | if (!node_online(node)) | |
1838 | continue; | |
1839 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, i); | |
1840 | } | |
1841 | ||
1842 | zonelist->zones[j] = NULL; | |
1843 | } | |
1844 | } | |
1845 | ||
1846 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ | |
1847 | static void __meminit build_zonelist_cache(pg_data_t *pgdat) | |
1848 | { | |
1849 | int i; | |
1850 | ||
1851 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1852 | pgdat->node_zonelists[i].zlcache_ptr = NULL; | |
1853 | } | |
1854 | ||
1855 | #endif /* CONFIG_NUMA */ | |
1856 | ||
1857 | /* return values int ....just for stop_machine_run() */ | |
1858 | static int __meminit __build_all_zonelists(void *dummy) | |
1859 | { | |
1860 | int nid; | |
1861 | ||
1862 | for_each_online_node(nid) { | |
1863 | build_zonelists(NODE_DATA(nid)); | |
1864 | build_zonelist_cache(NODE_DATA(nid)); | |
1865 | } | |
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | void __meminit build_all_zonelists(void) | |
1870 | { | |
1871 | if (system_state == SYSTEM_BOOTING) { | |
1872 | __build_all_zonelists(NULL); | |
1873 | cpuset_init_current_mems_allowed(); | |
1874 | } else { | |
1875 | /* we have to stop all cpus to guaranntee there is no user | |
1876 | of zonelist */ | |
1877 | stop_machine_run(__build_all_zonelists, NULL, NR_CPUS); | |
1878 | /* cpuset refresh routine should be here */ | |
1879 | } | |
1880 | vm_total_pages = nr_free_pagecache_pages(); | |
1881 | printk("Built %i zonelists. Total pages: %ld\n", | |
1882 | num_online_nodes(), vm_total_pages); | |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * Helper functions to size the waitqueue hash table. | |
1887 | * Essentially these want to choose hash table sizes sufficiently | |
1888 | * large so that collisions trying to wait on pages are rare. | |
1889 | * But in fact, the number of active page waitqueues on typical | |
1890 | * systems is ridiculously low, less than 200. So this is even | |
1891 | * conservative, even though it seems large. | |
1892 | * | |
1893 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1894 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1895 | */ | |
1896 | #define PAGES_PER_WAITQUEUE 256 | |
1897 | ||
1898 | #ifndef CONFIG_MEMORY_HOTPLUG | |
1899 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1900 | { | |
1901 | unsigned long size = 1; | |
1902 | ||
1903 | pages /= PAGES_PER_WAITQUEUE; | |
1904 | ||
1905 | while (size < pages) | |
1906 | size <<= 1; | |
1907 | ||
1908 | /* | |
1909 | * Once we have dozens or even hundreds of threads sleeping | |
1910 | * on IO we've got bigger problems than wait queue collision. | |
1911 | * Limit the size of the wait table to a reasonable size. | |
1912 | */ | |
1913 | size = min(size, 4096UL); | |
1914 | ||
1915 | return max(size, 4UL); | |
1916 | } | |
1917 | #else | |
1918 | /* | |
1919 | * A zone's size might be changed by hot-add, so it is not possible to determine | |
1920 | * a suitable size for its wait_table. So we use the maximum size now. | |
1921 | * | |
1922 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: | |
1923 | * | |
1924 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. | |
1925 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. | |
1926 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. | |
1927 | * | |
1928 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages | |
1929 | * or more by the traditional way. (See above). It equals: | |
1930 | * | |
1931 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. | |
1932 | * ia64(16K page size) : = ( 8G + 4M)byte. | |
1933 | * powerpc (64K page size) : = (32G +16M)byte. | |
1934 | */ | |
1935 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) | |
1936 | { | |
1937 | return 4096UL; | |
1938 | } | |
1939 | #endif | |
1940 | ||
1941 | /* | |
1942 | * This is an integer logarithm so that shifts can be used later | |
1943 | * to extract the more random high bits from the multiplicative | |
1944 | * hash function before the remainder is taken. | |
1945 | */ | |
1946 | static inline unsigned long wait_table_bits(unsigned long size) | |
1947 | { | |
1948 | return ffz(~size); | |
1949 | } | |
1950 | ||
1951 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1952 | ||
1953 | /* | |
1954 | * Initially all pages are reserved - free ones are freed | |
1955 | * up by free_all_bootmem() once the early boot process is | |
1956 | * done. Non-atomic initialization, single-pass. | |
1957 | */ | |
1958 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, | |
1959 | unsigned long start_pfn) | |
1960 | { | |
1961 | struct page *page; | |
1962 | unsigned long end_pfn = start_pfn + size; | |
1963 | unsigned long pfn; | |
1964 | ||
1965 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
1966 | if (!early_pfn_valid(pfn)) | |
1967 | continue; | |
1968 | if (!early_pfn_in_nid(pfn, nid)) | |
1969 | continue; | |
1970 | page = pfn_to_page(pfn); | |
1971 | set_page_links(page, zone, nid, pfn); | |
1972 | init_page_count(page); | |
1973 | reset_page_mapcount(page); | |
1974 | SetPageReserved(page); | |
1975 | INIT_LIST_HEAD(&page->lru); | |
1976 | #ifdef WANT_PAGE_VIRTUAL | |
1977 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1978 | if (!is_highmem_idx(zone)) | |
1979 | set_page_address(page, __va(pfn << PAGE_SHIFT)); | |
1980 | #endif | |
1981 | } | |
1982 | } | |
1983 | ||
1984 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1985 | unsigned long size) | |
1986 | { | |
1987 | int order; | |
1988 | for (order = 0; order < MAX_ORDER ; order++) { | |
1989 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1990 | zone->free_area[order].nr_free = 0; | |
1991 | } | |
1992 | } | |
1993 | ||
1994 | #ifndef __HAVE_ARCH_MEMMAP_INIT | |
1995 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1996 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1997 | #endif | |
1998 | ||
1999 | static int __cpuinit zone_batchsize(struct zone *zone) | |
2000 | { | |
2001 | int batch; | |
2002 | ||
2003 | /* | |
2004 | * The per-cpu-pages pools are set to around 1000th of the | |
2005 | * size of the zone. But no more than 1/2 of a meg. | |
2006 | * | |
2007 | * OK, so we don't know how big the cache is. So guess. | |
2008 | */ | |
2009 | batch = zone->present_pages / 1024; | |
2010 | if (batch * PAGE_SIZE > 512 * 1024) | |
2011 | batch = (512 * 1024) / PAGE_SIZE; | |
2012 | batch /= 4; /* We effectively *= 4 below */ | |
2013 | if (batch < 1) | |
2014 | batch = 1; | |
2015 | ||
2016 | /* | |
2017 | * Clamp the batch to a 2^n - 1 value. Having a power | |
2018 | * of 2 value was found to be more likely to have | |
2019 | * suboptimal cache aliasing properties in some cases. | |
2020 | * | |
2021 | * For example if 2 tasks are alternately allocating | |
2022 | * batches of pages, one task can end up with a lot | |
2023 | * of pages of one half of the possible page colors | |
2024 | * and the other with pages of the other colors. | |
2025 | */ | |
2026 | batch = (1 << (fls(batch + batch/2)-1)) - 1; | |
2027 | ||
2028 | return batch; | |
2029 | } | |
2030 | ||
2031 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) | |
2032 | { | |
2033 | struct per_cpu_pages *pcp; | |
2034 | ||
2035 | memset(p, 0, sizeof(*p)); | |
2036 | ||
2037 | pcp = &p->pcp[0]; /* hot */ | |
2038 | pcp->count = 0; | |
2039 | pcp->high = 6 * batch; | |
2040 | pcp->batch = max(1UL, 1 * batch); | |
2041 | INIT_LIST_HEAD(&pcp->list); | |
2042 | ||
2043 | pcp = &p->pcp[1]; /* cold*/ | |
2044 | pcp->count = 0; | |
2045 | pcp->high = 2 * batch; | |
2046 | pcp->batch = max(1UL, batch/2); | |
2047 | INIT_LIST_HEAD(&pcp->list); | |
2048 | } | |
2049 | ||
2050 | /* | |
2051 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
2052 | * to the value high for the pageset p. | |
2053 | */ | |
2054 | ||
2055 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
2056 | unsigned long high) | |
2057 | { | |
2058 | struct per_cpu_pages *pcp; | |
2059 | ||
2060 | pcp = &p->pcp[0]; /* hot list */ | |
2061 | pcp->high = high; | |
2062 | pcp->batch = max(1UL, high/4); | |
2063 | if ((high/4) > (PAGE_SHIFT * 8)) | |
2064 | pcp->batch = PAGE_SHIFT * 8; | |
2065 | } | |
2066 | ||
2067 | ||
2068 | #ifdef CONFIG_NUMA | |
2069 | /* | |
2070 | * Boot pageset table. One per cpu which is going to be used for all | |
2071 | * zones and all nodes. The parameters will be set in such a way | |
2072 | * that an item put on a list will immediately be handed over to | |
2073 | * the buddy list. This is safe since pageset manipulation is done | |
2074 | * with interrupts disabled. | |
2075 | * | |
2076 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
2077 | * | |
2078 | * The boot_pagesets must be kept even after bootup is complete for | |
2079 | * unused processors and/or zones. They do play a role for bootstrapping | |
2080 | * hotplugged processors. | |
2081 | * | |
2082 | * zoneinfo_show() and maybe other functions do | |
2083 | * not check if the processor is online before following the pageset pointer. | |
2084 | * Other parts of the kernel may not check if the zone is available. | |
2085 | */ | |
2086 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; | |
2087 | ||
2088 | /* | |
2089 | * Dynamically allocate memory for the | |
2090 | * per cpu pageset array in struct zone. | |
2091 | */ | |
2092 | static int __cpuinit process_zones(int cpu) | |
2093 | { | |
2094 | struct zone *zone, *dzone; | |
2095 | ||
2096 | for_each_zone(zone) { | |
2097 | ||
2098 | if (!populated_zone(zone)) | |
2099 | continue; | |
2100 | ||
2101 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), | |
2102 | GFP_KERNEL, cpu_to_node(cpu)); | |
2103 | if (!zone_pcp(zone, cpu)) | |
2104 | goto bad; | |
2105 | ||
2106 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); | |
2107 | ||
2108 | if (percpu_pagelist_fraction) | |
2109 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
2110 | (zone->present_pages / percpu_pagelist_fraction)); | |
2111 | } | |
2112 | ||
2113 | return 0; | |
2114 | bad: | |
2115 | for_each_zone(dzone) { | |
2116 | if (dzone == zone) | |
2117 | break; | |
2118 | kfree(zone_pcp(dzone, cpu)); | |
2119 | zone_pcp(dzone, cpu) = NULL; | |
2120 | } | |
2121 | return -ENOMEM; | |
2122 | } | |
2123 | ||
2124 | static inline void free_zone_pagesets(int cpu) | |
2125 | { | |
2126 | struct zone *zone; | |
2127 | ||
2128 | for_each_zone(zone) { | |
2129 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
2130 | ||
2131 | /* Free per_cpu_pageset if it is slab allocated */ | |
2132 | if (pset != &boot_pageset[cpu]) | |
2133 | kfree(pset); | |
2134 | zone_pcp(zone, cpu) = NULL; | |
2135 | } | |
2136 | } | |
2137 | ||
2138 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, | |
2139 | unsigned long action, | |
2140 | void *hcpu) | |
2141 | { | |
2142 | int cpu = (long)hcpu; | |
2143 | int ret = NOTIFY_OK; | |
2144 | ||
2145 | switch (action) { | |
2146 | case CPU_UP_PREPARE: | |
2147 | if (process_zones(cpu)) | |
2148 | ret = NOTIFY_BAD; | |
2149 | break; | |
2150 | case CPU_UP_CANCELED: | |
2151 | case CPU_DEAD: | |
2152 | free_zone_pagesets(cpu); | |
2153 | break; | |
2154 | default: | |
2155 | break; | |
2156 | } | |
2157 | return ret; | |
2158 | } | |
2159 | ||
2160 | static struct notifier_block __cpuinitdata pageset_notifier = | |
2161 | { &pageset_cpuup_callback, NULL, 0 }; | |
2162 | ||
2163 | void __init setup_per_cpu_pageset(void) | |
2164 | { | |
2165 | int err; | |
2166 | ||
2167 | /* Initialize per_cpu_pageset for cpu 0. | |
2168 | * A cpuup callback will do this for every cpu | |
2169 | * as it comes online | |
2170 | */ | |
2171 | err = process_zones(smp_processor_id()); | |
2172 | BUG_ON(err); | |
2173 | register_cpu_notifier(&pageset_notifier); | |
2174 | } | |
2175 | ||
2176 | #endif | |
2177 | ||
2178 | static __meminit | |
2179 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) | |
2180 | { | |
2181 | int i; | |
2182 | struct pglist_data *pgdat = zone->zone_pgdat; | |
2183 | size_t alloc_size; | |
2184 | ||
2185 | /* | |
2186 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2187 | * per zone. | |
2188 | */ | |
2189 | zone->wait_table_hash_nr_entries = | |
2190 | wait_table_hash_nr_entries(zone_size_pages); | |
2191 | zone->wait_table_bits = | |
2192 | wait_table_bits(zone->wait_table_hash_nr_entries); | |
2193 | alloc_size = zone->wait_table_hash_nr_entries | |
2194 | * sizeof(wait_queue_head_t); | |
2195 | ||
2196 | if (system_state == SYSTEM_BOOTING) { | |
2197 | zone->wait_table = (wait_queue_head_t *) | |
2198 | alloc_bootmem_node(pgdat, alloc_size); | |
2199 | } else { | |
2200 | /* | |
2201 | * This case means that a zone whose size was 0 gets new memory | |
2202 | * via memory hot-add. | |
2203 | * But it may be the case that a new node was hot-added. In | |
2204 | * this case vmalloc() will not be able to use this new node's | |
2205 | * memory - this wait_table must be initialized to use this new | |
2206 | * node itself as well. | |
2207 | * To use this new node's memory, further consideration will be | |
2208 | * necessary. | |
2209 | */ | |
2210 | zone->wait_table = (wait_queue_head_t *)vmalloc(alloc_size); | |
2211 | } | |
2212 | if (!zone->wait_table) | |
2213 | return -ENOMEM; | |
2214 | ||
2215 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) | |
2216 | init_waitqueue_head(zone->wait_table + i); | |
2217 | ||
2218 | return 0; | |
2219 | } | |
2220 | ||
2221 | static __meminit void zone_pcp_init(struct zone *zone) | |
2222 | { | |
2223 | int cpu; | |
2224 | unsigned long batch = zone_batchsize(zone); | |
2225 | ||
2226 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2227 | #ifdef CONFIG_NUMA | |
2228 | /* Early boot. Slab allocator not functional yet */ | |
2229 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; | |
2230 | setup_pageset(&boot_pageset[cpu],0); | |
2231 | #else | |
2232 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2233 | #endif | |
2234 | } | |
2235 | if (zone->present_pages) | |
2236 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2237 | zone->name, zone->present_pages, batch); | |
2238 | } | |
2239 | ||
2240 | __meminit int init_currently_empty_zone(struct zone *zone, | |
2241 | unsigned long zone_start_pfn, | |
2242 | unsigned long size) | |
2243 | { | |
2244 | struct pglist_data *pgdat = zone->zone_pgdat; | |
2245 | int ret; | |
2246 | ret = zone_wait_table_init(zone, size); | |
2247 | if (ret) | |
2248 | return ret; | |
2249 | pgdat->nr_zones = zone_idx(zone) + 1; | |
2250 | ||
2251 | zone->zone_start_pfn = zone_start_pfn; | |
2252 | ||
2253 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2254 | ||
2255 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
2256 | ||
2257 | return 0; | |
2258 | } | |
2259 | ||
2260 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
2261 | /* | |
2262 | * Basic iterator support. Return the first range of PFNs for a node | |
2263 | * Note: nid == MAX_NUMNODES returns first region regardless of node | |
2264 | */ | |
2265 | static int __init first_active_region_index_in_nid(int nid) | |
2266 | { | |
2267 | int i; | |
2268 | ||
2269 | for (i = 0; i < nr_nodemap_entries; i++) | |
2270 | if (nid == MAX_NUMNODES || early_node_map[i].nid == nid) | |
2271 | return i; | |
2272 | ||
2273 | return -1; | |
2274 | } | |
2275 | ||
2276 | /* | |
2277 | * Basic iterator support. Return the next active range of PFNs for a node | |
2278 | * Note: nid == MAX_NUMNODES returns next region regardles of node | |
2279 | */ | |
2280 | static int __init next_active_region_index_in_nid(int index, int nid) | |
2281 | { | |
2282 | for (index = index + 1; index < nr_nodemap_entries; index++) | |
2283 | if (nid == MAX_NUMNODES || early_node_map[index].nid == nid) | |
2284 | return index; | |
2285 | ||
2286 | return -1; | |
2287 | } | |
2288 | ||
2289 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID | |
2290 | /* | |
2291 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. | |
2292 | * Architectures may implement their own version but if add_active_range() | |
2293 | * was used and there are no special requirements, this is a convenient | |
2294 | * alternative | |
2295 | */ | |
2296 | int __init early_pfn_to_nid(unsigned long pfn) | |
2297 | { | |
2298 | int i; | |
2299 | ||
2300 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2301 | unsigned long start_pfn = early_node_map[i].start_pfn; | |
2302 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2303 | ||
2304 | if (start_pfn <= pfn && pfn < end_pfn) | |
2305 | return early_node_map[i].nid; | |
2306 | } | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ | |
2311 | ||
2312 | /* Basic iterator support to walk early_node_map[] */ | |
2313 | #define for_each_active_range_index_in_nid(i, nid) \ | |
2314 | for (i = first_active_region_index_in_nid(nid); i != -1; \ | |
2315 | i = next_active_region_index_in_nid(i, nid)) | |
2316 | ||
2317 | /** | |
2318 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range | |
2319 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. | |
2320 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node | |
2321 | * | |
2322 | * If an architecture guarantees that all ranges registered with | |
2323 | * add_active_ranges() contain no holes and may be freed, this | |
2324 | * this function may be used instead of calling free_bootmem() manually. | |
2325 | */ | |
2326 | void __init free_bootmem_with_active_regions(int nid, | |
2327 | unsigned long max_low_pfn) | |
2328 | { | |
2329 | int i; | |
2330 | ||
2331 | for_each_active_range_index_in_nid(i, nid) { | |
2332 | unsigned long size_pages = 0; | |
2333 | unsigned long end_pfn = early_node_map[i].end_pfn; | |
2334 | ||
2335 | if (early_node_map[i].start_pfn >= max_low_pfn) | |
2336 | continue; | |
2337 | ||
2338 | if (end_pfn > max_low_pfn) | |
2339 | end_pfn = max_low_pfn; | |
2340 | ||
2341 | size_pages = end_pfn - early_node_map[i].start_pfn; | |
2342 | free_bootmem_node(NODE_DATA(early_node_map[i].nid), | |
2343 | PFN_PHYS(early_node_map[i].start_pfn), | |
2344 | size_pages << PAGE_SHIFT); | |
2345 | } | |
2346 | } | |
2347 | ||
2348 | /** | |
2349 | * sparse_memory_present_with_active_regions - Call memory_present for each active range | |
2350 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. | |
2351 | * | |
2352 | * If an architecture guarantees that all ranges registered with | |
2353 | * add_active_ranges() contain no holes and may be freed, this | |
2354 | * function may be used instead of calling memory_present() manually. | |
2355 | */ | |
2356 | void __init sparse_memory_present_with_active_regions(int nid) | |
2357 | { | |
2358 | int i; | |
2359 | ||
2360 | for_each_active_range_index_in_nid(i, nid) | |
2361 | memory_present(early_node_map[i].nid, | |
2362 | early_node_map[i].start_pfn, | |
2363 | early_node_map[i].end_pfn); | |
2364 | } | |
2365 | ||
2366 | /** | |
2367 | * push_node_boundaries - Push node boundaries to at least the requested boundary | |
2368 | * @nid: The nid of the node to push the boundary for | |
2369 | * @start_pfn: The start pfn of the node | |
2370 | * @end_pfn: The end pfn of the node | |
2371 | * | |
2372 | * In reserve-based hot-add, mem_map is allocated that is unused until hotadd | |
2373 | * time. Specifically, on x86_64, SRAT will report ranges that can potentially | |
2374 | * be hotplugged even though no physical memory exists. This function allows | |
2375 | * an arch to push out the node boundaries so mem_map is allocated that can | |
2376 | * be used later. | |
2377 | */ | |
2378 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2379 | void __init push_node_boundaries(unsigned int nid, | |
2380 | unsigned long start_pfn, unsigned long end_pfn) | |
2381 | { | |
2382 | printk(KERN_DEBUG "Entering push_node_boundaries(%u, %lu, %lu)\n", | |
2383 | nid, start_pfn, end_pfn); | |
2384 | ||
2385 | /* Initialise the boundary for this node if necessary */ | |
2386 | if (node_boundary_end_pfn[nid] == 0) | |
2387 | node_boundary_start_pfn[nid] = -1UL; | |
2388 | ||
2389 | /* Update the boundaries */ | |
2390 | if (node_boundary_start_pfn[nid] > start_pfn) | |
2391 | node_boundary_start_pfn[nid] = start_pfn; | |
2392 | if (node_boundary_end_pfn[nid] < end_pfn) | |
2393 | node_boundary_end_pfn[nid] = end_pfn; | |
2394 | } | |
2395 | ||
2396 | /* If necessary, push the node boundary out for reserve hotadd */ | |
2397 | static void __init account_node_boundary(unsigned int nid, | |
2398 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2399 | { | |
2400 | printk(KERN_DEBUG "Entering account_node_boundary(%u, %lu, %lu)\n", | |
2401 | nid, *start_pfn, *end_pfn); | |
2402 | ||
2403 | /* Return if boundary information has not been provided */ | |
2404 | if (node_boundary_end_pfn[nid] == 0) | |
2405 | return; | |
2406 | ||
2407 | /* Check the boundaries and update if necessary */ | |
2408 | if (node_boundary_start_pfn[nid] < *start_pfn) | |
2409 | *start_pfn = node_boundary_start_pfn[nid]; | |
2410 | if (node_boundary_end_pfn[nid] > *end_pfn) | |
2411 | *end_pfn = node_boundary_end_pfn[nid]; | |
2412 | } | |
2413 | #else | |
2414 | void __init push_node_boundaries(unsigned int nid, | |
2415 | unsigned long start_pfn, unsigned long end_pfn) {} | |
2416 | ||
2417 | static void __init account_node_boundary(unsigned int nid, | |
2418 | unsigned long *start_pfn, unsigned long *end_pfn) {} | |
2419 | #endif | |
2420 | ||
2421 | ||
2422 | /** | |
2423 | * get_pfn_range_for_nid - Return the start and end page frames for a node | |
2424 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. | |
2425 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. | |
2426 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. | |
2427 | * | |
2428 | * It returns the start and end page frame of a node based on information | |
2429 | * provided by an arch calling add_active_range(). If called for a node | |
2430 | * with no available memory, a warning is printed and the start and end | |
2431 | * PFNs will be 0. | |
2432 | */ | |
2433 | void __init get_pfn_range_for_nid(unsigned int nid, | |
2434 | unsigned long *start_pfn, unsigned long *end_pfn) | |
2435 | { | |
2436 | int i; | |
2437 | *start_pfn = -1UL; | |
2438 | *end_pfn = 0; | |
2439 | ||
2440 | for_each_active_range_index_in_nid(i, nid) { | |
2441 | *start_pfn = min(*start_pfn, early_node_map[i].start_pfn); | |
2442 | *end_pfn = max(*end_pfn, early_node_map[i].end_pfn); | |
2443 | } | |
2444 | ||
2445 | if (*start_pfn == -1UL) { | |
2446 | printk(KERN_WARNING "Node %u active with no memory\n", nid); | |
2447 | *start_pfn = 0; | |
2448 | } | |
2449 | ||
2450 | /* Push the node boundaries out if requested */ | |
2451 | account_node_boundary(nid, start_pfn, end_pfn); | |
2452 | } | |
2453 | ||
2454 | /* | |
2455 | * Return the number of pages a zone spans in a node, including holes | |
2456 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() | |
2457 | */ | |
2458 | unsigned long __init zone_spanned_pages_in_node(int nid, | |
2459 | unsigned long zone_type, | |
2460 | unsigned long *ignored) | |
2461 | { | |
2462 | unsigned long node_start_pfn, node_end_pfn; | |
2463 | unsigned long zone_start_pfn, zone_end_pfn; | |
2464 | ||
2465 | /* Get the start and end of the node and zone */ | |
2466 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2467 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; | |
2468 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; | |
2469 | ||
2470 | /* Check that this node has pages within the zone's required range */ | |
2471 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) | |
2472 | return 0; | |
2473 | ||
2474 | /* Move the zone boundaries inside the node if necessary */ | |
2475 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); | |
2476 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); | |
2477 | ||
2478 | /* Return the spanned pages */ | |
2479 | return zone_end_pfn - zone_start_pfn; | |
2480 | } | |
2481 | ||
2482 | /* | |
2483 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, | |
2484 | * then all holes in the requested range will be accounted for. | |
2485 | */ | |
2486 | unsigned long __init __absent_pages_in_range(int nid, | |
2487 | unsigned long range_start_pfn, | |
2488 | unsigned long range_end_pfn) | |
2489 | { | |
2490 | int i = 0; | |
2491 | unsigned long prev_end_pfn = 0, hole_pages = 0; | |
2492 | unsigned long start_pfn; | |
2493 | ||
2494 | /* Find the end_pfn of the first active range of pfns in the node */ | |
2495 | i = first_active_region_index_in_nid(nid); | |
2496 | if (i == -1) | |
2497 | return 0; | |
2498 | ||
2499 | /* Account for ranges before physical memory on this node */ | |
2500 | if (early_node_map[i].start_pfn > range_start_pfn) | |
2501 | hole_pages = early_node_map[i].start_pfn - range_start_pfn; | |
2502 | ||
2503 | prev_end_pfn = early_node_map[i].start_pfn; | |
2504 | ||
2505 | /* Find all holes for the zone within the node */ | |
2506 | for (; i != -1; i = next_active_region_index_in_nid(i, nid)) { | |
2507 | ||
2508 | /* No need to continue if prev_end_pfn is outside the zone */ | |
2509 | if (prev_end_pfn >= range_end_pfn) | |
2510 | break; | |
2511 | ||
2512 | /* Make sure the end of the zone is not within the hole */ | |
2513 | start_pfn = min(early_node_map[i].start_pfn, range_end_pfn); | |
2514 | prev_end_pfn = max(prev_end_pfn, range_start_pfn); | |
2515 | ||
2516 | /* Update the hole size cound and move on */ | |
2517 | if (start_pfn > range_start_pfn) { | |
2518 | BUG_ON(prev_end_pfn > start_pfn); | |
2519 | hole_pages += start_pfn - prev_end_pfn; | |
2520 | } | |
2521 | prev_end_pfn = early_node_map[i].end_pfn; | |
2522 | } | |
2523 | ||
2524 | /* Account for ranges past physical memory on this node */ | |
2525 | if (range_end_pfn > prev_end_pfn) | |
2526 | hole_pages += range_end_pfn - | |
2527 | max(range_start_pfn, prev_end_pfn); | |
2528 | ||
2529 | return hole_pages; | |
2530 | } | |
2531 | ||
2532 | /** | |
2533 | * absent_pages_in_range - Return number of page frames in holes within a range | |
2534 | * @start_pfn: The start PFN to start searching for holes | |
2535 | * @end_pfn: The end PFN to stop searching for holes | |
2536 | * | |
2537 | * It returns the number of pages frames in memory holes within a range. | |
2538 | */ | |
2539 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, | |
2540 | unsigned long end_pfn) | |
2541 | { | |
2542 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); | |
2543 | } | |
2544 | ||
2545 | /* Return the number of page frames in holes in a zone on a node */ | |
2546 | unsigned long __init zone_absent_pages_in_node(int nid, | |
2547 | unsigned long zone_type, | |
2548 | unsigned long *ignored) | |
2549 | { | |
2550 | unsigned long node_start_pfn, node_end_pfn; | |
2551 | unsigned long zone_start_pfn, zone_end_pfn; | |
2552 | ||
2553 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); | |
2554 | zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type], | |
2555 | node_start_pfn); | |
2556 | zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type], | |
2557 | node_end_pfn); | |
2558 | ||
2559 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); | |
2560 | } | |
2561 | ||
2562 | #else | |
2563 | static inline unsigned long zone_spanned_pages_in_node(int nid, | |
2564 | unsigned long zone_type, | |
2565 | unsigned long *zones_size) | |
2566 | { | |
2567 | return zones_size[zone_type]; | |
2568 | } | |
2569 | ||
2570 | static inline unsigned long zone_absent_pages_in_node(int nid, | |
2571 | unsigned long zone_type, | |
2572 | unsigned long *zholes_size) | |
2573 | { | |
2574 | if (!zholes_size) | |
2575 | return 0; | |
2576 | ||
2577 | return zholes_size[zone_type]; | |
2578 | } | |
2579 | ||
2580 | #endif | |
2581 | ||
2582 | static void __init calculate_node_totalpages(struct pglist_data *pgdat, | |
2583 | unsigned long *zones_size, unsigned long *zholes_size) | |
2584 | { | |
2585 | unsigned long realtotalpages, totalpages = 0; | |
2586 | enum zone_type i; | |
2587 | ||
2588 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2589 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, | |
2590 | zones_size); | |
2591 | pgdat->node_spanned_pages = totalpages; | |
2592 | ||
2593 | realtotalpages = totalpages; | |
2594 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2595 | realtotalpages -= | |
2596 | zone_absent_pages_in_node(pgdat->node_id, i, | |
2597 | zholes_size); | |
2598 | pgdat->node_present_pages = realtotalpages; | |
2599 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, | |
2600 | realtotalpages); | |
2601 | } | |
2602 | ||
2603 | /* | |
2604 | * Set up the zone data structures: | |
2605 | * - mark all pages reserved | |
2606 | * - mark all memory queues empty | |
2607 | * - clear the memory bitmaps | |
2608 | */ | |
2609 | static void __meminit free_area_init_core(struct pglist_data *pgdat, | |
2610 | unsigned long *zones_size, unsigned long *zholes_size) | |
2611 | { | |
2612 | enum zone_type j; | |
2613 | int nid = pgdat->node_id; | |
2614 | unsigned long zone_start_pfn = pgdat->node_start_pfn; | |
2615 | int ret; | |
2616 | ||
2617 | pgdat_resize_init(pgdat); | |
2618 | pgdat->nr_zones = 0; | |
2619 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2620 | pgdat->kswapd_max_order = 0; | |
2621 | ||
2622 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2623 | struct zone *zone = pgdat->node_zones + j; | |
2624 | unsigned long size, realsize, memmap_pages; | |
2625 | ||
2626 | size = zone_spanned_pages_in_node(nid, j, zones_size); | |
2627 | realsize = size - zone_absent_pages_in_node(nid, j, | |
2628 | zholes_size); | |
2629 | ||
2630 | /* | |
2631 | * Adjust realsize so that it accounts for how much memory | |
2632 | * is used by this zone for memmap. This affects the watermark | |
2633 | * and per-cpu initialisations | |
2634 | */ | |
2635 | memmap_pages = (size * sizeof(struct page)) >> PAGE_SHIFT; | |
2636 | if (realsize >= memmap_pages) { | |
2637 | realsize -= memmap_pages; | |
2638 | printk(KERN_DEBUG | |
2639 | " %s zone: %lu pages used for memmap\n", | |
2640 | zone_names[j], memmap_pages); | |
2641 | } else | |
2642 | printk(KERN_WARNING | |
2643 | " %s zone: %lu pages exceeds realsize %lu\n", | |
2644 | zone_names[j], memmap_pages, realsize); | |
2645 | ||
2646 | /* Account for reserved DMA pages */ | |
2647 | if (j == ZONE_DMA && realsize > dma_reserve) { | |
2648 | realsize -= dma_reserve; | |
2649 | printk(KERN_DEBUG " DMA zone: %lu pages reserved\n", | |
2650 | dma_reserve); | |
2651 | } | |
2652 | ||
2653 | if (!is_highmem_idx(j)) | |
2654 | nr_kernel_pages += realsize; | |
2655 | nr_all_pages += realsize; | |
2656 | ||
2657 | zone->spanned_pages = size; | |
2658 | zone->present_pages = realsize; | |
2659 | #ifdef CONFIG_NUMA | |
2660 | zone->node = nid; | |
2661 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) | |
2662 | / 100; | |
2663 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; | |
2664 | #endif | |
2665 | zone->name = zone_names[j]; | |
2666 | spin_lock_init(&zone->lock); | |
2667 | spin_lock_init(&zone->lru_lock); | |
2668 | zone_seqlock_init(zone); | |
2669 | zone->zone_pgdat = pgdat; | |
2670 | zone->free_pages = 0; | |
2671 | ||
2672 | zone->prev_priority = DEF_PRIORITY; | |
2673 | ||
2674 | zone_pcp_init(zone); | |
2675 | INIT_LIST_HEAD(&zone->active_list); | |
2676 | INIT_LIST_HEAD(&zone->inactive_list); | |
2677 | zone->nr_scan_active = 0; | |
2678 | zone->nr_scan_inactive = 0; | |
2679 | zone->nr_active = 0; | |
2680 | zone->nr_inactive = 0; | |
2681 | zap_zone_vm_stats(zone); | |
2682 | atomic_set(&zone->reclaim_in_progress, 0); | |
2683 | if (!size) | |
2684 | continue; | |
2685 | ||
2686 | ret = init_currently_empty_zone(zone, zone_start_pfn, size); | |
2687 | BUG_ON(ret); | |
2688 | zone_start_pfn += size; | |
2689 | } | |
2690 | } | |
2691 | ||
2692 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2693 | { | |
2694 | /* Skip empty nodes */ | |
2695 | if (!pgdat->node_spanned_pages) | |
2696 | return; | |
2697 | ||
2698 | #ifdef CONFIG_FLAT_NODE_MEM_MAP | |
2699 | /* ia64 gets its own node_mem_map, before this, without bootmem */ | |
2700 | if (!pgdat->node_mem_map) { | |
2701 | unsigned long size, start, end; | |
2702 | struct page *map; | |
2703 | ||
2704 | /* | |
2705 | * The zone's endpoints aren't required to be MAX_ORDER | |
2706 | * aligned but the node_mem_map endpoints must be in order | |
2707 | * for the buddy allocator to function correctly. | |
2708 | */ | |
2709 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); | |
2710 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
2711 | end = ALIGN(end, MAX_ORDER_NR_PAGES); | |
2712 | size = (end - start) * sizeof(struct page); | |
2713 | map = alloc_remap(pgdat->node_id, size); | |
2714 | if (!map) | |
2715 | map = alloc_bootmem_node(pgdat, size); | |
2716 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); | |
2717 | } | |
2718 | #ifdef CONFIG_FLATMEM | |
2719 | /* | |
2720 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2721 | */ | |
2722 | if (pgdat == NODE_DATA(0)) { | |
2723 | mem_map = NODE_DATA(0)->node_mem_map; | |
2724 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
2725 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) | |
2726 | mem_map -= pgdat->node_start_pfn; | |
2727 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2728 | } | |
2729 | #endif | |
2730 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ | |
2731 | } | |
2732 | ||
2733 | void __meminit free_area_init_node(int nid, struct pglist_data *pgdat, | |
2734 | unsigned long *zones_size, unsigned long node_start_pfn, | |
2735 | unsigned long *zholes_size) | |
2736 | { | |
2737 | pgdat->node_id = nid; | |
2738 | pgdat->node_start_pfn = node_start_pfn; | |
2739 | calculate_node_totalpages(pgdat, zones_size, zholes_size); | |
2740 | ||
2741 | alloc_node_mem_map(pgdat); | |
2742 | ||
2743 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2744 | } | |
2745 | ||
2746 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP | |
2747 | /** | |
2748 | * add_active_range - Register a range of PFNs backed by physical memory | |
2749 | * @nid: The node ID the range resides on | |
2750 | * @start_pfn: The start PFN of the available physical memory | |
2751 | * @end_pfn: The end PFN of the available physical memory | |
2752 | * | |
2753 | * These ranges are stored in an early_node_map[] and later used by | |
2754 | * free_area_init_nodes() to calculate zone sizes and holes. If the | |
2755 | * range spans a memory hole, it is up to the architecture to ensure | |
2756 | * the memory is not freed by the bootmem allocator. If possible | |
2757 | * the range being registered will be merged with existing ranges. | |
2758 | */ | |
2759 | void __init add_active_range(unsigned int nid, unsigned long start_pfn, | |
2760 | unsigned long end_pfn) | |
2761 | { | |
2762 | int i; | |
2763 | ||
2764 | printk(KERN_DEBUG "Entering add_active_range(%d, %lu, %lu) " | |
2765 | "%d entries of %d used\n", | |
2766 | nid, start_pfn, end_pfn, | |
2767 | nr_nodemap_entries, MAX_ACTIVE_REGIONS); | |
2768 | ||
2769 | /* Merge with existing active regions if possible */ | |
2770 | for (i = 0; i < nr_nodemap_entries; i++) { | |
2771 | if (early_node_map[i].nid != nid) | |
2772 | continue; | |
2773 | ||
2774 | /* Skip if an existing region covers this new one */ | |
2775 | if (start_pfn >= early_node_map[i].start_pfn && | |
2776 | end_pfn <= early_node_map[i].end_pfn) | |
2777 | return; | |
2778 | ||
2779 | /* Merge forward if suitable */ | |
2780 | if (start_pfn <= early_node_map[i].end_pfn && | |
2781 | end_pfn > early_node_map[i].end_pfn) { | |
2782 | early_node_map[i].end_pfn = end_pfn; | |
2783 | return; | |
2784 | } | |
2785 | ||
2786 | /* Merge backward if suitable */ | |
2787 | if (start_pfn < early_node_map[i].end_pfn && | |
2788 | end_pfn >= early_node_map[i].start_pfn) { | |
2789 | early_node_map[i].start_pfn = start_pfn; | |
2790 | return; | |
2791 | } | |
2792 | } | |
2793 | ||
2794 | /* Check that early_node_map is large enough */ | |
2795 | if (i >= MAX_ACTIVE_REGIONS) { | |
2796 | printk(KERN_CRIT "More than %d memory regions, truncating\n", | |
2797 | MAX_ACTIVE_REGIONS); | |
2798 | return; | |
2799 | } | |
2800 | ||
2801 | early_node_map[i].nid = nid; | |
2802 | early_node_map[i].start_pfn = start_pfn; | |
2803 | early_node_map[i].end_pfn = end_pfn; | |
2804 | nr_nodemap_entries = i + 1; | |
2805 | } | |
2806 | ||
2807 | /** | |
2808 | * shrink_active_range - Shrink an existing registered range of PFNs | |
2809 | * @nid: The node id the range is on that should be shrunk | |
2810 | * @old_end_pfn: The old end PFN of the range | |
2811 | * @new_end_pfn: The new PFN of the range | |
2812 | * | |
2813 | * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node. | |
2814 | * The map is kept at the end physical page range that has already been | |
2815 | * registered with add_active_range(). This function allows an arch to shrink | |
2816 | * an existing registered range. | |
2817 | */ | |
2818 | void __init shrink_active_range(unsigned int nid, unsigned long old_end_pfn, | |
2819 | unsigned long new_end_pfn) | |
2820 | { | |
2821 | int i; | |
2822 | ||
2823 | /* Find the old active region end and shrink */ | |
2824 | for_each_active_range_index_in_nid(i, nid) | |
2825 | if (early_node_map[i].end_pfn == old_end_pfn) { | |
2826 | early_node_map[i].end_pfn = new_end_pfn; | |
2827 | break; | |
2828 | } | |
2829 | } | |
2830 | ||
2831 | /** | |
2832 | * remove_all_active_ranges - Remove all currently registered regions | |
2833 | * | |
2834 | * During discovery, it may be found that a table like SRAT is invalid | |
2835 | * and an alternative discovery method must be used. This function removes | |
2836 | * all currently registered regions. | |
2837 | */ | |
2838 | void __init remove_all_active_ranges(void) | |
2839 | { | |
2840 | memset(early_node_map, 0, sizeof(early_node_map)); | |
2841 | nr_nodemap_entries = 0; | |
2842 | #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE | |
2843 | memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn)); | |
2844 | memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn)); | |
2845 | #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */ | |
2846 | } | |
2847 | ||
2848 | /* Compare two active node_active_regions */ | |
2849 | static int __init cmp_node_active_region(const void *a, const void *b) | |
2850 | { | |
2851 | struct node_active_region *arange = (struct node_active_region *)a; | |
2852 | struct node_active_region *brange = (struct node_active_region *)b; | |
2853 | ||
2854 | /* Done this way to avoid overflows */ | |
2855 | if (arange->start_pfn > brange->start_pfn) | |
2856 | return 1; | |
2857 | if (arange->start_pfn < brange->start_pfn) | |
2858 | return -1; | |
2859 | ||
2860 | return 0; | |
2861 | } | |
2862 | ||
2863 | /* sort the node_map by start_pfn */ | |
2864 | static void __init sort_node_map(void) | |
2865 | { | |
2866 | sort(early_node_map, (size_t)nr_nodemap_entries, | |
2867 | sizeof(struct node_active_region), | |
2868 | cmp_node_active_region, NULL); | |
2869 | } | |
2870 | ||
2871 | /* Find the lowest pfn for a node. This depends on a sorted early_node_map */ | |
2872 | unsigned long __init find_min_pfn_for_node(unsigned long nid) | |
2873 | { | |
2874 | int i; | |
2875 | ||
2876 | /* Regions in the early_node_map can be in any order */ | |
2877 | sort_node_map(); | |
2878 | ||
2879 | /* Assuming a sorted map, the first range found has the starting pfn */ | |
2880 | for_each_active_range_index_in_nid(i, nid) | |
2881 | return early_node_map[i].start_pfn; | |
2882 | ||
2883 | printk(KERN_WARNING "Could not find start_pfn for node %lu\n", nid); | |
2884 | return 0; | |
2885 | } | |
2886 | ||
2887 | /** | |
2888 | * find_min_pfn_with_active_regions - Find the minimum PFN registered | |
2889 | * | |
2890 | * It returns the minimum PFN based on information provided via | |
2891 | * add_active_range(). | |
2892 | */ | |
2893 | unsigned long __init find_min_pfn_with_active_regions(void) | |
2894 | { | |
2895 | return find_min_pfn_for_node(MAX_NUMNODES); | |
2896 | } | |
2897 | ||
2898 | /** | |
2899 | * find_max_pfn_with_active_regions - Find the maximum PFN registered | |
2900 | * | |
2901 | * It returns the maximum PFN based on information provided via | |
2902 | * add_active_range(). | |
2903 | */ | |
2904 | unsigned long __init find_max_pfn_with_active_regions(void) | |
2905 | { | |
2906 | int i; | |
2907 | unsigned long max_pfn = 0; | |
2908 | ||
2909 | for (i = 0; i < nr_nodemap_entries; i++) | |
2910 | max_pfn = max(max_pfn, early_node_map[i].end_pfn); | |
2911 | ||
2912 | return max_pfn; | |
2913 | } | |
2914 | ||
2915 | /** | |
2916 | * free_area_init_nodes - Initialise all pg_data_t and zone data | |
2917 | * @max_zone_pfn: an array of max PFNs for each zone | |
2918 | * | |
2919 | * This will call free_area_init_node() for each active node in the system. | |
2920 | * Using the page ranges provided by add_active_range(), the size of each | |
2921 | * zone in each node and their holes is calculated. If the maximum PFN | |
2922 | * between two adjacent zones match, it is assumed that the zone is empty. | |
2923 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed | |
2924 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone | |
2925 | * starts where the previous one ended. For example, ZONE_DMA32 starts | |
2926 | * at arch_max_dma_pfn. | |
2927 | */ | |
2928 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) | |
2929 | { | |
2930 | unsigned long nid; | |
2931 | enum zone_type i; | |
2932 | ||
2933 | /* Record where the zone boundaries are */ | |
2934 | memset(arch_zone_lowest_possible_pfn, 0, | |
2935 | sizeof(arch_zone_lowest_possible_pfn)); | |
2936 | memset(arch_zone_highest_possible_pfn, 0, | |
2937 | sizeof(arch_zone_highest_possible_pfn)); | |
2938 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); | |
2939 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; | |
2940 | for (i = 1; i < MAX_NR_ZONES; i++) { | |
2941 | arch_zone_lowest_possible_pfn[i] = | |
2942 | arch_zone_highest_possible_pfn[i-1]; | |
2943 | arch_zone_highest_possible_pfn[i] = | |
2944 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); | |
2945 | } | |
2946 | ||
2947 | /* Print out the zone ranges */ | |
2948 | printk("Zone PFN ranges:\n"); | |
2949 | for (i = 0; i < MAX_NR_ZONES; i++) | |
2950 | printk(" %-8s %8lu -> %8lu\n", | |
2951 | zone_names[i], | |
2952 | arch_zone_lowest_possible_pfn[i], | |
2953 | arch_zone_highest_possible_pfn[i]); | |
2954 | ||
2955 | /* Print out the early_node_map[] */ | |
2956 | printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries); | |
2957 | for (i = 0; i < nr_nodemap_entries; i++) | |
2958 | printk(" %3d: %8lu -> %8lu\n", early_node_map[i].nid, | |
2959 | early_node_map[i].start_pfn, | |
2960 | early_node_map[i].end_pfn); | |
2961 | ||
2962 | /* Initialise every node */ | |
2963 | for_each_online_node(nid) { | |
2964 | pg_data_t *pgdat = NODE_DATA(nid); | |
2965 | free_area_init_node(nid, pgdat, NULL, | |
2966 | find_min_pfn_for_node(nid), NULL); | |
2967 | } | |
2968 | } | |
2969 | #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */ | |
2970 | ||
2971 | /** | |
2972 | * set_dma_reserve - set the specified number of pages reserved in the first zone | |
2973 | * @new_dma_reserve: The number of pages to mark reserved | |
2974 | * | |
2975 | * The per-cpu batchsize and zone watermarks are determined by present_pages. | |
2976 | * In the DMA zone, a significant percentage may be consumed by kernel image | |
2977 | * and other unfreeable allocations which can skew the watermarks badly. This | |
2978 | * function may optionally be used to account for unfreeable pages in the | |
2979 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and | |
2980 | * smaller per-cpu batchsize. | |
2981 | */ | |
2982 | void __init set_dma_reserve(unsigned long new_dma_reserve) | |
2983 | { | |
2984 | dma_reserve = new_dma_reserve; | |
2985 | } | |
2986 | ||
2987 | #ifndef CONFIG_NEED_MULTIPLE_NODES | |
2988 | static bootmem_data_t contig_bootmem_data; | |
2989 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2990 | ||
2991 | EXPORT_SYMBOL(contig_page_data); | |
2992 | #endif | |
2993 | ||
2994 | void __init free_area_init(unsigned long *zones_size) | |
2995 | { | |
2996 | free_area_init_node(0, NODE_DATA(0), zones_size, | |
2997 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); | |
2998 | } | |
2999 | ||
3000 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
3001 | unsigned long action, void *hcpu) | |
3002 | { | |
3003 | int cpu = (unsigned long)hcpu; | |
3004 | ||
3005 | if (action == CPU_DEAD) { | |
3006 | local_irq_disable(); | |
3007 | __drain_pages(cpu); | |
3008 | vm_events_fold_cpu(cpu); | |
3009 | local_irq_enable(); | |
3010 | refresh_cpu_vm_stats(cpu); | |
3011 | } | |
3012 | return NOTIFY_OK; | |
3013 | } | |
3014 | ||
3015 | void __init page_alloc_init(void) | |
3016 | { | |
3017 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
3018 | } | |
3019 | ||
3020 | /* | |
3021 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio | |
3022 | * or min_free_kbytes changes. | |
3023 | */ | |
3024 | static void calculate_totalreserve_pages(void) | |
3025 | { | |
3026 | struct pglist_data *pgdat; | |
3027 | unsigned long reserve_pages = 0; | |
3028 | enum zone_type i, j; | |
3029 | ||
3030 | for_each_online_pgdat(pgdat) { | |
3031 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
3032 | struct zone *zone = pgdat->node_zones + i; | |
3033 | unsigned long max = 0; | |
3034 | ||
3035 | /* Find valid and maximum lowmem_reserve in the zone */ | |
3036 | for (j = i; j < MAX_NR_ZONES; j++) { | |
3037 | if (zone->lowmem_reserve[j] > max) | |
3038 | max = zone->lowmem_reserve[j]; | |
3039 | } | |
3040 | ||
3041 | /* we treat pages_high as reserved pages. */ | |
3042 | max += zone->pages_high; | |
3043 | ||
3044 | if (max > zone->present_pages) | |
3045 | max = zone->present_pages; | |
3046 | reserve_pages += max; | |
3047 | } | |
3048 | } | |
3049 | totalreserve_pages = reserve_pages; | |
3050 | } | |
3051 | ||
3052 | /* | |
3053 | * setup_per_zone_lowmem_reserve - called whenever | |
3054 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
3055 | * has a correct pages reserved value, so an adequate number of | |
3056 | * pages are left in the zone after a successful __alloc_pages(). | |
3057 | */ | |
3058 | static void setup_per_zone_lowmem_reserve(void) | |
3059 | { | |
3060 | struct pglist_data *pgdat; | |
3061 | enum zone_type j, idx; | |
3062 | ||
3063 | for_each_online_pgdat(pgdat) { | |
3064 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
3065 | struct zone *zone = pgdat->node_zones + j; | |
3066 | unsigned long present_pages = zone->present_pages; | |
3067 | ||
3068 | zone->lowmem_reserve[j] = 0; | |
3069 | ||
3070 | idx = j; | |
3071 | while (idx) { | |
3072 | struct zone *lower_zone; | |
3073 | ||
3074 | idx--; | |
3075 | ||
3076 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
3077 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
3078 | ||
3079 | lower_zone = pgdat->node_zones + idx; | |
3080 | lower_zone->lowmem_reserve[j] = present_pages / | |
3081 | sysctl_lowmem_reserve_ratio[idx]; | |
3082 | present_pages += lower_zone->present_pages; | |
3083 | } | |
3084 | } | |
3085 | } | |
3086 | ||
3087 | /* update totalreserve_pages */ | |
3088 | calculate_totalreserve_pages(); | |
3089 | } | |
3090 | ||
3091 | /** | |
3092 | * setup_per_zone_pages_min - called when min_free_kbytes changes. | |
3093 | * | |
3094 | * Ensures that the pages_{min,low,high} values for each zone are set correctly | |
3095 | * with respect to min_free_kbytes. | |
3096 | */ | |
3097 | void setup_per_zone_pages_min(void) | |
3098 | { | |
3099 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
3100 | unsigned long lowmem_pages = 0; | |
3101 | struct zone *zone; | |
3102 | unsigned long flags; | |
3103 | ||
3104 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
3105 | for_each_zone(zone) { | |
3106 | if (!is_highmem(zone)) | |
3107 | lowmem_pages += zone->present_pages; | |
3108 | } | |
3109 | ||
3110 | for_each_zone(zone) { | |
3111 | u64 tmp; | |
3112 | ||
3113 | spin_lock_irqsave(&zone->lru_lock, flags); | |
3114 | tmp = (u64)pages_min * zone->present_pages; | |
3115 | do_div(tmp, lowmem_pages); | |
3116 | if (is_highmem(zone)) { | |
3117 | /* | |
3118 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't | |
3119 | * need highmem pages, so cap pages_min to a small | |
3120 | * value here. | |
3121 | * | |
3122 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
3123 | * deltas controls asynch page reclaim, and so should | |
3124 | * not be capped for highmem. | |
3125 | */ | |
3126 | int min_pages; | |
3127 | ||
3128 | min_pages = zone->present_pages / 1024; | |
3129 | if (min_pages < SWAP_CLUSTER_MAX) | |
3130 | min_pages = SWAP_CLUSTER_MAX; | |
3131 | if (min_pages > 128) | |
3132 | min_pages = 128; | |
3133 | zone->pages_min = min_pages; | |
3134 | } else { | |
3135 | /* | |
3136 | * If it's a lowmem zone, reserve a number of pages | |
3137 | * proportionate to the zone's size. | |
3138 | */ | |
3139 | zone->pages_min = tmp; | |
3140 | } | |
3141 | ||
3142 | zone->pages_low = zone->pages_min + (tmp >> 2); | |
3143 | zone->pages_high = zone->pages_min + (tmp >> 1); | |
3144 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
3145 | } | |
3146 | ||
3147 | /* update totalreserve_pages */ | |
3148 | calculate_totalreserve_pages(); | |
3149 | } | |
3150 | ||
3151 | /* | |
3152 | * Initialise min_free_kbytes. | |
3153 | * | |
3154 | * For small machines we want it small (128k min). For large machines | |
3155 | * we want it large (64MB max). But it is not linear, because network | |
3156 | * bandwidth does not increase linearly with machine size. We use | |
3157 | * | |
3158 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
3159 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
3160 | * | |
3161 | * which yields | |
3162 | * | |
3163 | * 16MB: 512k | |
3164 | * 32MB: 724k | |
3165 | * 64MB: 1024k | |
3166 | * 128MB: 1448k | |
3167 | * 256MB: 2048k | |
3168 | * 512MB: 2896k | |
3169 | * 1024MB: 4096k | |
3170 | * 2048MB: 5792k | |
3171 | * 4096MB: 8192k | |
3172 | * 8192MB: 11584k | |
3173 | * 16384MB: 16384k | |
3174 | */ | |
3175 | static int __init init_per_zone_pages_min(void) | |
3176 | { | |
3177 | unsigned long lowmem_kbytes; | |
3178 | ||
3179 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
3180 | ||
3181 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
3182 | if (min_free_kbytes < 128) | |
3183 | min_free_kbytes = 128; | |
3184 | if (min_free_kbytes > 65536) | |
3185 | min_free_kbytes = 65536; | |
3186 | setup_per_zone_pages_min(); | |
3187 | setup_per_zone_lowmem_reserve(); | |
3188 | return 0; | |
3189 | } | |
3190 | module_init(init_per_zone_pages_min) | |
3191 | ||
3192 | /* | |
3193 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
3194 | * that we can call two helper functions whenever min_free_kbytes | |
3195 | * changes. | |
3196 | */ | |
3197 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
3198 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3199 | { | |
3200 | proc_dointvec(table, write, file, buffer, length, ppos); | |
3201 | setup_per_zone_pages_min(); | |
3202 | return 0; | |
3203 | } | |
3204 | ||
3205 | #ifdef CONFIG_NUMA | |
3206 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, | |
3207 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3208 | { | |
3209 | struct zone *zone; | |
3210 | int rc; | |
3211 | ||
3212 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3213 | if (rc) | |
3214 | return rc; | |
3215 | ||
3216 | for_each_zone(zone) | |
3217 | zone->min_unmapped_pages = (zone->present_pages * | |
3218 | sysctl_min_unmapped_ratio) / 100; | |
3219 | return 0; | |
3220 | } | |
3221 | ||
3222 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, | |
3223 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3224 | { | |
3225 | struct zone *zone; | |
3226 | int rc; | |
3227 | ||
3228 | rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3229 | if (rc) | |
3230 | return rc; | |
3231 | ||
3232 | for_each_zone(zone) | |
3233 | zone->min_slab_pages = (zone->present_pages * | |
3234 | sysctl_min_slab_ratio) / 100; | |
3235 | return 0; | |
3236 | } | |
3237 | #endif | |
3238 | ||
3239 | /* | |
3240 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
3241 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
3242 | * whenever sysctl_lowmem_reserve_ratio changes. | |
3243 | * | |
3244 | * The reserve ratio obviously has absolutely no relation with the | |
3245 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
3246 | * if in function of the boot time zone sizes. | |
3247 | */ | |
3248 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
3249 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3250 | { | |
3251 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3252 | setup_per_zone_lowmem_reserve(); | |
3253 | return 0; | |
3254 | } | |
3255 | ||
3256 | /* | |
3257 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
3258 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
3259 | * can have before it gets flushed back to buddy allocator. | |
3260 | */ | |
3261 | ||
3262 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
3263 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
3264 | { | |
3265 | struct zone *zone; | |
3266 | unsigned int cpu; | |
3267 | int ret; | |
3268 | ||
3269 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
3270 | if (!write || (ret == -EINVAL)) | |
3271 | return ret; | |
3272 | for_each_zone(zone) { | |
3273 | for_each_online_cpu(cpu) { | |
3274 | unsigned long high; | |
3275 | high = zone->present_pages / percpu_pagelist_fraction; | |
3276 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
3277 | } | |
3278 | } | |
3279 | return 0; | |
3280 | } | |
3281 | ||
3282 | int hashdist = HASHDIST_DEFAULT; | |
3283 | ||
3284 | #ifdef CONFIG_NUMA | |
3285 | static int __init set_hashdist(char *str) | |
3286 | { | |
3287 | if (!str) | |
3288 | return 0; | |
3289 | hashdist = simple_strtoul(str, &str, 0); | |
3290 | return 1; | |
3291 | } | |
3292 | __setup("hashdist=", set_hashdist); | |
3293 | #endif | |
3294 | ||
3295 | /* | |
3296 | * allocate a large system hash table from bootmem | |
3297 | * - it is assumed that the hash table must contain an exact power-of-2 | |
3298 | * quantity of entries | |
3299 | * - limit is the number of hash buckets, not the total allocation size | |
3300 | */ | |
3301 | void *__init alloc_large_system_hash(const char *tablename, | |
3302 | unsigned long bucketsize, | |
3303 | unsigned long numentries, | |
3304 | int scale, | |
3305 | int flags, | |
3306 | unsigned int *_hash_shift, | |
3307 | unsigned int *_hash_mask, | |
3308 | unsigned long limit) | |
3309 | { | |
3310 | unsigned long long max = limit; | |
3311 | unsigned long log2qty, size; | |
3312 | void *table = NULL; | |
3313 | ||
3314 | /* allow the kernel cmdline to have a say */ | |
3315 | if (!numentries) { | |
3316 | /* round applicable memory size up to nearest megabyte */ | |
3317 | numentries = nr_kernel_pages; | |
3318 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
3319 | numentries >>= 20 - PAGE_SHIFT; | |
3320 | numentries <<= 20 - PAGE_SHIFT; | |
3321 | ||
3322 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
3323 | if (scale > PAGE_SHIFT) | |
3324 | numentries >>= (scale - PAGE_SHIFT); | |
3325 | else | |
3326 | numentries <<= (PAGE_SHIFT - scale); | |
3327 | ||
3328 | /* Make sure we've got at least a 0-order allocation.. */ | |
3329 | if (unlikely((numentries * bucketsize) < PAGE_SIZE)) | |
3330 | numentries = PAGE_SIZE / bucketsize; | |
3331 | } | |
3332 | numentries = roundup_pow_of_two(numentries); | |
3333 | ||
3334 | /* limit allocation size to 1/16 total memory by default */ | |
3335 | if (max == 0) { | |
3336 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
3337 | do_div(max, bucketsize); | |
3338 | } | |
3339 | ||
3340 | if (numentries > max) | |
3341 | numentries = max; | |
3342 | ||
3343 | log2qty = ilog2(numentries); | |
3344 | ||
3345 | do { | |
3346 | size = bucketsize << log2qty; | |
3347 | if (flags & HASH_EARLY) | |
3348 | table = alloc_bootmem(size); | |
3349 | else if (hashdist) | |
3350 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
3351 | else { | |
3352 | unsigned long order; | |
3353 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
3354 | ; | |
3355 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
3356 | } | |
3357 | } while (!table && size > PAGE_SIZE && --log2qty); | |
3358 | ||
3359 | if (!table) | |
3360 | panic("Failed to allocate %s hash table\n", tablename); | |
3361 | ||
3362 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
3363 | tablename, | |
3364 | (1U << log2qty), | |
3365 | ilog2(size) - PAGE_SHIFT, | |
3366 | size); | |
3367 | ||
3368 | if (_hash_shift) | |
3369 | *_hash_shift = log2qty; | |
3370 | if (_hash_mask) | |
3371 | *_hash_mask = (1 << log2qty) - 1; | |
3372 | ||
3373 | return table; | |
3374 | } | |
3375 | ||
3376 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
3377 | struct page *pfn_to_page(unsigned long pfn) | |
3378 | { | |
3379 | return __pfn_to_page(pfn); | |
3380 | } | |
3381 | unsigned long page_to_pfn(struct page *page) | |
3382 | { | |
3383 | return __page_to_pfn(page); | |
3384 | } | |
3385 | EXPORT_SYMBOL(pfn_to_page); | |
3386 | EXPORT_SYMBOL(page_to_pfn); | |
3387 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ | |
3388 | ||
3389 | #if MAX_NUMNODES > 1 | |
3390 | /* | |
3391 | * Find the highest possible node id. | |
3392 | */ | |
3393 | int highest_possible_node_id(void) | |
3394 | { | |
3395 | unsigned int node; | |
3396 | unsigned int highest = 0; | |
3397 | ||
3398 | for_each_node_mask(node, node_possible_map) | |
3399 | highest = node; | |
3400 | return highest; | |
3401 | } | |
3402 | EXPORT_SYMBOL(highest_possible_node_id); | |
3403 | #endif |